Study On Testing System Of Vehicle-body Dimension Based On Computer Vision

Accompanying with the development of the social economy,and with the quick increasing of the vehicle population , vehicle's oversize and overloading transportation has brought serious influence on state property and people's security ,endangered the social and economic order , which has become an outstanding social problem. People, enjoying the huge profits and traffic convenience, are more and more be subjected to the traffic jam and accident, pollution of the environment, the seriously damaged road and so on. These all result in tremendous economic loss. The state compulsory criterion, Used Vehicle's Body Dimension, Axle Load and Quantity Limitative Value, which is to solve the problem of vehicle's oversize and overload, is carried out formally on October 1 in 2004. It requests the automobile producers produce motorcars according to the criterion to put an end to the phenomenon of vehicles'carrying capacity over its marked load capacity. The law of road traffic safety and its implementing regulations make the strict limit value of load and forbid oversize. The length, width and height of the goods are required to abide by the loading rules. The behaviors of goods'spilling and scattering are not permitted. According to the The requests and examining methods for comprehensive performance of working vehicle (GB18565-2001), Vehicle-dimension parameters are required to be detected and the structure of the vehicle can not be refitted discretionarily .For identifying illegal oversize vehicle availably and perfecting detection technique of auto performance, improving the monitoring function of the highway traffic system and the administrant level of the vehicle, one feasible and effective technique is to be taken into account as a complement for the intelligent traffic system and the comprehensive performance detection technique of vehicle in this paper. In terms of above, the non-touch quick detection system of vehicle body dimension is studied in this paper in order to realize vehicle-type recognizing availably and body dimension measuring accurately, which is served as a effective means to monitor vehicle availably and solve oversize problem synthetically.Computer vision detecting system based on special sensors model is designed and developed in terms of the demand and the application background of vehicle-body large scale dimension. This paper puts forward a method combining the binocular stereo vision technique and the monocular vision technique, which can realize the identifying of the vehicle type effectively and the digital measure of the vehicle-body dimension. The techniques, such as feature extraction and analysis, type recognition, camera accurate calibration, stereo image matching and 3D reconstruction, are studied and discussed thoroughly. Some important questions faced in the process of algorithm implement are analyzed and some creationary research productions are also achieved. In this paper, many new sciences and techniques are mentioned and the main research work and its research production are as following:1. The design of the hardware and the software system . Computer vision detection theory is described in detail. The hardware and the software system project of vehicle body dimension measuring system based on computer vision is designed reasonably according to the design aim of vision detection. In accordance of production fact, the sensor model which combines the binocular stereo vision and the monocular vision is selected to design structure parameter for configurable measure. On the platform of PC, the hardware system is designed using multi-sensor data acquisition and other accessorial hardware. Measuring and controlling software module classifying and program flow are obtained with the combination of Computer vision theory and the actual detecting task.2. The research and improvement of image processing algorithm in the vision detection. The means of image processing for detecting and recognizing by the vision detecting techniques is discussed detailedly about Image segmentation, edge detection, feature extraction and Image Mosaic Algorithm. More image colorized information is used in this paper. It also puts forwards improved algorithm for accomplishing exacter edge feature extraction and corners. It offers a way for accomplishing to distill lines and circles based on improved Hough transform. It is proved effective and fast and robust by experiments. The Image Mosaic technique based on image matching is put forward with perfect image mosaic to solve the problem about small field of camera.3. Classifier based on image geometry character and classifier based on invariant moments are designed to recognize vehicle type automatically quickly and exactly. Taking advantage of monocular vision to get images, using background disparity to eliminate many interferential elements, the vehicle body is extracted accurately out of a complicated surrounding. The vehicle recognition system based on geometry character is designed on which a classifier based on invariant moments. Invariant moments and invariant moment vector, which keep invariant when the image is in different position, orientation and scale, were used to realize recognition of vehicle type as the main extracted feature. The experiment indicates that the system achieves automatic, fast and accurate recognition of vehicle type.4. The camera calibration method is summarized and the camera nonlinear imaging model and its calibrating parameter are mainly introduced. Putting forward high accurate sensor calibrating method based on binocular stereo vision sensor 3D measuring model. The chessboard is chosen to be the plane calibrating template mark. Intrinsic parameters, extrinsic parameters and sensor structural parameter are calibrated accurately using improved corners detection and Levenberg-Marquqrdt nonlinear iterative algorithm.5. Stereo matching technique is studied, and the optimal correspondence is achieved by combining feature matching with field matching, with multi-constraint and making use of various excellences of different methods. This algorithm mainly includes several algorithm steps like Feature points detection, image segmentation around feature point, invariant value computing and feature vector comparison and so on. A novel robust algorithm is designed to estimate the foundational matrix using random sampling algorithm. Furthermore, correspondence points matrix is reduced using epipolar geometry constraint, combination of feature matching and field matching, disparity constraint, adjacent domain relative disparity constraint, continuity constraint and exclusive constraint. The result indicates the stereo matching technology is a excellent and robust method with higher matching rate and less error matching points.6. On the basis of the other researchers'3D reconstruction algorithm , putting forward a method to increasing 3D reconstruction accuracy. For the vision detecting system, vehicle experiment is executed. It is indicated that measure system can accomplish the exact recognizing of the vehicle type and the fast and non-touched measuring of the vehicle-body dimension, and has higher 3D measure precision.The vehicle body dimension measuring system is a typical example which combining the computer vision detecting technology with the practical application about the research and the development work. Some novel research achievement is gained about algorithm and technique which has certain theory and application significance.