Dynamic gesture recognition for natural human system interaction

This paper addresses two problems: 3d dynamic gesture recognition and gesture misallocation. In order to solve these problems, we propose a new approach which combines Hidden Markov Models (HMM) and Dynamic Time Warping (DTW). The proposed approach has two main phases; first, recognizing gestures using a hidden Markov model. Second, avoiding misallocation by rejecting gestures based on a threshold computed using DTW. Our database includes many samples of five gestures obtained with a Kinect and described by depth information only. The results show that our approach yields good gesture classification without any misallocation and it is robust against environmental constraints.