Postural risk analysis for the manual demolition process leveraging a novel REBA assessment tool

Introduction: Manual demolition is essential to achieving resource circularity in the built environment, yet the literature lacks a comprehensive quantitative understanding of the postural risks facing manual demolition workers. Method: This study develops REBAPose, a novel computer vision (CV)-based automated Rapid Entire Body Assessment (REBA) tool. Drawing on biomechanics and anatomy, a dataset of 149,813 person images was created with precise annotations of the neck, head midpoint, and hip center, replacing the heuristic estimates prevalent in the existing literature. REBAPose was trained on this newly created dataset and applied to 2,802 postures of 48 workers across seven demolition sites. Results: REBAPose, based on the MMPose architecture, achieves a mean Average Precision of 0.80, outperforming several state-of-the-art algorithms. A comparative evaluation against heuristic estimates, using input from two industrial hygienists on 88 images each from demolition tasks, shows their preference to REBAPose over heuristic estimates for the neck (93%) and head (91%). Demolition postures show consistently high REBA scores (mean 9, maximum 13), with uniquely high risk for the lower arms (Relative Hazard Index 0.966) and neck (0.797). Conclusions: Manual demolition entails severe postural risk, concentrated in the lower arms and neck. REBAPose generalizes across datasets; however, wrist posture is not fully automated, and the load/force, coupling, and activity components of REBA require manual input. Practical Applications: Contractors, safety managers, and ergonomists can deploy the open-access REBAPose for automated postural risk assessment across industries and use its task-level results to prioritize interventions in high-risk demolition tasks.