Accuracy Improvement of Human Motion Recognition with MW-FMCW Radar Using CNN

Human Motion Recognition (HMR) using radar has been expected to be applied to human monitoring. Radar-based HMR can be realized by extracting features by deep learning from radar motion-sensing data. In this paper, three 2-D domains (Time-Doppler, Time-Range, and Range-Doppler) are analyzed as the motion data. The motion classification results using Convolutional Neural Network (CNN) shows that multiple domains improve classification accuracy. In addition, we show the motion analysis results with high range resolution including DOA, and show that further accuracy improvement can be expected.