An Intelligent Edge-AIoT Framework for Real-Time COVID-19 Safety Compliance: Integrating Deep Learning, RFID Authentication, and Precision Thermal Sensing

Abstract

Existing IoT-based access control systems for COVID safety, such as the referenced “Covid Safety Guidelines Detection” prototype, suffer from high inference latency, coarse ambient temperature sensing, and limited scalability. In this work, we propose an intelligent IoT-based autonomous access-control framework that overcomes these gaps through real-time deep learning, edge computing, and precise thermal sensing. The system integrates a camera for mask detection, an RFID reader for user authentication, and a contactless IR temperature sensor for fever screening. We develop a multi-stage decision model D(R,C,T) combining RFID validity (R), mask-classification confidence (C), and measured temperature (T) against thresholds, with D=1 granting entry only if all conditions are met. A convolutional neural network (CNN) is optimized for mask vs. no-mask classification with cross-entropy loss  On-device (edge) processing on a Raspberry Pi 4 reduces round-trip delay. In experiments with 8,000 images (50% masked), our system achieves >98% mask-detection accuracy, 0.95 precision, and a mean inference latency of ~50?ms – 3× faster than a cloud-based baseline. Thermal sensing with an MLX90614 IR sensor attains ±0.5°C accuracy. By integrating real-time CNN inference, RFID authentication, and accurate body-temperature validation, our framework significantly improves upon prior work with rigorous mathematical modeling and quantitative validation. Key outcomes include high detection accuracy (>96%), low end-to-end latency (~60?ms), and robust real-world performance, demonstrating the feasibility of a scalable smart surveillance gateway.