BCRP-YOLOv8n: A Traffic Sign Detection Model Based on Improved YOLOv8n Under Complex Scenarios

Traffic sign detection is an important functional component of autonomous driving systems and plays a crucial role in road environment perception, assisted driving decision-making, and traffic safety assurance. However, under complex scenarios such as occlusion, small objects, low-light conditions, and foggy weather, traffic sign detection is prone to missed detections and false detections. To overcome these challenges, this paper proposes an improved YOLOv8n model, named BCRP-YOLOv8n, for traffic sign detection in complex scenarios. First, the original PAN structure is replaced with the Concat_BiFPN bidirectional weighted feature fusion network to strengthen multi-scale feature information flow and weight allocation, thereby improving the model’s perception of target location information and edge details. Second, the CBAM attention mechanism is introduced to enhance the feature representation of key targets and suppress interference from complex backgrounds. Third, a RepC2f re-parameterization module is designed to replace part of the C2f modules. By integrating the feature concatenation mechanism with the re-parameterized convolution structure, the RepC2f module is used to enrich feature representation during training and is converted into an equivalent single-branch structure for inference, thereby avoiding the need to retain the training-time branches during deployment. Finally, a P2 detection layer is added to expand the perception range for small objects and more fully preserve their edge and texture information, thus significantly improving the detection performance for small traffic signs. Comparative and ablation experiments were conducted on the TT100K traffic sign dataset. The experimental results show that BCRP-YOLOv8n achieves a precision of 82.3%, a recall of 76.3%,an mAP@0.5 of 84.3%,and an mAP@0.5:0.95 of 64.9%,representing improvements of 5.6, 11.9, 12.2, and 8.6 percentage points, respectively, compared with the original YOLOv8n. In addition, visualization analysis of detection results was performed on complex-scenario samples from the TT100K dataset and the self-built RTSD-C dataset. The experimental results further demonstrate that the proposed model exhibits stronger robustness in complex scenarios such as occlusion, small objects, low-light conditions, and foggy weather.