Edge-Enabled Smart Traffic Surveillance System with Federated Learning Approaches

With urbanization accelerating and vehicular density rising, there is a growing demand for intelligent, scalable traffic surveillance systems. Traditional centralized approaches face challenges such as high latency, data privacy concerns, and limited network bandwidth, which hinder their effectiveness in real-time urban environments. This paper proposes an edge-based traffic monitoring framework that integrates Federated Learning (FL) with advanced object detection techniques tuned to indian traffic system. The system utilizes multiple versions of the YOLO (You Only Look Once) model specifically YOLOv8 through YOLOv12 trained on a GPU server and deployed on NVIDIA Jetson Xavier edge devices. These edge nodes perform real-time vehicle detection and periodically log summarized traffic data (class-wise) to a central server for future analysis. Federated Learning improves the global detection model by enabling local model updates without the need to transmit raw data, ensuring privacy and reducing communication overhead. This decentralized learning approach minimizes large-scale data transfers, making the system more efficient and secure. Experimental results show that the system achieves high detection accuracy and low inference latency, demonstrating its potential for real-time smart city applications.