Extremely Low Frequency Detection for Biometric Sensing
Date
2023-09-10
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Human detection is an important aspect of security and surveillance and is a highly researched area, typically using cameras, infrared sensors, or radar. Extremely Low Frequency (ELF) electromagnetic waves, ranging from 0.03 Hz to 300 Hz, offer unique advantages for human detection due to their ability to penetrate various materials, including walls and obstacles. The human body generates a weak magnetic field due to various physiological processes and is measurable within the ELF range. The goal of this thesis is to detect ELF signals emitted by human subjects using an ELF magnetometer and to rapidly distinguish the presence of a human from background noise. A large air-core magnetic induction coil is designed and constructed in conjunction with a low-noise, high-impedance amplification circuit to independently measure the ELF signals from background noise and two human subjects in a semi-remote, non-shielded location. A sampling frequency of 44.1 kHz was used, and the power spectrum density (PSD) was computed for each 3-second data segment. Using a Recurrent Neural Network (RNN) structure, several binary classification models were trained with supervised learning on the PSD sequences between 0.3 and 30 Hz to distinguish background data from human data. The candidate RNN models achieved an average accuracy of above 80% using K-Fold cross-validation on the limited dataset. Overall, the performance of the RNNs was high for one subject, and low for the other subject. It was found that a significant factor in distinguishing a human from background noise was the change in diurnal ELF background power levels between measurements. The results indicate the use of ELF data from humans has the potential for rapid human detection, however, requires a larger dataset for further consideration as a biometric device.
Description
Keywords
extremely low frequency (ELF), magnetometer design, human detection, recurrent neural networks (RNNs)
Citation
White, O. (2023). Extremely low frequency detection for biometric sensing (Master's thesis, Universty of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.