In this paper,a real-time human pose estimation method based on MediaPipe framework is proposed for accurate pose recognition of sports actions.In this study,we extract the depth information from a single frame image,and use BlazePose net-work to quickly detect the two-dimensional key points of the human body in sports actions,and map the two-dimensional coordi-nates to the three-dimensional space,so as to achieve the accurate human pose estimation of sports actions.The experimental re-sults show that the proposed method has low latency while maintaining high accuracy,and is suitable for sports training,virtual coaching and motion analysis.This study not only provides empirical support for the effectiveness of lightweight neural networks in real-time human pose estimation,but also can expand the application range and improve the performance through further research.
万方数据