Dual-Biometric Human Identification Using Radar Deep Transfer Learning.

Area of Science:

• Biometrics and Signal Processing
• Radar Technology
• Machine Learning for Security

Background:

• Current radar-based human identification relies on limited motion biometrics (micro-Doppler signatures).
• There is a need for more robust and accurate radar-based identification methods for applications like surveillance and access control.

Purpose of the Study:

• To propose and validate a novel human identification method using combined radar-based heart sound and gait signals.
• To achieve higher identification accuracy than existing radar-based techniques.
• To test the system's performance in a real-world access control scenario.

Main Methods:

• Extracting heart sound and gait signals from 18 subjects using radar.
• Converting biometric signatures into images and applying image augmentation.
• Utilizing deep transfer learning for subject classification.
• Combining heart sound and gait identification results using the joint probability mass function (PMF).

Main Results:

• Individual accuracies of 58.7% for heart sound and 96% for gait biometrics were achieved.
• The combined approach using joint PMF yielded a 98% identification accuracy, the highest reported to date.
• In an office access control test, the system achieved 76.25% accuracy.

Conclusions:

• Combining radar-based heart sound and gait biometrics significantly enhances human identification accuracy.
• The proposed method offers a promising, highly accurate, and novel solution for radar-based biometrics.
• The system demonstrates practical applicability in real-world security scenarios like access control.