Personal Identification Using Ultrawideband Radar Measurement of Walking and Sitting Motions and a Convolutional Neural Network

Abstract

This study proposes a personal identification technique that applies machine learning with a two-layered convolutional neural network to spectrogram images obtained from radar echoes of a target person in motion. The walking and sitting motions of six participants were measured using an ultrawideband radar system. Time-frequency analysis was applied to the radar signal to generate spectrogram images containing the micro-Doppler components associated with limb movements. A convolutional neural network was trained using the spectrogram images with personal labels to achieve radar-based personal identification. The personal identification accuracies were evaluated experimentally to demonstrate the effectiveness of the proposed technique.