Edge-Intelligent Deep Learning-Based Traffic Sign Detection for Autonomous Vehicles Under Real-World Conditions

Reliable traffic sign detection remains a critical unsolved challenge for autonomous vehicle perception systems, where failures under real-world conditions directly compromise passenger safety. This paper presents a deep learning-based traffic sign detection system using YOLOv8 fine-tuned on a carefully selected subset of the Mapillary Traffic Sign Dataset (MTSD), filtered and preprocessed into 16 safety-critical sign classes from over 100,000 street-level images. The model achieves 75.5% precision, 80.0% recall, and 61.9% mAP@0.5 across challenging conditions including variable lighting, adverse weather, motion blur, occlusion, and sign deterioration. The system is deployed as a real-time dashcam-style inference pipeline with bounding box overlays, confidence scoring, annotated video output, and adaptive day/night exposure modes. Validation on a Raspberry Pi and Pi camera confirms practical edge deployability, demonstrating a viable perception solution for intelligent transportation systems.