The Research on Traffic Sign Recognition Algorithm Based on Improved YOLOv5 Model

Abstract

Traffic sign detection is an important research direction in target detection. At present, the detection problem of small target traffic signs exists, therefore, a traffic sign detection algorithm based on YOLOv5s model is constructed, and some important improvements are proposed to solve the small target detection problem. In order to ensure lightweight, YOLOv5s model is selected, and a small target prediction head is added to detect small targets. Fuse module is proposed to supplement shallow information to the backbone network to increase the ability of small target detection. BIFPN idea is used and improved to solve the problems of network depth degradation and shallow information deficiency. Finally, the Loss function is improved, and Varifocal Loss function is used to improve the problem of unbalanced positive and negative samples. The experimental results show that the detection effect of the proposed algorithm is increased by 7.7% compared with the original algorithm, and the experimental results on multiple datasets show obvious improvement effect. This is a lightweight model that works well in the field of traffic sign detection.