3D Human behavior recognition based on binocular vision and face–hand feature

Abstract

To improve traditional video surveillance systems’ performance on human behavior recognition, a new method based on binocular vision was researched. This paper proposed a human behavior recognition method based on binocular stereo vision and human face–hand feature. Firstly, the paper got extrinsic and intrinsic parameters of the two cameras through binocular stereo vision calibration, then the paper located face and hands region through human face–hand feature segmentation method, and next then for getting the 3D information of face and hands by binocular stereo vision theory, the paper proposed a feature block matching algorithm based on local gray area correlation and epipolar constraint rule to match corresponding feature regional centers. And finally, the paper completed the analysis of the human upper body poses, 3D spatial moving and the measurement of moving velocity based on 3D information of face and hands. Experimental results showed that the method can successfully recognize human behaviors, and it can solve the occlusion and multi-direction movement problem of traditional video surveillance system. Our behavior recognition method is straightforward and robust. The uniqueness is different from traditional machine learning and classification framework about human behavior recognition.