SmartFinger: A Finger-Sensing System for Mobile Interaction via MIMO FMCW Radar

Abstract

In this paper, we present a finger-grained gesture recognition system that can be deployed on commodity Multiple Input Multiple Output Frequency Modulated Continuous Wave (MIMO-FMCW) radar platform as software, without any hardware modification. Firstly, we utilize the two-dimension fast Fourier transform algorithm (2D-FFT) to jointly estimate range-Doppler information. Secondly, by combining with binary phase modulation MIMO (BPM-MIMO) technique, a discrete Fourier transformation (DFT) based Multiple Signal Classification (MUSIC) algorithm is proposed to jointly measure range and angle of arrival (AOA) information without prior knowledge about the number of targets. Thirdly, a recurrent 3D convolutional neural network (R3DCNN) is employed to extract spatial-temporal fusion- features existing in range-Doppler and range-AOA map sequences. Next, we implement and evaluate this system utilizing commercial-off-the-shelf FMCW radar platform. The experimental results show that this system is able to achieve a high recognition rate of 93%.