3D human gesture matching via graph cut
Tianchu Guo, Xiaoyu Wu · 2013
In this paper, we propose a new method to match the human dance gesture, by using the human body gesture features which is extracted from the contour image obtained from depth image. The method is robust to noise and can avoid small fluctuations caused by the instability of the depth information. In the matching operation, we abstract the human gesture into a new 3D shape and extract the angles of the new 3D shape as the features of human gesture. Disparity matrix can be obtained to describe the distance between the templates' and the users' gestures. An energy function can be constructed by disparity matrix and minimized via graph cut.