The Detection And Tracking Of Moving Target In Video Surveillance

Today,the security in important areas has depended on video surveillance system increasingly.Its requirement becomes higher and higher. The traditional video surveillance is no longer applicable to the needs of the reality because of inseparable from the artificial.In the midst of the real scene, intelligent video surveillance system shows the inevitable trend of development. The system must own the function as follows:on real-time detection, tracking, analysis, determine and early warning. Focused on the complex and changing environment, the thesis researchs on the moving targets'detection and tracking in case of static camera.In the part of detection, using GMM into the inter-frame difference, the moving regions detected by the inter-frame difference are divided into the revealed background area and the movement existence region. Each of them is taken to different backgrouds update policy. It has solved the problem that classical GMM fails to treat the stationary object suddenly moving and the mass movement objects. It sets shadow elimination module.Firstly, we use highlights assessment criterion to detect the presence of shadows.Then, based on the minds of using color and gradient judgements step by step, the position of the shadow will be detected and eliminated.The result shows that the detection system can be applied to different change of the speed of moving objects.Different types of shade have been eliminated well in the indoor and outdoor scenes.In the part of tracking, combining the improved Mean-Shift algorithm and Kalman filter prediction. We use the adaptive tracking algorithm.Based on the positive and negative three incremental scale executing 3 times, we focuse on solving the problem how to choose the tracking box size in the case of the goal becoming bigger and changing largely, according the change to choose the size threshold. For the case of severe block,similarity coefficient is used to judge the degree of block. And Kalman filter will forecast trajectory areas that targets may appear, the Mean-Shift algorithm is searching and waiting for the target int these region.The result shows that the tracking system can meet the goal of the case of object changing fast. And it can be accurately tracked real-time in the case of serious block, inproves the system's ability to adapt the complex environment. It will be a certain degree of practical value.