Research On Key Technologies Of Auto-calibration And Distance Measurement For Speed Detection By Video In Highway

Vehicle speed detection is an important method for modern traffic management. The traditional vehicle speed detection tools need some special speed detection equipments, which are expensive and difficult for maintenance. Currently, as a basic surveillance method, the surveillance systems based on video image processing are installed widely on the highway. As the video detection technology is convenient for installation and has lots of information, it is the future direction in the area of vehicle speed detection. However, the existing detection methods can hardly meet the demand of the surveillance in highway as their poor detection accuracy and poor enablement. So it has important theoretical and practical role to investigate the video-based vehicle speed detection technology.This thesis researched the speed detection method for highway. We first analyzed the problems and difficulties in vehicle speed detection, and investigated the distance measurement technology and automatic calibration of ranging technology based on the lane line information of the image distance. The distance measurement technology is divided into two parts which are vehicle target detection and vehicle angular point feature extracting and matching. Finally, a novel of highway speed detection method based on video is established.In the aspect of distance automatic calibration, the median filter method is selected to deal with the noise. Then the Canny edge detection and Hough transform method are used to extract the lane edge line. An algorithm for extract the lane line is proposed which is based on a kind of self-tuning threshold binarization algorithm. At last, the distance image automatic calibration is realized by using lane area automatic extraction and lane line information. Experiments show that the image distance calibration method can transform the displacement with low error than 6.5%, and it build a good foundation for subsequent displacement conversion.In the aspect of objective extraction, the background of nonparametric kernel density estimation model was established firstly. Then the background update scheme is given by the characteristics of freeway with light mutation and gradient scene. On this basis, the vehicle target is well extracted by the method of the combining with morphology denoising and connected domain analysis method. In addition, a lane information and vehicle external rectangular body region extraction method is proposed for overcoming the vehicle division and integration problems. The experimental results show that the method can extract vehicle targets more preferably.In the aspect of features of vehicle detection and matching, a fusion USAN distribution information of corner detection algorithm is proposed for overcoming the leakiness part angular point in SUSAN algorithm. Then a new coarse matching algorithm is proposed with the combination of vehicle motion vector information corner. The relaxation iteration algorithm is selected to complete the essence of angular point matching vehicle range. Experiments show that the detection and matching algorithm proposed in this thesis can increase the angular point matching success rate and reduce the operation cost in the same time.Finally, the speed detection algorithm proposed in this thesis is verified with experiment by using the Chongqing highway video data from the field under the VC environment. Results show that the speed detection accuracy error can be controlled within 10%, and it demonstrated that the method has high accuracy and feasibility.