The study of target tracking method in video surveillance

Abstract

In intelligent video surveillance, the theories of Particle Filter and Mean-shift are the most classic in target tracking algorithm. But, Particle Filter algorithm exists degradation phenomena, large amount of calculation and some other disadvantages. Although Mean-shift algorithm has less calculation, it will result in tracking lost when it encounters occlusion. In allusion to the problems mentioned above, this paper adopts a method that Mean-shift is embedded to tracking framework of Particle Filter, when it encounters occlusion, the algorithm switches to Particle Filter for continuing tracking, and uses resampling methods to suppress the particle degradation phenomena. And when it encounters nonconclusion, it uses Mean-shift algorithm to improve the tracking real-time and robustness. The results show that the adopted algorithm has more real-time and robustness than any single traditional algorithm. © 2013 Springer-Verlag.