Multiple sub-spaces particle filtering for multi-target tracking

A new multiple target tracking method based on Bayesian filtering and Sequential Monte Carlo approximating method is proposed in this paper. The key principle of the proposed method is to decompose the multiple target tracking problem into multiple single target tracking problems by allocating Sub-Spaces which are sub-sets of single-target state space to targets. The computational cost of the proposed method is remarkably reduced by avoiding jointly estimating posterior probability distribution used in many conventional multi-target tracking methods. And compared with Finite Set Statistics based methods, the proposed method is more general, moreover, it could supply high level applications with trajectory of moving targets which is not available in Finite Set Statistics based method. The proposed method is tested by tracking pedestrian in video sequence captured from real-world and the tracking result shows that all targets are well tracked in real-time while the number of targets is unknown and varies with time.