Robust hand tracking using exemplar-based correlation filters

To deal with the hand tracking in unconstrained environment, an exemplar-based robust hand tracking algorithm was proposed. In the algorithm, multiple correlation filter based trackers were used to constitute a tracker ensemble. Each tracker was trained by a tracked hand as the positive sample, and the tracker ensemble would not be updated during the tracking process. Moreover, a color model based segmentation was exploited to adjust the tracking result in order to avoid model drift. Comparative experiments demonstrate that the proposed algorithm achieves better performance than single correlation filter or color based tracking algorithms and other tracking algorithms in complex environment. © 2018, Editorial Board of Journal of Huazhong University of Science and Technology. All right reserved.