Id
4.....qu
1 r 339

59:49! .15 .79
I1: . 1’... av...“
A . .. .

 

. ,1... c. .11
9 6.3.7.7..

no: ‘ . GA.

 

fififififiuj 1qu .4. . , imm? " 3. m6 *ugmogxfifik.

L ‘ : . agv. .nf.v\

Jimazmfifii. . ‘ H ‘, ‘ :fiaéfia

IlllllllllllllllllTllllllllllillllllll

3 1293 01555 9366

This is to certify that the

dissertation entitled

Development of an Interactive Traffic Accident
Geographic Information (ITAGI) System and a
Multimedia Left-Turn Control Strategies Expert
System

presented by

Shung-Kung Wu

has been accepted towards fulfillment
of the requirements for

 

 

Ph . D . degree in Civil Engineering

MW 41 724/4

Major professo

Date 7/2/96

 

MSU is an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

 

 

LIBRARY

Michigan State
University

 

 

 

PLACE ll RETURN BOX to move this chockout from your tocord.
TO AVOID FINES Mum on or before data duo.

DATE DUE DATE DUE DATE DUE

 

.fi-r

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Afflnnotivo Adlai/Equal Opponunity instituion
1

 

DEVELOPMENT OF AN INTERACTIVE TRAFFIC
ACCIDENT GEOGRAPHIC INFORMATION (ITAGI)
SYSTEM AND A MULTIMEDIA LEFT -TURN CONTROL
STRATEGIES EXPERT SYSTEM

By

Shung Kung Wu

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Civil and Environmental Engineering

1996

ABSTRACT
DEVELOPMENT OF AN INTERACTIVE TRAFFIC
ACCIDENT GEOGRAPHIC INFORMATION (ITAGI)
SYSTEM AND A MULTIMEDIA LEFT-TURN CONTROL
STRATEGIES EXPERT SYSTEM
By

Shung Kung Wu

The Intelligent Vehicle Highway System (IVHS) project in Oakland County,
Michigan, iS a major federally-funded demonstration project of the real-time traffic
adaptive signal system called SCATS (Sydney Coordinated Adaptive Traffic System).
SCATS was installed in a part of the City of Troy (in Oakland County, Metropolitan
Detroit, Michigan) in June 1992 and in the remainder of the city in November 1993. The
role of the universities involved in this project is to evaluate the impact of this system.
There are several criteria being analyzed in this evaluation including the number, type
and severity of traffic accidents, the level of service (LOS), and the average speed on the
major arterials.

A program, the Interactive Traffic Accident Geographic Information (ITAGI)
system was designed to assist the analyst in evaluating the safety and delay (travel speed)
effects of SCATS. The results of the accident analyses conducted in this project using the
ITAGI system are described. From this database system, statistical analyses of any
specific location or group of locations, both intersections and midblock areas in the city,

can be conducted. The ITAGI system consists of a traffic accident geographic

information system (GIS), quality control charts of accident rates developed by using the
Monte Carlo simulation algorithm, and a safest path (through the network) algorithm.

Finally, after integrating the ITAGI system and a knowledge-based system, a
multimedia left-tum control expert system was developed. The ITAGI system is a
decision support tool for the multimedia left-turn control expert system. This expert
system provides knowledge-based and user-friendly information to the user. It can help
the operator in the traffic operations center analyze conditions and select left-tum control
strategies in real time. In addition, it utilizes the computer-based multimedia techniques
to store video images, audio voices, pictures, and graphics for CD-ROM computer use.
The rules contained in the multimedia left-turn control expert system have been calibrated
with a consistency check and a Turing test.

This paper, an integration of GIS and multimedia left-turn control expert systems,
first describes the development of the ITAGI system, then highlights how the ITAGI
system supports the actual safety and travel speed analysis before and after installation of
SCATS. Second, this paper illustrates the development of the multimedia left-tum
control expert system. It focuses on how the ITAGI system is used as a decision support
tool to the multimedia left—turn control expert system, on how the expert system generates
inductive rules by using the ID3 algorithm, and on how the expert system is calibrated

with a consistency check and a Turing test.

I dedicate this dissertation to my parents, sister, and my friends.

ACKNOWLEDGMENTS

I would like to express my sincere appreciation to my advisor, Dr. William C.
Taylor, for his continuous advice and encouragement throughout the work as well as his
financial support during the study. I also want to thank my guidance committee: Dr.
Richard Lyles, Dr. Thomas Maleck and Dr. Raoul LePage for their comments during my
defense. I would like to give special thanks to Dr. Richard Lyles for his advice in the
expert system development, and Dr. Raoul LePage for his technical support in statistics.
Also thanks are due to my girl friend, Pei-Chun Hsieh, for her encouragement while I was
working on my dissertation.

I would also like to thank Dr. William C. Taylor, Dr. Richard Lyles, and Dr.
Thomas Maleck from Department of Civil and Environmental Engineering at Michigan
State University, Mr. James Neve who served as a district traffic engineer for the
Michigan Department of Transportation (MDOT), Dave Allyn and Mohammed Lutfi
from the Oakland County Road Commission, and William Savage, a traffic engineering
consultant, for their participation as experts in developing the left-tum control knowledge

base. Thank you all.

TABLE OF CONTENTS

LIST OF TABLES

 

LIST OF FIGURES

 

CHAPTER 1 INTRODUCTION

 

 

1.1 INTRODUCTION

1.2 OBJECTIVES

 

1.3 METHODOLOGY

 

 

1.4 ORDER OF PRESENTATION

 

CHAPTER 2 LITERATURE REVIEW

2.] LEFT-TURN CONTROL GUIDELINES
2.1.1 TYPE OF CONTROL
2.1.2 LEFT-TURN PHASING PATTERNS
2.1.3 GUIDELINES FOR LEFT-TURN PHASING
2.1.4 LEFT-TURN INDICATION SEQUENCES
2.1.5 GUIDELINES FOR LEFT-TURN INDICATION

SEQUENCES

 

 

 

 

 

 

2.2 EXPERT SYSTEMS AND TRANSPORTATION ENGINEERING ------

2.3 EXPERT SYSTEMS AND LEFT-TURN TREATMENTS

 

2.3.1 TRADE-OF F BETWEEN LEFT-TURN EFFICIENCY AND
SAFETY

 

2.3.2 THE VALUE OF A LEFT-TURN CONTROL EXPERT

 

SYSTEM
2.3.3 THE EXPERT SYSTEM APPLICATION TO THE LEFT-TURN
TREATMENTS

 

2.3.4 WHAT’S NEW IN THE ITAGI & MULTIMEDIA LEFT-TURN
CONTROL EXPERT SYSTEM?

 

vi

Page

xi

10
10
10
11
14
15
18
20
20
21
23

25

CHAPTER 3 DEVELOPMENT OF THE INTERACTIVE TRAFFIC
ACCIDENT GEOGRAPHIC INFORMATION

 

SYSTEM

3.1 THE ITAGI SYSTEM TOOL

 

3.2 TRAFFIC CHARACTERISTICS DATABASE
3.2.1 TRAFFIC ACCIDENT DATABASE
3.2.2 TRAFFIC VOLUME DATABASE
3.2.3 LEVEL OF SERVICE DATABASE
3.2.4 AVERAGE SPEED DATABASE
3.2.5 THE TRAFFIC ACCIDENT RATE DATABASE

 

 

 

 

 

 

3.3 GEOGRAPHIC INFORMATION SYSTEM
3.3.1 SINGLE INTERSECTION SELECTION
3.3.2 MULTIPLE INTERSECTION SELECTION
3.3.3 SINGLE MIDBLOCK SELECTION
3.3.4 MULTIPLE MIDBLOCK SELECTION

 

 

 

 

 

3.4 ROUTE GUIDANCE SYSTEM
3.4.1 SAF EST PATH PROBABILITY THEORY
3.4.2 DEMONSTRATION OF THE USE OF THE SAFEST PATH

PROGRAM

 

 

 

3.5 QUALITY CONTROL CHARTS FOR TRAFFIC ACCIDENTS ----------
3.5.1 DEFINITION OF QUALITY CONTROL
3.5.2 MONTE CARLO ALGORITHM

 

 

3.6 THE ITAGI SYSTEM TOOL USED IN THE CITY OF TROY ------------
3.6.1 THE CURRENT AND HISTORICAL TRAFFIC
CHARACTERISTICS DATA DISPLAY
3.6.2 TRAFFIC VOLUME DISPLAY
3.6.3 LEVEL OF SERVICE DISPLAY
3.6.4 TRAFFIC ACCIDENT RATE DISPLAY
3.6.5 SINGLE INTERSECTION SELECTION
3.6.6 SINGLE MIDBLOCK SELECTION
3.6.7 MULTIPLE MIDBLOCK SELECTION
3.6.8 ROUTE GUIDANCE SYSTEM

 

 

 

 

 

 

 

 

3.7 ACCIDENT REDUCTION BEFORE-AND-AFTER SCATS ..............
3.8 TRAVEL TIME SAVING BEFORE-AND-AFTER SCATS ................

3.9 ACCIDENT REDUCTION BEFORE-AND-AFTER PROHIBITING
LEFT-TURNS

 

vii

26

27

28
35
41
41
42
42

44
44
48
50
52

53
55

56

56
57
59

61

67
67
7O
7O
75
75
75
8O

8O

85

88

3.10 BEFORE-AND-AFTER PROTECTED LEFT-TURN CONTROL --------

3.11 SUMMARY
3.11.1 GIS CAPABILITY FOR TROY
3.11.2 GRAPHIC INTERFACE CAPABILITY FOR THE CITY OF

TROY
3.11.3 DETERMINING MULTIPLE INTERSECTION OR MIDBLOCK
CHARACTERISTICS
3.11.4 SAF EST (SHORTEST) PATH SEARCHING CAPABILITY ----

 

 

 

 

3.12 LIMITATIONS OF ITAGI
3.12.1 LACK OF DISAGGREGATE CAPABILITY
3.12.2 DATA AVAILABILITY
3.12.3 DATA ACCURACY

 

 

 

 

CHAPTER 4 DEVELOPMENT OF THE MULTIMEDIA LEFT-TURN
CONTROL EXPERT SYSTEM

 

4.1 KNOWLEDGE-BASED EXPERT SYSTEMS

 

 

4.2 OBJECTIVES

4.3 THE DESIGN PROCESS

 

4.4 PURPOSE

 

4.5 THE FACTORS USED IN DETERMINING THE APPROPRIATE
LEFT-TURN CONTROL
4.5.1 OVERALL LEVEL OF SERVICE AT THE INTERSECTION -----
4.5.2 LEFT-TURN LEVEL OF SERVICE AT THE INTERSECTION --
4.5.3 UPSTREAM AND DOWNSTREAM INTERSECTION LEFT

TURN LEVEL OF SERVICE
4.5.4 THE FREQUENCY OF LEFT-TURN ACCIDENTS PER
YEAR
4.5.5 THE LEFT-TURN VOLUME
4.5.6 THE OPPOSING THROUGH AND RIGHT-TURN VOLUME ----
4.5.7 THE AVERAGE SPEED IN THE DRIVER’S DIRECTION
(MPH)
4.5.8 THE AVERAGE SPEED IN THE OPPOSITE DIRECTION
(MPH)

4.6 SAMPLE SIZE

 

 

 

 

 

 

 

viii

98

100
100

102

104
105

105
105

107
107

108
108
111
112
112
112
114
115
115
115
115
116
116

116

117

4.7 KNOWLEDGE ACQUISITION AND REPRESENTATION ...............
4.7.1 METHODOLOGY
4.7.2 THE INITIAL SURVEY RESULT

 

 

4.8 MODEL VALIDATION
4.8.1 CONSISTENCY CHECK RESULTS
4.8.2 INDUCTIVE RULES’ VALIDATION RESULTS

 

 

 

4.9 REVISING THE PDC PROLOG PROGRAM

 

4.10 DEVELOPING THE MULTIMEDIA EXPERT SYSTEM

 

CHAPTER 5 SUMMARY

 

 

5.1 SUMMARY

5.2 THE RESULTS OF ANALYSES USING THE ITAGI SYSTEM ---------
5.2.1 BEFORE AND AFTER SCATS ACCIDENT CHANGES -----------
5.2.2 LEFT TURN POLICIES
5.2.3 TRAFFIC LANES IN MIDBLOCK LOCATIONS

 

 

5.3 DEVELOPMENT OF A MULTIMEDIA LEFT-TURN CONTROL
EXPERT SYSTEM

 

LIST OF REFERENCES

 

 

APPENDICES
A. THE EXPERT SYSTEM QUESTIONNAIRE (FIRST SURVEY) --------

B. THE EXPERT SYSTEM QUESTIONNAIRE (SECOND SURVEY) ----

119
119
123
128
134
136
136

137

139
139
141
141
142
143
144
146
151

151

157

LIST OF TABLES

 

 

 

 

 

 

 

 

 

 

 

 

Table Page
2.1 The Decision Table for the Left-Turn Control Strategies 24
3.1 The List of Intersection and Midblock Database File Names 35
3.2 The Definitions of Types of Traffic Accident 36
3.3 The List of Missing Data 43
3.4 A Comparison between Original Rank (i) and Monte Carlo Rank*(i)

(Intersection) 62
3.5 A Comparison between Original Rank (i) and Monte Carlo Rank*(i)

(Midblock) 64
3.6 The Average Speed on Rochester Corridor Before and After SCATS --------- 89
3.7 The Average Speed on John R Corridor Before and After SCATS ------------ 89
3.8 Intersections Implementing Prohibited Left-Tum Control in 1992 ------------- 93
3.9 Intersections Implementing Protected Left-Turn Control in 1992 -------------- 98
4.1 The Interviewed Traffic Experts List 119
4.2 The Percentage of the Traffic Experts Selecting Each Strategy

(Case 1-8) 124
4.3 The Percentage of the Traffic Experts Selecting Each Strategy

(Case 9-16) 125
4.4 The Percentage of the Traffic Experts Selecting Each Strategy

(Case 17-21) 126
4.5 The Selected Testing Questions for the Inductive Rules 131
4.6 The Model Validation Testing Rules for the Initial Survey 132
4.7 The Model Validation Results 133

 

LIST OF FIGURES

Figure

1.1

1.2

3.1

3.2

3.3

3.4

3.5

3.6

3.7

3.8

3.9

3.10

3.11

3.12

3.13

3.14

3.15

3.16 The AM Peak Total Traffic Volumes (Veh/Hour)

The Relationship Between the Interactive Traffic Accident
Geographic Information(ITAGI) System and Multimedia

 

Left-Tum Control Expert System

The Design Processes of the ITAGI System and the Multimedia
Left Turn Control Expert System

 

 

The Components of the ITAGI System

The Traffic Characteristics Database Flow Chart

 

The Geographic Information System Flow Chart

 

 

The Route Guidance System Flow Chart

The Quality Control Flow Chart

 

The Logo of Traffic Characteristics Database System

 

The Traffic Characteristics Database Menu Selection

 

List of Intersection ID Numbers

 

 

List of Midblock ID Numbers

 

The Midblock Mileage List

The Steps in Developing the Geographic Information System

 

 

The Steps in Developing the Route Guidance System

The Steps in Developing the Quality Control Chart of Traffic

 

Accident System

 

The Top Five High Intersection Accident Rate Areas

The Top Five High Midblock Accident Rate Areas

 

 

xi

Page

39
4o
41
43
44
47
48
50
51
55

56

68
73
76

78

 

3.17 The PM Peak Total Traffic Volumes (Veh/Hour)

3.18 The AM Peak LOS at the Intersections

 

 

3.19 The PM Peak LOS at the Intersections

3.20 The Intersection Accident Rate in 1994

 

3.21 The Midblock Accident Rate in 1994

 

 

3.22 The 1994 Top Five Angle Accident Intersections

3.23 The 1994 Total Intersection Accidents

 

3.24 The 1994 Top Five Angle Accidents Midblocks

 

3.25 The 1994 Total Midblock Accidents

 

3.26 The Number of Traffic Accidents & Injury Accidents (l989--1994)

 

(7x7 Arterials)

3.27 The Number of Traffic Accidents & Injury Accidents (l989--1994)

 

(5 Intersections)

3.28 The Number of Traffic Accidents & Injury Accidents (1989--1994)

 

(49 Intersections)

3.29 The Comparison of Intersection Accidents between 1991 and 1994 --------

3.30 The Comparison of Midblock Accidents between 1991 and 1994 ----------

3.31 Comparison of AM Peak Average Speed on Rochester & John R

 

Road between 1992 & 1995

3.32 Comparison of Off Peak Average Speed on Rochester & John R

 

Road between 1992 & 1995

3.33 Comparison of PM Peak Average Speed on Rochester & John R

 

Road between 1992 & 1995

4.1 The Steps in Developing the Multimedia Left Turn Control Expert

 

System

4.2 The Expert System Search Logic Flow Chart

 

xii

79

8O

81

83

83

86

87

89

90

102

103

105

106

107

110

111

112

138

147

4.3 The Model Validation Process for the Expert System 155

 

xiii

CHAPTER 1 INTRODUCTION

1.1 INTRODUCTION

The Intelligent Vehicle Highway System (IVHS) project in the city of Troy,
Michigan, is a unique federal demonstration project of the real time traffic adaptive signal
system called SCATS (Sydney Coordinated Adaptive Traffic System). This system
changes cycle length and the green and red phases based on current traffic demand. The
role of the universities involved in this project is to evaluate the impact of SCATS. There
are several measures of effectiveness (MOEs) being analyzed in this evaluation including
the number, type and severity of traffic accidents, the level of service (LOS) at each
intersection, and the average speed on the major arterials.

To facilitate the evaluation, the Department of Civil and Environmental
Engineering, Michigan State University has constructed an Interactive Traffic Accident
Geographic Information (ITAGI) System to analyze the impact of the SCATS on traffic
accidents. As a further development, the ITAGI system has been expanded to include a
multimedia left-turn control expert system. The relationship between the ITAGI system
and the multimedia left-turn control expert system is depicted in Figure 1.1. The
important features of the ITAGI system include accident location capability using a
limited geographic information system (GIS) for Troy, graphic interface capability,
determination of multiple (or group) intersections or midblocks characteristics and a

safest path searching capability.

The Multimedia Left-Tum The iTAGI System
Control Expert System

f__'\

 

Qualtty Control

Intersection I F; Ew-Bound I
; Midblock FLT NS-Bound i
\,_u-,_l.--- ___________ /‘

Route Guidance System

  

 

    

I Permissive I

/
Eastbound \l
8: NS-Bound

Traffic Characteristics Database

,1 Accident I Volume I {

Protected ‘ LO 3 d x Westbound
STMII ee

['4 ' I p /L——->::‘ ‘& NS-BOUI'Id

Geographic liformatlon System

‘\

l
//J

Figure 1.1 The Relationship between the ITAGI System and
Multimedia Left-Turn Control Expert System

 

 

 

I No Left Turn I

 

 

 

 

 

 

L J

The ITAGI system consists of computer programs used to assist in the analysis of
accidents and to display traffic accident characteristics, such as the type of accident at any
specified intersection or midblock location. Through a user-friendly PC-based
computerized database, the ITAGI system provides quick access to the data using
menu-selection. Through a geographic information system (GIS) function, the ITAGI
system allows the user to select data from any specified intersection(s) or road segment
(5) directly from the map. Thus, rather than being limited to textual queries, it is possible
to perform geographic queries. The user simply selects the desired points on the
computer screen, and acquires the desired information.

The ITAGI system has already embedded several hundred customized maps,
easily-understood color-coded graphics, and information associated with intersection or
midblock locations in the system. The analyst or the user can examine traffic accident
records (1989--1994) to determine high traffic accident locations (either at intersections
or at road segments). The data includes lane configuration, traffic accident data stratified
by location, type of accident and year. These and other data can be displayed on the
screen at the request of the user.

The ITAGI system also allows the user to see the aggregate decrease or increase
in the total number of accidents, or the percentage of various types of accidents on single
locations or a group of locations(intersection or midblock locations). This capability is
useful when a new traffic control strategy, such as SCATS, or when changing left-turn
control strategy from permissive to protected, or from protected to prohibited, is

implemented at specific locations.

In the FAST-TRAC federal demonstration program, Ali-Scout has been selected
as the route guidance and in-vehicle navigation system for improving travel efficiency
based on finding the minimum travel time path between any two points in the network.
The ITAGI system was programmed to select the route between any two points which
passes through intersections that had the lowest accident rates in the specified year. This
is named the safest path.

After integrating the ITAGI system with a knowledge-based system, it can be
used as an expert system for recommending left-turn control strategies. The procedure
used to develop the left-turn control expert system included conducting a survey of traffic
experts, and using the survey results to construct rules and facts to select the left turn
control strategy at Troy’s intersections. Integration of the knowledge-based expert
system (KBES) into the traffic operations center provides the engineer with a high level
of support in analyzing and recommending left-turn control strategies. In addition,
integration of KBES provides a tool to supplement the experience of individual operators
and reduce the potential for incorrect responses by traffic operations center staff. The
ITAGI system is fully integrated with the left-turn control expert system.

The ITAGI system was compiled by the PDC PROLOG (Programming in Logic)
program as a set of stand-alone executable files. PROLOG is a symbolic and artificial
intelligence (AI) language based on predicate calculus, and was developed in 1972 by
Colrnerauer and Roussel. PROLOG is a declarative language that uses simple relations
among facts, rules and queries [Teft, 1989]. A detailed introduction to the program is

provided in chapter 3.

1.2 OBJECTIVES
In this study, there are four objectives:

1. To produce a graphic interface capability for the City of Troy. The ITAGI
system has embedded several hundred customized maps, easily-understood color-coded
graphics and information associated with specified intersection or road segment locations
in the system. The data available to the user include: type of traffic accident, number of
traffic lanes, level of service, average travel speed, delay, accident rate, traffic volume,
and average travel speed on the major corridors. Any specified location(s) can be
selected and the specified type of accident (angle, turn, rearend, and others) can be
displayed in a bar chart, or pie chart format.

2. To develop a GIS capable of locating accidents in the City of Troy. A GIS is a
useful decision support tool for transportation applications because it provides a visual
display of a transportation network and associated spatial data. Therefore, via GIS, the
operator can examine accident characteristics over the past six years. The ITAGI system
provides a convenient format for the user to obtain the specified information from the
map. Rather than being limited to textual queries, it is possible to perform geographic
queries. The user simply selects the desired points on the computer screen, and acquires
the desired information.

3. To demonstrate the use of the ITAGI system in evaluating a new traffic control
system, such as SCATS. The ITAGI system provides the user with information on the
aggregate decrease or increase in the total number of accidents, or the percentage of type

of accidents at specified locations.

4. To use multimedia techniques to demonstrate how a knowledge-based expert
system (KBES) can be made more vivid and more effective than plain text in the expert
system or expert system shell. The ITAGI system provides the decision-maker
information on which traffic control strategies can be based. In this system, not only can
the expert knowledge be expressed by the computer in plain text format, but also in a

video (multimedia) format on the computer screen as well.

1.3 METHODOLOGY

In this section, the five steps used in constructing the ITAGI system and the
multimedia left—turn control expert system are described. The design process is shown
schematically in Figure 1.2. The first step is to write the SPSS and PDC PROLOG
program, then construct and verify all database files for intersections and midblock
locations from years 1989 through 1994. These database files are used by the geographic
information system, route guidance system, quality control chart of traffic accidents and
multimedia left-turn control expert system (the details will be explained in section 3.2).
The second step is to write the PDC PROLOG program so that the user can select any
specified points directly from the map (the details will be explained in section 3.3). The
third step is to write the PDC PROLOG program to develop the route guidance system so
that the user can specify two points on the map and obtain the safest path between two
points (the details will be explained in section 3.4). The fourth step is to write the PDC
PROLOG program to develop the quality control chart of traffic accidents so that the user

can identify high accident locations (the details will be explained in section 3.5). The

522m Sea .228 5:53.. 3352:: as ea 524$ .65. as .0

8382a .533 on... a; 2:9...

 

 

Sou-hm .396
.2280 E3 to... 3382::

 

A;

 

 

m amum

 

£31m .5203. .o .2200 >525

 

‘_

 

 

vaew

 

8296 852:0 Seam

 

1;

 

 

m .35

 

 

Sou-am 5:258... 2.32300

 

 

 

 

 

so:
.3325

came

 

 

 

.323:

cc:
.3385

coo w

 

 

 

 

330.88 5

«Sam Eo23<

2.020....-
33

 

 

.528.
3a.

 

 

 

20200.
no?

 

 

 

 

 

gaun— In EEO .2<
mag. .vmaw .noop .Naaw

 

 

.3on .32... .u><

Eugene
camp

 

 

 

20203
«amp

 

 

 

 

 

 

Segue-
wear

 

3c0200< x0032:

 

 

 

 

 

.30.; 2a
war .0 «our

 

 

3.01 §<
now? 3 «our

 

 

20203
33

 

 

 

3.25 .o .25..

.5203
ramp

 

 

 

2023'
naaw

 

 

 

 

 

ion an
"Gar .0 «car

 

 

ion 3
naaw .5 «our

.523-
33

 

 

 

 

 

 

3.3826: 9:53... 2:8...

2.323-
«3.

 

 

 

 

 

 

.523-
meow

 

8c.300( c0300¢50~5

 

rlf

$323 E0202 oEmt.

