Towards Sustainable Smart Transportation: Lightweight YOLO-Based Traffic Sign and Light Recognition with CLAHE-Aided OCR

Ensuring traffic safety remains a pressing challenge in Vietnam, where inconsistent signage and frequent violations continue to cause thousands of accidents annually. This study proposes a lightweight and practical edge-AI framework for real-time recognition of traffic signs and signals, integrating advanced YOLO-based detection with optical character recognition (OCR) and contrast enhancement techniques. We conduct a comprehensive evaluation of YOLOv8 and the latest YOLOv11 models (nano, small, and medium variants) on a custom traffic dataset of 1,689 annotated images comprising traffic lights and speed-limit signs collected from real-world Vietnamese roads. The YOLOv11s model achieves superior detection accuracy with mAP@0.5 of 97.8% and mAP@[0.5:0.95] of 0.683, while the application of CLAHE solely during inference further improves performance to 0.703, especially under low-light conditions. For speed-limit text recognition, EasyOCR significantly outperforms Tesseract, demonstrating greater robustness in real-world scenarios. Unlike prior studies focusing solely on object detection, our pipeline introduces a two-stage vision system that couples detection with OCR-based semantic interpretation, enabling more reliable speed sign reading and violation analysis. This approach provides a scalable and efficient solution for smart transportation systems, supporting driver assistance, automated enforcement, and broader deployment in resource-constrained urban environments.