A Lightweight Traffic Sign Recognition Model Based on Improved YOLOv5

Abstract

Traffic sign detection and recognition is a key technology for achieving automatic vehicle driving and maintaining road safety. The paper proposes a lightweight recognition algorithm based on YOLOv5 to address the problems of large model size, complex computation, low detection accuracy and high computational cost of existing traffic sign recognition algorithms. The algorithm is based on YOLOv5, replacing the convolutional structure in the original YOLOv5 neck network with Ghost Module and C3Ghost Module, thus reducing the redundant features in the feature fusion process, lowering the computational cost and the number of parameters; improving the PAN structure of the network and introducing the hybrid attention mechanism module CBAM, which ignores the unimportant information and capturing key information in traffic signs; adding cross-layer connections to the feature pyramid network shortens the path of information transfer, fuses more features, and improves the network feature recognition accuracy. In addition, the EIoU_Loss function is used as the bounding box regression loss function to improve the localization accuracy of the algorithm. The performance of the improved algorithm is also verified on the Chinese traffic sign dataset. The experimental results show that the improved algorithm improves the detection accuracy by 1.2% over the existing YOLOv5 algorithm, mAP@0.5 and mAP@0.5:0.95 by 1.5% and 3.4% respectively, and the overall number of parameters and computational cost of the model are reduced by 14.5% and 16%. The proposed algorithm has enhanced recognition capability for targets in multiple environments and can meet the requirements for real-time traffic sign recognition.