Traffic Lights Detection and Recognition Method using Deep Learning with Improved YOLOv5 for Autonomous Vehicle in ROS2

Abstract

One of the most significant uses of autonomous cars in recent years is the detection of traffic light signals. Deep learning technology, which has a number of benefits including high detection accuracy and quick response to changes, is supporting the development of traffic light recognition under various environmental situations. In this paper, we use two methods to improve the traffic light detection and recognition method. First, we speed up training time by using the K-means clustering algorithm to compress image data. Second, a real time traffic light signal (red, yellow, green) identity based on the You Only Look Once (Yolov5) model is introduced. We utilised a variety of datasets including a freely available Roboflow dataset, a set of data obtained from Gazebo simulator, and a traffic light of CanTho city dataset to train and evaluate the proposed system. Furthermore, our algorithm was validated on a vehicle model in a simulated environment Gazebo of Robot Operating System 2 (ROS2).