Research On Key Techniques Of Motor Vehicle Retrieval In Traffic Video

With the rapid development of economy and the expanding size of city, the number of motor vehicle has a large growth, and the demand for city transportation is increasing dramatically, but the road construction and people's awareness of traffic laws is dropping behind, so a variety of traffic violations and accidents frequently occur, which raises higher requirement to the traffic control department. The retrieval system of motor vehicle in traffic video analyze the traffic video, detect and track motor vehicles, extract several features and save them in database for users to retrieve. It not only provides the first hand data for traffic planning, but also provide considerable convenience for traffic control department and police to trace and investigate the accident vehicles.By doing research on four key techniques in retrieval system of motor vehicles, which are motion detection, moving vehicle tracking, feature extraction and classification, and key frame extraction of traffic video, we propose some new algorithms and application methods. Our research issues are follows:(1) Propose a background construction method based on adaptive on-line clustering, and combine it with background subtraction to do motion detection.(2) Propose vehicle tracking by combing motion detection with CAMShift to make up the defect of CAMShift when it is used in traffic video singly. And change the color model used in CAMShift to reserve the interface for the succeeding color extraction.(3) Propose the quantization and classification method of vehicle color, and do research on classifying vehicles with geometric features gained from front side of vehicles.(4) Propose applying the virtual detection line to the key frame extraction of traffic video, overcoming the problem that traditional methods are unfit for traffic video.This thesis performs experiments on the methods mentioned. Experiments show these methods are effective.