Human Gait Classification Based on Convolutional Neural Network using Interferometric Radar

Abstract

Doppler radars are capable of measuring the micro-Doppler signatures of a walking person for human gait analysis. However, when the person's trajectory is approaching 60° away from radar antenna broadside, the micro-Doppler signatures due to radial velocity diminish significantly, thus providing insufficient information for classification. To resolve this issue, we propose a new algorithm for human gait classification based on convolutional neural network (CNN) using interferometric radar. We implement the classification algorithm based on dual micro-motion signatures using interferometric radar which allows the measurement of both the interferometric and micro-Doppler signatures due to the angular and radial velocities of the walking person, respectively. Three human gait classes including full arm motion (FAM), partial arm motion (PAM) and no arm motion (NAM) at two different angles between the radar antennas baseline and direction of motion are considered for dataset generation. Time-varying Doppler and interferometric spectrograms are fed as input to train CNN. Simulation results validate the fact that the proposed classification algorithm using interferometric radar enhances the performance of human gait classification algorithm in terms of accuracy and robustness, regardless of the trajectory.