Gesture Recognition Using FMCW Radar Based on Deep Neural Network

In recent years, radar based gesture recognition technology has attracted more and more attention. Among different types of radars, frequency modulated continuous wave (FMCW) radar has the greatest application potential due to its high range and velocity resolution and low cost. However, insufficient gesture information extraction and interference signals affect the performance of FMCW radar gesture recognition. To address the problems, a gesture recognition method based on multi-dimensional features and deep neural network is proposed in this paper. First, the range-time map (RTM) and velocity-time map (VTM) of the gesture are constructed from the range-Doppler map (RDM) with two-dimensional fast Fourier transform (2D-FFT). Then, the angle-time map (ATM) is constructed with the multiple signal classification (MUSIC) algorithm. Subsequently, RTM, VTM and ATM are normalized and adaptively filtered to suppress interference signals, and the three maps are fused to construct the range-velocity-angle-time map (RV ATM) of the gesture. Finally, an improved VGG16 network with self-attention mechanism module is used for feature extraction and recognition. The experimental results show that the proposed method achieves an average accuracy of 98.3% for six gestures, and the improved VGG16 network outperforms other traditional convolutional neural networks.