Abstract. Classic meanshift algorithm is widely used in pattern recognition, but its accuracy decrease large when there is perturbation in background. In this paper, to solve this problem, we put forward an improved meanshift algorithm which is based on self-adaption weighting meanshift algorithm. At first, we catch moving object area. Secondly, we use capture to set new weighting coefficient. Then we track target model by the new coefficient to decrease effect of background. The experiment result of moving body in supervisory control video shows perturbation insensitivity, robustness and stability.

Abstract. Nowadays, moving body recognition method is used widely in all kinds of videos. But recognition accuracy of these methods is changed negatively because of complexity of background, e.g. void and noise. In this paper, we put forward a new recognition method with background robustness. Firstly, we get moving body by tripling temporal difference method. Then we eliminate noise of these images by mathematical morphology. Finally, we connect disconnected areas with quadruple directions connection method. The new method is more accurate and less computational in real time experiment by used less computation. The experiment result also shows its robustness of background.