Research on traffic sign recognition under complex meteorological conditions

Traffic sign recognition and classification is among the most significant technologies used in autonomous driving. In view of the fact that the identification accuracy of traffic signs is insufficient owing to multi-scale allocation of traffic signs under a variety of diverse meteorological circumstances, an improvement method of traffic signs based on Yolo model was proposed. Firstly, the first 16 traffic signs were selected according to the quantity of images for every type of traffic sign in LISA dataset. Considering the large requirements for the real-time recognition of traffic signs in unmanned driving, the Yolo v5 model with both efficiency and accuracy in the Yolo series model was used to recognize the image, and on this basis, the C3 module in the model was modified to the C2f module to enhance the integration of traffic sign features and the network's capability to express these features in complex environments, and the EIoU loss function was employed to refine results, and it was found that the mAP and other indicators were improved as well as recognition accuracy was higher. In addition, the paper also compares the previous work utilizing the LISA dataset and reveals enhancements in the accuracy, recall, and F1-score.