Deep Learning-based Open-set Person Identification using Radar Extracted Cardiac Signals.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Non-contact vital sign measurement using radar is gaining traction.
• Existing person identification methods often struggle with open-set conditions where test data includes unseen categories.

Purpose of the Study:

• To develop a novel deep learning algorithm for person identification using radar-extracted vital signs.
• To address the challenges of open-set person identification in real-world scenarios.

Main Methods:

• Extracting heart pulse signals from Doppler radar echoes.
• Employing transfer learning with two Convolutional Neural Network (CNN) models.
• Utilizing a distribution model for calibration and a strategic decision-making process for combining model outputs.

Main Results:

• Achieved 99.61% identification accuracy in close-set conditions.
• Demonstrated 94.35% identification accuracy in challenging open-set conditions.
• Outperformed existing person identification approaches.

Conclusions:

• The proposed deep learning algorithm offers a robust solution for radar-based person identification.
• The method shows significant promise for real-world applications, particularly in open-set scenarios.