Design of University Instrumental Music Performance Movement Recognition and Accurate Feedback System Integrating OpenPose and LSTM

Abstract

Accurate recognition of fine-grained human movements with low-latency feedback is fundamental to intelligent perception systems and real-time human–machine interaction in modern engineering applications. This study presents an end-to-end motion recognition and feedback framework for university instrumental music performance by integrating OpenPose-based pose estimation with a three-layer bidirectional long short-term memory (BiLSTM) network. The proposed approach extracts normalized skeletal keypoint sequences from performance videos and exploits bidirectional temporal modeling to capture both local motion evolution and long-range spatiotemporal dependencies. An adaptive feedback module further quantifies posture deviations through coordinate distance analysis and automatically generates structured correction instructions based on predefined performance standards. Experimental results demonstrate that the three-layer BiLSTM architecture achieves a motion recognition accuracy of 96.3%, while the optimized system reduces end-to-end latency to 48.3 ms and maintains 98.2% overall accuracy under standard teaching conditions. The framework also exhibits robust performance across challenging environments, including low illumination and complex backgrounds. By combining vision-based sensing with temporal sequence learning, the proposed method provides an effective solution for objective motion evaluation and intelligent feedback generation, while offering valuable methodological references for real-time signal interpretation, human-centered electromagnetic sensing, and next-generation intelligent monitoring systems requiring precise dynamic feature extraction.