The Feedback Tracking Based on the Sequence Likelihood Particle Filter

Given the low accuracy of visual tracking based on the routine particle filter caused by the particles poverty, this paper presents the novel sequence likelihood particle filter (SLPF) in which the particles are extracted from a discrete likelihood distribution. And the constructing procedure of the discrete likelihood is stated. And the weight updating formula of the sequence likelihood particle filter is proven. In this paper, two kinds of particle poverty caused by the occlusion and by the excessive clearness are analyzed, and two methods of switching working mode and the feedback adjustment are presented to deal with them. The experiment shows that the sequence likelihood particle filter and the feedback adjustment could improve the estimation accuracy.