 

N amum

Guam

 

 

 

 

 

final step is to build the multimedia left-turn control expert system (the details will be

explained in chapter 4).

1.4 ORDER OF PRESENTATION

The order of presentation in this paper is as follows:

Chapter 1 Introduction

Chapter 2 Literature Review

Chapter 3 Development of the Interactive Traffic Accident Geographic
Information (ITAGI) System

Chapter 4 Development of the Multimedia Left-Tum Control Strategies Expert
System

Chapter 5 Summary

CHAPTER 2 LITERATURE REVIEW

Left-tum control strategies have significant impacts on intersection capacity,
traffic operations, and safety of signalized intersections. Selecting the proper left-turn
Signal phasing can improve the level of service (or decrease intersection delay) and
reduce left-turn-related accidents. One of the objectives in this study was to provide
assistance in making the trade-off between left-turn efficiency and safety among three
left-turn treatments, permissive, protected, and prohibiting left turns. Often traffic
engineers rely on experience or trial-and-error until a suitable solution is found. From
the literature review, it was determined that most of the left-turn treatment guidelines
were developed by either before-and-after studies that compared a change from one type
of phasing to another or computer methods that optimize/simulate the network signal
timing for different left-turn treatments. Only a limited number of articles dealt with the
selection of a left-turn control strategy by expert systems, and none of the previous
studies discussed guidelines for selecting permissive, protected, and prohibiting left turn
control. In this chapter, the literature on the types of left-turn control, expert systems
applications to transportation engineering, and expert systems applications to left-turn

control treatments will be reviewed.

10

2.1 LEFT -TURN CONTROL GUIDELINES

2.1.1 TYPE OF CONTROL

A detailed review of previous research on the types of left-turn control and the
guidelines for the selection of left-turn phasing and indication sequences was
undertaken. Most of the guidelines were based on either a before-and-after study (Agent
[1987], Stonex et al. [1987], Asante et al. [1993], Warren [1985], Cottrell [1986],
Upchurch [1986, 1991], Greiwe [1986], Lalani et al. [1986], Shebeeb [1995], Florida ITE
[1982]) or simulation/optimization methods (Rouphail et al. [1994], Chang et al. [1988],
Cohen et al. [1985], Hummer et a1. [1991]). In this section, the left-turn phasing
patterns and guidelines will be introduced, then the left-turn indication sequences and

guidelines will be presented.

2.1.2 LEFT-TURN PHASING PATTERNS

According to the 1988 Manual of Uniform Traffic Control Devices (MUTCD),
there are three left-turn signal phasing patterns which are commonly used at signalized
intersections. These are the permissive (or permitted), protected (or exclusive), and
protected/permitted (or exclusive/permissive) phases. Permissive left turn means
vehicles are allowed to make a turn on a circular green indication but must yield to
opposing traffic. Protected left turn means vehicles are allowed to make a turn only on
a green arrow indication and have the right of way while the green arrow is displayed.
Protected/permissive means vehicles are allowed to make a turn either on a green arrow

indication or on the circular green after yielding to oncoming traffic.

2.1.3 GUIDELINES FOR LEFT -TURN PHASING

Upchurch [1986], through an analysis of time-lapse film at six intersections,
developed guidelines for selecting left-turn phasing based on the following decision
variables: cycle length, available Sight distance, product of left-turn and opposing
volumes (a volume cross product), Speed of opposing traffic, number of opposing
lanes, and left-turn accidents. This study concluded that protected left-turn phasing

should be provided when the left-turn volume during the peak hour is more than two

vehicles per cycle and the volume cross product is greater than 100,000 (Vph)2 and

144,000 (vph)2 for two and three opposing lanes, respectively. A combination of
protected/permissive phasing is recommended unless the opposing speed is greater than
45 mph, there are three or more opposing lanes, sight distance is restricted, or a severe
accident problem exists. However, in Upchurch’s study, the definition of a severe
accident problem is not clear and there are no general guidelines for the prohibiting left-
turn treatment.

Cottrell [1986] also developed guidelines for the use of protected/permissive
left-turn phasing. Data were collected at 45 sites in Virginia, including 20 with
protected/permissive lefts, 15 with protected-only lefts, and 10 with permissive-only
leits. Based on these field observations, specific guidelines for left-turn phasing were
developed. He concluded that protected/permissive phasing should be used when the
peak hour volume cross product is between 50,000 (vph)2 and 200,000 (vph)2 and the

left turning volume exceeds two vehicles per cycle; average peak hour delay for left-turn

12

vehicles exceeds 35 sec/veh and the total peak hour left-turn delay exceeds two
vehicle-hours (with permissive left-turn only); there are no more than two opposing
lanes; good intersection geometries and good access management exist; and annual
protected/permissive delay is less than that of protected-only delay. In this study,
prohibiting left-turn guidelines were not developed.

Another significant study in this field was undertaken by the Florida Section of
ITE [1982]. Accident analyses were conducted at 17 approaches that were converted
from protected-only to protected/permissive operation. They concluded that
protected/permissive phasing should be provided for all intersection approaches for
which protected lefis are provided unless there is a compelling reason for using
protected-only phasing. Conditions where protected-only phasing was recommended
include a double left-turn lane, restrictive intersection geometry requiring
protected-only, restricted sight distance, the leading left of a lead/lag phasing
sequence, speed limit of opposing traffic greater than 45 mph, three or more opposing
lanes, and more than six left-turn accidents in one year on the approach with
protected/permissive phasing. Once again the prohibiting left-turn guidelines were not
developed.

Asante et al. [1993] also developed guidelines for permissive,
permissive/protected, and protected left-turn treatments. He recommended protected or
permissive/protected left-turn control should be implemented when the sight
distance for the left-turning vehicle is restricted on the basis of the posted speed limit

for the opposing traffic, or more than eight left-turn-related accidents have occurred

13

within the last 3 years at any one approach with permissive left turn phasing, or
more than 450 left-tum-related conflicts per million (vph/lane)2 are observed at an
approach with permissive left turn phasing, or the plotted point representing the peak-
period volume in vehicles per hour (based on the peak 15 minutes) and the
corresponding opposing-traffic speed limit. He suggested that the protected only left-
tum control Should be used under any of the following conditions:
. Approaches with restricted sight distance, as determined on the basis of posted
speed limit on the approach opposing the left-turn traffic, or
. Approaches with four or more opposing lanes that must be crossed by the
left-turning traffic; or
. if any two of the following conditions are met:
. Peak-hour volume measured at lS-min intervals for the left-turning
traffic is greater than 320 vph,
. Peak-hour volume measured at 15-min intervals for the opposing traffic
is greater than 1,100 vph,
. The opposing speed limit is greater than or equal to 45 mph,
. There are two or more left-turn lanes exist; or
. when one of the following conditions or combination of conditions exist:
. Three opposing lanes and the opposing speed limit is greater than or
equal to 45 mph,
. Left-turn volume is greater than 320 vph and the percent of heavy

vehicles in the left-tuming traffic exceeds 2.5 percent,

l4

. Opposing volume is greater than 1,100 vph and the percent of heavy
vehicles in the left-tuming traffic exceeds 2.5 percent,

. Seven or more lefi-tum-related conflicts per million (vph/lane)2 are
observed for a permissive/protected approach, or

. Average stopped delay for left-turning traffic is acceptable for protected
only phasing and the traffic engineer judges that the use of
permissive/protected phasing will result in a greater number of left-turn
accidents.

The guidelines developed by the various before-and-after study are difficult to
synthesize because not all of the tested intersections were similar during the
before-and-after periods. The characteristics of the intersections used in the various
studies were not specified in the reports. Differences in these characteristics such as type
of phasing, number of opposing lanes, left-turn volume, opposing volume, and the

existence of a separate left-turn lane could lead to different results.

2.1.4 LEFT-TURN INDICATION SEQUENCES

Three left-turn indication sequences have been discussed in previous studies;
leading, lagging, or a leading/lagging sequence. Leading left turns are, from an efficiency
standpoint, more desirable since they are associated with lower delays and increased
intersection capacity. However, accidents were reported to be more frequent at
intersections with a leading left turn. Lagging left turns, on the other hand, appear to be

safer. Leading/lagging operation are often implemented for reasons other than the

15

reduction in delay, such as for progression purposes on an arterial to get the maximum
bandwidth in the time-Space diagram.

The Dallas phasing is not commonly used, but can be found in the recent literature
[Asante et al., 1993]. This is a modified leading/lagging, protected/permissive
sequence. During the portion of the cycle when one of the left turns is protected and its
adjacent through movements plus right turns are shown a circular green signal, the
opposing left turn is permitted. Because the throughs and rights adjacent to the
permitted left are shown a red signal (because the opposing left is protected), Dallas
phasing leads to a unique display. Motorist surveys have indicated that drivers
understand the Dallas phasing as well as or better than they understand other types of

left-turn phasing [Asante et al., 1993].

2.1.5 GUIDELINES FOR LEFT-TURN INDICATION SEQUENCES

Hummer [1991] developed some guidelines, by using the NETSIM simulation
program, for the use of leading and lagging phase sequences. He suggested that lagging
instead of leading phase sequences should be used when serving heavy pedestrian
traffic after checking for the possibility of trapping. Trapping will occur in intersections
where one approach has a permissive left turn and the opposing approach has a lagging
sequence. If trapping is possible, the phasing should be changed to eliminate that
possibility by eliminating the permissive turn (making it protected-only or prohibiting
the turn), by eliminating the lagging sequence, or by ensuring that the opposing

approaches both have lagging sequences with left-turn phases that begin simultaneously.

16

In this study, many of the decision variables are binary (such as is there heavy pedestrian
crossing volumes), but the paper does not specify how many pedestrians constitute a
heavy crossing volume.

Upchurch [1991] also compared the left-turn accident rates for different types of
left-turn phasing. A total 523 signalized intersections were included in developing
the accident statistics. This study concluded that the left-turn phasing type at a
signalized intersection affects the left-turn accident rate, with permissive left turn policies
being the most hazardous, and leading protected being less hazardous than lagging
protected. The accident rate is also influenced by the number of opposing lanes of
traffic, left-turn volume, and the volume of opposing traffic.

After collecting data from 216 intersection approaches, Asante et al. [1993]
used a before-and-after study method to construct the guidelines for the sequence of
phasing (leading, lagging, or leading/lagging). He used 10 decision variables which
affect the intersection performance. Those variables are left-turn volume, opposing
volume, volume cross product, vehicle mix, ratio of green to cycle length for left turn,
number of left-turn lanes, number of opposing lanes, volume-to-capacity ratio of the
approach, speed of opposing traffic, and sight distance. He indicated that a lagging
sequence is recommended when it is intended to improve the safety of an already
installed leading sequence under which more than 190 left-turn conflicts per million
(vehicles per hour per lane)2 are observed, or the lagging left-turn sequence is
necessary as part of an overall network progression scheme. A leading/lagging

sequence is recommended for intersections when there is inadequate space within the

intersection to safely accommodate a dual left-turn operation, or it is necessary for the
progression scheme.

Chang et. a1 [1988] used the MAXBAND program for optimizing the left-turn
phase sequence in multiarterial closed networks. In determining optimal signal timing
plans for signalized networks, Chang et al. used four decision variables to optimize the
signalized networks. They are offset, green phase time, cycle length, and left-turn phase
sequence. This study included a comparison of MAXBAND-produced timing plans with
and without phase sequence optimization and an analysis of the effects of using phase
sequence patterns given by MAXBAND in the TRANSYT-7F program. In other words,
first, the MAXBAN D program was executed to provide timing plans for each of the 10
arterials in the network, then TRANSYT-7F was used to optimize the timing plan for the
whole network, finally the NETSIM program was employed to simulate the network.
However, the delay-based programs, such as TRANSYT-7F, and bandwidth-based
programs, such as MAXBAND, measure the performance in terms of minimum delay and
stops. The delay-based programs optimize green phase time, offset, and cycle length; and
the bandwidth programs optimize offset, cycle length, and phase sequence. In fact, these
programs did not consider the safety and efficiency factors as their performance index
(PI). Often, the researchers just focus on minimizing the delay and stops, while safety is
not taken into account.

The guidelines developed by the various authors cited in this review are based on
many common variables, including:

. the product of left turning and opposing volumes,

l8

. speed of opposing traffic,

. left turn related accidents,

. number of opposing lanes and

. available sight distance.
In addition, Upchurch [1986] included cycle length as a variable. Cottrell [1986]
included the left turning volrune and delay as important variables. The Florida ITE
guidelines added dual left turn lanes as an additional variable to be included. Asante et
al. [1993] also listed delay and left turning volumes as important variables. All of these
variables, with the level-of-service serving as a surrogate for delay, are included in the
expert system developed in this study. Some of the variables are used as conditions
(adequate sight distance), as they are seldom an issue in an urban setting, and there are no
intersections in the City of Troy that have restricted sight distance. A limitation of each
of the cited studies is that they do not provide guidelines for determining when left turns
should be prohibited, as well as how the choice between protected and permissive should

be made. The expert system developed in this study addressing this deficiency.

2.2 EXPERT SYSTEMS AND TRANSPORTATION ENGINEERING

Expert systems are computer programs that consist of a collection of heuristic
rules and domain facts used by human experts in solving complex problems. The
development of knowledge-based expert systems (KBES) is one of the most
significant accomplishments in the field of artificial intelligence (AI). Since its initiation,

the KBES technology has drawn much interest because of its problem-solving capability.

19

Many prototype expert systems have been developed in a number of disciplines,
including agriculture, chemistry, computer systems, electronics, engineering,
geology, information management, law, manufacturing, mathematics, medicine,
meteorology, military science, physics, process control, and space technology.

The use of knowledge-based expert systems in transportation engineering has
also been increasing. For example, TREMEX is an expert system which offers advice to
the officer in charge of an incident, before the arrival of the expert [Lycett et al., 1989].
SRM (rollover threshold), YAWROLL (steering maneuver without braking), and
PHASE4 (steering and braking maneuver) are integrated as an expert system for
conducting stability analysis of heavy vehicles [Sankar et al.,1991]. There is also a
knowledge-based geographic information system for safety analysis at rail/highway grade
crossing [Panchanathan et al., 1995] and a system to address various aspects of heavy
truck size and weight impacts on pavements, bridges, truck safety, and economics
[Robinson, 1990]. In the field of traffic engineering, ISMIS is an intersection safety
management information expert system. This system provides users with measures
available for remedying a particular intersection safety problem. It permits the user to
perform an analysis of cost-effectiveness based on statistical decision theory to determine
the best alternative remedy [Seneviratne, 1989]. Other examples include expert systems
for roadside safety [Zhou et al., 1991], for design of highway safety structures [Roschke,
1991], for highway safety[Sayed, 1994], for acoustic design of highway noise barriers,
CHINA [Harris et al., 1985] and for selecting traffic analysis software [Chang, 1987].

Although much work has been done, only one expert system application to the left-turn

20

control treatments was found (Chang [1987]).

2.3 EXPERT SYSTEMS AND LEFT-TURN TREATMENTS

Several research studies on combinations of left-turn signal phase patterns and
indication sequences have been discussed. However, none of the previous study
considered the choice of permissive, protected and prohibiting left-turn control. One of
the purposes of this study was to develop an expert system that can assist the engineer in
making this three way choice. In this section, literature discussing the trade-off between
left-turn efficiency and safety will be presented, then the value of a left-turn control
expert system will be addressed, then an example (literature) dealing with an expert
system application to the left-turn treatment will be introduced, and finally the unique
features of this study (the ITAGI system and left-turn control expert system) will be

explained.

2.3.1 TRADE-OFF BETWEEN LEFT-TURN EFFICIENCY AND SAFETY

Shebeed [1995] and Warren [1985] studied the trade-off between left-turn
efficiency and safety. A total of 54 intersections comprising 179 approaches were
studied by Shebeed [1995]. The traffic volume counts, signal timing data, and left-turn
accidents were collected to calculate the accident rate and stopped delay. The
stopped delay for the left-turn group for each approach was calculated from the
Highway Capacity Software (HCS). Seven patterns for left-turn phasing were

considered: 1) permissive only, 2) lead-protected only, 3) lag-protected only, 4)

21

lead-protected permissive, 5) lag-protected permissive, 6) lead Dallas, and 7) lag Dallas.
The results show that a definite trade—off exists between left-turn accident rates and
delay. The protected approaches are less efficient; however, they offer a higher level of
safety than other phasing types. He concluded that protected phasing should be applied
as a means of enhancing left-turn safety when the expected delay is acceptable, based
on the desired level of service. Protected/permissive approaches are more efficient but
less safe than protected-only and Dallas approaches. Permissive-only approaches are
associated with the best efficiency but the highest accident rates. There is no significant
difference in efficiency and safety between lead and lag left-turn sequences.

Warren [1985] also concluded a study of the accident experience as a function of
the left turn phasing. He found that at the intersections that were converted from
protected to protected/permissive phasing, rearend and total accidents decreased, while
left turn accidents increased dramatically. From the survey result, he reported that
motorists overwhelmingly favor protected/permissive left turn phasing and want
additional protected-only signals converted to protected/permissive phasing. This means
that motorists are willing to accept the additional risk of protected/permissive left turns
in exchange for increased driving freedom, delay savings, and reduced driver frustration

from not having to wait at traffic lights for no apparent reason.

2.3.2 THE VALUE OF A LEFT-TURN CONTROL EXPERT SYSTEM
Expert system technology is a new approach to problem solving. It is designed

to provide the level of performance of a human expert in a specific professional domain

22

and enable a computer to assist people in analyzing specific problems using that
expertise [Waterman, 1986]. Because of the trade-off among the safety, efficiency,
and delay in left-turn control decision making, a prototype knowledge-based expert
system has been developed to address this problem.

The value of a left turn control expert system is based on the following conclusion
derived from the literature review.

1) Currently, there is no uniform and consistent guideline to assist traffic
engineers in selecting the appropriate phasing type. Engineers often rely on experience or
trial-and-error until a suitable pattern is found. Building an expert system can accelerate
the problem solving and decision making. The reasons for using expert systems in
transportation engineering are the same as for using any type of automation: the use of
less skilled personnel, quicker solutions, and more reliable solutions. An expert system
does not replace the human being, and it merely assists the human being solve a problem
quickly and efficiently.

2) There exists potential trade-offs between left-turn efficiency and safety.
Because of the trade-offs existing in the left-turn control treatments, an expert systems
can speed up the problem-solving .

3) There are many decision variables that need to be considered in selecting
left-turn treatments, such as sight distance, number of opposing lanes, traffic volumes,
delay and traffic accidents. Expert systems can help process the data and express
procedures in understandable rules because the expert system can store past decisions as

symbolic rules, search the symbolic rules and use these symbolic rules to assist the

23

engineer in making decisions.

2.3.3 THE EXPERT SYSTEM APPLICATION TO THE LEFT-TURN
TREATMENTS

Only one previous study of the use of an expert system to select left-turn control
treatments was done. Chang [1987] used TURBO PROLOG, PD PROLOG, and
INSIGHT 1 programs to construct an expert system to select left-turn control
strategies. He used six major decision variables in his expert system. The decision
variables were derived from Upchurch’s study as explained in section 2.1. They are
common traffic input information: the left-turn demand, number of opposing through
lanes, volume cross-product of the conflicting left-turn and through movement pairs,
opposing travel speed, sight distance restrictions, and history of severe left-turn
accidents. The factors used in his expert system were binary choices, such as: is the
left-turn demand greater than or less than two per cycle; is the number of opposing lanes
equal to two or three; is the opposing speed greater than or less than 45 mph; is the sight
distance adequate or not; and is there a history of severe left-turn accidents.
To clarify the basic relationships among the outcomes of different data input, three
different left-turn control treatments were combined with 15 different possible
situations. In Table 2.1, the first horizontal row lists the options considered in this
expert system (i.e., permissive phase, exclusive phase, and exclusive/permissive phase).
The second horizontal row lists the 15 possible conditions ranging from Conditions

“A”,”B”,”C” through “P” (refer to the second row in Table 2.1). In each column, X

24

Table 2.1 The Decision Table for Left-Tum Control Strategies

urn

Bertram: 2
Demand <= 2

Lanes = 2
Lanes = 3

> 144.000
<= 144.000
> 100,000
<= 100.000

> 45
<= 45

W/Restriction
No Restriction

Could Be Corrected
Exclusive Phase _
Could Not Be corrected

Exclusive Phase

 

Source: [Chang, 1987]

25

represents the requirements for fulfilling a certain decision criterion. For example, in
Condition M under the exclusive phase, there are three Xs, one representing left-turn
demand > 2 , the other standing for opposing lanes =3 and another the volume cross

product > 100,000. The existence of these three conditions causes the exclusive left-

turn phase treatment to be recommended.

2.3.4 WHAT’S NEW IN THE ITAGI & MULTIMEDIA LEFT-TURN CONTROL
EXPERT SYSTEM?

The factors used by Chang were based on Upchurch’s guidelines [1986] for the
choice between permissive and protected left turn treatments. The intent of this study
(the ITAGI system and multimedia left-turn control expert system) was to expand on
Chang’s expert system to include the choice of prohibiting left turns and to develop the
ITAGI system as a multimedia expert system which makes the knowledge presentation
more vivid and effective than the traditional plain text.

In addition, the binary choice used by Chang was improved by categorizing
several variables into different levels according to real traffic conditions found in the City
of Troy. All the knowledge was acquired from traffic experts by questionnaires, and an
ID3 inductive tool was employed to develop the rules and facts for selecting the left-turn
control treatments. The ID3 is a machine learning tool used to express the decision
tree (or decision table) with the minimum nodes and links. The ID3 methodology will be

explained in chapter 4.

 

 

 

 

.s..‘

’. De

K-

 

CHAPTER 3 DEVELOPMENT OF THE INTERACTIVE TRAFFIC

ACCIDENT GEOGRAPHIC INFORMATION SYSTEM

As outlined in chapter one, the ITAGI system has been integrated with the
multimedia left-turn control expert system. With the push of a button, the user can switch
between the ITAGI system and the multimedia left-turn control expert system. The
ITAGI system was designed to provide the user quick access to the traffic characteristics
database. This information, such as the number and type of accidents, link speed, level of
service, and traffic volume are important variables used to determine the appropriate
left-turn control strategy.

In this chapter, the ITAGI system and its application in Troy will be introduced.
The multimedia left-turn control expert system will be presented in chapter 4. There are
four components of the ITAGI system: The traffic characteristics database, the
geographic information system, the route guidance system, and the accident quality

control chart.

The traffic characteristics database provides the user with information on
historical and current records of traffic accidents, volrune, level of service at the
intersections, and average speed on selected corridors. The user can access relevant data

fiom the database by making selections from a menu.

The use of a geographic information system (GIS) will permit the user to select
data for any specified location within the city. Via GIS, information on the types of

traffic accidents at any specified intersections or midblocks (individual site or group of

26

‘1‘"

1:03} x

. ‘Im
L 7%..

.91 «v u

’\ \
.‘ .L.\ ’
1 -
1‘-~..
' ‘ I \
I. )1
n
.w.»
r

1.533.

e .

.- t‘;“,‘;1
5.“ - ~

1..

27

locations) can be displayed.

At the request of the user, the route guidance system will search for and display
the safest path through the network. The ITAGI system permits the user to input two
desired points (origin and destination) and obtains a display of the safest path on the

screen.

The quality control chart algorithm was used to identify the locations where the
average accident rate was higher than the statistically defined level of confidence over the
past six years (1989--l994). The Monte Carlo simulation method was used in this
analysis. The use of this algorithm can be used to identify high accident locations for the

Highway Safety Improvement Program (HSIP).

3.1 THE ITAGI SYSTEM TOOL

As described in the above section, the ITAGI system was designed as a tool to
assist the decision maker (traffic engineer) in retrieving and analyzing traffic
characteristic data in text or graphic format. It consists of computer programs used to
analyze and display traffic characteristics. The entire ITAGI system and multimedia left-
turn control expert system executable files are stored on a single CD. Upon entering the
ITAGI program, a logo will appear showing the title and the authors of the program.
After pushing the “right arrow” button, the main menu of the system will be displayed.

From this menu, the user can choose a desired item, and go directly to that specific topic.

In the main menu, there are five selections: “Introduction”, ”How to Use the

Program”, ”Purpose of the Program”, ”Interactive Geographic Information”, and

.h-
-r "‘2

5

s '1, ‘ ‘
"‘.~ ._.
o

.
-~'VC..

-.... u“

~- .... a

‘.

~. ‘A '

C‘\ I'.‘

.n, “1““-

‘n J. ‘1

\'&. l

e

A, ‘ _‘

:1: ml .
. l“ )1
.‘m '

“. 1“ -
“W“k

28

“Left-Turn Control Expert System”. The “Introduction” section explains the
objectives behind the program development; the “How to Use the Program” item tells
the user how to access and utilize the program; the “Purpose of the Program” illustrates
the objectives of the program; the “Interactive Geographic Information” item brings
the user to the ITAGI GIS system; the “Left-Turn Control Expert System” item brings
the user directly to the multimedia left-turn control expert system. The components of
the ITAGI system are depicted in Figure 3.1. A flow chart for each component of the
ITAGI system is shown in Figures 3.2 through 3.5. Figure 3.2 shows the flow chart of
the traffic characteristics database. Figure 3.3 shows the flow chart of the geographic
information system. Figure 3.4 shows the flow chart of the route guidance system.
Figure 3.5 shows the flow chart of the quality control algorithm. The details on the use of
the multimedia left-turn control expert system are described in chapter 4. This chapter

will introduce the ITAGI system.

