3D human pose estimation using aided constraints of physiological and action feature

3D human pose estimation is a hot research topic at present, and it also has a wide application potential. The inherent uncertainty and multiple solutions of 2D to 3D mapping based on a single image limit the accuracy of 3D human pose estimation. Considering that human posture is affected by physiological features and motion states, the network design in this paper uses physiological and motion features to provide constraints for posture estimation, in order to achieve better accuracy. Specifically, in the network design of this paper, three auxiliary judgment networks, namely gender, motion type and true false judgment, are used to further constrain the generated posture. Moreover, experiments on Human3.6M dataset show that the accuracy of mapping 2D joint coordinates to 3D pose coordinates can be effectively improved by introducing constraints of physiological features and motion states.