Abnormal Behavior Identification Of Vehicle Trajectory Under Video Sequence

Traffic management system is overall planning system which integrally manages all the features in traffic system.These features include personnel, vehicle, road condition and environment. It will cause congestion and accident once if these features become abnormal. And the earlier we access to information of traffic and then submit traffic management system for further processing, the more powerfully we could control the accident and congestion rate.Traffic monitoring system is an important means to dynamically access traffic information. As the core technique of intelligent traffic monitoring system, vehicle abnormal behavior analysis base on monitoring video has its significance and important application value in the field of effective, safe and sensitive traffic operation. In this paper, vehicle trajectory is thought of as the carrier of vehicle behavior information and the a new recognizable method is regarded as the the purpose of the study of detecting, tracking and recognizing moving vehicle abnormal behavior. A new recognizable method based on Adaptive Single Class Support Vector Machine is proposed to solve the problem about being lack of a large number of negative sample in the traffic monitoring video and samples need fast training. The vehicle’s track is mapped to high dimensional space. First and foremost, utilizing SVM of adaptive single class to learn the normal tack and establishing the scene supported vector. Subsequently, the discrimination function supported by vector is applied to detect the scene, identifying the vehicle’s abnormal behavior.What work we have done in this paper as follows:(1)In order to satisfy precision request in the real traffic scenes, Mixture Gauss model has been used for background modeling and detecting moving vehicle in the monitored scene.(2)The MeanShift algorithm is used to track the detected vehicle’s trajectory.And these trajectory will be regarded as the basic information of recognizing moving vehicle abnormal behavior.(3)Improved Hausdorff distance and comparison confidence are used to measure of weighted similarity between trajectories, then, in order to obtain the vehicle motion in the scene, Spectral Clustering is applied to classify with similarity. (4)Single Class Support Vector Machine is applied to recognize moving vehicle abnormal behavior, meanwhile, adaptive parameters are added to Vector Machine for implementing real time update and then meet the demand of long-term monitoring.