Automatic Radar-based Gesture Detection and Classification via a Region-based Deep Convolutional Neural Network

In this paper, a region-based deep convolutional neural network (R-DCNN) is proposed to detect and classify gestures measured by a frequency-modulated continuous wave radar system. Micro-Doppler (µD) signatures of gestures are exploited, and the resulting spectrograms are fed into a neural network. We are the first to use the R-DCNN for radar-based gesture recognition, such that multiple gestures could be automatically detected and classified without manually clipping the data streams according to each hand movement in advance. Further, along with the µD signatures, we incorporate phase-difference information of received signals from an L-shaped antenna array to enhance the classification accuracy. Finally, the classification results show that the proposed network trained with spectrogram and phase-difference information can guarantee a promising performance for nine gestures.