Multi-class Traffic Sign Recognition System Using One-Stage Detector YOLOv5s

Abstract

One of the crucial software elements in the upcoming generation of autonomous vehicles is image recognition. Traditional approaches to image recognition using computer vision and machine learning typically have a lengthy response time. Modern artificial neural network-based methods and designs, including the YOLOv5s algorithm, are able to tackle this issue without suffering precision losses. In this study, we demonstrate how to use the most recent YOLOv5s algorithm to identify traffic signs. We showed the reliability of the method by training the network for 4 traffic sign classes (speed limit, traffic light, crosswalks, stop,).