Object Tracking with a Novel Method Based on FS-CBWH within Mean-Shift Framework

Effective appearance models are one critical factor for robust object tracking. In this paper, we introduce foreground feature salience concept into the background modelling, and put forward a novel foreground salience-based corrected background weighted-histogram (FS-CBWH) scheme for object representation and tracking, which exploits salient features of both foreground and background. We think that background and foreground salient features are both crucial for object representation and tracking. Experimental results show that the proposed FS-CBWH scheme can improve the robustness and performance of mean-shift tracker significantly especially in heavy occlusions and large background variation scenes.