3.2 TRAFFIC CHARACTERISTICS DATABASE

The design process of the traffic characteristics database included writing the
PDC PROLOG program, then constructing and verifying all database files. The database
files include traffic accidents for intersections and midblock locations for years 1989
through 1994; traffic turning movements for selected intersections (AM & PM peak); the
level of service for selected intersections (AM & PM peak); average travel speed for
selected corridors (AM peak, Off-peak, and PM peak); and traffic accident rates for

intersection & midblock segments.

29

 

/, L -7. W- -. -
A Multimedia Left-Turn Control Expert System I

 

! interactive Geographic Information System
L (lTAGI System)

 

 

 

 

 

I - - , _ r I _ J
Trefllc Ctrerectedetice 3009"?th ”gm" ‘1
m... System ’ Route Guidance System Quality Control Chat

(5) <2“) (9 (9

(Figure 3.10) (Figure 3.11) (Figure 3.12) (Figure 3.13)

   

Figure 3.1 The Components of the ITAGI System

30

(MW Traffic Characteristics Database
/ - _

 

 

 

 

 

 

 

 

 

 

 

. Database 1 Graphics Comparison ‘ Guit '
.. “IT _ _ - u--. End
1 L H‘j\

 

 

 

 

 

on Peak PM Peak 1

 

 

 

__.___l_ __l_ .- ”71 ,_—l Traffic
Accident Data Volume on L“ °' . AW” 4 Accidents
_ , §°'VJ°°- ,, SPEL I
,/" “ . 1
L , 1 A1 1 AM Peak
L ,O

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1989 1990 1991 1 1992 AM Peak
. . l, , l ,* A l__
Choose a Year
_ 1 _ -1, . __~1k “Mg ,7 L ,
South Blvd 1 Square Lake . Long Lake 1 Wattles l ' 14 Mile 1
Choose a Location I
‘ W 1
Adam 1 Coolidge 1 Crooks Dequindre
@« 1...... W
/-N _
1’ A1 \1 fl L 7 J
/ 1 1994 8.1992 1 1994 8.1991 1993 8.1992 1993 & 1991

 

 

 

 

 

 

. hurl

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

e HT-T - * W w—i w .
1“ Accident . injury (Mum, Om Len-Tum Level or Approach Average 1
L Rates 1 Accidents __* L“ Bum 96__ Service LCS Speed 1
L1 w
intersection 1 Midblock 1 W T m ! mien—7 1
,5 l _J _ _ 7 _
(A2 >‘__‘Y
T I AM Peak For Peak PM Peak

 

 

 

 

 

 

Y

T
Cityorrroy ‘9. i 1 G2 ____m_l_
Intersections 1 Intersections ; \1

T4 _ ______.l

Figure 3.2 The Traffic Characteristics Database Flow Chart

31

il/ \‘

l l

1 Geographic Information System

 

Intersection- Midblock- amineaigng Midblock?”
Single 4 Multiple 1 Multiple

l

 

 

 
  
 
  

 

 

 

 

 

BarChart- 1 91.36915 ’Tops-In—tgc HighAccident
. (or Midblock) Location

 

 

lntec__1___ Intec__ _

 

 

 

 

 

1989
;__wl, ,

0W 3" 1No.ofAccidents
CM" _1(1939—1994)

 

 

Select a Point

Angle 1 Turn Rearend Others
, I 1 from the Map

 

 

 

 

Continue Main Menu '

 

 

 

 

, 1
Type of i lcddemi Display Pie

—— (Wcmi1 Chart 1 — _

_ L Continue Main Menu
(an
\7_7 /’ . Lfi __ _ ,__ 1

Continue Main Menu 1

,,__ A, , ,/-- / .-
I I 32 5 B1\
.\\> M I .\ I)

/' >\

1' 81".

\ /

 

 

 

 

 

1 Total '7 TAHgIe "1 Turn A T‘WREr—end’fo—IEMTT
1 [Accidents ‘ Accidents 1 Accidents V_Accidents_ ‘ AccfientLJ

.#

—1

1 Continue 1 Main Menu 5

Figure 3.3 The Geographic lnfonnation System Flow Chart

" arena.—

 

 

 

 

 

 

 

 

 

 

32

  
  
  

533“;

 

 

 

 

 

 

 

 

   
       

‘4. Chan ‘ Exit
1 Multiple Selection Multiple Selection
'—_‘> 5“” a Pam rTSeIect a Point 1
from the Map from the Map r” “\,
L ,__.____1 1 End
F "x- _ _ ,/
Add One More 1 Show Graphic Main Menu - Exit 1
1— . Add One More Graphic 1 Main Menu- Exit 1
1 Show the Aggregate 1 7 mm __ #1!‘ _7 1
‘ Result- Bar Chart 1
1 of} - —
. -/ Show the Aggregate Type of Accidents
| — ~— ——~« Result- Pie Chart (1989—1994)
Display the Locations Exit . ,
the User Selects /‘33 1
_ \__,/'
1 Display the Locations j—T
the User Selects
“at f. \A‘:
\y/ * ./

Figure 3.3 (cont'd)

l

33

 

1 Route Guidance System

7 _ I at

1
Eastbound . Westbound

 

 

 

 

    

Select Any Two 1
1 Points on the Map ‘
1 1
1' Show the Safest 7"
Path on the Map

 

1 Continue 1 Exit

L7,, a sh,

End

Figure 3.4 The Route Guidance System Flow Chart

34

1/ D\‘=
\\ /

1

1— Quality Control Chart

1

____ MEblook-EVJ Midblock-NS
1 Intersection ' Direction 1 Direction 1

 

 

i

 

 

Display the Outlier
Locations on the Map ‘

1" End 1
\g V 7/

 

\\

Figure 3.5 The Quality Control Flow Chart

35

3.2.1 TRAFFIC ACCIDENT DATABASE

The intersection accident database and the midblock accident database file names
are listed in Table 3.1. The type and number of accidents were derived from the MDOT

(Michigan Department of Transportation) master tape.

Table 3.1 The List of Intersection and Midblock Database File Names

 

 

 

 

 

 

 

 

Year Intersection Database File Midblock Database File
1989 _ W ,, aoc89£lL7v fl _ W 7 fl W b89.dba
1990 acc90.dba b90.dba
1991 acc91.dba b91.dba
1992 aocQZdba b92.dba
1993 acc93.dba b93.dba
1994 acc94.dba b94.dba

 

 

 

 

All directions and all types of traffic accidents are declared as an integer in the
domain of the PDC PROLOG program. The definitions of angle, turn, rearend, and other
traffic accidents as constructed from the Michigan State Police files are listed in Table
3.2.

When the “Traffic Characteristics Database” item is selected from the menu of
the ITAGI system, a logo of the traffic characteristics database system will appear (see
Figure 3.6), followed by an option menu. The user then selects the desired item from the
menu (see Figure 3.7). After selecting the “accident data” item and the desired
combination of attributes, the traffic accident output displays several items. They are the
year of the data, the intersection selected and its ID number, and the number of angle,

turn, rearend and other accidents. A map of the ID number for each intersection and

 

hill: 3

mm 1

E

II!
‘ul
0

,JJJMWE1]J1\11m\1..1\2\_1..\11.\-1\.-\ H: \ 5K \

‘.\
c
r
We
.V,

36

Table 3.2 The Definition of Types of Traffic Accidents

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Accident Type Meaning Angle Turn Rearend Others
000 Miscellaneous Vehicle X
010 Overturn x
020 Hit Train X
030 Hit Parked Vehicle x
048 Backing x
049 Parking X
050 Pedestrian x
060 Fixed Object X
070 Other Object X
080 Animal X
090 Bicycle X
141 Head-on X
144 Angle Straight X
147 Rear-End X
244 Angle Turn X
342 Side Swipe Same X
345 Rear End Left Turn X
348 Rear End Right Turn X
440 Other Drive X
444 Angle Drive X
447 Rear-End Drive X
543 Side-Swipe Opposite X
545 Head-On Left-Tum X
845 Dual Left-Tum X
646 Dual Rig-Turn X

 

Source: MDOT Traffic Accident Tape Format

 

37

Traffic Characteristics Database -A-
r“ Copyright (C) 1995 by
Dr. William c. Taylor ‘
& Shung-Kung Wu
TEL: (517) 353-7161
Civil Engineering
Michigan State University
Dec.13,1994

 

 

 

Press any key to continue ......

Figure 3.6 The Logo of Traffic Characteristics Database System

38

 

 

1 Database Graphics Comparison Quit

 

AccidentDeta

 

1 2 VoiumeData
LevelofService
AverageSpeed 1

 

 

l

r—-——Cheeaem1rearoera — _

 

 

 

 

 

 

1‘ raea
‘90 r— .ngmmw - 1
1&1 South Blvd 1
1*: Square LIRO _ mg. um-mmoouna Arterial _ ‘
Long Lake 1 ggm , _ ,7 _
read 1 Wattles “doe ‘ .
Big Beaver Crooks
MID" Livemoia
14 m. Rochester
—‘—‘—’* “T John R
Dequindre

. 1
i 1 Select with mouse or use upper case letter
L _ . . _- _ __ .

 

 

 

Figure 3.7 The Traffic Characteristics Database Menu Selection

v3.

firm-\‘(fllul‘

39

CITY OF TROY

SOUTH BLVD.

 

 

 

 

 

 

 

 

 

 
 
 

 

7 7 7
3
ATMzSCmo A
fwl11 .71.. 7-..... -. 1.1. 7A
fimzzosA_ a m _ .._.
_ AlA _. AA _ L m .. _
A A A _ 5 5 m 5 B A 5. A. 5 5
A G _ ..

)2. 1}... - ._ a w._-..41m.-7 a... .. $1.;
_ _. 1. (A... .7 ... _ K. 1../\
Amigo/ASA“ w. A AA. _ .A g

A A4 A ,

A A 4 A _ 4 m _ . 4 , 4 A4
. .1 . .5 .. _ . _
\slj .\»l/..2 ml__\\ 1 L“ 1%- ,/ 1.520 VlA/w/
. x. 17.-..- 11.7 ------ \ 1 ..,_- A -r x. -7 .
7A3 A A X - \mazmmzj .

A A l L _

. A . -

A 3 . 3 3 3 3 3
. .. . . 1 77.1; T1- . 7.. .\ L/

--\1 /\¥1.. 1 /l\. ..\ A) .\ flK.

. meOmo_A _ A

_ A

 

 

SQUARE LAKE

12

‘1

   

.7
1
1.
11:3. _
.1112 ~—
-A7 A-
TLA

 

 

List of intersection lD Numbers

Figure 3.8

40

CITY OF TROY

SOUTH BLVD.

1.1-Mmz1-1 NNmz -11 .11. 1.1 11 1..1_1.1_-1-

 

 

 

 

151 14mm:1

5191.75 -1'1A f1 awn

 

 

 

 

 

 

A 32.11831
m ”AA _ _
w. AW. _ _ W
E E
A... 111 A1 L 1- 1% 11 .1- 11- AA.
.17--. 81mz A 1Wm-mF1A 10821. 11513.} mom-121
A 1 . A A
AA a A...“ n A“. a .5. ._ w
1 1 e .
W WA W M m G W ._ AW
FA mmpmmzoomA . m AA .
6.0.2.1...» H.521 111?.sz Ewan-“118mm-
AA -A. MA A AA . A A
A4A 4A 4A 4 _M
1, 2 G 3 M 5
A... m m N. Am m _ m
. L
A 1mwzH1A-1-smzw-A A 5112.sz A 111....mzfl. 1%.... .
A A . 1 90235.. 1 1
m a. 1.3.. m... a m.

 

 

. A .
A. _mszAmMaz A. A mama-1.. A 11.8211 “82A A
A, MAAmeOmoAA 1
2_ E A . 2 A2 2 E A2.
A1 R _ .3 4, 5 P 1,6...
WA Wm _ ANA m m m AMA
A _ s A _ A A _ .A A
- 1 am:- A1.11. 1;. New 1A1 .1121” Meg-H. 18271..

AA woojoow

A

n W .1 Au m
m f 1. . Am m
11 11.511A1. :1amH1A111. MEAN

 

 

 

 

 

 

 

1 1

 

List of Midblock ID Numbers

Figure 3.9

4l

midblock location is shown in Figure 3.8 (for each intersection) and Figure 3.9 (for each
midblock location). The program automatically calculates the subtotal of each type of
accident, and displays the total number of traffic accidents for a specific location and

year.

3.2.2 TRAFFIC VOLUME DATABASE

The traffic turning movement database was constructed for each intersection in
the City of Troy. The data were provided by Oakland County. When “volume data” is
selected from the “Traffic Characteristics Database” item, and a specific intersection is
selected, the traffic volume output displays several items. They are the ID number of the
intersection, the name of the intersection, the year of the data, the time period of the data
(AM peak or PM peak), total volume for all direction at the specified intersection, and the
traffic volume and percentage of left-turn, through and right-turn movements in each

direction.

3.2.3 LEVEL OF SERVICE DATABASE

When the “level of service” option is selected from the menu, and a specific
intersection identified, the screen displays the ID number of the intersection, the name of
the intersection, the year of the data (1989-1993), the time period of the data (AM peak or
PM peak), overall level of service, overall delay (sec/veh), overall v/c, and level of
service and We in each direction and for each movement (left-tum, through, and right-

turn).

an
0"”. q
~l
A 5
4:9; .. ‘\
u...‘ u\

unq~ f"
P h . I'
w-...- \

5

Anna‘-

Ru”: 5“

. v u
“‘33."
durh§au

Warm“

gm .‘u ..‘.

314 A)

W ‘h ..
0 his .fl‘
, I5.

:71;
I 1,

v
f.)
V ‘45. 1"

M. V'-

42

The level of service data were derived from the Highway Capacity Software
(HCS). Some of the intersections are not included in this database because of missing
data. If one or more of the HCS input data items (geometry, volume, or timing plan) is
missing or incomplete, the level of service can not be calculated by the HCS. The
locations with missing data are listed in Table 3.3. Color codes are used to express

different levels of service (LOS) at each intersection on the map. Both approach and

intersection LOS can be displayed.

3.2.4 AVERAGE SPEED DATABASE

The average speed data were obtained from the average speed survey conducted
by the research team from Michigan State University. When the “average speed” option
is selected from the menu, the average speed database output shows several items. They
are the ID number of the corridor, the name of the corridor, the year of the data (1993
except for the Big Beaver and Rochester corridors which are 1992 data), and the average
speed and number of observations for each time period (AM peak, off peak, and PM
peak), Only selected corridors are shown because the survey was limited to these

locations.

31.5 THE TRAFFIC ACCIDENT RATE DATABASE

By combining the traffic accident database and the traffic volume database, the
intersection (or midblock) accident rate can be calculated and displayed on the map. The

definition of the intersection accident rate used in this study is:

Table 3.3 The List of Missing Data

43

 

 

 
  
  
   
 
 
   
   
 
   
  

l:ntersectlon Name Period of Time Volume Geometry Timing
N-SboundIE-Wbound) Data Data Data
damslBig Beaver 16:00—18:00 x
lidgel Long Lake 16:00—18:00 x
oolidgeIIBig Beaver 16:00—18:00 x
oolidgelSouth Blvd. x x
oolidgelWattles 07:00—09:00 x
lidgeIWattles 16:00—18:00 x
Crooks/Big Beaver 16:00—18200 x
rooksISquare Lake 16:00—18:00 x
CrooksISquare Lake 07:00—09:00 x
Dequindrel14 Mile x x
DequindreIBlg Beaver x x
DequlndreIMaple x x
DequlndreIMaple 07:00—09:00 x x x
DequindrelSouth Blvd. 07:00-09:00 x
DequindreIWattles 16:00-18:00 x
hn RI14 Mile x x
hn RlBig Beaver 16300—18200 x
hn RISquare Lake x
Livemolleong Lake x
Rochester/Big Beaver 16:00—18:00 x x
Rochester/Long Lake 07:00-09:00 x
Rochester/Long Lake 16:00—18:00 x
Rochester/South Blvd. 07:00—09:00 x
RochesterISquare Lake 07:00-09:00 x
tephensonl“ Mlle x x
tephensonIRochester 16200—18200 x
ray Center/Big Beaver 16:00—18:00 x
07:00—-09:00 x

 

 

 

 

 

 

 

44

(annual number of accidents x 1000000)
(annual number of entering vehicles)

ACC/MEV=

 

Where ACC/MEV: accidents per million vehicles entering the intersection
The definition of the midblock accident rate used in this study is:

(annual number of accidents x 1000000)
(annual vehicle - miles of travel)

ACCIMVM=

 

Where ACC/MV M: accidents per million vehicle-miles of travel

Note that the mileage in each road segment has been calculated and depicted on

the map (see Figure 3.10 for details).

3.3 GEOGRAPHIC INFORMATION SYSTEM

The design process of the geographic information system involved developing a
geographic information system which allows the user to select data from any specified
location(s) from a map. There were three activities used to develop the geographic

information system as shown in Figure 3.11.

The first was programming the computer to select any specified points(s) or
location(s) on the map. Second, the PDC PROLOG was programmed to display these
locations(s) on a map. Third, a method of summing the intersection(s) or midblock(s)

traffic accidents for multiple locations was developed.

3.3.1 SINGLE INTERSECTION SELECTION

In this section, the program (Map7.exe) for single intersection selection will be

45

 

   

 

 

 

 

 

 

 

 

 

TEES 03L: 1. 0.354-. -- M

 

1193

0:99 11.94
: 1.12: f. ; -
1138

m

I!

0

m

E

Q

 

 

N
CITY OF TROY A
SOUTH BLVD. ‘1 m}
"“ 1.31403; ”t: WT. +1135 f 43:92: i
I75 B T g a T mg
"’ SQUARELAKE‘ i p. 3 HI
" \ jm : ;fifi 113:: 1019514 095 13+
‘0 \“3‘ 32“. 8 e g
5 a g 0 B ‘: § 5 fig 2 §
3 '- 8 2 M “w §a I 8 .
< 01 0“ EiLONGLAKE 94L all
it 0'94 59—5 TEETJ‘EB l 095 W‘;
a ‘ s {"5 i la a
L______ c I WATTLES '. . ,
_0.94_ :: "“ ‘ gm

 

 

 

 

0.7a ““095; _...__.b.as:T

 

 

Unit: MILEP MAPLE
; 11.31315; 1

 

.
0.90
11311.

,m' ;
1:0,.
LIES

 

 

 

Figure 3.10 The Midblock Mileage List

46

629$ co:mE.2c. 023380 9: 9.30650 5 32m 9:. 3d 239”.

 

2305 on: con: «5
:2: 35208 2:2. .8032... 3 3280825 2: =o Eam

 

 

 

 

b.

 

>236 can .535 «a: .3: e5 .2: 28:30.
2: 398062 2 0:35.02 a 3.26.. 8950.6
00.5mm 00a 05 S 3323 2:556 2: as»:

 

 

 

 

b.

 

 

e265 .8353 a as»: 5
3E 2: :0 26:80. 823 2 89393 a 3.260

 

 

 

 

’

m amuw

N nouw

F 35

 

£21» 3.3.525 023503 05 3.26“. oh

 

x530

 

 

47

introduced and explained. When the “Single Intersection” item is selected from the
menu of the ITAGI system, there are five major selections available in this program.
They are “Bar Chart-Intersection”, “Pie Chart-Intersection”, “Top 5—Intersection”,
“High Accident Location-Intersection”, and “Exit-Intersection”.

This first item in the menu is “Bar Chart-Intersection”. When the user selects
this item, a map of the City of Troy will appear and wait for the user to select an
intersection. After the user selects an intersection, a bar chart showing the number of
traffic accidents for the years 1989 to 1994 at the selected intersection will be displayed.

The second item in the menu is “Pie Chart-Intersection”. When the user selects
this item, a menu selection showing “1989”, “1990”, “1991”, “1992”, “1993”, and
“1994” will appear. After the user selects a year, a map of the City of Troy will be
displayed. When the user selects an intersection, a pie chart will appear showing the

percentage of accident types at the intersection.

The third item in the menu is “Top 5-Intersection”. When the user selects this
item, a menu selection showing “1989”, “1990”, “1991”, “1992”, “1993”, and “1994”
will appear. After the user selects a year, another menu selection showing “Angle”,
“Turn”, “Rearend”, and “Others” will appear. When the user selects the type of traffic
accident, the five intersections with the highest frequency of the selected type of accident

will appear on the map.

The fourth item in the menu is “High Accident Locations-Intersection”. When
the user selects this item, a menu selection showing “1989”, “1990”, “1991”, “1992”,

“1993”, and “1994” will appear. After the user selects a year, another menu selection

...

3m.

Ll.

48

showing “Total Accidents”, “Angle Accidents”, “Turn Accidents”, “Rearend
Accidents”, and “Other Accidents” will appear.

If the user chooses “Total Accidents”, the screen will display the total traffic
accidents at each intersection, stratified by three selected frequency levels. If the user
chooses “Angle Accidents”, “Turn Accidents”, “Rearend Accidents”, or “Other

Accidents”, the screen will display the data for the selected accident type.
The fifth item in the menu is “Exit-Intersection”. The user can exit the program

by selecting this item.

3.3.2 MULTIPLE INTERSECTION SELECTION

In the above section, a program called Map7.exe used to analyze a single
intersection was introduced and explained. If the user wants to analyze a set of several
intersections, a program called Map77.exe was designed for this purpose.

This type of analysis can be performed using the PDC PROLOG dynamic
database function. With PDC PROLOG, dynamic databases are stored entirely in the
computer’s memory. Special standard predicates are used to add new facts to the
database or remove facts from the database: asserta(fact), assertz(fact), and retract(fact).
The asserta(fact) predicate inserts facts at the beginning of a database, while the
assertz(fact) inserts facts into the end of the database. The entire dynamic database in the
PDC PROLOG can be viewed as a push-down stack and the new fact can be placed either
on the top of the database or at the end of the database depending on what the user needs.

This fimction is an important feature for developing the expert system. Since the rules

I
.q4 ’6
.3! "‘l

.',’,l
"1

liter!

9

177‘»
10;;

Grip

a“...
V“.

ran

“‘5“;

‘f-ar
“A,

‘-

.'__
'7‘

49

and facts stored in the computer memory can be retracted or added as the rules and facts
are obsolete or new rules are learned from the traffic experts. In the Map77.exe program,
the dynamic database function in the PDC PROLOG program was used to memorize the
locations that the user has chosen, and to make the corresponding calculation.

When the “Multiple Intersections” item is selected from the menu of the ITAGI
system, a menu will show three items to be selected: “Bar Chart-Multiple
Intersection”, “Pie Chart-Multiple Intersection”, and “Exit-Multiple Intersection”.

If the user chooses the “Bar Chart-Multiple Intersection”, a map of the City of
Troy will appear and wait for the user to select an intersection. After the user selects a
location from the map, the next menu will display three items: “Add One More”, “Show
Graphic”, and “Main Menu-Exit”. If the user wants the computer to display the bar
chart of the current selected intersection, the “Show Graphic” item is selected from the
menu. If the user wants to add one more intersection and analyze the sum of the
accidents at the first two selected intersections, the user selects the “Add One More”
menu item. Note that this program has a special feature to prevent the user from selecting
the same intersection twice. For example, if the user initially chooses the Adams/Long
Lake intersection, this intersection will be displayed in red to indicate that this
intersection is already in the selected data set. The user can continually select

intersections until the computer memory is exhausted.

The same operating procedure applies for the “Pie Chart-Multiple Intersection”.
The “Pie Chart-Multiple Intersection” item can display six years of pie charts of the

types and percentage of accidents. The user can choose as many locations as desired until

50

the computer memory is exhausted.
The third item in the menu is “Exit-Multiple Intersection”. The user can exit the

program by selecting this item.

3.3.3 SINGLE MIDBLOCK SELECTION

In the above section, two programs called Map7.exe and Map77.exe used to
analyze a single intersection or multiple intersections were introduced and explained. If
the user wants to analyze a singe midblock location, the program called Map7l.exe was
designed for this purpose. When the “Single Midblock” item is selected from the menu,
there are six major selections available. They are “Geometry-Midblock”, “Bar
Chart-Midblock”, “Pie Chart-Midblock”, “Top 5 - Midblock”, “High Accident
Locations-Midblock”, and “Exit-Midblock”.

The first item in the menu is “Geometry-Midblock”. When the user selects this
item, a map of the City of Troy showing the number of traffic lanes on each road segment
will appear. On the map, color codes are used to express different numbers of traffic
lanes. This can be used to study the relationship between the number of traffic accidents

and the number of traffic lanes.

The second item in the menu is “Bar Chart-Midblock”. Again, like the “Bar
Chart-Intersection” selection described in the preceding section, when the user selects
this item, a map of the City of Troy will appear and wait for the user to specify a
midblock location on the map. After the user chooses a midblock location, a bar chart

showing the number of traffic accidents over the past 6 years (1989-1994) will be

51

displayed. Note that the midblock number is the ID number which was shown in Figure
3.9.

The third item in the menu is “Pie Chart-Midblock”. When the user selects this
item, a menu selection showing “1989”, “1990”, “1991”, “1992”, “1993”, and “1994”
will appear. After the user selects a year, a map of the City of Troy will appear and wait
for the user to specify a location. The pie chart will then display the types of accidents
and the percentage of accidents (by type) at the specific midblock location.

