Machine Learning UHF-RFID to Support Video Tracking and Recommendation for Attendance System

This paper presents a machine learning-enhanced UHF-RFID system aimed at improving video tracking and attendance systems. Building on prior research, to mitigate doppler effects we propose a novel UHF-RFID reader configuration and evaluate its performance across various setups, including different tag types, covers, and reader power levels. Additionally, we also propose a synchronized attendance framework that integrates UHF-RFID and camera data collection analysis to enhance overall system efficiency. This comprehensive approach provides a robust solution for accurate and secure real-time attendance tracking, addressing the limitations of traditional methods and highlighting the importance of overcoming challenges posed by physical obstructions. We utilized supervised learning, Convolutional Neural Networks (CNN), and Light-weight CNN models to categorize the UHF-RFID tag dataset into distinct two classes based on the power levels. Specially, we consider and evaluate tags with school bags and personal physics applied fine-tuning CNN and Light-weight CNN models. Our experiment results for this scenario relatively boosted classification up to \(94.05\%\) and \(93.69\%\) in accuracy, respectively.