Improved KCF Long-term Gesture Tracking Based On GMM And ELM

Considering the shape, color and motion characteristics of gesture in human-computer interaction, and aiming at the problems of KCF algorithm in the case of scale change, fast movement and non-rigid deformation of gesture target, a long-term gesture tracking method combining GMM skin color detection, ELM and KCF is proposed. Firstly, the GMM skin color model is obtained by cross-validation training method, and the GMM skin color filter result is refined by Bayesian estimation. According to the search strategy, ELM is used to further determine the gesture area and gesture category. Then, the KCF tracker is improved by combining the onedimensional scale filter, and is initialized by the detection result of the gesture detector. Finally, the tracking result is evaluated by the designed confidence model. If the confidence is less than the threshold, the gesture detector is used to reinitialize the improved KCF filter and restart tracking. Otherwise, continue to track. The experimental results show that the method can track the deformed gesture in human-computer interaction quickly and accurately in real time, and has good adaptability to scale changes.