Safety Helmet Wearing Detection Based on YOLOv7

Safety helmets are essential protective equipment in construction sites, so it’s imperative to implement immediate detection and alert systems for monitoring construction workers’ adherence to wearing safety helmets. Traditional methods for helmet wearing detection often overlook the impact of complex backgrounds on detection results, and small background objects can disrupt the detection outcomes, leading to a high false positive rate. However, Deep learning-based helmet wearing detection methods exceed in adapting to environmental variations but often neglect the relationship between safety helmets and individuals. This paper introduces a safety helmet detection system utilizing the YOLOv7 algorithm, and it’s aimed at detecting hazardous situations that people are not wearing safety helmets in the monitoring area. The detection process involves using the YOLOv7 algorithm to extract features from the input image, followed by predicting the objects in the test image, next assigning scores to each bounding box, and finally outputting the bounding box with the highest score and its corresponding category onto the test image. Experiment results on the self-constructed dataset demonstrate that using the proposed model for the safety helmet wearing detection achieves an mAP of 96.33%, which outperforms the current state-of-art models.