The fourth item in the menu is “Top S-Midblock”. When the user selects this
item, a menu selection will display “1989”, “1990”, “1991”, “1992”, “1993”, and “1994”.
After the user selects a year, another menu selection showing “Angle”, “Turn”,
“Rearend”, and “Others” will appear. Again, after the user selects the type of traffic
accident, a map showing the top five midblock accident locations will appear.

The fifth item in the menu is “High Accident Locations-Midblock”. When the
user selects this item, a menu selection showing “1989”, “1990”, “1991”, “1992”,
“1993”, and “1994” will appear. After the user selects a year, another menu selection
showing “Total Accidents”, “Angle Accidents”, “Turn Accidents”, “Rearend
Accidents”, and “Other Accidents” will appear.

If the user selects “Total Accidents”, the screen will show the specific year and
total midblock accidents broken down into three pre-selected categories. If the user
selects “Angle Accidents”, “Turn Accidents”, “Rearend Accidents”, or “Other
Accidents”, it will show the selected accident type in the specific year.

The sixth (last) item in the menu is “Exit-Midblock”. The user can exit the

52

program by selecting this item.

3.3.4 MULTIPLE MIDBLOCK SELECTION

As described under the section on intersections, it is often desirable for the user to
analyze multiple midblock locations. The ITAGI system provides a mechanism for
conducting this type of analysis.

When the “Multiple Midblock” item is selected from the menu of the ITAGI
system, there are three major selections available. They are “Bar Chart-Multiple
Midblock”, “Pie Chart-Multiple Midblock”, “Exit-Multiple Midblock”.

If the user chooses the “Bar Chart-Multiple Midblock” item, a map of the City
of Troy will appear and wait for the user to select a location. After the user selects a
midblock location from the map, the next menu will display three items: “Add One
More”, “Show Graphic”, and “Main Menu-Exit”. If the user wants the computer to
display an accident bar chart of the current selected midblock, they can select the “Show
Graphic” item from the menu. If the user wants to add one more location and look at the
sum of the traffic accidents of the first two selected midblock locations, the user selects
“Add One More” item. The user can continually select midblock locations until running
out of computer memory. The midblock locations that have been chosen will be

displayed on the map.

The same operating procedure is used for the “Pie Chart-Multiple Midblock”
selection. The “Pie Chart-Multiple Midblock” can display six years of pie charts of the

type of accident and accident percentage.

53

The user can click the “Exit-Multiple Midblock” item from the menu to exit the

Map777.exe program.

3.4 ROUTE GUIDANCE SYSTEM

As an example of an Advanced Traveler Information System (ATIS), the route
guidance system in the ITAGI system was programmed to select the safest path. In the
federal demonstration project in Troy, the Ali-Scout system has been selected to provide
the road user with the shortest travel time route. The ITAGI system could be designed as
a dynamic route guidance system like Ali-Scout if the travel time on all arterials were
available in a real time. The ITAGI system offers the ability to search for the safest
(shortest) path through the network. It provides a convenient device for the user to input
two desired points (origin and destination), by simply selecting two desired points on the
map, and the user can immediately obtain the safest (shortest) path.

In this section, the design process, the safest path probability theory and the two
programs (Shortew2.exe and Shortwe2.exe) are explained and demonstrated. With these

two programs, the safest (shortest) path between any two specified points can be found.

The route guidance system was designed so the user can input the origin and
destination and obtain the safest (shortest) path displayed on a map. There were three

activities required to develop the route guidance system as shown in Figure 3.12.

The first task was collecting the intersection (accident rate) data from the database
for each intersection. Second, a distance (traffic accident rate) database file which

contained the traffic accident rate between nodes was constructed. Third, a PDC

54

623w 35230 230m 9: 9.30350 E 3on of.

 

a2: .5 c0 52.. 28.62. .5 >233
new 5!. =33:- 2: ace 2 6202.. 00.55. 00.... 3:3

 

 

 

 

neuoc

c0223 .2: 32.23 3:32.. 2: Baa-coo

 

 

 

 

2.9 2:22,

a... 3:023. gaze-:35 .o 32.5: 2: 80:00

 

 

 

.9

m 85

N aoam

F 85

N; 9:9”.

 

 

8a.. .528- 0503 .5 co ten-n $3.». 3.3an .52 no.0»... o»

6030

 

 

55

PROLOG program was written to find the safest (shortest) path and display the path on

the map.

3.4.1 SAFEST PATH PROBABILITY THEORY

If PiJ denotes the probability of an accident happening when traveling from node i
to node j, then l-PiJ represents the probability of not having an accident which traveling
from node i to node j. Assume at each link, the probability of an accident happening is
independent. Then the probability of an accident happening along the path from node i to

node 2 is:

1-[ (1- PiJ)(1- Pj’k) (1- PM) ...... (l-Px‘y) (1- Py,z)]

= l- [1- (PiJ+Pj,k+Pk,.+ ........ P”) + (Phi Pj‘k +Pj’k PM + Phi +PM + ........ + P,“y P”)
- (PU PjJ: PM ............. PM Py,z)]

(PM PM, +Pj,k PM + PiJ +PM + ........ + PM, P”) s 0 and (PiJ PM PM ............. P,"y P”) a 0
5 1- [1- (Phi + Pik + PM + ......... + Py,z)]
= PiJ + Ptk + PM + ................ + P”
Therefore, the cost from node i to node 2 can be defined as ‘
PiJ + P“, + PM + ................ + P”
The cost is enlarged 1,000,000 times for analysis convenience
as : 1,000,000 (PU + PM + PM + ................ + P”)
Therefore, the final cost from node i to node 2 can be defined as:

1,000,000 (PU + pj,k + PM + ................ + P”)

which is equal to (Rii + R“, + .............. + R”)

56

where RM, is the accident rate per million vehicles from node or to node [3

Therefore, from the probability standpoint, the cost should be defined as an
accident rate between two nodes. To demonstrate this program, the route that minimizes
the probability of vehicle collisions will be found and displayed using the accident rate
between two nodes (intersections) as a measure of distance or cost from node

(intersection) A to node (intersection) B.

3.4.2 DEMONSTRATION OF THE USE OF THE SAFEST PATH PROGRAM
There are two programs for the safest path: one is for the safest path from east to
west (from north to south or from south to north included) called Shortew2.exe, while
the other is for the safest path from west to east (from north to south or from south to
north included) called Shortwe2.exe. When the “Eastbound” or “Westbound” item is
selected from the menu of the ITAGI system, a map of the City of Troy will appear.
The user then specifies any two points on the map. Note that the input direction must
follow from east(origin) to west (destination) in the Shortew2.exe program, and from
west (origin) to east (destination) in the Shortwe2.exe program. A mistake in specifying

an origin and a destination will cause the program to fail.

3.5 QUALITY CONTROL CHARTS FOR TRAFFIC ACCIDENTS
The purpose of using the quality control chart in this study is to determine the
high accident locations and road segments (outliers) necessary for a Highway Safety

Improvement Program (HSIP) based on accident rates and changes in accident rates.

57

Safety improvement programs require an effective means for identifying hazardous or
potentially hazardous intersection or midblock locations. Hazardous locations are those
at which accidents happened at an abnormal rate when compared with similar locations
elsewhere. The accident rate is a good index to use in quality control analyses because
the traffic volumes are already taken into account in this measure. In this study, a
statistical method known as the Monte Carlo method was used to decide the critical high
accident locations.

The four activities used to develop the quality control chart of traffic accidents is
shown in Figure 3.13.

First, the traffic accident rate data was obtained from the traffic accident database
and geographic information system for intersections and midblock locations. Second, the
MATHEMATICA program was used to perform the Monte Carlo simulation for 1000
iterations to get the 1000th sample. Third, the top five high accident locations were
embedded in the PDC PROLOG program. The final step was developing a program to

display the intersection and midblock quality control charts.

3.5.1 DEFINITION OF QUALITY CONTROL

According to the Branverrnan [1981] definition, quality control is that set of
function and activities whose purpose is to assure the achievement of established
standards of quality. These functions and activities can involve both qualitative and
quantitative factors. In the ITAGI system, the traffic accident rate is a quantitative factor.
Therefore, the scope of this study is quality control using a quantitative measurement.

The primary objective of process quality control is to achieve and maintain an

58

E3»? 2023.4. 052... ea :20 .9200 >530 9: 9.32900 5 33m 9:. mfn e59...

 

2:20 .923 >523 5032:. new 20.80825 05 :25“.

 

 

 

I?

 

ESaoa 0040mm
one as 25 2.2.32 .8208 .3... 2: no. as 325

 

 

 

9

 

5:22....» 0.50
2:22 83:3 2 883.3 <o_h<_2w:h<s. a 2:3

 

 

 

a

 

c.3328 25
5:03.25 some .8 5226 c2353.... 023303
2: Ea: Sun 28 2.3an 05a: 05 .00

 

 

 

9

v .35

m 85

N new

a gem...

 

E223 Eoufioa use: .0 tone .223 2:26 82.28 8.

 

“$.30

 

 

i0.

Fl

59

acceptable and stable level of some process quality parameter. The principal statistical
tool used for this purpose is the control chart. A control chart is simply a graph designed
to present information. This information is obtained from a series of samples, called
subgroups, drawn periodically from the process output.

A control chart is simply a device for testing a hypothesis about the state of a
process on a continuing basis. There are two hypotheses displayed on a quality control
chart, the null hypothesis is that the process is in control. The alternative hypothesis is
that the process is out of control. The control limit can not be calculated directly by the
traditional quality control method because the intersections selected for accident analysis
will not be randomly selected (do not meet the requirement of the Central Limit Theory).
In addition, each intersection has its own distribution over the six year time period.
Therefore, it is not reasonable to treat the 45 intersections as if they were random samples
drawn from the same population. Instead, the Monte Carlo simulation algorithm was

used to construct the top 5 accident locations.

3.5.2 MONTE CARLO ALGORITHM

In statistical data analysis, the classical approach requires a strict distributional
assumption. If this distributional assumptions can not be met, the classical analysis can
not be performed. Modern electronic computation has enabled the analyst to use other
statistical methods, such as the Monte Carlo algorithm, that require fewer assumptions
than their predecessors (classical statistics), and can be applied to more complicated

statistical estimators. Monte Carlo sampling is very straightforward, yet it provides

y ,
‘ n
“no-

.'I_'

.0!

(A...
ll

b.r

6O

sampling distributions for a given parameter. Monte Carlo sampling does require a
population distribution. One knows the population distribution so it is easy to simulate
sampling from it.

In building the a? quality control chart, the six year average accident rates at each

intersection (£1,362 ,5c'3 , .......... E45) were not treated as random samples (variables) drawn

from the same population. Each individual intersection was assumed to have its own
distribution(Poisson distribution). Therefore, the critical accident locations were formed
by using the Monte Carlo algorithm rather than the traditional statistics method. The
critical accident locations were determined by the following steps:
$129.11
Combining accident counts and traffic volumes over six years (1989-«1994) for
each intersection produced observed accident counts n(i), traffic volumes
v(i), and accident rates r(i) for intersections, i =1 ,2,...45.
Slim—2:
Ranking intersections according to accident rates. The accident rate r(i) obtained
from the traffic volumes v(i) and observed accident counts n(i) for intersections
i=1,2,, ..... 45 were ranked from 1(highest) to 45(lowest).
51ml:
For each intersection i independently, generate a random sample n*(i) from the
Poisson distribution having mean n(i). Using these n(i) and v(i) re-rank
intersections according to accident rates, obtaining ranks*(i) for intersections,

i=1,2,....45.

61

SRM:

Repeat step 1 to step 3 1000 times.

$1341.51

Find the top 5 intersections from 1000 rank*(i) observations.

The major purpose of performing the above steps is to see if the intersections
ranked in the top five are due to “chance”. That is to see if rank*(i) varies with
“replication”. After performing the Monte Carlo simulation through 1000 iterations,
those intersections ranking in the top five are assumed to be high accident locations. To
perform the above procedures, the MATHEMATICA program was employed. The result
is shown in Table 3.4. The location of the top five high accident intersections is shown in
Figure 3.14. These results were then built into the PDC PROLOG program as the top 5
critical accident locations for intersection traffic accidents. Steps 1 to 5 were repeated to
get the top 5 critical midblock traffic accident locations. The result is shown in Table 3.5.

The top five midblock locations are shown in Figure 3.15.

3.6 THE ITAGI SYSTEM USED IN THE CITY OF TROY

The ITAGI system was designed as a tool to graphically display the current traffic
characteristics data and to be used in the evaluation of new traffic control systems. In this
study, two such systems changes was evaluated. The first was the implementation of
SCATS, and the second was the change in the left-turn control strategy from permissive
to protected, or from protected to prohibited, implemented at specific locations in the City

of Troy. Several examples are provided to compare before-and-after data where these

62

Table 3.4 A Comparison between Original Rank(i) and Monte Carlo Rank*(i)
(Intersection Locations)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

lnteraec. Rank(i) 4 Rank*(i) 95% c1

ID No. 6-year average "TTTM T *Hfor Rank*(i)
55 ' 1 1 [1,21
36 g 2 _ 22.2.1151
66 3 3 [2,6] “
45 4 4 [2,6]
35 5 5 [2.6]
34 6 6 [3,7]
53 7 9 [6,11]
46 8 10 [7,15]
63 9 7 [7,13]
24 1o 12 [7,16]
56 11 16 [8,16]
44 12 11 [9,21]
65 ___1_9 , 22 [10,21]
37 14 T5 [10,231 A
43 15 17 [10.23]
67 16 13 [10,22]
15 17 14 [10,24]
17 1e 8 [9.26]
25 19 20 [12.24]
62 20 19 [12,25]
64 21 18 [14,24]
51 22 27 [14.27]
26 23 23 [14.25]
77 “my 25 [17,25] -
42 ' 25 ' “21‘ T1851?
54 26 26 [22,29]
21 27 24 [21,351
76 25 28 [24.33]
52 29 31 [27,35]
47 30 33 [26.36]
32 31 30 [26.371
23 32 35 [27,38]
12 33 29 [24,39]
33 34 37 [28.38]
16 35 32 M[26.3_91fi
22 ” 36 _.___ 3f E27391
57 37 36 [31,38]
27 35 38 [31.40]
31 39 41 [36,41]
14 4o 39 [36.41]
13 41 40 [40,42]
11 42 42 [41,43]
41 43 43 [42,44]
74 44 44 [43.44]
75 45 45 [45,45]

 

 

 

Note: Monte Carlo Rank“(i) was derived from the Mathematica simulation results (1000 iterations)
l=highest accident rate; 45=lowest accident rate

63

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

666.2 26”. EoEoo< cozooemzc. :91 62“. no... 9:. 36 659".
4
was. 3 _a
,7..- .111-) was; 5252:
F _
mw><mm 05 ‘.
_
I 1 a 1
\ .2
8:23 W a m
N _I
m m
5 m
m m 93 029 , w
m M / O
m H / WW I
a 11 II I II \ I I I
Lfl W I I
% mxfi mm<acm 1,
w
a...
Lu
2. .o>._m150w
>0m... “_O >._._0

SWVCIV

/.
m:

64

Table 3.5 A Comparison between Original Rank(i) and Monte Carlo Rank*(i)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(Midblock Locations)

'ia'tdblock iankm 13561611) 95% c1
ID No. 6-year average for Rank‘fl)
N855 1 1 [1,1]
EW53 2 3 [3,3]
N854 3 4 [4,4]
EW75 4 2 [2,2]
EW52 5 5 [5,8]
N835 6 7 [6,10]
N866 7 9 [7,10]
N832 8 8 [5,10]
EW65 9 6 [5,8]
N845 10 12 [10,14]
EW51 11 14 [12,22]
EW54 12 15 [13,22]
EW33 13 10 [10,16]
N863 14 16 [11,21]
EW62 15 18 [13,24]
EW64 16 44 [33,44]
EW34 17 19 [14,31]
EW36 18 24 [16,32]
N853 19 21 [19,31]
EW63 20 23 [19,32]
EW32 21 22 [17,33]
EW35 22 34 [19,35]
EW55 23 30 [20,32]
N864 24 28 [20,36]
N876 25 13 [13,26]
N862 26 26 [21 .37]
N865 27 25 [18,33]
N844 28 27 [14,30]
N841 29 29 [20,38]
N843 30 31 [30,44]
N842 31 39 [25,41]
EW23 32 33 [25,44]
N856 33 20 [12,26]
N834 34 35 [32,45]
EW41 35 32 [21,49]
N822 36 11 [5,12]
EW76 37 50 [42,52]
N825 38 41 [33,46]

 

 

 

 

 

Note: Monte Carlo Rank“(i) was derived from the Mathematica simulation results (1000 iterations)

l=highest accident rate; 45=lowest accident rate

 

65

Table 3.5 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

WbIOCk Wank") Rank*(i) 95% c1
ID No. 6-year average for Rank*(i)
N861 39 17 [11,28]
EW56 40 49 [49,57]
EW24 41 40 [33,54]
N852 42 36 [35,46]
EW31 43 45 [37,55]
EW43 44 51 [36,55]
EW44 45 52 [34,52]
N851 46 38 [[35,47]
EW66 47 47 [41 ,53]
EW46 48 48 [42,57]
EW45 49 53 [39,55]
N831 50 37 [23,39]
N875 51 46 [46,56]
EW42 52 43 [36,56]
N811 53 56 [56,67]
N824 54 54 [43,56]
EW25 55 42 [38,57]
N846 56 58 [49,59]
N872 57 55 [50,60]
Ew26 58 57 [50,66]
EW16 59 65 [55,67]
EW21 60 67 [51,67]
W11 61 60 [52,68]
N874 62 61 [57,64]
ew13 63 59 [56,70]
N823 64 68 [[58,69]
N814 65 64 [60,70]
N821 66 69 [58,74]
N873 67 70 [64,72]
EW22 68 63 [61,74]
N833 69 73 [68,75]
N871 70 66 [60,70]
EW15 71 71 [66,75]
N813 72 62 [65,74]
EW14 73 74 [[67,75]
EW12 74 75 [69,75]
N812 75 72 [67,75]
EW74 76 76 [76,76]

 

 

 

 

 

 

Note: Monte Carlo Rank*(i) was derived from the Mathematica simulation results (1000 iterations)
l=highest accident rate; 45=lowest accident rate

66

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6662 26m $6284. 5.8.62.2 :9: 62". 66F 9: m; 659”.
m.__s_ E .3
.2
mm><mm 9m
wm4t<>> w m
N _l
m m
80, m
0 m
w W 92.. 023 , w W
n H O V
m N o w
H 9.5 $238 / /.
q. n:
a
a
3. dim :58
>0m... ”.0 >._._o

67

system changes were implemented. The data analyses include the aggregate decrease or
increase in the total number of accidents, and the percentage of type of accidents.
Statistical tests were conducted externally to determine if the expected number of
accidents were different than the actual number of accidents after implementing SCATS

or a new left-turn control strategy.

3.6.1 THE CURRENT AND HISTORICAL TRAFFIC CHARACTERISTICS

DATA DISPLAY

As described in a previous section, the current and historical traffic characteristics
data can be displayed to the decision maker (traffic engineer) in text or graphic format
(such as bar chart or pie chart). In addition, by identifying high, middle and low accident
locations, the ITAGI system can assist the analyst in identifying roadway deficiencies or
areas needing improvement. To demonstrate the use of the ITAGI system, selected

examples using each component were conducted.

3.6.2 TRAFFIC VOLUME DISPLAY

With the ITAGI system, the traffic volumes at intersections can be presented on a
map. Figure 3.16 and Figure 3.17 show the total traffic volumes at intersections for AM
peak and PM peak hours, respectively. From these two graphics, it is clear that during
peak hours, Big Beaver and Maple, and John R Road near I-75 have high traffic volumes,
while Adams Road has a lower traffic volume than the other roads in the city. This

information can be used to identify links with unused capacity.

68

361262 66526> 2th _93 5.666 52 6.: a; 65?.

 

 

m.___)_ 3

 

 

 

 

 

 

I mica—2

83.6 A‘
ognéoan . .
83 R Gem!

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SINVGV

 

 

 

 

 

2-.

 

 

 

 

 

i

 

mm><mm OE \
_
_
i . i .. 4
8:23 w 2 m
\ _I
8 m
w 9
a I >
a — (V
m m 9.5 623 , w
n H / O
m N x M
a llllll £ / I l 2‘
H m¥5m¢<30m .
3
S
a
¢ .o>._m 150m
>Om._. “_O >._._0

69

961263 8526> 65E. .28 5.666 .26 6.: ta 659“.

 

 

was. 3 2

 

 

 

 

 

 

5

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mm><wm 05
I . a. - t... 1
mmntss w m
a m
a r .6.
I w. 5 I
m 0 9.5 023 , w
n W / O
m M
G H S .11..I
u 1 1 1 1 x x
0 ¢‘4 51/
WW. mg mm<30w
3
_d
‘ .oZm :58
>Om._. “.0 >._._0

 

 

 

 

 

 

SWVGV

on.

70

3.6.3 LEVEL OF SERVICE DISPLAY

With the ITAGI system, the level of service can also be depicted on the map.
Figure 3.18 and Figure 3.19 show the level of service for AM and PM peak hours. As
shown in these two graphics, nearly all the intersections have LOS F or worse during the

peak hours. The intersections with no symbols represent missing data.

3.6.4 TRAFFIC ACCIDENT RATE DISPLAY

The 1994 intersection accident rates are shown in Figure 3.20. The intersection
accident rates for other years are contained in the ITAGI system. As with the accident
frequency data, the ITAGI system provides a method of displaying the intersection
accident rate for each intersection on a map. From Figure 3.20, it can be seen that the
high accident rate locations are concentrated on Big Beaver, Rochester, and John R Road.

The 1994 midblock accident rates are shown in Figure 3.21. The midblock
accident rates for other years are contained in the ITAGI system. Figure 3.21 shows that
the high midblock accident locations are also concentrated on Big Beaver and Rochester
Road. Big Beaver, Crooks and 14 Mile Road all have interchanges with I-75. Thus, the
midblock accident rate is high for these segments because all the ramp terminal accidents
are attributed to these links. This menu item in ITAGI can be used to determine the high
accident locations and road segments necessary for the Highway Safety Improvement

Program (HSIP) required of all cities.

71

26:86.9... 65 .6 won 3.666. s_< 9:

 

 

 

 

 

 

w; 556.”.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

>01... ”.0 >._._O

 

 

4.
was. 3 .. 5.9... 64 G
b 04 D
.3 a... [165: m 0
x1111111 1 36s). 65.56;.
D .9 D
6—41 41.. \
mm><mm 06
D. P . »
e e a. v a. 4 1 2—5 6) / fl
I AN
wmnt<>> w m
m n
I G
C .4. _ 5 -
G P. V
m O 9.5 928 . w G
n H O V
m N w. w
m a 1 11111 1111/7
0 . m 7i. fl
m 5.5 $.38 /.
8 m5-—
.. m. 1r
< a? «v r
.D>._m IFDOw

72

26.8629... 9.. .6. m9 5.666. .26 6.: a; 656.".

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

w 519:5
m.__s. 3 . 04 Q
. a. D
. .MW We mAAU
»
x1151111 1 56$... «56.5“. +
.9 9
6?. v 4 6
mw><wm gm
4' P .
4+4 4*. 4 4 Jr ,1) .1 4
! ‘ 1 IV
$4622. w m
m n
I 0
e S .Er mu. 0
.5 . + . <1 ..
O 4 P . 1 1
m 0 5.5 626. . w
n H / O
m N x
O H -I3111.\17111.
a “[J I I a
o .1. x
mm gum—(Dow
g
a
HIH
/\ n a o
¢
¢ >05 “.0 E0 .96 :56...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SWVOV

3-.

73

32 c. 6.6m 26.6.8.4 8.68.9... 6.: own 6.56.“.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SWVCIV

o'

/4
ms;

4 3.05
m....2 3 ..
m
4 .3 ._1 - o .4 w .4
/ I I 1 I I I I I m._n_<_2 >m_2 ”ta:
/
D .—. m .N .o .. _. cd in.
mw><wmm=m
.o n. .4 be. emf m: .01. B.
aunts... w. m
N _I
m m
s 19..
F mv m ._. sh: QC m .o: N
m P 9.5 620.. . O
o o u
n H O
M N m
m a 1. /11191111.-1111
me w "m a .. .6 .5 1
m Am. 9.5 magnum
S
I.
a
.4 “31 .6 4 .o .36
A. dim 150m

>Om._. “_0 >150

74

vmmw 5 9mm EmEoo< x0032.)— 9:.

 

 

 

 

 

3m 656.“.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mm; 3m 6
1. 3:21 5 . z .n
H S _ F.
L 7 l_ I .
r04. flm 1. wk-.. MON _ ch
.1 c / 1 I far I I I 2 r 9.922 u .222 2:5
.1 .1 . a 1
89 Nwd Km and. #3. 5d
L I $555055. “1. . 7. L _W.
m m. m _ m n m
are 2: mmé and mm; mm c
.
W 4 mmntss 7. u . .n. 1. n
9 w _w. u . w m .m
vmd and ovd 2m; vwé EN
6 P 9.5 029 u 0 V
1. a Z W 7. 9 c m/ _L W Z w n. W
mm m N um m m / m o m n m w
mu. 8 w my / i 11 W 1 1| \ 11 m I / l. S
Mm.“ SW17 E15 E mWQ dud
.0 Z L. c 7. 9.5 magnum .u .n./.mn-_
m. w... m. w m m m.
Exp . - 1 o :5 New Ed . wwd
¢ 9.5 :58
2 >05 “.0 E0

75

3.6.5 SINGLE INTERSECTION SELECTION

The 1994 top five angle accident intersections are shown in Figure 3.22. In
addition, the 1994 total intersection accidents is shown in Figure 3.23. The other years
and accident type data are contained in the ITAGI system. The other years total

intersection accident data are contained in the ITAGI system.

