Mean Shift Tracker With Grey Prediction for Visual Object Tracking

Abstract

In this paper, the mean shift (MS) tracker embedded with grey prediction is proposed for visual object tracking. As the basic model of grey prediction, grey model [GM(1,1)] is employed to predict object location with few historical information. The predicted location is taken as the initial point of MS iteration instead of the previous tracking result in the original MS tracker. The prediction equation of GM(1,1) is simplified to reduce computation, and the occlusion degree is determined by the Bhattacharyya coefficient and a set threshold. If the degree exceeds a certain limit, the MS iteration may not converge to the true result and the object location is replaced with the predicted location to prevent failure tracking. The experimental results show that the proposed approach can effectively deal with the problems of fast-moving and serious occlusion and has a better performance than the original MS tracker and the MS tracker with particle filter.