The development of large scale structured light based dimensional measurement systems is introduced in this study. Based on different applications, there are generally two different research directions for a structured light system: accuracy 3D inspection and fast environmental reconstruction. The first one is emphasized on the accuracy of the results and the second one is mainly concerned about reconstruction speed. Both of them become challenge tasks when the systems' scales increase.Quality inspection is a process to evaluate the quality of a manufactured work piece's 3D shape. Compared with traditional Coordinate Measurement Machine (CMM), structured light based optic measurement has the merits of fast speed, high inspection sample rate, and overall simplicity of operation, and becomes a popular alternative inspection method for CMM. The measurement result, point cloud of the test work piece, is compared with the Computer Aided Design (CAD) model to find the error distribution map. The color-code error distribution map will be further evaluated with the acceptable tolerance. There are basically three main issues, when the system scale is enlarged.Calibration of such a large system with long standoff distance, large Field Of View (FOV), deep Depth Of View (DOV), and multi-sensor became a challenge task. The calibration errors are enlarged when a large scale system is treated. In order to maintain high accuracy, an innovative 3D sensor model with less calibration parameters was developed. Instead of employing the incident light in projector frame, 3D point recovery was achieved within the camera frame via plane induced parallax information, in which projector's intrinsic and orientation are avoided in the 3D model. Precision of the large scale optic system was also simulated and was tested against the random image noise in the system. Multiply-plane strategy was developed and implemented to calibrate the sensor.As the system scale increased, more work pieces could be inspected at the same time. The optical properties became more complicated. Same with all the other vision based measurement system, structured light systems is usually weak against surface optical properties. Material exhibiting different color textures, reflection ratios, and especially a mixture between specula reflection and diffuse reflection, fails the optic sensor to correctly acquire incident light information. Traditional structure light method is only valid for parts with diffuse reflection property. Therefore pre-treatment of the surface have to be added before inspection. Aiming on this problem, a structured sensor with a robust surface decoding method for industry application was developed. The coding strategies were properly designed against variously test surface optic properties. Monochromatic light was utilized against different object color textures. The illumination of the projector was adjusted pixel by pixel based on the optic properties of the test material to composite different reflection coefficients and internal reflection. Furthermore, an extrapolation model to solve the internal reflection problem and a sub-pixel interpolation model to increase measurement accuracy were proposed too. The proposed system was capable to inspect various materials with different shapes and different optic properties, from black, dark, to shiny.Registration among sensor frame to the common was achieved based on Iterative Closest Point (ICP) method. The final point cloud joined by each individual point cloud could represent the 3D shape of a large work piece. Object-oriented on-line calibration was developed based on ICP and Geometry Dimension and Tolerance (GDT) information to register the final point cloud and the CAD model. Several modifications to traditional ICP are applied to speed up the registration process due to the large amount of points in both sets.Environmental reconstruction for navigation is a process to quickly acquire surrounding information around the vision sensor and presents a 3D fusion display for the operator. Traditional navigation system only employed a camera to view the environment, in which the depth information is lost. Operator is often confused with the object distance, the distance between the object and the camera. Structured light system based on infrared light could quickly rebuild objects depthes and fuse it into the displayed images without influence the operator's normal vision. The sensed points on the objects are highlighted by the color-codes: from red to blue, which are used to indicate the object distances.Fast environmental reconstruction emphasizes the acquisition time. In order against the moving objects, an one-shot surface coding algorithm was developed. Only one projection image is needed to acquire the 3D information. The codeword is determined in a single pattern because the code of each primitive depends on the values of the primitive and its neighbors. Compared with the previous patterns, this pattern is more robust because it can avoid the influence of the ambient light and the inspected part reflective property. Moreover, the requirement of the accuracy performance is achieved by the pattern primitive which is similar to the corner of the checker board since it can provide high accuracy performance even when the occlusion occurs.In order to reconstruct the environment without blind areas, an omni-direction panoramic structured light sensing system was developed to increase the system field of reconstruction. Hyperbolic mirrors are put in front of a projector and a camera. 3D reconstruction model was build up associated with the hyperbolic mirror. Task level calibration is conduct for the system. At last, a 360 image fused with depth information is achieved by the designed system.In summary, the study developed large scale structured light systems based on two different applications: accurate inspection for industry quality control and fast environmental reconstruction for mobile robot navigation.
Read
