Traffic sign recognition for complex traffic condition based on YOLOv5

Aiming at the traffic sign recognition problem of complex road conditions in intelligent driving scenarios, a traffic sign recognition network based on the improved YOLOv5 was proposed. By changing the neck structure of the network, BiFPN is used to delete the single input node in the feature transfer mode, which improves the recognition speed, and adds paths between the feature nodes of the same layer, which enriches the semantic information of the feature map, and has a good recognition effect on traffic signs in complex road conditions. The mAP of the improved recognition network reaches 90.2% and the recognition speed is 69.3 FPS, which is 3.6-percentage-point and 8.7 FPS higher than that of the original network.