Traffic Sign Detection Based on Improved YOLOv5-S

This paper proposes an improved YOLOv5-S traffic sign detection model to address the issues of missed and false detections in existing traffic sign detection models in complex backgrounds. Firstly, a small-sized detection layer was added to enhance the model's utilization of detailed shallow feature information; Secondly, a C3-Res2Block module that integrates multi-scale information is proposed to improve the network's multi-scale perception ability and better adapt to targets and scenes of different sizes, thereby improving recognition accuracy and robustness; Once again, introduce the CBMA attention mechanism to focus on key information and enhance the feature expression ability of traffic signs; Finally, to address the challenge of localizing small targets in loss calculation, the NWD(Normalized Wasserstein Distance) loss function was introduced. Experiments confirm the improved model's efficacy in [email protected] and [email protected]:0.95, it achieved 85.2% and 64.0% respectively, while achieving a detection speed of up to 108FPS, meeting the real-time and reliable detection requirements.