Robust Object Tracking Method Dealing with Occlusion

Abstract

This paper proposes a robust object tracking algorithm dealing with the occlusion problem and illumination change. In particular, we judge whether each frame image is occluded by the confidence map, and two object spatial-temporal context models are established, one is updated every frame in the process of tracking, and the other is set up before the frame that the object is occluded. This method can correct the tracking drift caused by the model update during the occlusion process. Besides, the target image's pixel value is normalized to remove the effect of illumination change, and the phase of the object image's Fast Fourier Transform is used as the texture of the image. Experimental results show that the proposed method is robust to complex situation such as heavy occlusion, dramatic illumination change and surface variation.