Measurement of Shoulder Abduction Angle with Posture Estimation Artificial Intelligence Model

Substantial advancements in markerless motion capture accuracy exist, but discrepancies persist when measuring joint angles compared to those taken with a goniometer. This study integrates machine learning techniques capture, an aim enhance this accuracy. Two artificial intelligence-based libraries—MediaPipe and LightGBM—were employed executing shoulder abduction angle estimation. The of ten healthy volunteers was captured using smartphone cameras right ranging from 10° 160°. were set diagonally at 45°, 30°, 15°, 0°, −15°, or −30° relative the participant situated distance 3 m. To estimate angle, models developed considering data goniometer as ground truth. model performance evaluated coefficient determination R2 mean absolute percentage error, which 0.988 1.539%, respectively, for trained model. approach could even if camera positioned respect object. Thus, proposed can be utilized real-time estimation during rehabilitation sports motion.