Workers and Safety Helmets Detection in Day and Night Scenes based on improved YOLOv5

Source abstract

Abstract Safety helmets, as crucial protective equipment, significantly contribute to the head safety of workers. Adherence to safety helmet regulations is integral to construction site safety management. Recognizing the limitations inherent in manual supervision methods, we have developed a vision-based framework for the detection of workers and their safety helmets. This framework features enhancements to the YOLOv5s model, resulting in the advanced YOLOv5-Pro. The enhanced YOLOv5-Pro model achieved a mean Average Precision (mAP) of 95.4% on the validation set, marking an improvement of 3.6% over the original model. Furthermore, we expanded the utility of the YOLOv5-Pro model by incorporating nighttime data augmentation. The augmented YOLOv5-Pro model demonstrated robust performance in both daytime and nighttime conditions, as evidenced by our experimental results.