Traffic Sign Detection in Autonomous Driving: Optimization Choices for YOLO Models
Junyao Wu · 2024
Autonomous driving technology is advancing at an unprecedented rate, yet its immaturity has led to a series of traffic accidents. As a critical component of autonomous driving technology, traffic sign detection is important for ensuring vehicle safety. Improving the accuracy of traffic sign detection not only reflects technological progress but also demonstrates a commitment to safety. This paper identifies the most suitable YOLO model for traffic sign detection by training and comparing the performance of three YOLO models. It was found that YOLOv8 outperforms YOLOv9 and YOLOv7 in terms of precision, recall, average precision at an IoU threshold of 0.5, and mean average precision across IoU thresholds from 0.5 to 0.95, with YOLOv8 being 0.6, 0.1, 0.4, and 2.1 percentage points higher than YOLOv9, and 12.0, 25.3, 19.8, and 29.8 percentage points higher than YOLOv7, respectively. However, YOLOv8 exhibits detection omissions when predicting overlapping targets. Therefore, it is concluded that YOLOv8 is optimal for non-overlapping traffic signs, while YOLOv9 is preferable for overlapping signs.