Position constrained network for 3D human pose estimation

Human pose estimation is a challenging research task in the field of computer vision. The current mainstream works have made great progress in pose estimation, but these works still have weakness in two aspects: first, the feature extraction module is not competent for representation learning; second, the training process does not take fully advantage of the projection model from 3D space to 2D plane. In this work, we propose a human pose estimation framework which exploits 3D root coordinates as subordinate input to 2D joint coordinates to provide positional reference to the recovered 3D joint coordinates, and employs inner camera parameters to construct additional projection constraints for recovering 3D joint coordinates. Moreover, we enhance the feature learning through residual branch. We tested our method on two benchmark datasets for human pose estimation, and the experimental results show that the proposed model is superior to current mainstream algorithms.