Human actions segmentation and matching based on 3D skeleton model
Huang Shuzi, Jing Yuan, Huan Chen · Chinese Control Conference · 2013
Kinect is a motion sensing device developed by Microsoft Company. It can provide data of human skeleton in order to achieve motion tracking of the people. This paper proposes a method to segment and match human actions based on 3D human skeleton node data. This method can be used to segment and match complex movements such as dance and gymnastics and so on,and it can also detect the deviation between the measured action and the standard action.We divide the human skeleton into five parts and calculate kinematics information of each part. Using the naturally separable characteristic of human actions, we segment human actions based on the kinematics information of the skeleton model in order to acquire action pieces. Then, analysis of the human action behaviors and matching of the action pieces are finished. The rotation matrix is used to calculate the characteristics of the joint angles of the skeleton, which are adopted to achieve skeleton poses matching. The experimental results show that 97% accuracy is achieved when matching simple actions. Moreover, segmentation of the actions can improve the matching accuracy of long and complex actions.