The Automatic Tracking And Counting Of Moving Target In Video Sequences

The automatic tracking and counting of moving objects in video sequences is one of the most important topics in computer vision field. It mainly uses image processing method to detect moving objects of interest in the video, find the exact map location in current image, feedback it to the tracking system and count the targets passing through some specific area. It is wildly used for intelligence, flexibility, and simple equipment, etc.The automatic tracking and counting of moving objects in video sequences is mainly composed of targets detection, tracking and counting. Targets detection is the foundation, which affect the accuracy of following processing. In moving target tracking, the objects must not be lost when it is covered. While the purpose of targets counting is mainly to statistic the number of some passing objects, which must be accurate and real-time.In this article, we described several classical algorithms of detection and tracking. In moving objects detection, we studied three classical methods, such as optical flow, frame difference and background difference. According to the simulation results of frame difference and background difference method, we analyzed their talent and disadvantages. In the part of target tracking, we introduced the relevant algorithm and Snake algorithm. Simulation results of Snake algorithm and improved GVF-Snake one proved that the traditional Snake algorithm is sensitive to the selection of initial points, which can not well convergent in the depressions, while the GVF-Snake algorithm can overcome the insufficiency of traditional one. However, although the GVF-Snake method has high accuracy of target tracking, its time consumes is large for Iteration.In view of that, we proposed a method by combining the largest connected region with the Kalman filter in this article. The largest connected region method can improve the accuracy of target detection for removing a large area of noise in the image. With Kalman filter prediction algorithm, we are able to predict the target location when the object is covered. Finally the creating dynamic window-based method is proposed to complete the objects counting of traffic flow. In this article, we realized the target tracking and counting for several actual video series, and analyzed the algorithm's effect. The experimental results indicate that the method in this article can automatically track and count moving objects accurately, and satisfy the requirement of real-time, so it is validity.