Real-Time Object Detection for Road Safety A Yolo-Based Approach for Sustainable Cities

This paper presents the development and deployment of a high-performance road safetydetection system using the YOLOv8 architecture to enhance urban traffic safety, directlysupporting Sustainable Development Goal 11 (Sustainable Cities and Communities). Thesystem provides real-time detection of vehicles, pedestrians, and traffic infrastructure fromroad scene images. A comparative analysis of YOLOv8 model variants (Nano, Large, andExtra-Large) was conducted to determine the optimal balance between inference speed anddetection accuracy. The YOLOv8x model was selected for the final deployment, achieving anaverage inference time of 1,227ms while detecting an average of 32 objects per scene.However, a critical analysis revealed a significant challenge in pedestrian detection, with a57.1% high-confidence detection rate (8 out of 14 pedestrians), meaning 42.9% require humanverification, highlighting the limitations of current computer vision technology for safetycriticalapplications. The system was deployed as a web application using Streamlit and hostedon Hugging Face Spaces, demonstrating a modern MLOps workflow. This research providesvaluable insights into the practical application of deep learning for road safety, emphasizingthe ethical considerations and the need for multi-modal sensor fusion to overcome thelimitations of purely vision-based systems.