ABSTRACT
Human gait classification using µ-D signals has numerous potential civilian and military applications. This chapter presents a µ-D feature extraction method for radar human gait classification. The local dynamics of human motions embedded in the Doppler signal are captured using a joint time–frequency (T–F) representation. The arm and leg motions, which are periodic, induce µ-D modulations near the torso frequency shift. Therefore, local T–F patches centered on the torso frequency shift are extracted for feature extraction, thereby introducing some tolerance to variations in target speed. Then, each patch is convolved with a set of log-Gabor filters to detect discriminative features, which are then compressed by a dimensionality reduction technique. The chapter presents the basic mathematical description of µ-D effect from a point scattering target, followed by the description of three T–F analysis techniques, namely Wigner–Ville distribution, short-time Fourier transform, and S-method used to analyze Doppler signals. Then, it reviews several existing methods for classifying human µ-D signals.
