Research on gesture-recognition method in video based on the sparse representation theory

摘要

Gesture recognition is an important research topic in computer vision. Existing gesture recognition methods are generally based on single image and lack spatiotemporal continuity in the analysis of image content. In order to deal with this problem, a new gesture recognition method in video based on sparse representation theory is proposed in this paper. Firstly, the foreground image of the hand region is obtained by using the skin color segmentation of the YCbCr color space for a continuous video. Secondly, the center of gravity of the foreground image for the hand region is extracted as feature vector for recognition. Gesture dictionary is further constructed, and a sparse representation model of certain kind of gesture is established. Then, gestures in video are classified by determining the sparse representation error for a new sample to be identified. Finally, experiments on the collected video sequences are performed. Experimental results show that the proposed method can recognize four kinds of gestures such as moving up, down, left and right in video. The proposed method would be used to recognizing more complex gestures in future work.