Navigating the Future with YOLOv9 for Advanced Traffic Sign Recognition in Autonomous Vehicles

Abstract

In the realm of autonomous and self-driving vehicles, accurate traffic sign detection is critical for ensuring road safety, efficient navigation, and compliance with traffic regulations. This paper presents an advanced traffic sign detection system based on YOLOv9, an enhanced form of the YOLO (You Only Look Once) architecture. YOLOv9 offers significant enhancements over its predecessor, YOLOv8, through advanced feature extraction, multi-scale feature fusion, and optimized detection heads. The suggested YOLOv9 variant provides a notable accuracy of 95.0%, surpassing YOLOv8's 90.5%. This improvement is complemented by enhanced performance metrics, including a precision of 93.0%, recall of 94.0%, and an F1 score of 93.5%, compared to YOLOv8's precision of 88.0%, recall of 87.5%, and F1 score of 87.7%. The mean Average Precision (mAP) also increases from 85.5% in YOLOv8 to 91.0% in YOLOv9, reflecting superior detection and classification capabilities. The YOLOv9 model demonstrates superior efficiency with reduced training time (12 hours compared to YOLOv8's 15 hours) and faster inference (30 ms compared to YOLOv8's 40 ms). It utilizes a more comprehensive dataset with a greater number of images, traffic sign classes, and varied conditions, enhancing its robustness and generalization in real-world scenarios. Key parameter adjustments, including a lower learning rate, smaller batch size, and refined IoU threshold for non-maximum suppression, contribute to YOLOv9's improved performance. These enhancements make YOLOv9 a highly effective solution for real-time traffic sign detection in autonomous driving systems, offering a safer and more efficient driving experience. This work demonstrates the potential of YOLOv9 in advancing traffic sign detection technologies and provides a solid structure for further R&D in autonomous vehicle systems.