How to detect vehicle violation effectively has been the focus of the traffic management department. Conventional vehicle violation detection methods are basically based on the principle of electromagnetic induction to work and have the shortcomings of low reliability, inconvenient installation and maintenance. With the development of computer vision and digital image processing technology and the appearance of the concept of Intelligent Transportation System, researchers have proposed a vehicle violation detection method based on video analysis which generally consists of three main modules, that is vehicle detection, vehicle tracking and behavior understanding, vehicle violation can be detected automatically and compared with conventional methods, this method has many merits. So this paper makes some research on object detection and tracking algorithm and new vehicle detection algorithm and vehicle tracking algorithm have been proposed, and based on the vehicle detection and tracking algorithm, a vehicle violation detection algorithm has been realized. The main work and innovation are as followed. (1) Introduce the existing object detection algorithm based on static images and video sequences. A two-level vehicle detection algorithm has been proposed. Firstly, a globally search has been made which uses the edge template trained before to give the approximate location of the vehicle. Secondly, with the pre-trained vehicle classifier model, a second detection on the coarse locations given by the first step has been made to achieve an accurate and stable vehicle detection results. (2) Introduce the existing object tracking algorithm. A vehicle tracking algorithm based on support vector machine has been proposed. Firstly, vehicle's candidate region in this frame is identified based on vehicle's position in the last frame. Secondly, each position within the candidate region is validated using the support vector machine classifier model trained in vehicle detection part to get the best position of the vehicle. Background model of the traffic video scene is achieved simultaneously and vehicle’s best position is corrected by using the result of motion foreground detection to get the final tracking result which is robust and can be achieved quickly. (3) Vehicle violation detection. Three basic modules of vehicle violation detection algorithm are vehicle detection, vehicle tracking and violation behavior identification. A judgment that whether lane violation...
修改评论