Tiny-YOLOv8 long-distance electric vehicle helmet detection model

Aiming at the problem of false positives and false negatives of electric vehicle helmets worn at the edge, far away, or with a small proportion in complex road scenes by traditional target detection networks, an electric vehicle helmet detection model, Tiny-YOLOv8, based on the improved YOLOv8n algorithm, is proposed. A tiny object detection layer module is added to enhance the model’s utilization of detail features, and Coordinate Attention (CA) mechanisms are embedded at the end of the backbone network to enhance the network’s ability to obtain small target location information. The EIOU loss function is used to replace the traditional bounding box regression loss, and the convergence speed and detection accuracy of the model are further improved by optimizing the actual differences in center point distance, overlapping area, and aspect ratio. This study optimizes the model parameters based on the training dataset of the Helmet-Plate-Detection (HPD) project, and finally constructs the TinyYOLOv8 long-distance helmet detection model. To prove the accuracy of the prediction model, comparative experiments are conducted on the mainstream networks SSD, YOLOv5s, HPD, and YOLOv8n. According to the experimental results, the TinyYOLOv8 model achieves $98.3 \%, 98.9 \%$ and $99.4 \%$ in precision, recall, and mean average precision (mAP), respectively. Compared with the original basic network, the improved model shows a $1.2 \%$ increase in mAP, and its parameters have been reduced by $\mathbf{0. 1 M}$. It is verified that the Tiny-YOLOv8 model can effectively optimize the detection performance, realize multi-scale feature fusion, significantly improve the positioning accuracy and recognition effect of small targets, and provide strong protection for road traffic safety.