Pose Comparison Based on Part Affinity Fields
Chujin Zhou, Weilong Li · 2019
Inspired by the massive demand for pose-comparison technology in today's society especially athletes and police; A real-time human body pose comparison method based on part affinity fields neural network is proposed. The neural network is used to extract the feature map. The CNN network and the L2 loss calculation are used to extract the detected confidence map and the part affinity field. The maximum weight bipartite graph is used to calculate the optimal connection between the limb parts. The example shows that the method can extract the angles of various parts of the human body more accurately as the reference basis for the pose comparison.