Contrast Stretching For Automatic Road Marking Detection at Night With YOLOv5

The ability to recognize and keep track of road signs is one of the most significant obligations that a visual driver assistance system must fulfill. The study of deep learning and recognition of road markings has made tremendous progress in recent years thanks to advances in technology. Road surface markings cover a wide range of features, including crossings, directional arrows, zebra crossings, and other signs. The surface of the road is immediately painted with these marks. In this study, to detect the road marking sign, we implement Contrast Stretching (CS), as well as YOLOv5s. Often referred to as "normalization," contrast stretching is a straightforward method for boosting an image's contrast by artificially expanding its intensity range. We build the dataset by ourselves and focus on Taiwan road marking sign dataset at night (TRMSDN). Our results show that when compared to YOLOv5s trained on the original image, YOLOv5s trained on the CS image performs better in both training and testing. Furthermore, YOLOv5s exhibits 87.25% mAP during training and 83.83% mAP in testing phase.