An Improved YOLOv5-Based Method for Crane Operator Absence Detection

Preventing crane operator absence during construction process is of vital importance for ensuring construction safety. The currently used manual supervision methods are subjective and lack real-time capabilities. Thus, this paper introduces a method utilizing an enhanced version of YOLOv5 to achieve real-time and precise detection of crane operator absence. Firstly, to achieve real-time detection, the module named C3-Ghost is introduced to lightweight the original feature extraction network. Then, to tackle the issue of large-scale variations of detected objects in construction scene images, a detection head specifically designed for small object detection based on fused shallow network features is proposed to retain detailed features of small objects that are difficult to detect, thus enhancing small object detection capability. Finally, the NWD loss function is utilized to improve the model’s ability to detect small objects. Experimental results demonstrate that the improved model meets the requirements for real-time detection of crane operator absence.