Precise feedback system for motion posture based on computer vision

Abstract

Against the backdrop of national fitness programs, the standardization of motion posture in daily public exercise directly affects training effectiveness and sports safety. Traditional offline coaching has disadvantages such as high cost and significant spatiotemporal constraints, while standardized online courses fail to provide real-time and personalized corrective feedback on movements. An automated and lightweight posture feedback system has become a key solution to this problem. This study constructs a precise feedback system for motion posture with a five-layer series architecture, which integrates the AlphaPose posture extraction network, Long Short-Term Memory (LSTM) temporal network and Attention mechanism to build a full-process closed-loop link from video input to posture scoring, compliance screening and correction advice output. Relying on keypoint detection, temporal correlation mining and key feature enhancement technologies, the system realizes automated precise discrimination and personalized feedback generation of motion postures. Experimental results show that the system achieves a motion posture discrimination accuracy of 94.2% and an F1-Score of 0.938, and the performance of its core algorithm is significantly better than that of multiple baseline schemes. In an ordinary CPU terminal environment, the end-to-end feedback delay of the system is stably controlled at 0.28 seconds with an average CPU occupancy rate of only 13.6%, and it can run steadily without relying on GPU acceleration. The system can meet the real-time posture guidance needs of non-professional users in daily sports scenarios and provides an implementable technical solution for automated correction of public motion postures.