A Novel Bagged Particle Filter for Object Tracking

In this paper, we propose a novel bagged particle filter framework to filtering the noise information from object trackers using generative model as well as the discriminative model. The framework makes use of objectness measurement for modeling observation likelihood and two powerful object detectors: the real-time L1 tracker and the TLD tracker combined to bagged trackers. By maxmazing the posterior of the proposed inference, inaccuracy information is filtered and more accuracy result from varying samples returned by different trackers is provided by the bagged particle filter. The experiment results suggest that the proposed particle filter is effective in combing the complementary nature of either the sparse tracking approach and the discriminative learning approach.