Collaborative tracking in video sequences using corners and gradient information

In this paper the problem of the simultaneous tracking of multiple video objects is addressed. In the proposed approach, each tracker behaves independently using corners and gradient-based information until an interaction with other trackers is reported. During the interaction, a new Bayesian method that allows the exploitation of the information of each tracker in a collaborative way is used. By using this method, it will be shown that it is possible to improve the global correctness of the tracking and targets model estimation by fusing the information owned locally by each tracker in a collaborative way. The reported experimental results indicate good performances of the algorithm in crowded scenes.