Advancing construction safety: YOLOv8-CGS helmet detection model

Abstract

In the context of construction site safety management, real-time object detection is crucial for ensuring workers’ safety through accurate detection of safety helmets. However, traditional object detection methods often face numerous challenges in complex construction environments, such as low light, occlusion, and the diverse shapes of helmets. To address these issues, we propose an improved helmet detection model, YOLOv8-CGS, which is based on the YOLOv8 architecture and integrates optimization modules such as CBAM (Convolutional Block Attention Module), GAM (Global Attention Mechanism), and SLOU (Smooth Labeling Loss Function). The goal is to enhance the model’s detection accuracy and robustness in complex scenarios. Specifically, GAM improves the model’s attention to key regions, CBAM enhances its ability to perceive important features, and SLOU optimizes the accuracy of bounding box predictions, particularly in complex and occluded environments. Experimental results show that YOLOv8-CGS achieves accuracy rates of 94.58% and 92.38% on the SHD and SHWD datasets, respectively, which represent improvements of 5.9% and 5.94% compared to YOLOv8. This enhancement allows YOLOv8-CGS to provide more efficient and accurate helmet detection in practical applications, significantly improving the real-time monitoring capabilities for construction site safety management.