VR-TSD: A Real-World Dataset and a Lightweight YOLOv8n Model for Traffic Sign Recognition in Vietnam

Accurate and real-time traffic sign recognition is a key component for Advanced Driver Assistance Systems (ADAS) and autonomous vehicles, particularly when deployed on resource-constrained edge devices. However, existing solutions often face challenges in Vietnam’s traffic environment due to the lack of datasets that truly reflect local conditions. To address this gap, we introduce the Vietnamese Real-World Traffic Sign Dataset (VR-TSD), a novel dataset comprising 58 common traffic sign categories collected under diverse and challenging scenarios, including low-light conditions, rainy weather, and partial occlusions. Based on this dataset, we propose a lightweight recognition system built on the YOLOv8n architecture, optimized to balance accuracy and efficiency. Experimental results demonstrate strong performance, achieving a mAP@0.5 of 0.9434, processing speeds of 168.86 FPS on an NVIDIA A100 GPU and 77.52 FPS on a smartphone (iPhone 11 Pro Max), with a compact model size of just 6 MB. These findings confirm the feasibility and effectiveness of the proposed system for real-world applications, paving the way for ADAS development tailored to Vietnam’s traffic conditions.