Research On Vehicle Dimensions Measuring System Based On Machine Vision

Vehicle dimensions and passing capacity geometrical parameters are the obligatory items of military vehicle test and the important contents of safety check for vehicles operating on roads as specified by both the national standards and national military standards for automobiles. But the conventional measuring methods that used by the automobile test institutions now are inefficient, so it is significant to develop a more automative and advanced vehicle dimensions measuring system. After comparing with types of the large-scale coordinate measuring systems, two design proposals based on machine vision are proposed: camera array system and large-scale vision coordinate measuring machine, both characterized by automation, non-contact, high accuracy, flexibility and expansibility on functions. The design proposals are innovational in the domain of vehicle dimensions and passing capacity geometrical parameters measurement. The camera array system, as preferred, has passed the examination and comment, and is coming into practical construction.The application of machine vision to dimension measuring methods is researched deeply. After analyzing the theories of machine vision techniques, the researches put emphasis on the practical technical problems, and the construction and measuring methods of the vehicle dimensions measuring system are expatiated. All the involved algorithms are tested repeatedly to validate the feasibility. The major innovational work is as follows:1. The camera array system based on stereo vision is proposed. In this system, several digital cameras are set around the vehicle; two of the cameras qua basic measuring unit reconstruct the 3D points using binocular stereo vision theory; then the dimensions of vehicle can be calculated. The system components, construction, measuring principle and method are described in detail, some problems in constructing the system such as software designing, camera setting, cameras distribution and fixing are presented.2. The approaches of stereo vision technique are researched and realized. The algorithms of feature detection used in the measuring system are ameliorated based on practicality, by means of which the performance and effectiveness of the algorithms are improved. A high accuracy corner detecting method for the image of calibration model plane is proposed. A cross correlation criterion of images based on gray scale and gradient is proposed, and a stereo matching algorithm is designed according to several constraints. Good results have been obtained in testing the algorithm.3. The large-scale vision coordinate measuring machine based on visual servoing is proposed. In this system, a video camera is set on two rails perpendicular to each other, so the video camera can move in 2D; the feature point can be detected and aimed at automatically, and the coordinates can be acquired from the magnetic grid rulers on the rails. The system components, construction, measuring principle and method are described, the systemcalibration method is presented, and the measuring accuracy is analyzed.4. The auto-focusing algorithm for motorized lens is researched. Some representatives of image definition criterions are analyzed at first, and the limitations of those criterions are indicated. After analyzing the gradient histograms of variational-definition images, a criterion based on gradient transform is proposed, which is more robust and has better performance on image definition evaluating. Aiming at the characteristic of motorized lens, a new focusing control algorithm is presented, which combines mountain-climbing-servo and global searching method, and avoids the limitations of conventional methods. The auto-focusing system has been successfully tested and shows such characteristics as good reliability, high sensitivity and high speed.5. Automobile headlamp test method is introduced, including the detection of far and low beam direction and the detection of luminous intensity. The test method of luminous intensity distribution performance of headlamp is also analyzed.In addition, the following work is achieved:1. The technique for camera calibration is researched systemically, the binocular vision calibration method and the global calibration method for whole camera array are depicted in detail. In this part, the approaches of camera calibration are described amply; some practical problems and calibration results are analyzed; image undistortion method is introduced; the process of system calibration and the method of dimensions calculation are presented.2. Sorts of the experiments for binocular stereo vision are analyzed detailedly, including the accuracy comparison of 6 kinds of 3D point reconstruction algorithms, the influences on accuracy by moving the cameras or calibration model plane, measuring vernier caliper, measuring with 4 cameras, 3D surface recovering and stereo visualization. Substantive experiment data are listed.The experiments for the camera array system show that the dimension measuring error by using two cameras is less than 1.0mm in the range of 2m from the cameras, and the error of measuring large dimension by using 4 cameras is less than 2.0mm. The measuring accuracy analysis for the large-scale vision coordinate measuring machine also shows that the error is less than 2.0mm. Therefore, the two systems can both meet the target of plan.