A Moving Object Tracking Method Based on Mean Shift Rectification

The traditional mean shift tracking method can not acquire high accuracy when the object undergoes partial occlusion. An improved tracking method called mean shift rectification is proposed. The candidate matching points in target region searched by mean shift method are filtered using a log-likelihood ratio function, and the target region is divided into subregions. Then the spatial matching restrictions are considered to compute the difference displacement through partial histogram matching. Finally all the difference displacements between reference subregions and target subregions are syncretized to compute the rectification displacement. The experiment results show the improvements of the proposed method in robustness and accuracy.