A Study on AI-Driven Motion Capture and Pose Estimation for Physiotherapy

AI-driven motion capture and pose estimation have revolutionized physiotherapy by enabling real-time, markerless motion tracking for rehabilitation. Traditional physiotherapy relies on manual observation, which is subjective, inconsistent, and inaccessible for remote patients. This study presents a scalable AI-based system using deep learning models such as BlazePose and OpenPose for pose estimation, biomechanics-based movement analysis, and real-time corrective feedback. The system achieves 85-95% accuracy in detecting human movement, with latency below 100 milliseconds, ensuring real-time response. Physics-based constraints enhance pose estimation accuracy by 10-15%, while a custom physiotherapy dataset and multi-camera validation improve robustness. Performance is evaluated through physiotherapist-guided ground truth data, multi-camera motion capture, and real-world rehabilitation testing, demonstrating automated movement assessment, improved rehabilitation adherence, and cost-effective remote physiotherapy monitoring. This research highlights AI’s potential in enhancing physiotherapy outcomes, reducing long-term healthcare costs, and bridging the gap between clinical supervision and independent recovery. Index Terms - Biomechanics, Deep learning, Motion capture, Physiotherapy, Pose estimation, Real-time tracking