Construction workers are at high risk of work-related musculoskeletal disorders (WMSDs) due to prolonged awkward postures. Although computer vision approaches provide an objective solution for risk assessment, existing studies primarily rely on static image data, and the influence of camera angles on assessment accuracy remains insufficiently investigated. This study aims to conduct WMSD assessment based on multiangle computer vision technology and to develop an automated assessment method as an alternative to traditional, inefficient observational techniques. An open available construction work posture video (CWPV) data set was developed in this study, comprising multiangle video recordings of eight common awkward postures in construction, collected from 21 subjects. A computer vision–based model was developed to extract key body points, calculate joint angles, and evaluate WMSD risk according to rapid upper limb assessment (RULA), rapid entire body assessment (REBA), and Ovako working posture analysis system (OWAS) assessment criteria. Scores from two ergonomic experts, with a mean intraclass correlation coefficient greater than 0.9, were used to evaluate the accuracy of the assessment algorithm. The highest evaluation accuracies achieved for RULA, REBA, and OWAS were 93.3%, 93.3%, and 86.7%, respectively, demonstrating the method’s effectiveness in WMSD risk assessment. Differences in scoring trends among the three methods reflected their distinct evaluation criteria, with REBA’s inclusion of lower limb posture resulting in greater score variability. Further analysis revealed that camera position significantly influences assessment accuracy, with frontal and rear views generally outperforming lateral views. Compared with previous studies, a multiview working posture data set was created to enable comparison of WMSD assessment performance across different viewing angles. The proposed approach provides a reliable and efficient tool for assessing WMSD risk, facilitating a shift from passive observation to proactive, data-driven safety management in smart construction sites.
Kou et al. (Sun,) studied this question.