Measurement of Shoulder Abduction Angle with Posture Estimation Artificial Intelligence Model

Area of Science:

• Biomechanics
• Computer Vision
• Machine Learning

Background:

• Markerless motion capture shows advancements but struggles with joint angle accuracy compared to goniometers.
• Existing markerless systems require further refinement for precise clinical and athletic motion analysis.

Purpose of the Study:

• To improve markerless motion capture accuracy for joint angle measurement using artificial intelligence.
• To develop and validate machine learning models for precise shoulder abduction angle estimation.

Main Methods:

• Integrated MediaPipe and LightGBM for markerless motion capture and AI-driven angle estimation.
• Captured shoulder abduction motion (10°-160°) of 10 healthy volunteers using smartphone cameras at various diagonal angles.
• Trained machine learning models using goniometer data as ground truth for shoulder abduction angles.

Main Results:

• Achieved high accuracy with a coefficient of determination (R²) of 0.988.
• Demonstrated a low mean absolute percentage error of 1.539% for the trained model.
• Validated the model's ability to estimate shoulder abduction angles even with diagonally positioned cameras.

Conclusions:

• The proposed AI-integrated markerless motion capture system significantly enhances joint angle measurement accuracy.
• This technology offers a viable solution for real-time shoulder motion analysis in rehabilitation and sports.
• The method is robust to varying camera perspectives, increasing its practical applicability.