3.6.6 SINGLE MIDBLOCK SELECTION

The top 5 locations based on angle midblock accidents in 1994 is shown in Figure
3.24, and the 1994 total midblock accidents is shown in Figure 3.25. Detailed
information about the top 5 locations for each year (1989--1994) and each type of traffic
accident is contained in the ITAGI system. The other years total midblock accidents are

contained in the ITAGI system.

3.6.7 MULTIPLE MIDBLOCK SELECTION

An example of the potential use of this program would be to determine the
relationship between the number of traffic lanes and midblock traffic accidents. For
example, if the traffic engineer wants to examine the number of accidents occurring on
2-lane roads, 5-lane roads, and boulevards, it is easy to obtain these results. Five road
segments on Wattles Road were selected as a 2-lane road sample, five road segments on
Maple Road as a 5-lane road sample, and five road segments on Big Beaver as a

boulevard sample. The results show that:

1. The boulevard segments had the highest number of traffic accidents, followed

76

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2286225 E02094. 292. 62.... no... 33 62h «Nd 659m
.
m....2 E . £4 4
__ 5 v4 4
ii 2 n4 I
2 5.. N4 4.
/|I.1III I w..n_<s_ .34
«95303
.1 m .1
wWFSE. w o
N .I
m m
s m
G P. g _ H
m. 0 92.. 023 w
n H O
m N o
a 8 x I 1
.5... 1 . 1s. 1 1 1 1 1
411 w! I I
H m¥59226
a
I.
3..
1“ .o>..m :Som
>Om._. ”.0 >120

SWVCIV

mm;

4’.

9:66.85. 5.8629... .28 33 6.:

 

1... ..
19...... 3 .

mm...” 656.“.

 

 

 

 

 

 

 

 

 

 

 

U

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

77

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 F
j fi+l4

SINVCIV

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9.2.9. 65 .
.
.
. i
$9.23 w m
N n
m a
#. N > {Id-d
a P 9.5 020.. . mm
m... o o
n H O
m N x
O a S \ I I
m IIIIII r .1 / /
111 l
W mxfi mm<30w /
S
l.
a i H

tr

 

>Om._. m0 >._._0

a:

78

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9.83.2.2 £528.... 292 oz”. gob .52 9: «mm 2:9...
~
5.... E . .3 A. ml»
m. _ f A. B
. En4_M
/ / I can A a
/ I.
/ l a ..... I / m 3.25. .3 4 [IL
. H
mw><mm 9m x.
mm.E<>> w m
N n
m G
S? B
a .
w W 9:35. . m
m w 0
m H / m I
m a / IS a .. x I / x r
m wxs mm<DCm
«a.
m
m

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

>Om._. m0 >._._0

.0>.._m 150w

SINVCIV

/.
mu...

79

3coEoo< x0032} _th mev 9:.

.

 

 

 

 

 

mud 9:9".

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

mj=21 _ or u 0 § _U
om .. :4 7A
x1: I\.. 1.. Has... RuAAI
M W mm><wm9m . /
/
W W m __ w
m mm??? W .. m
M m. m
\ a
//////////x ////////////fl///.////////j //V///////A/A//////////////////
/ 0 / P 923.67 0 I V
W m M m M ,, w M m
m m M N w / o M w
”M an M - w----fl--/- s
m i m M, m 5.38 x L, /.
M M a 2..
fl / m...
U
4. .03.. I58
>Om._. “_O >._._O

80

by 5-lane roads, and then 2-lane roads;

2. 5-lane roads have the highest percentage of angle accidents, then 2-lane roads,
and then boulevards;

3. After implementing SCATS in 1993, the 2-1ane roads showed a significant
increase in the percentage of rearend accidents. The change on the boulevard
and 5-lane road segments were not significant; and

4. After implementing SCATS, the percentage of angle accidents on 2-lane roads
decreased from 20% (1992) to 18% (1993), to 9% (1994), while the
percentage of angle accidents on 5-lane roads increased slightly from 21%
(1992) to 27% (1993), to 23% (1994). On boulevards, the angle accidents

increased from 7% (1992) to 8% (1993), to 9% (1994).

3.6.8 ROUTE GUIDANCE SYSTEM
The ITAGI system can be used to determine if the safest path changes from year
to year. This type of information would be useful in tracking trends in accident rates as a

city or region develops, or to evaluate the effect of spot improvements.

3.7 ACCIDENT REDUCTION BEFORE-AND-AFTER SCATS

Data necessary to develop the trend of all accidents occurring in the City of Troy
is contained in the ITAGI system. SCATS control was initiated in a part of the city in
June 1992 and in the remainder of the city in November 1993. The trend analysis was

constructed to differentiate these areas, and to consider total accidents and injury

81

accidents separately.

Figure 3.26, Figure 3.27 and Figure 3.28 display the results of this analysis.
Figure 3.26 displays all accidents in the City of Troy. This accident summary includes
intersection accidents and non-intersection accidents in the six mile by six mile area of
the city. Figure 3.26 shows that the total number of accidents in the city was higher in
1994 (after SCATS installation), than in the three prior years, including the year of
installation. The number of accidents in the last six months of 1994 was less than the
number in the first six months. These same observations are true of injury accidents as
well.

Figure 3.27 presents a similar analysis for the five intersections where SCATS
was installed in June 1992. At these intersections, total accidents were lower in the three
six-month intervals following installation than in the six month period which included the
installation. However, in the first six months of 1994 the accidents increased
significantly and then declined again in the last six months of the year. Injury accidents
were higher in each of the after period six-month intervals except the last six months of
1994.

Figure 3.28 depicts the accident history for the entire 49 major intersections in the
city. The same phenomenon is present as was observed in the first two figures. That is, a
large increase in both total and injury accidents in the first six months of 1994 followed
by a decrease in the last six months of the year.

Another example of how the ITAGI system can be used to assist the analyst is in
identifying locations where the change in accident fi'equency between the before period

(1991) and the after period (1994) at an intersection or midblock location was large. The

82

1,600

 

1,400 - ..............................
1,200 - .......................
1,000 ..................
800 i ..................
600 ..................
400 .............

200

 

 

 

Traffic Accident Injury Accident

 

 

1989(1-6) 1:11989(7-12) m? 1990(1-6)
I 1990(7-12) 1.1991(1-6) I.1991(7-12)
I1992(1-6) 131992042) 3? 1993(1-6)
I 1993(7-12) I 1994(1-6) I 1994(7-12)

 

 

 

 

Figure 3.26 The Number of Traffic Accidents & Injury

Accidents (1989--1994)
(The City of Troy -- 7x7 Major Arterials)

 

160

140

120

100

80

60

40

20

83

 

 

 

 

Traffic Accident Injury Accident

 

 

1989(1-6) 1:] 1989(7-12) % 1990(1-6)
I 1990(7-12) [1991(1-6) I 1991(7-12)
1992(1-6) 1:11992(7-12) a 1993(1-6)
I 1993(7-12) I 1994(1-6) I 1994(7-12)

 

 

Figure 3.27 The Number of Traffic Accidents & Injury

Accidents (1989-4994)
(5 intersections in the City of Troy)

500

400 -

300 -

200 -

100 -

 

 

 

 

 

Traffic Accident Injury Accident

 

 

 

I1989(1-6) D1989(7-12) 1990(1-6)
I1990(7-12) I1991(1-6) I1991(7-12)
I1992(1-6) 1:11992(7-12) 1993(1-6)
I 1993(7-12) I 1994(1-6) I 1994(7-12)

 

 

 

Figure 3.28 The Number of Traffic Accidents & Injury

Accidents (1989--1994)
(49 Intersections in the City of Troy)

85

ITAGI system has the ability to display intersection and midblock comparisons for
1991& 1993, 1992 & 1993, 1991 & 1994, and 1992 & 1994. For demonstration
purposes, the intersection and midblock comparisons for 1991 and 1994 are shown in
Figure 3.29 and Figure 3.30. From Figure 3.29, it is clear that the majority of the
intersections experienced an increase in traffic accidents in 1994 compared to 1991.
Almost all the intersections experienced more accidents in 1994 (after SCATS). This was
not the expected result when implementing SCATS.

Similar comparison were made for midblock locations for 1994 and 1991 (refer to
Figure 3.30). Unfortunately, in 1994 (after SCATS) almost all the midblock locations
experienced more accidents than in 1991. Again, this was not the expected result when
implementing SCATS.

It should be noted that there are frequency comparisons and do not necessarily
represent statistically significant differences. That is, there can be no before-and-after
statistical analysis because the data is not controlled for the effect of extraneous factors,
such as changes in geometry, changes in control strategy, changes in signal phasing,
changes in volume, changes in weather, changes in the enforcement level, changes in the
economy of the region, changes in driver performance and vehicle design, and changes in
reporting and coding of accident data. These are all known to affect accident data. It is
difficult to attribute changes in accidents over time to a single factor, and at this level of

analysis, it is not possible to determine which of the changes caused any recorded effect.

3.8 TRAVEL TIME SAVING BEFORE-AN D-AFT ER SCATS

The average speed data were obtained from the survey conducted by the research

86

39 can 5m? c0923 $52091 5:03.25 ho coatmano on...

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. J]? idw
.
m.=_2 3 . S v 3 4 AV
. .. m? mm: 5 u 3 4 D
A
/iiiii..i 11/ 5%.... Sides?
_
._. 9 ma; .. v
mm><mm 9m .
_
+ Li
v .1 - .v I
N ~ IN
mm.E<>> a . m
N .I
m . m
S. w
. \ ‘ > dd: '
a r _ r\
w O 9.5 020.. / w
H
n O
m a / S .l I
U 1 T: .I I .1 \ I / /
mm mV—S 9.2de
v...
I.
3 41? v
.. J& <. z
¢ >05 ”.0 Eu .9... as...

an...” 9:9“.

SINVGV

we.

87

32 2:... $2 8923 35.202 0.832.... s 8250.58 2:. and 2:9“.

 

 

 

 

 

-.////////////l

_I..I 
m I l 1 Eu 3 A %
1i- I ..

U”. mw><mm gm '— II

 

         
 

 

  

   

          

  

mm.t<>> -—.
I. m i I

m 5.5 023 W
D
m V
m m
a S I. .I ..I \
n. I

I////7// /él l2”7///////Ial

9.5 $228 /
mm;

 

 

2%..

2 >05 ".0 E0

Wr/////V%///V//////l
.0)..m IPDOw

 

88

team from Michigan State University. After data reduction, the average speeds are
summarized in Table 3.6 and Table 3.7. These two tables show the average speed on the
Rochester and John R corridors before and after SCATS. The before SCATS data was
collected in 1992 or 1993, and the after data was collected in 1994 or 1995. With the
ITAGI system, the effect of SCATS on travel time can be displayed on the map. As
shown in Figure 3.31, Figure 3.32 and Figure 3.33, during the AM peak, southbound
Rochester road and northbound John R road show an increase in average speed afier
SCATS, while northbound Rochester and southbound John R show a negative effect after
SCATS. During off-peak, southbound John R road shows a negative effect after SCATS.
During the PM peak, only northbound John R has a better average speed after SCATS.
Note that only selected corridors are shown because the survey was limited to these

locations.

3.9 ACCIDENT REDUCTION BEFORE-AND-AFTER PROHIBITING

LEFT-TURNS

This section compares the accident frequency before-and-after prohibiting

left-turns. In Troy, lefi-tums were prohibited at several intersections in 1992 (refer to
Table 3.8). With the multiple intersection selection function in the ITAGI system (as
described in section 3.3.2), the aggregate results were obtained to determine whether
there was an increase or decrease in the number or type of accidents. The results show
that the percentage of turn accidents at these locations increased from 4% in 1991 to 7%
in 1993, then decreased to 4% in 1994. However, the total number of accidents increased

from 110 in 1991, to 137 in 1993, and to 151 in 1994.

89

Table 3.6 The Average Speed on Rochester Corridor Before and After SCATS

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

IRochester Road 19 92 (before) 19 95 (after)

I Avg. Speed No. of Obs. Avg. Speed No. of Obs.

INB AM Peak 30.60 7 15.80 47
Off Peak 29.90 17 NIA N/A
PM Peak 20.20 12 15.90 51

SB AM Peak 30.80 5J| 33.36 67
Off Peak 35.30 14“ 14.84 51
PM Peak 32.30 9“ 23.42 44

Unit: mph

Table 3.7 The Average Speed on John R Corridor Before and After SCATS

 

 

 

 

 

 

 

 

 

 

IJohn R Road 19 93 (before) 19 94 (after)

I Avg. Speed No. of Obs. Avg. Speed No. of Obs.

INB AM Peak 31.90 1481' 32.56 204
Off Peak 42.20 117" N/A NIA
PM Peak 28.80 238 29.78 161

88 AM Peak 36.70 349 31.19 189'
Off Peak 35.60 131" NIA N/Al
PM Peak 31.40 49“ 22.49 27'

 

 

 

 

 

 

Unit: mph

90

mm? 2.0 Nam? c0053 umom m 5.2. a .2353. cc “.0on 3903. xmmn. .2< .6 52.3500 Fwd 0.59“.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

G. -
m...s. 3E .~ 3 v a. a
_ mm N Na B
L. 3 A 3 I
/
/ I i i i .. i I I 30$...
,
.
‘mmsfimoa \ .
.
¥ .
r k h .9 m
mm..t<.<. w . m
m. m
F s a
.
m I W 93626.. . O
o H / a
m N / mw
m a H / M \ I. I
m i T 91 i I -1
.. W 1..
.1... $59.58 4.
a
J U
4 .
.98 1.58
>Om._. “_O >._._0

 

 

MW”!

2335....

Q

 

0.53.2.8

MOO.“

 

SINVCIV

/

wk.-.

 

91

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. T 3 «a s I
m._=2v_.
_ 8 S _B
. 11 mm 3 I
/
xiiiiii...) 9.22.2
.
.
mm><mm 0.0 .
n .. 1m
mm..t<>> w . m
m. m
3 3
.
O «I
m 0 9.5 020. . m
n H / O
N , m
M a m / I IS, II \ ‘1 I I / I
J m 9.5 92:00 1.
a
U
‘ O 0 0301500

30.. 0:0 ~03 000500 000m .2353. :0 noeaw 000.0>< 0.000 to .0 coatmano wad 0.59”.

 

 

2325...:

c

poem-I3

11>

9.5035300

Whom-0M

 

SINVGV

I

mm.-.

 

92

003 0:0 «00.. c0050“. 000K x 5.0.. .0 0.00501 :0 0008 000.03.. 0.000 in. .0 coutmano mud 0.50.“.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

§ 0.535....
.. 3 v 3 A I o
w._=2 I. m. m
. 3 u S _ B a m
. m m
. mm A um I
>
I l l II Ii II .1 Ii 1) WA; 2535.50
.
.
$55.0 0.0 .
.
_
n . 4w.
00.522, W . m
m. m
S. 3
.
m m 9.5 020. . w W
m H > o W
M N / O
a a a / x \\ ‘1 5| I I S
N w I 1 .81 I / -
E 9.5 05300 I. x
I.
E whi—

‘. .0>..m 5:00

2 >05 “.0 5.0

93

Table 3.8 Intersections Implementing Prohibited Left-Tum Control in 1992

 

 

 

 

 

 

 

 

 

 

Ilnteraectlon Name 19 91 19 93

I Total Acc. Turn Acc. Total Acc. Turn Ace.
[319 Beaver 8. Crooks W} 37 (WW4 ..-- 41 3
[Big Beaver 8. leemols 23 2 23 2
IBlg Beaver 8. Rochester 50 4 73 4
Iroul Accidents 110 10' 137 9|

 

 

 

 

A statistical test was conducted to determine if the prohibition of left turns
resulted in more accidents in 1993 than in 1991.

Suppose AiJ denotes the number of accidents for all independent intersections i
and independent years j, NiJ denotes the yearly traffic volumes for these same
intersections and years and P0 denotes the probability of an accident happening for the

same i,j. Assume there are NiJ Bernoulli trials with accident probability P, and the

J 9
number of accidents AiJ is approximately binomially distributed. If NU is large and PiJ
is small, the number of accidents Au is approximately Poisson distributed with mean NiJ
Pi.) and variance NiJ PiJ . If NU PM 2 5, the number of accidents Aid is approximately

normally distributed with mean NiJ PiJ and variance NiJ PiJ- The formula can be written

as described above in the following:

AU 2 Binomial (N13, P1,] )

Ill

POISSOHWiJ,PiJ ), ifNiJ zoo andPiJ z 0

III

Nomal (NiJPiJ’NiJPiJ)9 if NiJPiJ Z 5

V independent intersection i and independent year j

94

In 1991, the distribution of the total number of accidents is :

3

zNormal(Ni,P9r 1'91, N191 191) ~ Normal(ZN,9, i9l’ZNi,P9l 1,)91

i=1 i=1 i=l

In 1993, the distribution of the total number of accidents is :

3 3 3
Z Nor ma] (N193 13.93 1 N1,93P1,93 ) z Nor ma] (2 N1,93P1',9sa Z N1‘,93P1‘,93 )

i=1 i=1 i=1
Therefore, the distribution of the difference in the total number of accidents between

1991 and 1993 is:

3 3 3
N0rmal(ZN,9_,P, 93 ‘2 N1 9IP1' 91’ N1 9313.93 + Z Ni,91P1',91 )
i=1 '

3 3

Wherethe ENLPM 1,93 ENLPM 191 and ZNI93 1,93+ZN1',P9I 1,91 areunknown

i=1 i=1 i=1 is]
For the purpose of setting confidence limits on the difference in the expected total

number of accidents in 1993 and the expected total number of accidents in 1991,
p.193 — 119, can be estimated by observing the total number of accidents (2X,93 — 2X,91 )

for all intersections for both years combined where:
XiJ denotes the number of accidents at intersection i , in year j , \7’ i=1,2,3

u 131' denotes the expected total number of accidents (2 X” ) at intersection i , in year j ,
V i=1,2,3

The value of (XXL93 — 2X19: ) may or may not be a good estimator for the
1,193 —p.9,. If x1= 2X,” , x2 = 2X,“ , u, = 1193 , and u, = pg], atest of whether

x1 and x2 are good estimators for u, and p, is to determine whether R is a standard

95

normal distribution of the form n, — u, , where

 

xl— - x2—
R=( l1(113145112) p2)1x1~N(u11u1)1x2 ~N(Hzau1)

Note that with very small probability x1+x2 is not positive and so ,/(x1 + x2) will be

imaginary or zero. This simply means there is a very small probability that the ratio R
upon which the confidence interval is constructed is not well defined, but the probability
is very small and causes no problem. Based upon independent N(0,1) random variables

Z and Z’,

wetakea =‘/E:,A= u,,b= Ju—z,andB= H2. Hence,Rcanbeexpressedas:

_(A+aZ—A—B-bZ'+B)

R
J(aZ+A+bZ‘+B)

 

 

(aZ—bZ)
J(aZ+A+bZ+B)

 

 

:>R=

Define the normal random variables:
U=aZ-bZ’
V=aZ+A+bZ’ +B
then
Var(U) = Var(a z - b Z’) = 112 Var(Z) + b2 Var(Z’) = a2 + b2 = A+B.
Var(V)=Var(aZ+A+bZ’ +B)=a2Var(Z)+b2Var(Z’)=a2+b2=A+B.
VarCU) = Var(V) = A+ B.
Cov(U,V) = E(UV) - E(U) E(V)
= 13([11222 + AaZ + abZZ’ + aBZ - abZZ’- AbZ’ - bZZZ’ - BbZ’]) - E(aZ-

96

bZ’)E(aZ + A + bZ’ + B)
Cov(U,V) = a2 E(Zz) - b2 E(ZZ’) = a12 - b2 = A - B.

Cov(U, V) ___ (A — B)
JVar(U) JVar(V) (A + B) '

 

Corr(U,V) =

In terms of a standard normal random variable W, U can be expressed as:

U= e W, where c=,/(A + B),

c W W

c \/1<aZ+ b!) \/H<a_z__+ L?)
C C

 

 

So, R:

 

 

VIC

where C= A+B.

The normal random variable W and (a Z + b Z’) have correlation equal to the correlation

between (a Z - b 2’) and (a Z + b 2’). This correlation is equal to gig). The random

A+B)

variable W’ is standard normal too, where

W,=(a Z + b 2')
C

 

Therefore, R= —W— , where W, W’ are each standard normal with
(1 + m)

C

(A-B)

correlation (W, W’)= (A B) . For large c, R is approximately standard normal. The
+

correlatron really does not matter, srnce the Important tlnng rs that —— rs small.
c

Therefore, the 95% confidence interval for the expected total number of

accidents for year 1993 minus the year 1991 ( 11,33 — “1.91 ) is:

97

 

(ZX,,,-ZX,,, )i 1.96 x (#22993 +2)(,,,) ,v i=1,2,3

= (137-110) i (1.96 x ./(110 + 137))
= 27 :t (1.96 x 15.72)

= {-3.81, 57.81]

The 95% confidence interval for the expected total turn accidents for year 1993

minus the expected total turn accidents for year 1991 ( 11,93 - pkg, ) is:

 

(Exam-2X,” )i 1.96x $296,, +2112...) ,Vi=1,2,3

= (9-10) i (1.96 x W)
= -1 :t (1.96 x 4.36)
= {-9.54, 7.54]
where the total number of accidents in 1993 2X1.” is approximately normal with
unknown mean and unknown variance, and
the total number of accidents in 1991 ZXW is approximately normal with
unknown mean and unknown variance.
Hxnnthesis

H0 : 0m - 9L9! = 0 (null hypothesis)
H1 :9i,93 - 9L9! at 0 (alternative hypothesis)

If you accept H0, there is no significant evidence to say the expected total number

of accidents and the expected total turn accidents in 1993 is lesser or greater than 1991.

98

Since 0 is inside the 95% confidence interval [-3.81,57.81] and [-9.54,7.54], these data
are sufficient to accept the null hypothesis. Therefore, the conclusion can be drawn that
the process in 1993 did not produce a statistically significant change than the before year

(1991) at significance level on = 0.05 (two-tailed statistics test).

3.10 BEF ORE-AN D-AFT ER PROTECTED LEFT-TURN CONTROL

In this section, an example to demonstrate the use of the ITAGI system in
analyzing before-and-after protected left-turn control in accident reduction will be
demonstrated and explained. For example, in the City of Troy, left-turns were protected
at several intersections in 1992 (refer to Table 3.9). Table 3.9 shows the accident figures
for the intersections of Maple/Livemois and Maple/Rochester where protected lefi-turn
control was implemented in 1992. The performance after implementing this control
strategy is that he percentage of turn accidents decreased from 13% in 1991 to 7% in
1993, and to 14% in 1994. However, while the total number of accidents decreased in

1993, they increased again in 1994.

Table 3.9 Intersections Implementing Protected Left-Tum Control in 1992

 

 

 

 

 

 

 

 

 

 

llntercccflon Name 19 91 19 93

I Total Ace. Tum Ace. Total Acc. Tum Ace.
[Maple 8. Livernois 34 6 23 3
lMaple & Rochester 28 5 22 0|
ITotal Accidents 62 11 45 3|

 

The 95% confidence interval for the expected total number of accidents for year 1993

99

minus the expected total number of accidents for year 1991 ( “1.93 — “1.91 ) is:

 

(2X... —ZX.-,9. )i 1.96 x ./(ZX.-,.. +226...) ,v i=1,2

= (45-62) i (1.96 x ./(45 + 62))

-17i(1.96 x 10.34)

{-37.27 , 3.27]

And the 95% confidence interval for the expected total turn accidents for year 1993

minus the expected total turn accidents for year 1991 ( “1.93 - p1,.91 ) is:

 

(2X... -ZX.-,91 ): 1.96 x ([(ZX... + 2X6.) ,v i=1,2

= (3-11) .1: (1.96 xm)
= -8 i (1.96 x 3.74)
= {-15.33, -O.67]
where the total number of accidents in 1993 2X,93 is approximately normal with

unknown mean ”93 and unknown variance 1193 , and

the total number of accidents in 1991 Exam is approximately normal with
unknown mean and unknown variance
Hypothesis

H0 : 9,93 - 9L9! =0 (null hypothesis)
H1 29,,” - 9L9! #- 0 (alternative hypothesis)

If you accept HO, there is no significant evidence to say the expected total number

100

of accidents in 1993 is fewer than 1991. Since 0 is inside the 95% confidence interval
{-37.27, 3.27] for total accidents, these data are sufficient to accept the null hypothesis.
However, we must reject H0 with respect to the expected turn accidents since 0 is outside
the 95% confidence interval {-15.33, -0.67]. Therefore, the conclusion can be drawn that
the process in the after year (1993) did not produce fewer total accidents than the before
year (1991), but it did produce fewer turn accidents than the before year (1991) at

significance level a = 0.05 (two-tailed statistics test).

