Human Action Recognition With Raw Millimeter Wave Radar Data

Abstract

Prediction of human movements is a current hot topic that researchers are working on. The most common technologies such as image or wearable inertial sensors are not preferable due to privacy issues or subject discomfort. In this study, we propose to use FMCW (Frequency Modulated Continuous Waves) Millimeter Wavelength (MmWave) Radar technology in raw format for this task. We recorded human movements using FMCW MmWave, and fed our CNN network by transforming a multi-channel 2D structure with our new method called Video Of Radar (VIDAR). We tested our model with both real-time human movements and offline test datasets. The results were promising with over 90% accuracy. © 2022 Elsevier B.V., All rights reserved.