Improved mean shift target tracking algorithm

An improved mean shift (MS) target tracking algorithm is proposed to improve the robustness and accuracy of tracking under template drift and large-area occlusion. Firstly, the method predicts whether the target is in occlusion. If the target is not in occlusion, the original MS algorithm is used to track targets and the target template update strategy based on selected component is used to reduce the influence of template drift; when the target is occluded, the target candidate model is corrected by an asymmetric kernel model to reduce the influence of occluded pixels on MS vector and target tracking stability. The experiment result shows that the proposed algorithm can steadily track non-rigid and large-area occlusion targets.