Real-time Tracking of Non-rigid Objects
Sheng Wei, Jianxin Ren · 2016
Currently, pose variations and irregular movements are the main constraints in the tracking of the non-rigid object. In order to avoid the inaccurate location or the failure of tracking the non-rigid object, a novel tracking method combining particle filter and Mean Shift algorithm is proposed. The motion segmentation is used to correct particle filter's estimation error of the non-rigid target, which improves the efficiency, as well as the robustness of the algorithm against noises. The normalized correlation coefficient is calculated to determine whether to update the template of Mean Shift algorithm. We also test the algorithm on the open popular datasets. Results prove that the algorithm presented in this work shows better results in both aspects of effectiveness and efficiency than the method combining CAMShift algorithm with Kalman filter.