Human Motion Capture Technology Based On Multi-Binocular Vision And Image Recognition
Weile Zhang, Kaizhu Yang, Yi Tang, Junpeng Yang, Xin Liang, Yiliang Fan, Qintao Du · 2020
This paper proposes a human motion capture system based on multi-binocular vision and image recognition. The system which relies on the principle of binocular imaging and stereo matching of the same name points of the human joints in the left and right views to obtain the 3D coordinates of the human joint points in real time, uses multiple binocular cameras placed at different angles for simultaneous shooting, expanding the integrity of the original binocular vision system. At the same time, the detection model of 3D abnormal estimation points of human joints under multi-binocular vision is established, so that the system can more accurately identify the 3D abnormal estimation points of human joints due to occlusion. Finally, the average coordinates of the non-abnormal human 3D joint points obtained by other binocular cameras are dynamically selected as system's final estimated coordinates, thereby further improving the accuracy of human motion capture.