Mean shift using novel weight computation and model update

The tracking accuracy of mean shift algorithm employing the Bhattacharyya coefficient as similarity metric is low. The reason is analyzed on the basis of effect identity of the histogram bins in the target model and the candidate. A novel weighting method which is derived from the enhanced histogram intersection is proposed. Several improvements are considered in the implementation of the algorithm to handle rapid object motion or scene interference with similar color to target. Then we give a novel model update method based on the prior information of the tracked target. Experimental results illustrate that the proposed method outperforms the standard mean shift algorithm.