The improvement of YOLOv8 for detecting railway workers’ safety helmets and safety vests based on GhostConv

In this paper, we present an enhanced YOLOv8 model specifically tailored for the detection of railway workers’ safety helmets and safety vests. The proposed improvement incorporates GhostConv layers into the YOLOv8 architecture, aiming to reduce the computational cost while maintaining high detection accuracy. GhostConv, known for generating more feature maps with fewer parameters, helps in capturing fine-grained features relevant to safety gear detection. Extensive experiments conducted on a custom dataset of railway workers demonstrate that the enhanced model outperforms the baseline YOLOv8-n in both detection accuracy and processing efficiency. Notably, the proposed model achieves a 3.1% increase in mAP50 and a 2.6% improvement in recall, while also reducing the parameter count by 1.4M and the model weight by 2.5MB. Additionally, it surpasses SSD with a 15.8% improvement in mAP50 and significantly lower computational demands. These advancements offers the practical solution for real-time monitoring systems to ensure the safety of railway workers by reliably detecting essential safety equipment.