Short-Duration Doppler Spectrogram for Person Recognition with a Handheld Radar

Abstract

This paper examines the classification of walking, standing and mirrored persons based on radar micro-Doppler (m-D) measurements to resolve ambiguities in thermal infrared (TIR) mirror images in firefighting. If the walking or standing person is observed directly, its m-D is measured. In the case of a person mirrored on a reflecting object, only the m-D of the reflecting object is measured. Their spectrogram is differentiable which enables a classification. One difficulty is the random movement of the handheld radar which leads to short observation durations and Doppler blurring. A classification based on short spectrograms is proposed, where the influence of the short-time Fourier transform window length is investigated. Furthermore, a regularization is proposed to improve the classifier interpretability for this safety application.