XÂY DỰNG HỆ THỐNG CẢNH BÁO ÙN TẮC GIAO THÔNG ĐÔ THỊ THEO THỜI GIAN THỰC

Abstract

This study presents the design and development of a real-time urban traffic congestion alert system. Motivated by severe congestion and limitations of existing solutions in Vietnam, the study proposes a distributed four-layer architecture comprising camera data acquisition, AI processing, backend services and multi-platform applications. Using a YOLOv8n model trained on a dataset of 20,000 traffic images from Da Nang, the system achieves a mean average precision (mAP@0.5) of 92.4% and a processing speed of 45 FPS on an RTX 3060 GPU. The backend, built with Node.js, PostgreSQL and Redis, concurrently handles up to 25 cameras with an average API latency of 145ms. The web application (React/Vite) delivers 11 functional pages, while the mobile app (Flutter) provides 16 screens on iOS and Android. A simulation-based experiment on 15 camera streams in Da Nang recorded over 800 alerts with accuracy exceeding 90%. These results confirm the system's feasibility within an undergraduate research context and its practical potential to reduce congestion, save travel time and enhance urban traffic safety.