Estimation of 3D human pose using prior knowledge

Estimating three-dimensional (3D) human poses from the positions of two-dimensional (2D) joints has shown promising results. However, using 2D joint coordinates as input loses more information than image-based approaches and results in ambiguity. To overcome this problem, we combine bone length camera parameters with for input. This combination is discriminative that it can improve accuracy model’s prediction depth alleviate ambiguity comes projecting 3D into space. Furthermore, introduce direction constraints, which better measure difference between ground truth output proposed model. The experimental on Human3.6M show method performed other state-of-the-art pose estimation approaches. code available at: https://github.com/XTU-PR-LAB/ExtraPose/.