Multi-Window Target Tracking

Because of the movement under complex environment, the characteristic of tracking target is influenced by illumination, camera angle. And to make the matter worse, the target may be occluded by some part of itself or by some similar objects in the scene in the tracking process. Thus the tracking accuracy is greatly influenced by those elements. To solve those problems, a multi-window tracking method is proposed, which uses some unfixed windows, with each window corresponding to a tracker that may be mean shift algorithm, to represent the tracking target in varied circumstance. It is worth noting that there is no restriction on selecting these windows. That is to say, there could be overlaps among the windows, or the windows could be selected within the target, or the windows could include the object around the tracking target. For an easy example, in the face tracking, some windows could contain the color of the neck or the frock. Thus, the target position is easily estimated by the relation of the windows and the similarity in the successive tracking procedure. Experimental results show that the novel tracking method is robust to the partial occlusion and the change of illumination in the complex background.