Real-time Detection of Change of Human Motion by Analyzing Millimeter-wave Doppler Radar Signals Using Deep Learning Techniques

Abstract

A system based on a millimeter-wave radar module is presented in this paper. After detecting the change of human motion, the changes in the point cloud are observed by analyzing the Doppler signal. Then, the change of human motion is classified in real-time through deep learning (DL) techniques that include long short-term memory (LSTM) and 1D time distributed convolutional neural network (CNN) methods. Temporal continuity and scalability are considered for the techniques. Measuring 100 mm wide, 40.8 mm long, and 52.8 mm high, this millimeter-wave radar module features Frequency Modulated Continuous Wave (FMCW) in the 60 to 64 GHz frequency range with 3 transmit (15 dBm) and 4 receive antennas (14 dB) on a package (AOP), 120° azimuth field of view (FoV), and 40° elevation FoV. An additional 5V/2A DC power supply is required during operation, and 1843200bps communication is used through the USB serial port.