Traffic Sign Detection System Using YOLOv11 Within Intelligent Transport Systems for Enhanced ADAS Performance with Deep Learning

Abstract

Real-time accurate traffic sign detection is among the key affluences to Advanced Driver Assistance Systems (ADAS) safety and efficiency. The paper presents a new Traffic Sign Detection System based on the latest YOLOv11 architecture, benchmarked against the 23,623-image CCTSDB. YOLOv11 provides superior detection with precision at 98%, recall at 98.8%, F1-score at 98.4%, and superior mAP scores of 99.3% at mAP@50 and 85.7% at mAP@50- 95. The key improvements to the architecture are the pre-training of the CSPDarknet backbone, the optimization of PANet feature fusion across scales, and depthwise separable convolutions, which far surpass the performances of the earlier models of YOLO. These improvements illustrate the potential of YOLOv11 to promote reliability and safety in real-time driver-assisting systems.