Vision Based Target Tracking Algorithm in Natural Complex Scene

Vision based target tracking under complicated background is difficult for image analysis. Cam-shift (continuously adaptive mean-shift) algorithm which is based on color cue can track objects with certain color in a simple scene. But under complicated background, tracking converges to another local maximum and loses the real target because of the varying illumination, occlusion, etc. To deal with these problems, an improved cam-shift algorithm is proposed. By using three-dimensional background-weighted histogram in the HSV color space and combining with motion cue, it can track different color objects in real time. And a new method is also proposed to detect and deal with the case of occlusion factors. Experimental results indicate it is robust and has good performances for target tracking in natural complex scene.