Beyond Contact: An Open-Set Biometric Identification System Using Radar-Extracted Heart Signals.

Area of Science:

• Biometrics and Human-Computer Interaction
• Signal Processing and Machine Learning
• Radar Systems Engineering

Background:

• Traditional biometrics (fingerprint, facial recognition) face privacy, spoofing, and proximity limitations.
• Non-contact identification methods are needed for secure and privacy-conscious scenarios.
• Radar technology offers potential for remote physiological signal extraction.

Purpose of the Study:

• To propose and evaluate a novel radar-based framework for non-contact biometric person identification.
• To reconstruct electrocardiogram (ECG) signals from radar-detected cardiac motion for identification.
• To implement a robust open-set identification system capable of distinguishing known from unknown individuals.

Main Methods:

• Integration of ECGReconNet for reconstructing ECG signals from chest wall displacement data captured by radar.
• Utilized an enhanced InceptionTime model with fixed-Class Anchor Clustering (fixed-CAC) loss for feature extraction.
• Employed a hypersphere-based method for open-set recognition, differentiating known and unknown subjects.

Main Results:

• Achieved state-of-the-art performance on a public dataset.
• Demonstrated 99.61% accuracy in closed-set identification (27 subjects).
• Obtained 93.97% accuracy in challenging open-set conditions (14 known, 13 unknown subjects).

Conclusions:

• The proposed radar-based framework enables accurate non-contact biometric identification via ECG signal reconstruction.
• The system shows high efficacy in both closed-set and open-set identification scenarios.
• Limitations include sensitivity to motion artifacts, noise, severe cardiac irregularities, and a large number of unknown identities.