RCW-YOLO: Improved traffic sign detection model based on YOLOv8 for complex scenes

Aiming at the low accuracy of traffic sign detection as well as the problem of leakage and false detection in complex scenes at this stage, this paper proposes an improved RCWYOLO traffic sign detection model based on YOLOv8s. First, the Receptive-Field Convolutional Block Attention Module Convolution (RFCBAMConv) is used to replace the standard convolution in the original model’s backbone and neck networks to help the model come to focus on the detailed features of the traffic sign images. Second, in order to make the model better equalize traffic sign samples with different qualities, the Wise Intersection over Union (WIoU) loss function is used instead of the original Complete Intersection over Union (CIoU) to further improve the generalization of the model for traffic sign detection. The experimental results show that compared with the original YOLOv8s on the challenging traffic sign datasets TT100K and CCTSDB-2021, the detection accuracies of RCW-YOLO are improved by 3.5% and 1.3%, respectively, while maintaining the real-time performance.