A YOLO-Powered Edge Computing Framework for Low-Latency, Privacy-Preserving Virtual Environment Monitoring

Abstract

Modern urban environments rely heavily on real-time monitoring systems for traffic management, security, and smart city operations. Traditionally, these systems followed a cloud-centric approach where surveillance cameras continuously captured video and transmitted raw data to centralized cloud servers for processing. Although effective, this approach suffers from high latency, excessive bandwidth consumption, and privacy concerns due to the transfer of sensitive video data. These limitations make traditional systems inefficient for time-critical applications and difficult to scale in resource-constrained environments. Unlike traditional methods, the system performs data processing locally on an edge device, reducing dependency on cloud infrastructure. The core of the system is the YOLOv7 (You Only Look Once) deep learning model, which is based on Convolutional Neural Networks (CNN). This model is trained on traffic datasets to detect, classify, and count vehicles in real time. The performance of the model is evaluated using metrics such as accuracy, precision, recall, and F1-score, ensuring reliable detection results. Instead of transmitting raw video, the system sends only processed data such as vehicle count, location, and timestamp to the cloud using socket-based communication. A comparison between cloud-based and edge-based approaches demonstrates that the proposed system significantly reduces latency and bandwidth usage. The project is implemented using Python and deep learning frameworks, providing an efficient, scalable, and privacy-preserving solution for smart city applications.