Recognizing Human Actions Using 3D Motion Trail Model
Song Wang, Jianwu Dang, Yangping Wang, Lixia Liu, Zhenhai Zhang · 2018
In this work, a novel descriptor is proposed for 3D motion trail model. The proposed descriptor is used to describe the motion pattern contained in the three dimensional motion trail. Key poses are selected in order to improve the performance of the proposed method. Specifically, we employ three dimensional motion trail model to represent movements. In order to speed up the computation of 3DMTM, the key poses are selected according to the kinetic energy of joint motion. Finally, Kernel extreme learning machine is used to classify the depth sequences. Experimental results show that the proposed method achieves better results on MSR Action3D dataset than the existing method.