Multi target tracking using multiple independent particle filters for video surveillance

Tracking a group of targets in a video sequence is a common problem in many video surveillance applications. In this paper, multi target tracking method that uses multiple independent particle filters (MI-PF) for video surveillance systems is proposed. We apply mutual exclusion constraint that uses intuitive and simple method consisting of updating a weight of samples by the comparison of the prior probability between the target to track and marginal targets. To solve occlusion problem, a dynamic mode consisting of tracking and occlusion modes is used. When the occlusion occurs, the mode is switched to the occlusion mode and samples are propagated by the Gaussian distribution without resampling process. Experiments on the soccer video demonstrate the independent multi-target tracking and robustness of occlusion between players.