Un sistema de puntuación de acciones en tiempo real para el análisis de movimiento y retroalimentación en...

Human Pose Estimation (HPE) has become an essential tool in physical therapy, enabling automated movement analysis and rehabilitation monitoring. However, existing HPE techniques often suffer from limitations such as motion blur, occlusions, inconsistent keypoint visibility, and sensitivity to variations in camera angles and subject positioning. This makes the movement assessment very challenging, particularly in home-based rehabilitation settings where real-time supervision is limited. To address these issues, a novel action-scoring algorithm that integrates angular-based movement analysis with keypoint normalization techniques is presented in this study. Specifically, the proposed method employs Dynamic Time Warping (DTW) and Normalized Cross-Correlation (NCC) for precise movement comparison, alongside a fixed bounding box strategy to improve tracking stability. Additionally, a new repetition counting mechanism based on angular calculations is introduced to ensure accurate assessment of repetitive exercises. The proposed approach primarily aims to minimize the angular noise such as motion blur. Hence, the proposed system can be easily integrated with existing human pose estimation systems. Experimental validation demonstrates that the proposed approach achieves high accuracy in joint angle measurements and repetition detection while offering increased robustness against occlusions. A comparative evaluation with RepNet, a state-of-the-art video-based repetition counting model, shows that the proposed method outperforms RepNet in both accuracy and computational efficiency, making it more suitable for real-time rehabilitation feedback. These findings highlight the potential of the proposed design to improve movement analysis reliability, optimize rehabilitation outcomes, and expand access to automated physical therapy assessment tools.