A MIMO Radar-based Few-Shot Learning Approach for Human-ID

TL;DR

This paper introduces a MIMO radar-based few-shot learning method for human identification using micro-motion spectrograms, achieving high accuracy with minimal training data in real-time scenarios.

Contribution

It proposes a novel combination of micro-Doppler and elevation angular velocity spectrograms with adaptive segmentation and metric learning for efficient human ID.

Findings

Achieves 11.3% classification error with only 2 minutes of training data per subject.

Demonstrates effectiveness of concatenated spectrograms for improved classification.

Validates approach on a dataset of 22 subjects with various aspect angles.

Abstract

Radar for deep learning-based human identification has become a research area of increasing interest. It has been shown that micro-Doppler ($\mu$-D) can reflect the walking behavior through capturing the periodic limbs' micro-motions. One of the main aspects is maximizing the number of included classes while considering the real-time and training dataset size constraints. In this paper, a multiple-input-multiple-output (MIMO) radar is used to formulate micro-motion spectrograms of the elevation angular velocity ($\mu$-$\omega$). The effectiveness of concatenating this newly-formulated spectrogram with the commonly used $\mu$-D is investigated. To accommodate for non-constrained real walking motion, an adaptive cycle segmentation framework is utilized and a metric learning network is trained on half gait cycles ($\approx$ 0.5 s). Studies on the effects of various numbers of classes (5--20), different dataset sizes, and varying observation time windows 1--2 s are conducted. A non-constrained walking dataset of 22 subjects is collected with different aspect angles with respect to the radar. The proposed few-shot learning (FSL) approach achieves a classification error of 11.3 % with only 2 min of training data per subject.