Visual Target Tracking Based on Multiple Cues and Particle Filter

The problem of target tracking in video sequences is addressed in this paper. The framework we choose is particle filter which can deal with non-linear/non-Gaussian models effectively. A particle filter can create several hypotheses in state space by randomly sampling and evaluate the true state by weighing them respectively. Color is a powerful feature in target tracking, but with color only it may be difficult to distinguish the object from background when they have similar color distribution or the illumination varies. We try to conquer the problem by using multiple cues extracted from video sequences. Here the color information is fused with the motion information in particle filter. The fusion of these two kinds of information can make the tracking process more robust. Several tracking results are given to illustrate the effectiveness of our proposed method.