Multi-Vehicle Tracking And Traffic Incident Recognition In Intersection Area

With the sustainable and rapid development of economy in our country, the number of vehicles in city is increasing. Traffic congestion and safety have become more and more serious. Video based traffic incident detection is an important part of ITS (intelligent transportation system). Due to high complexity of the traffic condition in the urban intersection, it leads to the difficulty for video based traffic incident detection.Aiming to study the tracking problem of multiple moving vehicles in intersection under vehicle occlusion, the main work of this dissertation is as follows:(1) The performances of several main target detection methods in the intersection area are addressed. And the background difference method is used to detect moving vehicles. The advantages and disadvantages of several kinds of background modeling methods are analyzed. A background modeling method is presented to deal with the ghosting problem of VIBE(Visual Background extractor) effectively. This method can model the background quickly.(2) In order to eliminate pedestrian disturbance, aspect ratio is used to classify moving targets in intersection. A moving vehicle tracking method, which combining Kalman filtering and improved SURF(Speed-up robust features) algorithm, is proposed to handle vehicle occlusion problem. Based on the center position of vehicle, this method tracks moving vehicle and predicts vehicle occlusion in intersection using Kalman filter. An improved SURF algorithm is introduced to realize feature matching for occlusion region. Thus, the occlude vehicle can be tracked continuously.(3) The traffic parameters of moving vehicle are extracted to detect traffic incidences is such as vehicle converse running illegally, vehicle parking illegally, vehicle collision and road congestion.