Hierarchical structure correlation inference for pose estimation

Abstract

Human pose estimation is a challenging and important basic subject in computer vision. At present, mainstream methods have achieved remarkable progress through the powerful representation capabilities of convolution neural networks. However, these methods don’t pay enough attention to the correlation of the structure of each part of the human body in pose estimation. To solve this problem, the present paper proposes the hierarchical structure correlation inference algorithm, composed of correlation inference modules, which increases the ability of the network to infer the correlation information of each part of the body. Moreover, this algorithm promotes the understanding of the human body structure. The proposed algorithm is effectively verified on the MPII Human Pose Dataset and Leeds Sport Pose Dataset, and the experiment shows that our results are better than current mainstream algorithms.