Skeleton-based Human Keypoints Detection and Action Similarity Assessment for Fitness Assistance

Human pose detection refers to finding the position of important joints of human body such as head, hands and feet in image or video, which is a frontier topic in computer vision and is widely used in many fields such as human activity analysis, advanced human-computer interaction and virtual reality. The human pose similarity metric refers to the measurement of similarity between different human poses by metric, which is crucial for the research of human pose recognition based on database retrieval. Currently, in the online fitness situation, video movement instructions are commonly used in a lecture-style with limited teaching effect. In this paper, we propose to use OpenPose and BPE algorithms to analyze and compare fitness movements based on human pose estimation and movement similarity assessment, which can provide real-time movement modification for learners and obtain movement comparison analysis and evaluation results. The experimental results show that the method has higher accuracy, shorter time consumption and better robustness.