Particle filter based on multi-window

Abstract

A novel particle filter based on multi-window is proposed for target tracking, which represents the tracking target with several windows that each one corresponds with a particle filter. Because tracking performance is usually influenced by factors such as camera angle, illumination. Meanwhile the target may be occluded by obstacles in the clutter background or be occluded by other target. A multi-window particle filter tracking method is proposed. And the multi-window may overlap or contain the scene around the target. In the tracking process, the different particles have different tracking cues of the targets. In the novel algorithm, through the spatial relation of the non-overlap windows and overlap windows, the spatial configuration of the tracking target is included and considered. All particle filter trackers are used to estimate the target status parameters. Using this algorithm, we can overcome the tracking failure because of the object spatial features considered. And experiment results clearly demonstrate the effectiveness of the novel method on illumination influence or self-occlusion problem in a complex background.