Image Enhancement Method Utilizing YOLO Models to Recognize Road Markings at Night

Abstract

Artificial intelligence has been rapidly developed and implemented across numerous industries in recent years. A notable advancement is the enhancement of transportation modalities. Vehicle accidents often lead to a significant number of deaths and cause substantial economic harm. Self-driving automobiles can utilize road detection as one of their applications. While traffic accidents are unfortunate, numerous countries employ artificial intelligence to develop intelligent urban areas and applications for autonomous vehicles. Extensive road sign detection and analysis research has been conducted using public road sign datasets. Consequently, these datasets hold immense importance in the training of autonomous cars. This study documents the road networks of different cities in Taiwan through road driving. The process involves manually collecting traffic signs throughout Taiwan to compile a dataset that includes road signs in both daylight and nighttime settings. This study is crucial for creating a thorough data collection for traffic signs in Taiwan due to the need for existing ones. This work utilizes the YOLO model to recognize road signs in Taiwan. The techniques of Contrast Stretching (CS), Histogram Equalization (HE), and Contrast Limited Adaptive Histogram Equalization (CLAHE) are assessed and contrasted with the original nighttime image. The experimental data indicates that YOLOv8l performs the best during nighttime, achieving an accuracy of 90%, precision of 89%, and recall of 90%. This study recommends using the CLAHE picture technique with the YOLOv8l model for the best results. The proposal suggests using CLAHE’s YOLOv8l as the optimal model for nighttime road sign recognition to improve its efficiency. This study makes a valuable contribution to the overall objectives of autonomous driving and intelligent traffic management systems by improving the precision and dependability of road sign identification. Enhancing the ability to recognize road signs is of utmost importance for the advancement of fully autonomous vehicles. In conclusion, the results of this study can contribute to the development of safer and more effective roads, which will have positive impacts on both individual drivers and the wider community.

Keywords

• CLAHE
• CNN
• deep learning
• YOLOv8
• road sign detection