3.11 SUMMARY

3.11.1 GIS CAPABILITY FOR TROY

Geographic Information Systems (GIS) play an important role in Intelligent
Transportation Systems (ITS). The term ITS, (previously known as IVHS), is composed
of six major areas: Advanced Traffic Management Systems (ATMS), Advanced Traveler
Information Systems (ATIS), Advanced Vehicle Control Systems (AVCS), Commercial
Vehicle Operations (CVO), Advanced Public Transportation Systems (APTS), and
Advanced Rural Transportation Systems (ARTS) [Euler, 1990]. The definition of GIS
according to Simkowitz [1988] is a computerized database management system for the
capture, storage, retrieval, analysis, and display of spatial (i.e., locationally defined) data.
A GIS is a useful decision support tool for transportation applications because it provides
a visual display of a transportation network and the associated spatial data. GIS has been
used primarily for planning activities by transportation agencies. With the development

and availability of larger and faster computers, GIS appears to have applications for

l01

traffic management operations in real time. In ATIS and ATMS, it would be desirable to
capture real time traffic data and apply it to a GIS database to provide roadway
information that the operator at the traffic operations center can use to make decisions.
Most of the traffic characteristics data, such as traffic volume, travel speed and delay data
can now be captured through detectors and transmitted on to a control center in real time.
With a GIS, the operator in the traffic operations center could monitor the road conditions
in real time. By adding a static accident database to the GIS, the operator could identify
where high hazard locations are and can associate these locations with real time traffic
characteristics data. GIS offers the decision maker information on which traffic control
strategies can be based. The ITAGI system was developed as a GIS-based tool for use in
the traffic operations center in Troy. The operator in the traffic operations center can
monitor traffic data in real time, and with ITAGI examine the accident records over the
past six years at specified locations, and make decisions based on real time data on traffic

conditions combined with historical accident data.

The ITAGI system provides a convenient format for the user to obtain the
specified information. More importantly, rather than being limited to textual queries, it is
possible to perform geographic queries in the ITAGI system. The user simply selects the
desired points on the computer screen, and acquires the desired information. This is an
important feature of a GIS which is not included in a general spreadsheet program, such
as Excel. Because transportation-related data usually have a spatial component, the most
natural way to associate elements from different data sets is through a consistent spatial

referencing system. The ITAGI system is capable of topological operations; that is, it

102

understands how elements contained in the database are related to each other spatially and
it can perform spatial manipulations on these elements. In other words, what
distinguishes a GIS from a traditional database is that the attribute data are associated
with a topologic object (point, line, or polygon) that has a position somewhere on the
surface of the earth. Using the ITAGI system, easily understood data can be shown on a
screen. A convenient graphic interface program provides quick access to traffic
characteristics information and easily understood presentations, improved updating
capabilities, and a more flexible and comprehensive source of information than general

spreadsheet programs.

3.11.2 GRAPHIC INTERFACE CAPABILITY FOR THE CITY OF TROY

A major advantage of the ITAGI system is quick access. The ITAGI system has
already embedded several hundred customized maps, easily-understood color-coded
graphics and information associated with their locations in the system. The data includes
lane configuration, traffic accident data stratified by high, medium and low levels, or as a
display of the top five locations by number of angle, turn, rearend, or other accident
types, and a comparison of specified differences in yearly accident frequencies. These
and other data can be displayed on the screen at the request of the user. This is an
important feature of the ITAGI system.

The second important feature of the ITAGI system is graphical presentation.
Specifically, a graphic representation of the data can serve as an effective “filing system”

because any specified location(s) in Troy can be instantly accessed by pointing to a map

103

rather than by searching through records. With little computer knowledge, the user or
traffic operator can easily get the information by simply selecting the desired location.
This fimction can be very useful in ATIS when information is provided on an in-vehicle
screen to the general road user via the traffic operations center. At this time, a quickly
understood graphic presentation definitely is more useful than raw data.

The third important feature of the ITAGI system is the provision for possible
modification. A GIS data storage system is able to respond to the changing needs of
transportation agencies. The ITAGI system was designed for this purpose. The
separation of the database and source program allows a more flexible data input and
output process. The program codes need not be rewritten when the coded data are
revised. Thus, the system is easily maintained. There are some exceptions, such as the
aggregation and disaggregation of data, and the introduction of other types of accidents,
or other variables. These limitations will be explained in the “limitations of ITAGI”

section.

Finally, the ITAGI is a more flexible source of information than a general
spreadsheet program. GIS can be formulated to meet the many objectives of the
transportation agency. Inevitably, every transportation agency that implements a
roadway information system (RIS) will have some GIS data requirements that will be
unique to that agency. In this system, a traffic accident database, GIS capability of
selecting and aggregating single or multiple intersections and midblock locations, the
display of the safest path, quality control of traffic accidents, and a multimedia left-turn

control expert system are integrated into a single system. The operator in the traffic

104

operations center is not required to switch or retrieve files and perform routine
calculations with the ITAGI system.

In summary, the ITAGI system provides quick access and response to traffic
characteristics information, more easily understood presentations, improved updating
capabilities, and a more flexible and comprehensive source of information than general

spreadsheet programs.

3.11.3 DETERMINING MULTIPLE INTERSECTION OR MIDBLOCK

CHARACTERISTICS

The ITAGI system has the ability to combine data from multiple intersections (or
groups of intersections) or midblock locations. With mathematics-performing capability
(note that currently, the program has aggregation ability only, but it can be rewritten to
perform other mathematics-performing functions), the ITAGI system allows the user to
do mathematics on the map. As described in section 3.3.2 and section 3.3.4, by
combining traditional conditional query capabilities with spatial queries and adding
mathematical manipulation, statistics, and charting, a full assessment of the road system
is possible from within the ITAGI system. By using the “add one more” function, the
user can exam accident records over the past six years and to incrementally add desired
locations (intersection or midblock) and look at the incremental aggregate accident data
for these locations. This capability is useful when a new traffic control strategy, such as a
left-turn control strategy is implemented on specific locations. The ITAGI system allows

the user to see the aggregate change in the total number of accidents, or the percentage of

105

any specified type of accidents at these locations.

3.11.4 SAFEST (SHORTEST) PATH SEARCHING CAPABILITY

In the FAST-TRAC federal demonstration program, Ali-Scout has been selected
as the route guidance and in-vehicle navigation system for improving travel efficiency
based on finding the minimum travel time path between any two points in the network.

The ITAGI system has been constructed to find the minimum path between any
two points based on other characteristics contained in the database. For example, if
current data on the level-of-service (LOS) at each intersection approach were available, it
would be possible to select the route between two points that maximizes the LOS, thus

avoiding congested areas of the city.

To demonstrate this capability, the ITAGI system was programmed to select the
route which passes through intersections that had the lowest accident rates in the most
recent year for which accident data are available. This is named the safest route. The
safest route is drawn on the screen and the analyst can see how the logic directs the route

away from certain major intersections.

3.12 LIMITATIONS OF ITAGI
3.12.1 LACK OF DISAGGREGATE CAPABILITY

Like most analysis programs (not including general spreadsheet programs),
aggregation and disaggregation problems also exist in the ITAGI system. When

developing a program, a decision must be made on the number of years to use for a

106

historical base and on how many variables to include in the system. The ITAGI system
was designed primarily for the evaluation of SCATS since this evaluation was
programmed to last three years, an annual database (rather than a daily or weekly
database) was selected, and the prescribed graphical results are shown on a yearly basis
before and afier SCATS. Because of its intended use, certain types of accidents (angle,
turn, and rearend) and limited variables (traffic accidents, traffic volumes, level of
service, and average travel speed) that were necessary for the evaluation of SCATS were
captured in the ITAGI system. These variables were selected in consultation with the
Road Commission of Oakland County. While the system was not designed to test other
types of accidents or to analyze changes occurring over a shorter period of time (less than
one year), this is not a limitation of the ITAGI system as the source program could be
rewritten for the user’s specified purpose. However, it is a limitation on the application
of the system as developed for this project. If a user wants to stratify accident data for a
shorter period of time, such as a month or a day, or to use variables such as weather
conditions (rain, clear, snow), road conditions (dry, wet, icy), or driver conditions
(drinking), the database could be rewritten to include these variables. For example, if the
user wanted to study the effect of converting traffic signals to flashing operation at night
on alcohol involved accidents, the original database could be structured to include these
variables. However, there is always a trade-off between the size of the database and the
processing time for the program. The decision was made to include only these variables

expected to be influenced by the SCATS system in the database.

107

3.12.2 DATA AVAILABILITY

Only selected intersections have turning movement volume data and only selected
corridors have travel time data because the data were not available for other intersections
and corridors. A study of the effect of geometric changes that occur in the system would
how to be conducted external to the ITAGI system, as was done with the left turn control
evaluation. The accident data is available in the ITAGI system, but the geometry could
be coded to represent either the before condition or the after condition. Thus, the system

is not designed to be used by an analyst not familiar with the network.

3.12.3 DATA ACCURACY

The traffic accident data were derived from the Michigan Department of
Transportation master tape by writing SPSS program. Thus, any coding errors or location
error included in the master tape will be carried forward into the ITAGI system. It is
known that the 1992 master tape contains location errors, and these will be found in the
ITAGI database. For this reason, comparisons were made for the years 1991 and 1993 or
1991 and 1994 in this study. The criteria used to separate the intersections from the
midblock locations were the physical road (PR) number and mileage. The definition of
the intersection used in the ITAGI system is the location of the intersection of the two
centerlines plus 0.03 miles in each direction. This criteria may or may not be the same as

other agency’s definitions of an intersection.

CHAPTER 4 DEVELOPMENT OF THE MULTIMEDIA LEFT-TURN

CONTROL EXPERT SYSTEM

This chapter describes how the multimedia lefi-turn control expert system was
developed (knowledge acquisition and knowledge representation) and calibrated (model
validation). This expert system provides knowledge-based and user-fiiendly information
for the user. In addition, this expert system utilizes the computer multimedia technique to

store video images, audio voices, pictures, and graphics for CD-ROM computer use.

4.1 KNOWLEDGE-BASED EXPERT SYSTEMS

Mishkoff [1985] defined expert systems or knowledge-based systems as computer
programs that contain both declarative knowledge (facts about objects, events, and
situations) and procedural knowledge (information about courses of action) to emulate
the reasoning processes of human experts in a particular domain or area of expertise.
Harmon and King [1985] also defined expert systems as artificial intelligence programs
designed to represent human expertise in a specific domain. A knowledge-based expert
system is a computer program that emulates human behavior in solving problems. It
includes a separate mechanism that mimics the reasoning process of the human brain.
Expert systems are composed of at least three basic entities: the knowledge base,
inference engine, and a user interface.

The knowledge base contains rules expressing an expert’s heuristics for the

domain. The collection of facts, rules, and computational procedures that represent this

108

109

domain is called its knowledge base; it is the power base of the expert system.

The inference engine is made up of rules that are used to control how the rules in
the knowledge base are used or processed. Inference is the derivation of new facts from
known facts. An inference engine is the processor of the known facts. The set of
procedures for manipulating the information in the knowledge base to reach conclusions
is called the inference engine. The objective of the inference engine is to find one or
more solutions for the goal (or for a subgoal) of the consultation. It searches the facts and
rules in the knowledge base and identifies and stores conclusions.

The user interface allows communication or interaction between the expert system
and an end user. This can be textural, or the text can be augmented by visual or audible
communication.

A knowledge-based expert system is one of the applications of artificial
intelligence (AI) research to software programming. Artificial intelligence is a specialty
area of computer science that attempts to make computers behave in a way that mimics
logical human behavior [Ramamoorthy et al., 1987]. Robotics, image processing, and
pattern recognition, as well as knowledge-based expert systems, are branches of artificial
intelligence.

Expert systems provide advice for problem solving that is derived from the
knowledge of experts. Expert systems typically make recommendations based on a
knowledge base stored as a set of rules and facts. The objective of an expert system is to
help the user choose a solution based on expert opinions.

As described above, the components of the expert system must contain at least

110

three parts: the knowledge base, the inference engine, and a user interface. The
knowledge base in the current application is composed of rules obtained from traffic
experts responses to a questionnaire. The inference engine is the component that
separates an expert system from a system using experts. If the knowledge base were
stored in the computer without any inference mechanism and there was a good user
interface to search the knowledge base, it could be considered a system using experts.

To construct a computerized expert system, the program must contain the
knowledge base, the inference engine, and a user interface. It is possible to write an
expert system in a computer language, such as C or FORTRAN. However, certain
languages have been developed with features which make them particularly suitable for
artificial intelligence work. PROLOG is a language which is designed for this purpose.
The ITAGI system and multimedia left-turn control expert system were developed using
an AI language PDC PROLOG, which stands for programming in logic. The reasons
for selecting this particular language are its descriptive nature, ease in describing a
relationship, and capability in symbolic processing.

By far, the most popular knowledge representation technique used in expert
systems is rule-based. The production rule is simply a combination of IF and THEN as
shown below:

Rule : IF Condition,
THEN Condition.

A rule-based system uses production rules to specify a set of conditions. The
left-turn control expert system is a rule-based expert system. It uses IF-THEN

statements to represent a production rule.

111

The PROLOG language provides a natural way to represent production rules. In
addition, PROLOG’s built-in pattern matching (unification) and backtracking

facilities make it easy to implement a backward-chaining inference mechanism.

4.2 OBJECTIVES
As described in chapter 2, the purpose of a lefi turn control expert system is to
expedite problem solving and decision making in the potential trade-offs between left-
tum efficiency and safety, and express the rules and facts in understandable “English”
rules. The objective of this study is to provide assistance in making the trade-off between
left-turn efficiency and safety among the three left-turn treatments, permissive turns,
protected turns, and no left turns. In this study, the object is divided into several subgoals
listed below:
1. to expedite problem solving and decision making in the potential
trade-offs between left-turn efficiency and safety,
2. to express the knowledge base (rules and facts) in understandable “English”
rules,
3. to use multimedia techniques to demonstrate how a knowledge-based expert
system (KBES) can be made more vivid and more effective than plain text, and
4. to demonstrate that the ITAGI system can be integrated with the expert system
to form a decision support tool to assist the decision maker (traffic engineer) in
retrieving and analyzing the available traffic characteristic data before

determining the appropriate left-turn control treatment.

112

4.3 THE DESIGN PROCESS

There were seven activities used to develop the multimedia left-turn control
expert system (Figure 4.1). These are: determining the variables for the lefi-turn control
expert system; designing a questionnaire for acquiring knowledge from traffic specialists
to build the knowledge base; video taping traffic experts and digitizing the video tape
using Intel Video Recorder Pro; recompressing and editing the video file by Asymetn'x
Digital Video Producer; using the ID3 method to build decision trees and embedding
rules and facts into the PDC PROLOG program; performing model verification and
validation for the questionnaire rules and inductive rules; and finally employing
Multimedia Toolbook 3.0 to integrate video, audio and picture files as a multimedia left-

tum control expert system.

4.4 PURPOSE

The purpose of this part of the study was to select guidelines for determining
whether permissive turns, protected turns or no left turn control should be used at an
intersection using an expert system. As noted in the literature review, only a limited
number of articles dealt with the selection of left-turn control by expert systems, and none

of the previous studies contained guidelines for the option of no left turn control.

4.5 THE FACTORS USED IN DETERMINING THE APPROPRIATE
LEFT-TURN CONTROL
As described in section 4.2, one of the subgoals for building a multimedia expert

system was to demonstrate the integration of the ITAGI system as described in chapter 3

113

829$ toaxm .9200 :5... :3 «60:53.2 9: 9.32650 5 32m 9:. fiv Bani

 

.3532 3.: 230:. an. .053. .033
3203:. 2 0.0 xoo£ooh 30062.32 .5 >o_nEm

 

 

 

.9

 

3...: 2526:. van 3:: 2352.333
2.: .2 £232.: use sea-oat: _ouoE .Eotoa

 

 

 

up

 

69.02.. 00.55.
can 25 so! u:- 33. none.
as. 35.5 no. 3 .2. 8.263 2:5

 

 

 

 

.836... 82> .335 5:25;
an 0.: can; .5 Eco uco 30.5833.

 

 

 

9

 

8a .0230:
00!) toEm .3:— >n 9.50. 0:. 002) 2.: ONEQO

 

 

 

up

 

253 we. goo-E. :2: 05.882
uco 33.302... 2:2. 0:30.325 one 03232:.
3:33. .2 83:55.30 .223 5:33. a c930

 

 

 

1»

 

£223 toaxo
.2280 52-..! 3. 33...: c0303 2: 05.582.

 

 

 

p

A 86

0 Q05

m 8%

v 9.6

m Q05

N .35

v 3%

 

822? £098 .223 95.-.... £025.55 a 23.. ch

 

 

 

66.30

114

as a decision support tool to assist the decision maker (traffic engineer) in retrieving and
analyzing traffic data necessary to determine the appropriate left-turn control treatment.
Considering the decision variables used in the previous studies (literature) and the desire
to integrate the ITAGI system in the selection process, the factors to be used in

determining the left-turn control treatment were selected. They are:

1. overall level of service at the intersection,

2. left turn level of service at the intersection,

3. upstream and downstream intersection lefi turn level of service,
4. the frequency of left-turn accidents per year,

5. the left-tum volume (veh/hr),

6. the opposing through and right-turn volume (veh/hr),

7. the average speed in the driver’s direction (mph), and

8. the average speed in the opposite direction (mph).

4.5.1 OVERALL LEVEL OF SERVICE AT THE INTERSECTION

The level of service (LOS) at the intersection is an important factor which will
affect the choice of the left-turn control strategy. If the LOS at the intersection is too low,
such as E or F, there is often insufficient time in the cycle for a protected left-turn phase.
There may also be insufficient Opportunities for a permissive left-turn movement. An
option is to prohibit left turns for the approach. If the LOS is D, permissive left-turn
control might be considered, while if the LOS is A, B, or C, with high left-turning

volume the protected left-turn control becomes an option.

115

4.5.2 LEFT -TURN LEVEL OF SERVICE AT THE INTERSECTION
For the same reasons as above, the left-turn level of service at the intersection is
an important variable. If the left-turn LOS is lower than the LOS of the other

movements, protected left-turn control may be considered, depending on other factors.

4.5.3 UPSTREAM AND DOWNSTREAM INTERSECTION LEFT-TURN LEVEL
OF SERVICE
If the upstream and downstream lefi-turn movements are already at LOS E or F, a
protected or permissive left-turn control is desirable at the intersection, since there may

not be acceptable alternatives for the driver if left turns are prohibited.

4.5.4 THE FREQUENCY OF LEFT-TURN ACCIDENTS PER YEAR

According to ITE Traffic Engineering Handbook [1991], protected left-turn
control phasing should be considered when on any approach, there are four or more
left-turn accidents within one year. Therefore, if the left-turn traffic accident frequency is

high, protected left-turn control or prohibiting lefi turns should be considered.

4.5.5 THE LEFT-TURN VOLUME

The left-turn volume is also an important factor in selecting the appropriate
left-turn control strategy. Historically, the product of left-turning volume and the
opposing through and right-turning volume has been considered when deciding on a

left-turn control strategy. If the demand for left-turning vehicles is high and the opposing

116

through and right-turn volumes are high, permissive left-turn control will not function

satisfactorily.

4.5.6 THE OPPOSING THROUGH AND RIGHT-TURN VOLUME

According to the Traffic Engineering Handbook [1991], protected left-turn control
should be considered when the product of left-turning vehicles and opposing volumes
during peak hours exceeds 100,000 on a four-lane street or 50,000 on a two-lane street,
and the left-turn volume is greater than two vehicles per cycle with an average delay per

left-turning vehicle of at least 35 seconds.

4.5.7 THE AVERAGE SPEED IN THE DRIVER’S DIRECTION (MPH)
Higher speed in the driver’s direction increases a hazard for left-turning vehicles.
The speed differential increases the risk of a collision between through traffic and traffic

slowing to make a left turn. Therefore, prohibiting left turns should be considered on

high speed roads.

4.5.8 THE AVERAGE SPEED IN THE OPPOSITE DIRECTION (MPH)

The speed of the traffic opposing the left-turning vehicles is an important factor in
the safety of left-turns. As the speed of opposing traffic increases, it becomes more
difficult for left-turning vehicles to judge the gap, and traffic accidents may increase.
According to the Bottom et al. [1983] studies, most drivers underestimate time gaps at

high speed. Therefore, high-speed opposing traffic argues against permissive left-turn

117

control.

4.6 SAMPLE SIZE

The selection of the expert(s) is an important task in the development of an expert
system. Unlike a general poll, or marketing survey, the appropriate sample size for
building a knowledge-based expert system can not be easily determined by assuming (or
testing for) a statistical distribution. The quality of the knowledge is much more
important than the quantity of the knowledge. A large number of non-experts is worse
than a few good experts who can contribute valuable knowledge to the expert system.
None of previous studies reviewed for this study discussed the sample size needed in an
expert system. However, Gonzalez and Dankel [1993] indicated the relationship between
the size of an expert system and the number of experts interviewed. They assumed a
direct relationship between the size of the knowledge-based system and its complexity.
They concluded that small knowledge-based systems (containing less than approximately
200 rules) could be obtained from one expert or from the printed literature. Medium-
sized knowledge-based systems (containing between 250 and 1,000 rules) are generally
developed using specialized workstations or larger computers. The system may employ
multiple experts in their development because of the need for multiple fields of expertise
to build the system. The large expert systems (containing more than 1,000 rules) used to
solve intricate and complex problems need multiple experts and a workstation or special
purpose computer. These complex problems need multiple fields of expertise to build the

system, and the problem itself can not be well defined.

118

The assumption about the size of the knowledge-based system and its complexity
may not always be valid. In fact, the number of rules is not necessarily the only factor
that should be used in deciding the number of experts. Other factors, such as the breadth
of the knowledge based involved, the complexity of the rules (well or ill-defined),
experts’ competence and availability are also important factors for deciding the number
of experts that should be involved.

In this study, the expert system is a small knowledge-based expert system
containing less than 200 rules, and only one field of knowledge is involved
(transportation engineering or traffic engineering). The rules used in this expert system
were derived from a questionnaire, and the rules were well-defined (simple) and case by
case (case-based) as shown in the questionnaire. The selected experts’ competence were
measured by a combinations of several factors. First, experts were selected from related
organizations, such as Michigan State University (MSU), Michigan Department of
Transportation (MDOT), and Oakland County Road Commission (OCRC). Second,
years of employment in the field (transportation engineering or traffic engineering) were
considered. The last factor was the experts’ availability. The experts needed to be
available and willing to participate throughout the expert systems development processes.
After considering these factors, traffic experts for the first and second survey (model
validation and consistency check) were selected (Table 4.1). In the first survey, six traffic
experts were asked to complete the questionnaire. Among the traffic experts, three are
professors, Dr. William Taylor, Dr. Richard Lyles, and Dr. Thomas Maleck from

Michigan State University. The fourth expert from Michigan State University was Mr.

119

James Neve who served as a district traffic engineer for the Michigan Department of
Transportation (MDOT) for many years. The other two experts were from the Oakland
County Road Commission (OCRC), they are Mr. Mohammed Lutfi, a senior traffic
engineer, and Mr. Dave Allyn, the director of traffic safety at OCRC. In the second
survey, five traffic experts were asked to complete the questionnaire for model validation.
Among the traffic experts, four out of five were the same experts as in the first survey. In
addition, Mr. William Savage, who directed the traffic signal section of the Michigan

Department of Transportation for many years, was invited to participate.

Table 4.1 The Interviewed Traffic Experts List

 

 

 

 

 

 

 

 

 

 

 

IExport'c Name I Organmtion I Position Years In the

I I I rune-rem Field
IDr. William Taylor W [Michigan State University [Professor 38

IDr. Richard Lyles m IMichigan State University IProfessor 29

Ior. Thomas Maleck "- IMichigan State University [Professor 31

IMr. James Neve " IMichlgan State University [Senior Traffic Engineer 38

IMr. Mohammed Lutii “‘ IOakland County Road Commission ISenior Traffic Engineer 11

IMr. Dave Allyn ‘ IOakiand County Road Commission IDirector. Traffic Safety Department] 21

IMr. Willem Savage " [Private Consultant (self-employed) IPresioent I as

 

Note: * interviewed in the first survey
** interviewed in the second survey

*** interviewed expert in both surveys

4.7 KNOWLEDGE ACQUISITION AND REPRESENTATION
4.7.1 METHODOLOGY

A left-turn control questionnaire was designed for acquiring knowledge and for

120

use in developing the inductive learning tool. Because not all combinations of all the
variables were included in the questionnaire (by design), the knowledge-based expert
system was forced to develop inductive rules. The definition of inductive reasoning,
according to Webster’s dictionary [Webster, 1960] is “reasoning from particular facts or
individual cases to general conclusions”. In artificial intelligence terms, inductive
learning is performed by presenting to a machine a series of examples and the
conclusions drawn from each example. The machine, through a set of redefined
induction heuristics, analyzes each example and builds an internal representation of the
domain characteristics that are present.

ID3 is an inductive reasoning tool used to develop rules and facts by learning
from examples. ID3 was developed by Quinlan in the late 19708 to learn object
classifications from labeled training examples (a machine learning method). This tool is
very useful when a significant quantity of knowledge exists as examples or cases. The
detailed computation of the ID3 algorithm can be found in most artificial intelligence (or
expert system) books. The ID3 algorithm has the ability to derive a minimal classification
tree from examples and to construct the necessary decision tree.

