An Edge-Based Multimodal Framework for Quantitative Human Motion Assessment with Biomechanical Applications

Objective assessment of human movement quality has significant clinical importance in rehabilitation medicine, sports medicine, and the early screening of neurodegenerative diseases. However, existing assessment methods rely heavily on subjective expert judgments and are constrained by high equipment costs and limited real-time feedback capabilities. To address these challenges, this study developed a multimodal human movement assessment framework based on the Internet of Things (IoT) and Edge Computing (EC). The proposed framework adopts a three-layer IoT architecture and enables localized video data processing through EC devices. The system first performs identity authentication using face recognition, followed by a MediaPipe-based pose estimation module that extracts 33 skeletal keypoints from each video frame on edge devices. Multimodal data fusion is then applied to achieve quantitative analysis of movement features. The system performs frame-by-frame comparisons between the subject's movement postures and predefined standard templates, automatically generating an objective movement quality score ranging from 0 to 100. The proposed framework was validated in a sports dance training scenario involving 18 participants. The results showed that the recognition accuracies of the four movement categories reached 96.59%, 97.25%, 96.85%, and 90.84%, respectively, with corresponding F1 scores of 96.5%, 97.3%, 96.0%, and 90.8%. The Pearson correlation coefficient between the system-generated scores and evaluations from two experts was r = 0.87 (p < 0.001), with a mean absolute error of 4.82 points. The EC architecture reduced the end-to-end latency to 780 ms and decreased data transmission volume by approximately 99.97%. Experimental results demonstrate that the proposed multimodal assessment framework can provide objective and real-time analysis of human movement quality, showing strong potential for broader applications in biomedical scenarios such as clinical rehabilitation assessment and motor function monitoring.