Research On Vision-Based Traffic Vehicle Detection And Tracking

Video-based vehicle detection and tracking is one of the key technologies in intelligent transportationsystem, and it is the foundation of traffic flow statistics, behavior analysis of dynamic target intraffic monitoring system. This paper mainly studies vision-based target detection and target trackingtechnologies in traffic monitoring video, and has a deep research and discussion on vehicleextraction, particle filter, multi-object relevance and occlusion detection, and on the basis of itsimulate a whole system to realize multiple dynamic object detection and tracking. This paperprincipally included four portion:1. In moving vehicle extraction part, under the background difference framework,use the method ofCodeBook to get the background modeling and update the model. Remove the shadow in binaryobject by using the method based on HSV color space. To deal the binary image obtained byshadow removal with the post processing morphology. Then do region grow and new object reasonabledetection to check whether the measure is a target or not. Initialize the target which is pass thereasonable detection.2. In particle filter part, study the single object tracking under the framework of particle filter, inorder to solve the problem of particle degradation, apply the particle resampling algorithms,contrast the commonly used four resampling methods, and further discuss the template updating ofthe target. Finally compare this four resampling method in different angle and realize a real-timesingle object tracking.3. In the multi-object association part, introduce the common multi-object correlation algorithm,then research the combination of particle filter and the Probability Data Association to realize multiobjecttracking. The associated probability makes up the incidence matrix, and is calculated basedon the overlap area of the target. Through a variety of judgements, distinguish object state into fourcategories, under the condition of no occlusion, realize multi-object tracking.4. In occlusion detection part, introduce the common occlusion examination method, and have adeep research on how to use particle filter to detect object in occlusion. According to the differentoccurrence time of the occlusion, apply different scheme to resolve the occlusion happened beforeobject initialized and after object initialized correctly. After a deep discussion on occlusion detec tion and removal, realize the occlusion removal method. Apply the method to actual traffic scenevideo, the experiment results show that the occlusion removal method based on particle filter canbasically resolve the occlusion in object tracking.