Multitarget Human Motion Recognition via Beamforming based on Millimeter-Wave MIMO Radar

This study presents an advanced radar-based multitarget human motion recognition (M-HMR) approach using multiple-input-multiple-output (MIMO) frequency-modulated-continuous-wave (FMCW) millimeter-wave (mmWave) radar. First, leveraging the scattering point cloud obtained from radar, a multitarget localization algorithm is proposed, which iteratively filters the scattering points to improve the accuracy of target localization. A signal separation algorithm is then developed to obtain motion signals for each target, wherein an adaptive beamforming algorithm with dynamic beamwidth and empowered by linearly constrained minimum variance (LCMV) is proposed to alleviate the interference among different targets. Based on the results of signal separation, the M-HMR problem is converted into single-target HMR, and a convolutional neural network (CNN) is used to achieve the classification of different motions. Tested on real datasets collected using the mmWave radar system, MMWCAS-RF-EVM, from Texas Instruments, the proposed approach demonstrates high accuracy in isolating target signals and classifying motions, showcasing its potential for complex HMR applications.