Traffic Sign Detection and Recognition Based on Improved YOLOv4 Algorithm

Abstract

Traffic sign detection and recognition play an important role in intelligent transportation. In this paper, a traffic sign detection framework based on YOLOv4 is proposed. The original CSPDarkNet53 backbone network model is replaced by RepVGG, and the SPP module is added in the feature pyramid part to improve the expression ability of information. The CCTSDB traffic sign data set is used to detect three categories of indication signs, prohibition signs and warning signs. In order to further improve the performance of YOLOv4 network, K-means++ algorithm was used to perform cluster analysis on the experimental data to determine the size of the priori box suitable for CCTSDB dataset. The experimental results show that the map value of the improved framework is increased by 4.1%, which indicates that the improved YOLOv4 network has a high practical value in traffic sign detection and recognition.