Research On Multi-target Detecting And Tracking Technology In The Video Surveillance System

In the intelligent surveillance system, the moving multi-target detecting and tracking in videos are key technologies, and they are widely applied for both military and civilian. They are also basic technologies to realize video scene analyzing and behavior understanding.The goal of moving target detection is to abstract the targets we need from the scene. If the background is simple, we can detect targets quickly. But based on the assumption of an unstable background, it is hard to realize effective object detection. In the second chapter, an improved three-frame-difference method is proposed. Combined with morphological filter and connected component detecting, this method can effectively realize multi-target detecting. Mean-Shift is a good algorithm in computer visual target tracking area. The algorithm has many advantages, such as good real-time, robust for occlusion and target distortion, so researchers pay more and more attention to this, and it is widely used in various tracking systems. However, there's no template update module of Mean-Shift. Traditional tracking algorithm is used by getting some fixed features, and it might lead to the failure of tracking. In the third chapter, a new method based on the similarity evaluation and Kalman filter is proposed to realize template update in real time, and it proves to be robust. In addition, combined with Kalman filter, traditional algorithm can realize target tracking in case of target occluding.In the forth chapter, another important algorithm used in target tracking is particle filter. Because multi-target tracking is very difficult, particle filter algorithm based on nonuniform sampling and kernel-histogram of HSV space is proposed. New algorithm makes good use of old value of particle filter to realize particle resampling, so the complexity of algorithm isn't added, and the particles are more close to the real state. Experiment result shows that new algorithm can track target effectively when target is occluded.In the fifth chapter, target tracking technology is used in the precision location and strike system. Based on exact position of the ground martial target acquired from aerial image, we can realize precise strike against martial target.At the end of the paper, we summarize all the research work and give the future work..