Vision Based Intelligent Shoulder Joint Movement Recognition Using Feature Extraction

Accurate assessment of shoulder joint movements is essential for rehabilitation and musculoskeletal health management. This study presents a human centered intelligent system for recognizing shoulder joint movements using computer vision and machine learning techniques. Visual motion features are extracted using a lightweight feature extraction approach and used for classification of different shoulder joint movement patterns. Several supervised Machine learning models, including XG-Boost, Random Forest, Light-GBM, and Logistic Regression, were evaluated using hold-out validation and stratified 10-fold cross-validation. Among all models, XG-Boost achieved the best performance with an accuracy of 87.38%, demonstrating superior stability and robustness. To further evaluate generalization capability, the proposed framework was also tested on publicly available rehabilitation benchmark datasets, namely UI-PRMD and KIMORE, where it maintained competitive performance. The results indicate that the proposed method effectively captures subtle motion variations across different datasets while maintaining computational efficiency, making it suitable for real-time and resource-constrained rehabilitation applications.