Highway Vehicle Speed Detection And Tracking System Based On Video

With the development of intelligent transportation system, the application of video surveillance in traffic management is increased. As the frequency of highway accidents becomes greater, the traditional artificial detection methods cannot meet the actual demand. Through video analysis to obtain the traffic information on road surveillance cameras is the future aim. Vehicle speed is an important part of traffic information, the focus of current research is extracting the vehicle speed accurately and effectively.This thesis firstly introduces the current classic speed detection methods based on video, and analyzes the advantages and disadvantages of these vehicle speed detection methods. For the requirement of highway vehicle speed detection system, proposes a video-based corner feature matching vehicle speed detection method for vehicle speed abnormal detection. Based on highway surveillance video, this method uses Gaussian mixture model for detecting motion objects, then detects the harris corner. Improves the traditional normalized cross-correlation matching method through combining motion estimation and NCC template matching, optimizes matching area search strategy to achieve corner coarse matching, and uses random sample consensus to achieve corner fine matching, finally achieves multi-vehicle speed measurement by single view metrology coordinate. Experimental results show that by improving the NCC template matching method, corner coarse matching speed increases four hundred percent, corner fine matching speed increases two hundred percent, the vehicle speed accuracy reaches more than ninety percent. Compares with the old corner matching vehicle speed detection method, this method successfully improves the real-time of the algorithm and the accuracy of speed detection, can satisfy the system requirements of the actual highway vehicle speed abnormal detection and illegal parking.This thesis introduces the typical tracking algorithms and analyzes the Mean Shift algorithm, then uses the Mean Shift method based on color and edge feature for vehicle tracking, this method can successfully achieve highway vehicle tracking. For the requirement of cross-camera tracking, this thesis proposes a moving target tracking method based on Mean shift and SIFT matching. Experimental results show that, this method can achieve cross-camera vehicle tracking under certain conditions.