The major components of the ID3 algorithm are:

1. attributes,

2. decisions,

3. examples, and

4. inductive decision tree (or table)

There are two sets of rules imbedded in the left-turn control knowledge based

121

expert system. First, all rules developed from the questionnaire. The second source is the
inductive rules developed by the ID3 algorithm. The search logic for this program is
shown in Figure 4.2. Given a set of input data, the search will first determine if an
answer already has been derived by the traffic experts for an identical data set. If the
identical data set does not exist, then the rules and facts created by the ID3 inductive
machine learning tool are applied to the intersection. If neither of these sources produces
a rule, the message “Rules can’t be found” will be shown on the screen. If a rule can be
found, it will immediately display the rule on the screen and explain what conditions that
led to this conclusion.

To narrow the factors the users had to consider, some assumptions were made and

presented to the experts when they were asked to complete the questionnaire:

1. the cycle length is equal to 80 seconds,
2. good intersection geometry,

3. there are no visual obstructions for drivers that want to make a left
turn, and

4. only two opposing lanes must be crossed.

The questionnaire for the left-turn control expert system contains nine attributes,
three decisions and twenty-one examples. The nine attributes are overall LOS, left-turn
LOS, upstream left-turn LOS, downstream left-turn LOS, total left-turn related accidents
at the intersection per year, left-turn volume (veh/hour), opposing through and right turn

volume (veh/hour), average speed in the driver’s direction (mph), and average speed in

122

 

 

 

 

 

 

 

 

Start
The ITAGI System
I rratnc I Treliic Level or Average I
Find Y” Accident 1 Volume Service Speed
Questionnaire I I
Rule? I
A i , 2 _—\_, V ___" , i
No
- _ A, Y m_
/ Find Yes I
Inductive Display the Rule
Rule 7 1
No
I _ /// \\
I ' , \
' ‘3‘"; Fm" u” ls Data Available?
newer

    

No

 

 

Display Left Turn
Policy

 

Figure 4.2 The Rule Search Logic Flow Chart for the Expert System

123

the opposing direction (mph). Each case presented in the questionnaire contained a
different set of attribute values. For example, in the overall LOS attribute, there are three
attribute values: C or better, D or E, and F or worse. The three decisions include
permissive turns, protected turns, and no left-turn control. The 21 examples were
developed using data from intersections in the City of Troy, Michigan. A copy of the

questionnaire is shown in Appendix A (first survey) and Appendix B (second survey).

4.7.2 THE INITIAL SURVEY RESULT

The initial survey results were collected and prepared as input files for the ID3
program. The program then constructed the inductive decision tree (table) for the
left-turn control expert system. The questionnaire based rules are shown as cases 1
through 21 in Table 4.2, Table 4.3 and Table 4.4. The inductive rules developed from the

responses to the questionnaire are:

Inductixemlefih
Implement permissive left-turn control
if left-turn LOS is C or better, and

if downstream left-turn LOS is C or better.

Inducfimmlflfl:
Implement permissive left-turn control
if left-turn LOS is C or better, and

if downstream left-turn LOS is D or E.

124

Table 4.2 The Percentage of the Traffic Experts Selecting Each Strategy (Case 1 - 8)

 

ItlonandcaaeNumber 'I 2 3 4 5 5 7

CorBetter

ME I l lTTxl I 1 [fl

ForWorse X X

 

CorBetter

DorE X X X X
ForWorse
NoLd‘tTum

CorBdter l l

DorE
ForWorse
NoLdiTum X X

 

 

 

 

 

 

CorBetter

DorE X X X X
ForWorse X
NoLei'tTum X X

 

 

 

 

 

 

 

 

 

 

 

 

 

0—5

6-‘° I l l I l 1

11mm

 

0-100
101—150
151— 200
201andabove

 

 

 

 

 

 

 

 

 

 

 

 

 

0- 800

601— 800

801mdabove
20~29.9 IXI IXI IXIXIXI
30andabove X X X

 

 

—
‘

15—19.9
20—29.9 l I X J I X I X I X I I X I
SOmdabove X X X

 

Permissive

Protected J 331 WI 61 0] GI oI war
331

No Ldt Turn

 

,.
Q
_o
a
_o
_9
2
u:

125

Table 4.3 The Percentage of the Traffic Experts Selecting Each Strategy (Case 9 - l6)

 

ltionandCaseNumber

CorBetter

'IU

11 13 '14 15

 

DorE
ForWorse

Cornetter

T1 H?
X

 

DorE

 

ForWorse

 

No Ldt Turn

CorBetter

 

DorE

 

ForWorse

 

 

No Left Turn

CorBetter

 

 

 

 

 

 

 

 

 

 

DorE

 

ForWorse

 

No Left Turn

 

 

 

 

 

 

 

 

 

 

 

6—10

 

I 118mm Ix I
0— 100

 

101 —- 150

 

151— 200

 

 

201mm

0-600

 

 

 

 

 

 

 

 

 

 

601—300
801andabove

15— 19.9

 

20—29.9
scandabove

15—19.9

 

20-29.9
soaridabove

Permissive

 

I Protected

 

I No Ldt Turn

126

Table 4.4 The Percentage of the Traffic Experts Selecting Each Strategy (Case 17 - 21)

 

 

 

r we 4mm 1 r4

F orWorse X X X

CorBetter
DorE
ForWorse I X
NoLeflTum X X X X

 

 

 

CorBetter

DorE I
ForWorse I X X X X
NoLeftTum

 

 

 

CorBetter
DorE I X
ForWorse I X X X

No Left Turn
0- 5

I 6'10 l I I I I I

11andabove

 

 

 

 

 

 

 

 

 

0—100

101 —150 I

151—200 I X
201andabove

 

 

 

 

 

 

 

 

 

0—600

 

 

I ““800 1 I I x I I XJ

801 and above

 

 

 

30 and above X X

15-19.9

20-29.9 I IXIXTXIH

30andabove X

 

Permissive
Protected
No Left Turn

 

o o o 33
so] 17

 

_—

 

 

 

 

810
8

 

 

127

WWW

Implement permissive left-turn control

if left-turn LOS is D or E,

if upstream left-turn LOS is C or better and

if left-turn volume is less than 100 (veh/hour).

Marietta:

Implement permissive left-turn control

if left-turn LOS is D or E,

if upstream left-turn LOS is F or worse and

if left-turn volume is less than 100 (veh/hour).

lndnctixemlfifi:

Implement protected left-turn control

if left-turn LOS is D or E,

if upstream left-turn is prohibited (no left turn) and

if left-turn volume is more than 201 (veh/hour).

Minimum
Retain prohibited left-turn control
if overall LOS at the intersection is D or E,

if left-turn is prohibited (no left turn),

128

if upstream left-turn is C or better and

if downstream left-turn LOS is F or worse.

[Mum

Retain prohibited left-turn control

if overall LOS at the intersection is F or worse,
if left-turn is prohibited (no left turn),

if upstream left-turn is F or worse and

if downstream left-turn LOS is C or better.

The percentage of the experts selecting various left-turn control strategies is
shown in Tables 4.2, 4.3 and 4.4. There was general agreement on the use of permissive
left turns and the prohibition of left turns (cases 4, 5, 6, 11, 12, 13, 18 and 19), while the
experts disagreed on the use of protected left turns (cases 1, 9, 15 and 16). The facts and
rules from the survey results and from the ID3 inductive tool form the left-turn control

knowledge base expert system.

4.8 MODEL VALIDATION

Because of the trade-off required in the left-turn control treatment decision, it was
necessary to use traffic experts to build rules. Since each individual traffic expert may
have a different concept of the most desirable trade-off between delay and safety, some
variation in the response to the questionnaire was expected. To determine whether the

inductive rules developed by the expert system were consistent with views of the experts

129

when they were presented with new cases, a model validation procedure was developed.
The model validation process is depicted in Figure 4.3. The overall purpose of model
validation is to test if the expert system mimics the thought process of the experts. One
form of validation is a check for consistency. The purpose of a consistency check is to
determine whether the experts give the same answer to identical questions asked at
different times. There were two consistency checks used in this study. First, two
identical questions were included in the original questionnaire (first survey) to determine
if the answer presented by each expert was the same. Second, after constructing the
expert system, a second questionnaire was used in which several questions used in the
first survey were presented to the same experts again to see if the results were consistent.
Another type of validation was a check on the accuracy of the inductive rules developed
by the ID3 algorithm. Banning [1984] reported that system validation is ideally
accomplished by means of a Turing test (Figure 4.3) in which the same questions are
presented to experts and the computer expert system. The expert system is validated if
the results show no difference between the two sources. In the second questionnaire,
some of the questions from the first survey and some new questions which required use
of the inductive rules were mixed together. Six questions from the first questionnaire
were repeated on the second questionnaire as a consistency check (two questions for each
left-turn control strategy). Fifteen new scenarios selected to represent each of the
inductive rules were used as the ID3 inductive rule check (refer to Table 4.5). The
questions were mixed together and presented to five traffic experts including four experts

who completed the first questionnaire. Please refer to Table 4.6 for rule numbers and

130

     

 

   

 

 
  
 

  
   

: ' Obtal
Obtain Rules V18 Questiongaire
‘ ID3 Method RuIOS

  
     
    

 

        
       

KL- [M ,. LA...
I Model
Model
I Validation °°"‘"”"°V

      
    

Check

    

 

 
    

Check the
Questionnaire
Rule

Check the
I inductive Rule

T’M\

  
  
 
 

    

 

   

/_ - __
I Extract 6
Questions

Find 1 5
Questions
among Cases

      
   
    

Turing Test . - —————— .. - - . . ........

/ Mix Those

Questions
Together

V / Present to

 

      
   
 
  

 

 

  
  
  

    
  

   
 

Present to
Computer Traffic
Expert System Experts

flan the

Result ?

 
   

 
 

Computer
Expert System

  

Figure 4.3 The Model Validation Process for the Expert System

131

Table 4.5 The Selected Testing Questions for the Inductive Rules

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

is No. 17111.1 "111.! Ind. 3 ind.‘ lnd.‘ indJ Ind]
C or Better X X
X
l 6... l i I I x I I I I I
F or Worse x 7 7 x X
C or Better
D or E I X X X
F or Worse I
No Left Turn X X
C or Better 7 x
D or E I x
F or Worse I X 7 x X
No Left Turn X 7 X
C or Better 7 7 7
D or E I X x
F or Worse I 7 x 7 x
No Len Turn 7 7

 

0—5 X X X X X X X

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I
I °-'° I I I I I I I I
11endebove
0-100 x x x x
101—150 I x
151-200 I
201endabove 7 7 X
0400 x x x 7 x 7 7
l a-.. 1 1 1 1 x 1 1 1 x I
sound-hove 7 7 7 7 7 x 7

 

 

 

 

 

Pennlsslve

FPMW l I I I ITI I I

 

    

No Left Turn

Total: 15 paths

Note: ? represents the selected testing questions for inductive rules in the second survey; X and X
represent the questions which were asked in the first survey.

132

Table 4.6 The Model Validation Testing Rules for the Initial Survey

 

No. 0.3 0.6 0.0 0.7 I 0.18 0.1! Ind.f Ind.2 I Ind: Ind.‘ lnd.‘ M8 M7

 

Cormmu X X X 1 x

 

DorE )1 1x I ‘ .x

 

ForWorse . I x . I x

 

 

 

 

 

Coraeuu X I X

 

DorE X

 

Favmru

 

NeLettrun I x x ' . x x

Cormmu X X . X X

 

DorE X

 

ForVWxse X X

 

 

 

NoLett Tu'n x I 1 x

Cormmu' X X
DorE I x x

ForWorae I x ' I . x

NeLeitrirn . x

 

 

 

0—5 x x x x x x 1
6—10 I .
iimabove

 

 

 

 

 

0.100 X X X X X X
101—150 . 1

 

 

151—200 I ‘

 

 

 

 

 

601—800 I x I
aciuuaune X

 

15—199
20n299 X X X X x
30an1dmwe I x ‘

 

 

 

 

 

 

 

 

151-19.9 3 I I

 

20~29,9 I x 1 x ‘ x I
30andabmn X X

 

 

Permissive x x x x ' x x

Protected x x 1 _ x
No Len Tun ' x x I x x

I I

 

 

 

 

 

 

 

I 1
*lote: Q.5 means questionnaire rule number 5 and 1nd.3 means inductive rule number 3. etc.

Table 4.7 The Model Validation Results

133

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Case Expert #1 Expert 82 Expert 83 Expert #4 Expert #6 Rule Rule Rule
No. (First time) (First time) (F lrst time) (First time) Validation Consistent No.
1 1 2 2 1 2 2 x Ind. 5
2 1 1 1 1 2 1 x Ind. 2
3' 1(1) 1 (1) 1 (1) 1 1 (1) 1 x 0.5
4" 1(2) 2(2) 1 (2) 1 1 (2) 2 Q. 3
s 3 3 3 3 1 3 x Ind. 7
e 1 1 1 1 1 1 x Ind. 3
1 1‘ "_1 +1 T11 i 1 ,_ ___—‘1 x Ind.3
8‘ 1 (1) 1 (1) 1 (1) 1 1 (1) 1 x 0.6
9'" 3 (3) 3 (3) 3 (3) 3 1 (1) 3 x Q. 19
10 3 3 1 1 1 1 x Ind. 4
11 1 1 1 1 1 1 x Ind. 4
12 1 1 1 1 2 1 x ind. 2
13 1 1 1 1 1 1 x Ind. 1
14 1 1 1 1 1 1 x Ind. 4
1s 1 1 1 1 1 1 x Ind. 4
1e 3 1 1 1 1 1 x Ind. 3
17 3 3 3 3 2 3 x Ind. 7
1s 1 3 3 3 1 3 x Ind. 6
19 2 1 2 1 2 2 x ind. 5
20'" 1 (3) 3 (3) 1 (3) 3 2 (1) 3 0.18
21" 2 (2) 1 (2) 2 (2) 1 1 (2) 2 0.7
ISecond TimeI —T L— LI

Note: 1 represents permissive left-turn control See

2 represents protected left-turn control Table 4'6
3 represents no left-turns fl?"
definitions

“ represents permissive left-turn control consistency check
" represents protected left-turn control consistency check
I""“"represents no left-tums consistency check

 

 

 

 

134

Table 4.7 for the model validation results.

4.8.1 CONSISTENCY CHECK RESULTS

In the first survey, one out of six interviewed experts failed to give the same
answer when presented identical cases (case 10 and case 18 in Appendix A). From Table
4.7, the consistency check result shows that the experts did not always select the same
strategy when the identical data set was presented in the first and second questionnaire.
In the second survey, three cases (which then are treated as questionnaire rules) were
answered differently by at least one expert. The logical reason for the traffic experts to
change their mind is that the difference between the two strategies is small, and either of
the strategies would be acceptable. For example, the choice between permissive and
protected left-turn control must have been close for the questionnaire cases 3 and 7
because of the low number of left-turn accidents (0-5 per year), the high left-turn volume
(201 and above), low opposing through and right-turn volume (0-600 veh/hour), high
opposing speed (30 and above mph) and D or E left-turn level of service. From the safety
and delay viewpoint, the low number of left-turn accidents per year and D or E left-turn
level of service may cause the traffic experts to choose the permissive left-turn strategy.
However, from the safety point of view, the high left-turn volumes and high opposing
speed may cause the traffic experts to choose the protected left-turn control strategy. On
the second survey, all experts except one agreed that the choice for questionnaire rule 18
was between permissive and no left-tums. This scenario included prohibiting left-turns at

the intersection, heavy opposing through and right turn volume (801 and above

135

veh/hour), high traffic speed in driver’s direction (30 and above mph), low left-turn
accidents (0-5 per year), and D or E overall intersection level of service. These
conditions caused some traffic experts to vary their choice between permissive and no
left-turn treatments. From the safety point of view, fewer traffic accidents should occur
at the intersection when left turns are prohibited. Therefore, retaining the prohibited left-
turn control may be a good choice. However, from the delay and convenience viewpoint,
when left-turns are prohibited at the downstream intersections and there is D or E overall
level of service at the current intersection, it may be a good choice to implement
permissive left-turn control because the traffic has nowhere to make left turns.

The fact that there is a choice between safety and delay in the reason for
considering the development of an expert system. If the traffic engineer merely relies on
computer software to find the lowest delay solution for the left-turn control strategy, the
protected left-turn control will always be the worst choice because it takes too much
green time from the other vehicles (directions). However, the use of permissive control
may cause a safety problem, and it is not reasonable to prohibit left turns at all
intersections. In both surveys, the safety factor was kept constant with the number of the
left-turn accidents between 0 and 5 per year. However, potential safety considerations
were embedded in other variables, such as the opposing traffic speed, the cross product of
left-turn volumes and opposing through and right turn volumes. These safety factors can
not be measured by the computer software. Therefore, if the objective function for the
left-turn control treatment includes safety, efficiency, and delay factors, the expert system

can provide an acceptable answer because the experts are expected to take all these

136

factors into account.

Some inconsistency between the results of the first and second survey was
expected because of the sample size. The fewer the number of interviewed experts, the
more likely it is to get an inconsistent result. In fact, in the first survey, one expert
selected two different treatments when the same conditions were presented twice. In the
second survey, if two out of the four experts changed their mind, it could lead to the
conclusion that the questionnaire rules are not consistent, even if the rule was based on a
minimum 67 percent majority. On the other hand, if there were a 67 percent majority for
one solution in a sample of 20 and two experts changed their mind, the rule would not

change, and the solution would meet the consistency check.

4.8.2 INDUCTIVE RULES’ VALIDATION RESULTS

In Table 4.7, where the inductive rules are marked with asterisks, it can be seen
that the ID3 inductive rules exactly matched those of the experts. All expert system
answers were the same as the traffic experts. Therefore, the ID3 inductive rules are

considered to be valid representations of the experts.

4.9 REVISING THE PDC PROLOG PROGRAM

After calibrating the expert system, there were three questionnaire rules which
were not consistent with the first survey results. All of these changes were the result of
two or more experts changing their recommendations when provided with the same

scenario. Therefore, the PDC PROLOG program was revised by removing these three

137

rules under the assumption that they represent close choices. The elimination of rules
based on only one response to these close choice scenarios is one of the benefits of

conducting a consistency check.

4.10 DEVELOPING THE MULTIMEDIA EXPERT SYSTEM

A multimedia program is the integration of two or more media effects (text,
graphics, sound, video, and animation) to express or illustrate a message or tell a story,
and to be viewed and interacted with on a computer. In this study, Multimedia Toolbook
3.0 was employed as a media integration tool to combine video, audio, and graphics.
Intel Smart Video Recorder Pro was used as a digitizing tool to convert from video tape
to video file (.avi) and the Asymetrix Digital Video Producer was utilized as an editing
video file (.avi) tool.

Multimedia Toolbook 3.0 is the product of Asymetrix Corporation. It runs in two
different modes: author mode, in which the application is created; and the reader mode,
where the user interacts with the creation. Multimedia Toolbook 3.0, as its name implies,
operates like a book that has pages, and the reader can interact with each page, move
from page to page, or jump to any page.

Multimedia Toolbook 3.0 is an object-oriented program. Therefore, while
objects are easy to create, the relationship among objects needs to be explained by the
user. The language used to explain this relationship is called script. The script contains
information which explains what happens when the user clicks a button or moves a scroll

bar from one place to another.

138

The first step in the process is to capture the video as a digital computer file (.avi).
After digitizing the video, the video must be trimmed and reproduced or recompressed to
size the image (.avi file) to fit the Multimedia Toolbook 3.0 requirements.

The multimedia left-turn control expert system was constructed using the tools
described above. It contains two parts: one is the input window used as a data entry
allowing users to select any combination of the input variables. Note that these
selections are exclusive (i.e. single selection). Once the user selects any combination of
inputs and pushes the “result” button, the suggested left-turn control strategy and a data
entry summary will be displayed. If the user would like an explanation of why these
variables are related to the result or what the left-turn control guidelines are, the user can
push the “right arrow” button to see the video tape. Note that the video stage in the
program looks like a VCR which can be rewound and paused repeatedly. The video
tape’s source was created by taping Dr. William C. Taylor of Michigan State University.
The contents of Dr. Taylor’s comments have been summarized in the text window located
below the video stage. In addition, pictures showing experts who contributed their

knowledge to this KBES are posted on the right side of the video screen.

CHAPTER 5 SUMMARY

5.1 SUMMARY

This paper first described the development of the ITAGI system, then illustrated
the use of the ITAGI system to conduct a safety analysis of SCATS. Second, this paper
described the development of a multimedia left-turn control expert system. This expert
system was designed to use the ITAGI system as a data source. Using data from
intersections in the City of Troy, and the results of a questionnaire, the expert system
generated a set of questionnaire rules and developed additional inductive rules by using
the ID3 algorithm. These rules were tested using both a consistency check and a Turing
test. These exercises have demonstrated that both the ITAGI system and the
multimedia left-turn control expert system are useful tools in conducting traffic
engineering studies. The following objectives of this study have been met:

1. Production of a graphic interface capability for the City of Troy. The
ITAGI system has embedded several hundred customized maps, color-coded graphics and
information associated with intersection(s) or road segment(s) in the City of Troy,
Michigan. The data available to the user for any location include: number and type of
traffic accident, number of traffic lanes, level of service, intersection delay, accident rate,
traffic volume, and average travel speed on the major corridors. Once a location or a set
of locations is specified, the number and type of accident (angle, turn, rearend, and

others) for the years 1989 through 1994 can be displayed in a bar chart, or pie chart

139

140

format. Detailed information concerning this graphic interface capability is contained in
section 3.2. Several examples of the use of this tool are provided in section 3.6. The
traffic accident rate data available in the database were used in developing the route
guidance system and quality control charts. As an example of an ATIS, the ITAGI
system is programmed to select the safest path between any two locations in the network.
The details of the safest path theory used in the route guidance system is described in
section 3.4. Another system contained in the ITAGI system is the quality control charts
for traffic accidents. The use of the quality control chart to determine the high accident
locations and road segments (outliers) necessary for a Highway Safety Improvement
Program (HSIP) is illustrated in section 3.5.

2. Development of a GIS capable of locating accidents in the City of Troy. A
GIS is a useful decision support tool for roadway information systems because it provides
a visual display of a transportation network and associated spatial data. The ITAGI
system provides a convenient format for the user to obtain the specified information from
the map. Rather than being limited to textual queries, it is possible to perform geographic
queries. The user simply selects the desired points on the computer screen, and acquires
the desired information. This capability is described and illustrated in sections 3.3.1 and
3.3.2 for intersection selection(s) and sections 3.3.3 and 3.3.4 for midblock selections(s).
Examples based on this capability are contained in section 3.6.

3. Development of a multimedia left-turn control expert system. Multimedia
techniques were employed to demonstrate a knowledge-based expert system (KBES) for

selecting the most appropriate left turn control strategy at an intersection. The ITAGI

141

system was integrated with the expert system to demonstrate how data can be accessed to
assist the decision maker (traffic engineer) determine the appropriate left-turn control

strategy.

5.2 THE RESULTS OF ANALYSES USING THE ITAGI SYSTEM

The ITAGI system was designed as a tool to graphically display traffic
characteristics data that can be used to evaluate changes in the traffic control system.
Studies were conducted to evaluate the effect of SCATS and left-turn control strategy

changes implemented at specific locations in the City of Troy, Michigan.

5.2.1 BEFORE AND AFTER SCATS ACCIDENT CHANGES

In the evaluation of SCATS, the ITAGI program was used to conduct an analysis
of traffic accidents and injury accidents before and after SCATS was installed. The
conclusion reached is that while certain accident types decreased after the installation of
SCATS, overall there was no significant reduction in the accident frequency or severity.

The intersection accident comparison shows that the number of traffic accidents in
1993, when the majority of the intersections in the city were equipped with SCATS
controller, was lower than the number of traffic accidents in 1991 or 1992. In 1994 and
1995 the intersections experienced more accidents than in 1991, 1992, and 1993. This
was not the expected result when implementing the SCATS system. There was a large
increase in both total and injury accidents at intersections in the first six months of 1994

(after installation of SCATS). The midblock accident comparison showed similar results,

142

where the number of traffic accidents in 1993 was lower than the number of traffic

accidents in 1991 followed by an increase in 1994 and 1995.

5.2.2 LEFT TURN POLICIES

The effect of prohibiting left-turns at specific intersections was analyzed by using
the aggregate bar chart and pie chart properties of ITAGI. These results show that left
turns were prohibited starting in 1992 for northbound and southbound traffic at the
intersections of Big Beaver/Crooks, Big Beaver/Livernois, and Big Beaver/Rochester.
After implementing this control, the total number of accidents increased from 110 (1991)
to 137 (1993), to 151 (1994). The result shows that the percentage of turn accidents
remained relatively stable: 4% (1991), 7% (1993) and 4% (1994).

A statistical test was conducted to determine if the prohibition of left turns
resulted in an increase in accidents in 1993 compared to 1991. The statistics test shows
that the null hypothesis could not be rejected for either total accidents or turn accidents at
significance level or=0.05.

