Small-target traffic sign detection based on improved YOLOv8

To address the challenges of high proportions of small traffic signs and significant environmental interference in road traffic scenarios, an improved YOLOv8 network specifically designed for small traffic sign detection is proposed. We first integrated BRA based on the Transformer architecture into the C2f network structure of YOLOv8n. This integration aims to reduce missed detections of small objects and improve the network’s perception of small targets. In addition, CARAFE was employed to retain complex image feature information and reduce the loss of target information without significantly increasing model complexity and parameters. This approach enhances the model’s accuracy and generalization capability. The inner-IOU loss function based on auxiliary bounding boxes was also introduced. The improved method was trained on the public traffic sign dataset CCTSDB, achieving a detection accuracy of 97.5%, which is 1.8% more than the original algorithm, and it also increases detection speed by 11 FPS. Experimental results indicate that the improved YOLOv8 network effectively detects traffic signs in complex road scenarios.