HUMAN ACTION DETECTION AND ERGONOMIC RISK ASSESSMENT AT CONSTRUCTION SITES, BY USE OF MACHINE VISION AND DEEP LEARNING

Abstract

The research work described herein focuses on the realtime detection and pose analysis of human activities at construction sites, as well as on the evaluation of the ergonomics of these activities. The pose detection and ergonomic analysis utilize machine vision (MV) and deep learning technologies for the processing of images and/or video streams, and a “skeletonization” mechanism that upon detection of a worker pose, measures the geometric properties of the pose’s keypoints in the skeletal shape and then calculates the corresponding scores according to the Rapid Entire Body Assessment (REBA) methodology. The utilized approach, which was successfully tested on several typical construction activities, (1) has the potential of providing fast ergonomic assessment at construction sites; and (2) it contributes to the knowledge of occupational safety and health in the construction industry, by providing a low-cost and accurate approach for assessing the risk factors of Work-related Musculoskeletal Disorders (WMSDs).