Traffic Sign Recognition Based on Improved YOLOv5s
Jin Zhang, Fan Cao, Jianming Zhang, Ying Wai Li, Sheng Yi Wu · 2023
In the field of autonomous driving, it is of great significance for autonomous driving to detect traffic signs accurately. This paper proposes a method named MR_YOLO. This method addresses the issue of inadequate accuracy that plagues current systems. More specifically, to address varying sizes of traffic signs in images, our strategy refined anchor box dimensions using K-Means++ clustering and fine-tuned the model to enhance traffic sign recognition efficacy. Next, a multi-scale feature extraction model, C3Res2Net, is formed by combining the C3 and Res2Net models to address the different scales and sizes of traffic signs. This approach enables the network to accurately capture comprehensive information and intricate details of traffic signs, thus enhancing recognition capabilities. Experimental data reveal competitive outcomes when tested on the CCTSDB dataset. The method achieves values of 92.8% and 88.4% in precision and mAP, respectively.