The effect of implementing protected left-turn control was also analyzed by using
the aggregate bar chart and pie chart properties of ITAGI. Protected left turn control was
implemented in 1992 at the intersections of Maple/Livemois and Maple/Rochester. As a
result of this change, the percentage of turn accident decreased from 13% (1991) to 7%
(1993). The total number of accidents also decreased in 1993, but both numbers

increased again in 1994.

143

A statistical test was conducted to determine if the protected left turn strategy
resulted in fewer accidents in 1993 than in 1991. The statistics test shows that the null
hypothesis could not be rejected for total accidents, but it was rejected for turn accidents

at significance level or=0.05.

5.2.3 TRAFFIC LANES IN MIDBLOCK LOCATIONS

Using the ITAGI program, differences in the number of traffic accidents occurring
on 2-lane roads, 5-lane roads, and Boulevards were analyzed. From this analysis, the
results show that:

1. The Boulevard segments had the highest number of traffic accidents, followed
by S-Iane roads, and then 2-lane roads.

2. 5-lane roads have the highest percentage of angle accidents, then 2-lane roads,
and then Boulevards.

3. After implementing SCATS in 1993, 2-lane roads showed a significant
increase in the percentage of rearend accidents.

4. After implementing SCATS, the percentage of angle accidents on 2-lane roads
decreased from 20% (1992) to 18% (1993), to 9% (1994), while the percentage of angle
accidents on 5-lane roads increased slightly from 21% (1992) to 27% (1993), to 23%
(1994). On Boulevards, the angle accidents increased from 7% (1992) to 8% (1993), to

9% (1994).

144

5.3 DEVELOPMENT OF A MULTIMEDIA LEFT-TURN CONTROL EXPERT
SYSTEM

The objective of the multimedia left-turn control expert system is to provide
assistance in making the trade—off between left-turn efficiency and safety among the three
left-turn treatments: permissive turns, protected turns, and no left turns.

The knowledge base was constructed through a questionnaire survey of experts.
The results were used to identify the questionnaire rules, which are shown in Table 4.2,
Table 4.3 and Table 4.4. There were 21 questionnaire rules developed. Because not all
combinations of all the variables were included in the questionnaire (by design), the
knowledge-based expert system was forced to develop inductive rules using the ID3
algorithm. The inductive rules are shown in Table 4.6. To determine whether the
questionnaire rules and inductive rules developed by the expert system were consistent
with views of the experts when they were presented with new cases, a model validation
procedure was developed. The overall purpose of model validation is to determine if the
expert system mimics the thought process of the experts. There were two tests involved
in the model validation stage: a consistency check and a Turing test. There were two
consistency checks used in this study. First, two identical questions were included in the
original questionnaire (first survey) to determine if the answer presented by each expert
was the same. One of the six interviewed experts failed to give the same answer when
presented identical cases (case 10 and case 18 in Appendix A). Second, when the same
case was presented on both the first and the second survey, three out of six selected
questions were answered differently by at least one expert. The consistency check results

shows that the experts did not always select the same strategy when presented with

145

identical data sets. The logical reason for the traffic experts to change their mind is that
the difference between the two strategies is small, and either of the strategies would be
acceptable. Finally, the Turing test was performed to test if cases where the inductive
rules obtained from the ID3 algorithm were consistent with the left turn control strategy
selected by the traffic experts. All the cases derived from the ID3 algorithm passed the
Turing test as all of the inductive rules matched those of the experts. Therefore, the ID3

inductive rules are considered to be valid representations of the experts.

LIST OF REFERENCES

I46

LIST OF REFERENCES

Agent, K. R., “Guidelines for the Use of Protected/Permissive Left-Turn Phasing”, ITE
Journal, v.57, no.7, July 1987, pp.37-42.

Asante, S. A., Ardekani, S. A., and Williams, J. C., “Selection Criteria for Left-Turn
Phasing and Indication Sequence”, Transportation Research Record 1421, 1993, pp.
11-20.

Bihan, P., Turbo Prolog: An Introduction to Artificial Intelligence, John Wiley &
Sons, 1987.

Bonneson, J. A. and McCoy, P. T., “Operational Analysis of Exclusive Left-Turn Lanes
with Protected/Permitted Phasing”, Transportation Research Record 1114, 1987,
pp.74-85.

Bottom, C. G., and Ashworth, R., “Factor Affecting Variability of Driver Gap
Acceptance Behavior”, Ergonomics, Vol. 21(9), 1983.

Braverman, J. D., Fundamentals of Statistical Quality Control, Reston, Va.: Reston
Pub. Co., 1981.

Bruce, E. L. and Hummer, J. E., “Delay Alleviated by Left-Turn Bypass Lanes”,
Transportation Research Record 1299, 1991, pp. 1-8.

Chang, E. C. P., “Selection Traffic Analysis Software Using Expert Systems”, paper
presented at the 66th annual Transportation Research Board meeting, Washington, DC,
January 1987.

Chang, E. C. P., “Expert Systems Application on Left-Turn Signal Treatment”, paper
presented at 66th annual Transportation Research Board meeting, Washington, DC,
January 1987.

Chang, E. C. R, Cohen, S.L., Liu, C., Chaudhary, N. A. and Messer,C.,
“MAXBAND-86: Program for Optimizing Left-Turn Phase Sequence in Multiarterial
Closed Networks”, Transportation Research Record 1181, 1988, pp. 61-67.

Chang, G. L., Zhuang, L., and Perez, C., “Permitted Left-Turn Capacity of Exclusive
Lanes: Simulation-Based Empirical Method”, Transportation Research Record 1457,
1994, pp. 96-103.

Cohen, S. L. and Mekemson, J. R., “Optimization of Left-Turn Phase Sequence on
Signalized Arterials”, Transportation Research Record 1021, 1985, pp. 53-58.

Cottrell, B. H., “Guidelines for Protected/Permissive Left-Turn Signal Phasing”,
Transportation Research Record 1069, 1986, pp. 54-61.

Craus, J. and Mahalel, D., “Analysis of Operation and Safety Characteristics of Left-
Turn Lanes”, ITE Journal, v.56, no.7, July 1986, pp.34-38.

Datta, T. K., Bowman, B. L., and Opiela, K. 8., ”Evaluation of Highway Safety
Projects Using Quality-Control Technique”, Transportation Research Record 672,
1978, pp. 9-16.

147

Deacon, John A., Zegeer Charles V., and Deen, Robert C., “Identification of
Hazardous Locations on City Streets”, Traffic Quarterly, v. 31, no. 4, Oct 1977, pp.
549-570.

Essen, I. V., Hanchette, C., and Dildine, G., “Evaluation of GIS Workstation
Performance Within a Distributed Network Enviromnent”, Transportation Research
Record 1261, 1990, pp. 52-61.

Florida Section ITE, “ Left-Turn Phase Design in Florida”, ITE Journal, September
1982.

Gazis, D. C., “Congestion Abatement in ITS Through Centralized Route Allocation”,
IVHS Journal, v. 2, no. 2, 1995, pp. 139-158.

Goldenthal, N., Turbo Prolog Programmer’s Guide, Weber System, Inc., 1987.

Gonzalez, A. J. and Dankel, D. D., The Engineering of Knowledge-Based Systems
Theory and Practice, Prentice-Hall, Inc., 1993.

Greiwe, R. R., “Intersection Management Techniques for the Left-Turning Vehicle: The
Indianapolis Experience”, ITE Journal, v.56, no.6, June 1986, pp. 23-28.

Grubba, J. and Barbaresso J. C., “IVHS on the FAST-TRAC”, IVHS Review, summer
1994, pp. 5-12.

Hancock, K. and Abkowitz, M., “The Use of Geographic Information Systems for
Customer Service in Urban Public Transportation”, preprint Paper No. 920896 presented
at Transportation Research Board, 71St Annual Meeting, January 12-16, 1992,
Washington, D. C.

Harris, R. A., Cohn, LE, and Bowlby, W., “An Application of Artificial Intelligence in
Highway Noise Analysis”, Transportation Research Record 1033.

Hummer, J. E., Montgomery, R. E., and Sinha, K. C., “Guidelines for Use of Leading
and Lagging Left-Turn Signal Phasing”, Transportation Research Record 1324,1991,
pp.ll-20.

ITE, Traffic Engineering Handbook, 1991.

Joshua, S. C. and Saka, A. A., “Mitigation of Sight-Distance Problem for
Unprotected Left-Turning Traffic at Intersections”, Transportation Research
Record 1356, 1992, p.73-79.

Kikuchi, S., Chakroborty, P., and Vukadinovic, K., “Lengths of Left-Turn Lanes at
Signalized Intersections”, Transportation Research Record 1385, 1993, pp.l62-171.

Klein, L. A., Rantowich, N. A., Jacoby, C. C. and Mingrone, J.,“IVHS
Architecture Development and Evaluation Process”, IVHS Journal, v.1, no.1, 1993,
p.13-34.

Lalani, N., Cronin, D., Hattan, D., and Searls, T., “A Summary of the Use of Warrants
for the Installation of Left-Turn Phasing at Signalized Intersections”, ITE Journal, v.56,
no.4, April 1986, pp.57-59.

148

Lee, J. C., Wortrnan, R. H., Hook, D. J. P., and Poppe, M. J., “Operational
Comparison of Leading and Lagging Left Turns”, Transportation Research Record
1421,1993, pp.l-lO.

Levine, J. and Underwood, 3., “Stakeholder Preferences in Urban Transportation
Evaluation: A Multiattribute Analysis of Goals for Intelligent Transportation System
Planning”, preprint Paper No. 950406 presented at the Conference of the
Transportation Research Board, Washington, D. C., January 1995.

Levinson, H. S., “Critical Movement Analysis for Shared Left Turn Lanes”,
Transportation Research Record 1194, 1988, pp.l67-178.

Lin, F. B., “Left-Turn Signal Phasing for Full-Actuated Signal Control”,
Transportation Research Record 1324, 1991, pp.53-58.

Lin, F. B, “Left-Tum Adjustment Factors for Saturation Flow Rates of Shared
Permissive Left-Turn Lanes”, Transportation Research Record 1365, 1992, pp.62-70.

Lycett, J .E., Maudsley, D., and Milner, D. L., “TREMEX: Transport Emergency
Expert System, Reliability on the Move: Safety and Reliability in Transportation”, New
York: Elsevier Applied Science, 1989, pp.l60-l80.

Machemehl, R. B., “An Evaluation of Left-Turn Analysis Procedures”, ITE Journal,
v.56, no.1], Nov. 1986, pp.37-41.

Mannering, F. L. and Kilareski, W. P., “The Common Structure of Geo-Based Data for
Roadway Information Systems”, ITE Journal, v.56, no.7, July 1986, pp.43-49.

Marcellus, Daniel H., Expert Systems Programming in Turbo Prolog, Prentice Hall,
1989.

McCoy, P. T. and Malone, M. 8., “Safety Effects of Left-Turn Lanes on Urban four-
Lane Roadways”, Transportation Research Record 1239, 1989, pp. 17-22.

McCoy, P. T. and Navarro, U., “Additional Lost Time Caused by Permitted Left
Turns”, Transportation Research Record 1112, 1987, pp.39-44.

McCoy, P. T., Ballard, J. L., Eitel, D. S., and Witt, W. E., “Two-Way Left-Turn Lane
Guidelines for Urban Four-Lane Roadways”, Transportation Research Record 1195,
1988, pp] 1-19.

McGraw, K.L., Knowledge Acquisition Principles and Guidelines, Prentice Hall,
1989.

Melhem, Hani G. and Wentworth, James A., “F ASTBRID: An Expert System for
Bridge Fatigue”, Public Roads, v.53, no.4, March 1990, pp.109-1 17.

Miller, C., Spyridakis, J ., and Haselkom, M., “A Development Tool for Advanced
Traveler Information Systems Screen Designs”, IVHS Review, summer 1994, pp.75-97.

Miller, G. A., “The Magical Number 7”, Psychological Review, vol. 63, No. 2, March,
1956.

149

Morris, M. and Hord, K. B., “Communications and Intelligent Transportation
Systems for Freeway Service Patrols”, preprint Paper No. 950428 presented at
Transportation Research Board, 74th Annual Meeting, January 22-28, 1995, Washington,
D. C.

Oppenlander, J. C. and Oppenlander J. E., “Storage Lengths for Left-Turn Lanes with
Separate Phase Control”, ITE Journal, v.64, no.1, January 1994, pp.22-26.

Oppenlander, J. C. and Bianchi, J. C., “Guidelines for Left-Turn Lanes”, ITE 1990
Compendium of Technical Papers, pp.l91-196.

Ome, D. E. and Humphrey, T. F ., “In Search of IVHS Benefits”, IVHS Review,
Winter/Spring 1994, pp.5-12.

Panchanathan, S. and Faghri, A., “A Knowledge-Based Geographic Information System
for Safety Analysis at Rail/Highway Grade Crossings”, preprint Paper No. 950717
presented at Transportation Research Board, 74th Annual Meeting, January 22-28, 1995,
Washington, D. C.

Petzold, R. G. and Freund, D. M., “Potential for Geographic Information Systems in

Transportation Planning and Highway Infrastructure Management”, Transportation
Research Record 1261, 1990, p.1-9.

Prassas, Elena S. and Roess, Roger R, “Left-Turn Adjustment for Permitted Turns fi'om
Shared Lane Groups: Another Look”, Transportation Research Record 1398, 1993,
pp.75-81.

Ramarnoorthy, C., Shekhar S., and Garg, V., “Software Development Support for AI
Programs”, Computer, Jan. 1987, pp. 30-40.

Ricci, R. C. and Gazda, W. E., “Transportation Modeling: an Essential Tool for
Evaluation of IVHS”, IVHS Review, Winter/Spring 1994, pp.43-52.

Roberts, Ralph, The Power of Turbo Prolog, TAB Books, Inc., 1987.

Robinson,J. B. L., “Highway Infrastructure, Heavy Trucks, and Expert Systems”,
Journal of Transportation Engineering, v.116, no. 6, November/December, 1990,
pp.780-788.

Rogers, J. B., A Turbo Prolog Primer, Addison-Wesbey Publishing Company, 1987.

Roschke, P. N., “Advisory System for Design of Highway Safety Structures”,
Journal of Transportation Engineering, v.117, no.4, July/August, 1991, pp.418-434.

Rouphail, N. M., “Analytical Warrant for Separate Left-Turn Phasing”,
Transportation Research Record 1069, 1986, pp.20—24.

Rouphail, N. M. and Khatib, Z. K., “Arterial Signal Optimization Considering Left-
Tum Control”, Transportation Research Record 1456, 1994, pp.72-82.

Sankar, S., Rakheja, S., and Aubin, S., “A First Step Towards an Expert System
Advisor for Stability Analysis of Heavy Vehicles”, paper prepared for
presentation at the 1991 TAC Annual Conference, Winnipeg, Manitoba, v.4, p. A63-A76.

150

Sayed, T., Navin, F., Abdelwahab, W., and Holowachuk, L., “A Knowledge-based
Expert System for Highway Safety”, paper prepared for presentation at the 1994
Intersectional Road Federation Conference, Calgary, Alberta, Canada, v.5, p. D49-D67.

Seneviratne, P. N., “Managing Traffic Safety Using a Knowledge-Based System”,
paper prepared for presentation at the 1989 RTAC Annual Conference, Galgary,
Alberta, Canada, v.5, p. C19-C37.

Shebeeb, 0., “Safety and Efficiency for Exclusive Left-Turn Lanes at Signalized
Intersections”, ITE Journal, v.65, no.7, July 1995, pp.52-59.

Siegfried, R. H., Vaidya, N., Avila, S. J., McCasland, W., and Levine, S., “Incident
Management Using Geographic Information Systems”, ITE 1992 Compendium of
Technical Papers, pp.433-436.

Simkowitz, H. J ., “Geographic lnfonnation Systems: An Important Technology for
Transportation Planning and Operations”. Office of Planning, FHWA, US. Department
of Transportation, Aug. 5, 1988.

Stokes, Robert W., Messer, Carroll J ., and Stover, Vergil G., “Saturation Flows of
Exclusive Double Left-Turn Lanes”, Transportation Research Record 1091, 1986,
pp.86-95.

Stonex,A. and Upchurch,J. E., “Conversion from Permissive to Exclusive/Permissive
Phasing: A Before-and-After Evaluation”, Transportation Research Record 1114,
1987, pp.63-73.

Teft, L., Programming in Turbo Prolog with an Introduction to Knowledge-based
Systems, Prentice Hall, 1989.

Upchurch, J. E., “Guidelines for Selecting Type of Left-Turn Phasing”,
Transportation Research Record 1069, 1986, pp.30-38.

Upchurch, J. E., “Comparison of Left-Turn Accident Rates for Different Types of Left-
Tum Phasing”, Transportation Research Record 1324, 1991, pp.33-40.

Upchurch, J. E., “Status of Intelligent Vehicle/Highway Systems Activities in Japan”,
[TE 1993 Compendium of Technical Papers, pp.55-59.

Upchurch, J. E. and Warne, T. R., “Reducing Delay to Left-Turning and Through
Vehicles: A Study of Three Signal Conditions”, Transportation Quarterly, v.46, no.1,
January 1992, pp.81-98.

Warren, D. L., “Accident Analysis of Left Turn Phasing”, Public Roads, v.48, no.4,
March 1985, pp.121-127.

Wright, C. R. and Upchurch, 1., “Before and After Comparison of Leading Exclusive and
Permissive/Exclusive Lagging Left Turn Phasing”, Transportation Research Record
1368, 1992, pp.3l-38.

Zhou, H. and Layton, R. D., “Development of Prototype Expert System for Roadside
Safety”, ITE Journal, v.117, no.4, July/Aug. 1991, pp.435-443.

APPENDICES

Appendix A

The Expert System Questionnaire (First Survey)

151

Dr. William C. Taylor

Transportation program

Department of Civil and Environmental Engineering
Michigan State University

June 28, 1995

01’ I' i! I'

The purpose of this questionnaire is to help us build an expert system for use
in determining the appropriate left-turn control strategies at an urban intersection.
In order to narrow the context that we have to consider, there are some assumptions
to be made in this questionnaire:

0 Assume good intersection geometry.

0 Assume there are no visual obstructions for drivers making a left turn.
0 Assume the number of opposing lanes that must be crossed are not a problem.
LeftiurnfiontmLStrategieslaBeSeiected

1. Permissive Left Turn Control

2. Protected Left Turn Control

3. No Left Turn

EactorsjaBLCnnsideLed;

1. Overall Level of Service at the Intersection

2. Left Turn Level of Service at the Intersection

3. Upstream and Downstream Intersection Level of Service

4. The Frequency of Left-Turn Accidents Per Year

152

5. The Left-Turn Volume (V EH/HR)
6. The Opposing Thru and Right-Tum Volume (VEH/HR)
7. The Average Speed in the Driver’s Direction (MPH)

8. The Average Speed in the Opposite Direction (MPH)

 

 

 

'-:_~ 7i r"'~t_-‘tit..‘_‘i:.\ I l‘l"'t-tel Mil
strategiesjhankmu.
I B . I I I.
Name:
Education:
Discipline
Degree

How long have you been working in a transportation-related job?

(years)

 

Employer Name:

 

Position:

 

Employer’s Address

 

153

 

City State Zip Code

Your Phone: ( )

 

We appreciate your time. Thank you.

Part II. Questionnaire

154

 

ndttlon and Case Number

CorBetter

 

DorE
ForWorse

CorBetter

 

DorE

 

F orWorse

 

No Left Turn

CorBetter

 

DorE

X X X

 

ForWorse

 

No Left Turn

CorBetter l

 

DorE

 

ForWorse

 

 

No Left Tum

 

 

 

 

 

 

 

 

 

m

6—10

 

11andabove

0-100

 

101 — 150

 

151-- 200

 

 

601 -- 800

—

801 and above

15— 19.9

201andabove X X X
I 0-600 I l l l l l l l I

 

 

 

 

 

 

 

 

 

I I I I I I I I

‘

 

20 - 29.9
30 and above

15— 19.9

 

20 - 29.9
30 and above

T

Permissive

 

Protected
No Left Turn

_JL
H
I

155

Part II. Questionnaire

 

 

 

 

Cchetter

DorE X
ForWorse X X X
NoLeftTum X X

 

 

 

CorBetter

DorE X
ForWorse X X X X X
NoLeflTum X X

 

 

 

CorBetter
DorE
Fchorse X X X
NoLeftTum X X

I 6-10 I I I I I I I I l

11andabove

 

 

 

 

 

 

 

 

 

 

 

 

 

o--1oo
101-.150 X X X X
151—zoo X

201andabove

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I N° L9" Tum I I I I I I I | I

156

Part II. Questionnaire

 

 

CorBetter

 

I WE IXI IXI I II

F or Worse X X X X

CorBetter
DorE I

ForWorse I X

NoLeftTum X X X X

 

 

 

CorBetter

DorE I X
ForWorse I X X X
NoLeftTum

 

 

 

CorBetter

DorE I X
ForWorse I X X X X
NoLeftTum

 

 

 

 

 

 

 

 

 

 

0—5

I 6‘10 l I I I I I

1iandabove

 

0—100
101—150 I X
151—200 I X
201andabove

 

 

 

 

 

 

 

 

0—600

601-80° L L I IXI I

801 and above X X X X

 

 

 

 

 

 

I 20—29.9 IfIXI IXI IXI
30andabove X X

 

   

 

 

 

 

I 20—29.9 IJXIXI IX X]
BOand above X
Permissive
Protected I
No LeftTurn I

 

 

 

 

 

 

 

 

Appendix B

The Expert System Questionnaire (Second Survey)

157

Dr. William C. Taylor

Transportation program

Department of Civil and Environmental Engineering
Michigan State University

East Lansing, MI 48824

Phone: (517) 353-7161

Fax: (517) 432-1827

January 26, 1996

01' I' l! I'

The purpose of this questionnaire is to help us build an expert system for use
in determining the appropriate left-turn control strategies at an urban intersection.
In order to narrow the context that we have to consider, there are some assumptions
to be made in this questionnaire:

0 Assume the cycle length at each direction is less than 80 seconds, and the left-
turn capacity won’t be affected by the cycle length

0 Assume good intersection geometry.

0 Assume there are no visual obstructions for drivers making a left turn.

0 Assume the number of opposing lanes that must be crossed are not a problem.

W

1. Permissive Left Turn Control

2. Protected Left Turn Control

3. No Left Turn

158

EafinmttLBeLQnsidered;

1. Overall Level of Service at the Intersection

2. Left Turn Level of Service at the Intersection

3. Upstream and Downstream Intersection Level of Service
4. The Frequency of Left-Turn Accidents Per Year

5. The Left-Turn Volume (VEH/I-IR)

6. The Opposing Thru and Right-Turn Volume (VEH/HR)
7. The Average Speed in the Driver’s Direction (MPH)

8. The Average Speed in the Opposite Direction (MPH)

 

 

 

' l, lm‘ir'Iliii‘ ‘110\ i 1'1a"t-rut tilt
strategieslhankm
E II B . I E I.
Name:
Education:
Discipline
Degree

How long have you been working in a transportation-related job?

(years)

159

Employer Name:

 

Position:

 

Employer’s Address

 

 

City State Zip Code

Your Phone: ( )

 

 

We appreciate your time. Thank you.

Part II. Questionnaire

160

 

ndition and Case Number

CorBetter

 

DorE
ForWorse

CorBetter

 

DorE

 

F orWorse

 

No Left Turn

CorBetter

 

DorE

 

ForWorse

 

No Left Turn

CorBetter

 

DorE

 

F orWorse

 

No Left Turn

 

0—5

X

 

 

 

 

 

 

 

 

 

 

I 6—10

I I I I I I I l

I 11andabove I I
0-100

 

101--150

 

151 -- 200

 

201 and above

 

—
a)
O
_a

 

 

 

 

 

 

 

 

 

 

.—
N
o

i
to
o
o

 

—
M
O

l
N
O
D

 

No Left Tum

i—I—

161

Part II. Questionnaire

 

 

 

 

CorBetter

DorE X X X X
ForWorse
NoLeflTum X

 

 

 

CorBetter
DorE
ForWOtse X X X X X
NoLeflTum X

 

 

 

CorBetter

DotE X
ForWorse
NoLeftTum X X

 

 

 

 

 

 

 

 

 

 

 

 

 

On!)

I °--‘° I I I I I I

11andabove

 

0--100
101—150
151 —- 200

201mdabove X X

 

 

 

 

 

 

 

 

 

 

 

 

 

0-600

 

I 601—3“ I x I I I I I I I?

801 and above

 

 

I 20-29-9 I x I I I I I I I 1

 

 

I 20-29-9 I I I I I I f I 1

 

 

L
I

No Left Turn J ] | T

162

Part II. Questionnaire
ondltlon and Case Number

 

 

 

 

ForWorse X

 

 

F or Worse
No Left Turn X X X

 

CorBetter

DorE I
ForWorse I X
NoLeflTum X X

 

 

 

C or Better
0 or E I
F or Worse I X X X
No Lefl Turn X

 

 

 

 

 

 

 

 

0—5

I 6'10 I I I I

11andabove

 

 

0—100
101—150
151—200

201 and above X X

 

 

 

 

 

 

 

 

 

0—600

 

I 501-5” I I I I I I

801 and above

 

15—19.9

I 20—29.9 | j x I I I X‘l

30 and above

 

 

 

 

Permissive

 

Protected I
No Left Turn |

 

 

 

 

 

 

 

 

"IIIIILIIIIIIIIIIIIIIIIII