Two-Stream Time Sequential Network Based Hand Gesture Recognition Method Using Radar Sensor

Abstract

This paper proposes a deep learning based twostream time series hand gesture recognition method using the frequency modulated continuous wave (FMCW) radar. Firstly, we collect the hand gesture data by the FMCW radar, and the range and Doppler of the hand gesture are estimated by the 2 dimensional Fast Fourier Transform (2D-FFT). Then, the angle of hand gesture is estimated by Multiple Signal classification (MUSIC) algorithm. Afterward, we construct the Range- Doppler Map (RDM), and generate the Angle-Time Map (ATM) via multiframe accumulation. The interference in RDM is filtered out by peak interference cancellation, and the hand gesture feature in RDM and ATM are enhanced by wavelet transform. A systematic of two-stream time series neural network is designed for gesture feature extraction and classification. The experimental results show that the recognition accuracy rate for each type hand gesture of the proposed method is higher than 95%.