Dynamic Optimization in AI-Powered Traffic Prediction Models for Smart Cities

Traffic congestion poses significant challenges in rapidly urbanizing areas, especially on Indian roads with diverse vehicle types and mixed traffic conditions. This paper presents "Urban Predict," an AI-powered traffic management system utilizing deep learning for real-time object detection and traffic prediction. The system combines YOLO (You Only Look Once) for detecting vehicles, pedestrians, and road hazards from camera feeds with a Spatio-Temporal Graphical Model (STGM) to forecast traffic patterns. Urban Predict enhances prediction accuracy by integrating external factors, such as weather conditions from APIs. It is deployed in a scalable cloud environment and employs AWS for real-time data streaming, Docker for containerized deployment, and Kubernetes for microservices management. Security features detect hazardous road conditions and facilitate dynamic traffic rerouting, promoting safer urban travel. Performance metrics, including detection, prediction accuracy and latency, demonstrate the system’s effectiveness as a real-time solution for urban traffic congestion. Urban Predict represents a significant advancement in intelligent traffic management, providing valuable insights for future smart city applications tailored to the unique demands of urban environments in India.