Research On Abnormal Vehicle Behavior Detection And Application In Traffic Video

With the rapid economic development, the significant position of Intelligent Transportation System (ITS) in human economic and social activities is increasing. As the core function of ITS, the abnormal vehicle behavior detection algorithm plays an important role in people's daily life, social progress and economic development.This thesis tries to get insights on some key issues of abnormal vehicle behavior in traffic video, including vehicle detection, vehicle tracking, trajectory pattern learning, and abnormal vehicle behavior detection. It proposes some new algorithms and application methods. The research issues are as follows:(1) Aiming at the vehicle detection under complex traffic scene, adaptive Gaussian mixture model was used to extract the moving object. Based on the above work, it proposes the multi-structure multi-scale morphological operators to obtain the complete vehicle edge. The well done of these works promoted the credibility of the following works.(2) To solve the time-comsuming problem that the big time cost problem in traditional optical flow tracking algorithm, this thesis take advantage of the L_K optical flow tracking method based on image pyramid, it can minimize the possibility that the moving targets in video don't meet the movement hypothesis, so as to achieve the rapid and accurate tracking. Then it pre-process the trajectory obtained by tracking, laid the foundation for later trajectory model learning.(3) To solve the problem that behavior pattern recognition only depend on the spatial characteristics, this thesis proposes a trajectory spatial model learning methods based on spectral clustering; using GMM model for the trajectory direction learning.(4) The paper proposes an abnormal vehicle behavior detection method based on the normalized Bhattacharyya distance and the origin-destination direction pattern matching, the methods are used on the detection of abnormal location and direction in traffic video.This thesis performs experiments on the methods mentioned. Experiments show that these methods are effective.