Performance Evaluation of Advanced YOLO Models for Road Marking Detection

Road marking identification is a highly essential part of intelligent transportation systems and autonomous vehicles, as it directly influences route guidance and road safety. However, due to varying light sources, weather conditions, faded paint, and complex road geometry, their identification is a challenging task for computers and machines to perform precise analysis and deduction. This work conducts a comparative analysis of some new YOLO-based techniques for tracking and identifying objects, YOLOv8, YOLOv9, YOLOv10, and YOLOv11, on a road marking identification dataset containing varied real-time situations on roads. The experiment on all techniques is conducted under similar settings for consistent analysis and comparison. The performance evaluation criteria for assessing their efficiency include precision, recall, F1-score, mAP value with a 0.5 IoU requirement, and accuracy. The accuracy for road marking identification reaches 93% with YOLOv11, which has notably superseded its predecessors by significant margins, thanks to improvements in architecture, the use of anchor-free techniques, new learning methods, and end-to-end testing and analysis for enhanced accuracy and precision. This analysis clearly predicts that YOLO-based techniques, especially YOLOv11, are exceptionally efficient and well-suited for real-time, precise road marking identification in advanced driver assistance systems, automated vehicles, and other road vehicles, thereby enhancing road safety and efficiency.