Working-at-high operation safety protection recognition based on target detection and spatial relationship

Construction workers of long tunnel projects are confronted with numerous safety hazards such as fall from height (FFH) and object strike due to the harsh jobsite environment, limited space, and complex working conditions. And the absence of protective guardrails is identified as the primary cause of falling accidents from height. In order to automatically detect the safety protection status of working-at-high workers, a computer vision-based recognition method for working-at-high operation safety protection according to target detection and spatial relationship was proposed in this study. Firstly, the Cycle-consistent Generative Adversarial Networks (CycleGAN) was used to preprocess construction site images to enhance the image quality. Secondly, a YOLOv8 model integrated with the coordinate attention (CA) module was established to rapidly detect targets such as workers, trolleys, and guardrails in the tunnel. Furthermore, an identification method for working-at-high operation safety protection is proposed based on the detected targets and their spatial relationships. Finally, a case study was conducted, revealing that the model achieves an accuracy and recall rate of 95.89% and 97.22%, respectively, in identifying the safety protection status of working-at-high workers. The result indicates that the proposed method provides a new way for intelligent identification of working-at-high operation safety protection and assisting on-site management personnel to prevent the risk of FFH in the tunnel.