Real-time 3D bare-hand gesture recognition using binocular vision videos

Current bare-hand based gesture recognition algorithms generally have the problems of low recognition accuracy and being prone to be affected by skin-like objects. In this paper, a 3D bare-hand gesture recognition algorithm is proposed using binocular vision videos. Firstly, a relationship between the depth and area of the gesture is achieved according to the principle of binocular vision, on the basis of which fast 3D gesture recognition is realized. To further speed up the method, a fast stereo matching algorithm is proposed following the epipolar line constraint rule, which regards the gesture's centroid as the matching point. Experimental results have demonstrated that compared with existing algorithms the proposed algorithm significantly improves the performance in processing speed, recognition accuracy, and robustness. It should be noted that the proposed algorithm is open, where more 3D gestures can be easily added upon requirement.