Convolutional Neural Network Model in Human Motion Detection Based on FMCW Radar Signals

Abstract

The detection of body movements is the essential step for sleep quality analysis. Contactless approaches for sleep motion recognition are unobtrusive and are easier to use in comparison to wearable technologies. In this paper, two contactless sensors based on Frequency-Modulated Continuous Wave (FMCW) radar technology were positioned on the side of, and underneath the bed on which the participant was lying. FMCW data from 10 participants were acquired during the experiment scenario that included the following three states: resting state, movement, and cough. Magnitude-phase coherency method was applied to FMCW data for finding optimal phase signals. Finally, a one-dimensional convolutional neural network was used for the classification based on optimal phase signals. The best classification results were obtained using only FMCW data from the radar positioned underneath the bed: 72% accuracy for differentiating between the resting state, movement, and cough class, and 89% accuracy for the resting state and movement class.