Spatiotemporal neural radiance fields for AI driven motion quality analysis

Abstract Accurate evaluation of mobility quality is necessary for rehabilitation. Still, the techniques already at use rely on either low-fidelity skeleton-based models or expensive motion capture (MoCap) technology. This work presents a framework for Spatiotemporal Neural Radiance Fields (NeRF) allowing for markerless, high-fidelity 3D motion reconstruction and analysis Our solution effectively handles occlusions and models temporal motion flow, while dynamically capturing fine-grained movement deviations surpassing conventional pose estimation and graph-based approaches. Combining NeRF-based motion synthesis with deep learning, we present explainable artificial intelligence feedback for real-time physiotherapy intervention. Our method makes rehabilitation more accessible and less expensive since it allows one to monitor it without using wearable sensors. Particularly with complex rehabilitation activities, experimental data indicate that this approach is NeRF-MQA outperforms conventional skeleton-based techniques in measuring mobility quality, laying the foundation for highly accurate AI-powered rehabilitation systems scalability for usage in both home and clinical environments, and power source.