Multi-target Tracking Based on Fusion-Feedback Particle Filter

Aiming at the tracking problem of soccer players in soccer videos, a tracking algorithm based on fusionfeedback particle filter is proposed. By using a mixture model that incorporate information from the dynamic models of players and the results of the detection block, the proposal distribution is constructed and the state of the players in the next frame are sampled. And then, the similitude degrees between the samples and the reference players are calculated. At last, the player states are estimated and the player under tracking is attained. Experiment results show that this present method is effective for coping with the occlusion of two players and attaining the tracking players, and has better realtime and robustness.