Research on Safety Helmet Detection for Construction Site

In the complex construction environment, unknown danger is everywhere. As an important safety gear, safety helmets can effectively protect the head. After analyzing and summarizing the existing safety helmet detection methods, this paper proposes an improved lightweight safety helmet detection model based on YOLOv4. Firstly, the backbone network of YOLOv4 is replaced with its lightweight version CSPDarknet53-Tiny, the SPP structure was abandoned, and the PAN structure is replaced with FPN structure to improve the model detection speed. Secondly, in order to make up for the loss of detection accuracy, the attention mechanism model CBAM is introduced to improve detection accuracy. Finally, helmet detection experiment was carried out. The experimental results show that the detection speed of the lightweight model is up to 114.26FPS, and the detection accuracy of the helmet is 90%, which is suitable for deployment in actual construction scene.