Identifying Struck-By Hazards in Unknown Lifting Loads: A Hybrid Computer Vision Approach

Struck-by hazards frequently result in severe injuries during lifting operations. Computer vision (CV) approaches have experienced rapid development in recent years, enabling automatic monitoring of hazard scenarios and enhancing on-site safety management. By detecting lifted loads and preventing workers from entering the fall hazard zone, the frequency of struck-by accidents in lifting operations can be reduced. However, existing CV approaches primarily focus on detecting known loads, rendering them ineffective in the presence of unknown loads during lifting operations. This study proposed a novel hybrid CV-based approach for the automated identification and tracking of struck-by hazards in lifting operations, specifically designed to handle loads of indeterminate categories and irregular configurations. The approach integrates object detection and optical flow methods to identify unknown loads in addition to well-defined entities, such as workers. The inclusion of binocular vision method permits real-time spatial detection in three dimensions (3D). In the case study, the proposed approach successfully detected unknown lifting loads with a 90.10% F1 score, operating at 11.90 frames per second (FPS). Moreover, it identified hazards with an 88.93% F1 score at a consistent frame rate of 4.81 FPS during lifting operations. The proposed approach holds promise as a valuable tool for supporting on-site safety management and preventing accidents.