Real-time multi-cue mean-shift hand tracking algorithm in cluttered background

Abstract

This paper proposes a multi-cue real-time hand tracking algorithm effective for skin color cluttered background. Traditional color based mean-shift algorithm often fails in color confusable background. To deal with this problem, we first construct a novel dynamic motion-color joint probabilistic distribution with optical flow feature based motion model to clearly distinguish the target moving object from visually similar background or other moving objects. We then apply the multi-cue mean-shift tracking by combining the motion-color joint distribution with the mean-shift iteration process. A motion detection process is also performed to avoid the target centers' erroneously shifting to color similar static background when no motion exists. Representative experiments validate that the proposed method improves the accuracy of mean-shift tracking algorithm under cluttered background environment.