mmWave Radar-based Hand Gesture Recognition using Range-Angle Image

The radar sensing on fine human-motion/hand-gesture provides further human-computer interaction (HCI) experience. Most of the studies about gesture recognition with mmWave frequency modulated continuous wave (FMCW) radar adopts the range and the velocity estimated from the raw data, such as the time-frequency spectrogram, micro-doppler spectrogram, or range-Doppler image (RDI). Besides, the angle estimated using multiple receive antennas also contains rich information of gesture, especially in the discrimination among the horizontal movement of the gesture. Thus, we propose to use the range-angle image (RAI) as the input and train a model consisting of the convolutional neural network and long short term memory that is capable of recognizing hand gestures. We validate the proposed scheme based on the collection of hand gestures by several subjects in different classrooms using 77 - 81GHz mmWave radar of Texas Instrument. Based on the configuration of one transmit antenna and four receive antennas, we show that the hand gesture recognition using RAI outperforms that using RDI. Also, we adopt the fusion strategy to consider both RDI and RAI to further improve accuracy.