Gait extraction from short radar dwells collected on dismounts

A technique is presented for estimating fundamental gait frequencies from a dismount observed by a radar. The proposed method differs from gait extraction schemes [1-2] that use the fast Fourier transform (FFT) to compute one fundamental gait frequency that gives rise to the dismount´s stride rate. In this approach, two fundamental gait frequencies are computed via a very versatile signal processing scheme, i.e., the state space method [3]. The first fundamental gait frequency corresponds to the side to side movement of the body of the walking dismount, whereas the second gait frequency describes the shift of the body weight from one leg to the other that induces forward and vertical displacements. The radar is assumed to have a very limited time to stare at the target while maintaining wide area surveillance and performing other tasks. Range-range-rate maps formed from data collected at few coherent processing intervals (CPIs) are used to generate a spectrogram snapshot at each dwell. The spectrogram snapshot is therefore passed to a limb mitigation filter that extracts the torso motion which is fed to a one-dimensional (1-D) signal processing scheme that computes a set of state matrices to estimate the gait frequencies and other parameters associated with the dynamic of the dismount. As time evolves, more spectrogram snapshots are generated and the gait frequencies are estimated to obtain the “fingerprint” of the dismount. The effectiveness of the gait extraction algorithm is confirmed using simulated radar data from the Thalmann human motion model [4] of a six foot tall man at six different aspect angles.