Non-contact human identification through radar signals using convolutional neural networks across multiple...

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Contactless identification methods are increasingly important for security and convenience.
• Radar-based identification is a novel, non-intrusive technology that uses heart dynamics.
• Limited research exists in radar-based subject identification.

Purpose of the Study:

• To investigate the feasibility of secure subject identification using continuous wave radar-acquired heart dynamics.
• To explore identification across multiple physiological scenarios.
• To compare traditional machine learning with deep learning approaches for radar-based identification.

Main Methods:

• Proposed and compared a traditional machine learning pipeline with a deep learning approach (CNN-SVC).
• Extracted features from scalograms using a Convolutional Neural Network (CNN).
• Assessed system generalizability across various physiological scenarios, evaluating performance with known and unknown physiological states.

Main Results:

• The deep learning method achieved 97.70% accuracy in the Resting scenario, outperforming the traditional pipeline.
• Across diverse physiological scenarios, 82% of predictions met confidence thresholds, yielding 98.6% accuracy in this subset.
• Demonstrated high accuracy and generalizability of radar-based identification.

Conclusions:

• Radar-based identification systems can achieve performance comparable to established biometrics like ECG and PPG.
• Offers the advantage of contactless, hygienic, and seamless identification.
• Radar heart signal analysis is a promising solution for secure human identification in various conditions.