부산대학교 도서관

Work-related musculoskeletal disorders (WMSDs) have been the major cause of occupational injuries among construction workers. The traditional observational assessment is time-consuming and subjective, while the sensor-based postural analysis is usually associated with high setup costs and intrusiveness. This study proposed an automated ergonomic risk assessment method based on computer vision and machine learning focusing on lower body postural risks. It provided a comprehensive risk dashboard, including posture detection and rule-based extreme flexion examination. Specifically, with raw video input, the postural feature extraction module can identify skeleton coordinates frame by frame by adopting a state-of-art 3D pose estimation algorithm. In the ergonomic assessment module, the knee angles can be calculated using skeleton coordinates, and the support vector machine (SVM) classifier was trained for posture recognition. The illustration based on a real-life example demonstrated the applicability and reliability of the proposed method, with nearly 95% accuracy for posture detection. In summary, the study provided a more comprehensive and in-depth postural analysis of construction activities, which has great potential to facilitate intervention strategies for WMSD prevention with quantified evidence.