Deep Learning for Automatic Road Marking Detection with Yolov5

Abstract

One of the most important responsibilities of a visual driver aid system is recognizing and tracking road signs. In recent years, tremendous progress has been made in both deep learning and the identification of road markings. Pedestrian crossings, directional arrows, zebra crossings, speed limit signs, and similar signs and text are all road surface markings. These markings are painted directly onto the surface of the road. This paper implements YOLOv5s and YOLOv5m to identify the road marking sign. We built a dataset and focused on the Taiwan road marking sign. According to the findings of our experiments, YOLOv5m contains eleven categories of whose training accuracy is superior to that of YOLOv5s. It has been discovered that the YOLOv5m model is the most accurate, scoring 87.30 percent overall throughout testing, while the YOLOv5s model scores an average of 83.60 percent.