Robust Object Tracking via Hierarchical Particle Filter

A potential weakness of generic particle filters discussed above is that the particle-based approximation of filtered density is not sufficient to characterize the tail behavior of true density, due to the use of finite mixture approximation; To alleviate this problem, In this paper we propose a general hierarchical particle filtering framework for designing an optimal proposal distribution. The essential idea is to augment a second filterpsilas estimate into the proposal distribution design. We shall see that several existing improved particle filters can be unified into our general framework. Based on this framework we further propose novel variant algorithms for robust and efficient visual tracking.