Grid features based visual tracking

Abstract

Vulnerability to occlusion is one of the main issue in visual tracking. In this proposal, we exploit the local grid features to build a robust tracker. To improve performance under occlusion, local and global features are modeled for a target tracking. Cooperating with the novel features, a new segmentation and similarity measurement are proposed for exploring the local grid advantages. Experimental results show that our tracker outperforms other two effective visual tracking methods under occlusion.