A Method of Target Recognition for Visual Surveillance

In this paper a method of real-time target recognition is proposed. When the target is moving, the features of the target are changed. So the features for BP neural network training were obtained according to the moving direction and the target´s position in the video Scene. And the recognition results of the previous frames were also considered to get the result of the current frame. The experiments showed that the probability of the correct decision was increased. In order to increase the efficiency of target detection, Camshift Algorithm was used to track the target. Then the seeking range was reduced, the seeking time and the probability of misrecognition were also decreased.