A Pose-Estimation Physical Movement Training System with User-Created Content

Physical movement training systems can offer structured, accessible, and personalized fitness or rehabilitation. While many existing systems pose estimation for real-time feedback and movement assessment, they often lack support for user-generated content. This study presents an online physical training system enabling trainers to upload custom videos and utilize pose estimation for real-time feedback and posture similarity assessment. A pilot feasibility study with five healthy participants yielded a System Usability Score of 78.5 and high usefulness ratings (4.8/5 for rehabilitation, 4.6/5 for frozen shoulder diagnosis). Posture similarity scores were closely aligned with expert ratings (0.16 difference). However, several challenges remain for clinical use. Future work should focus on improving 3D pose estimation accuracy, conducting broader clinical validation, developing vision-based usage guidelines, and integrating human expertise with AI to foster ecosystem adoption.

Recommended citation: Tharatipyakul, Atima, Suporn Pongnumkul, Dulyawat Wiriyaphong, Sarunya Kanjanawattana, and Gun Bhakdisongkhram. "A Pose-Estimation Physical Movement Training System with User-Created Content." In 2026 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), pp. 808-813. IEEE, 2026.
Download Paper