Application of FMCW Radar for the Recongnition of Hand Gesture Using Time Series Convolutional Neural Networks
Desheng Liu, Hongfu Meng · 2020
Gesture recognition methods based on traditional optical camera and sensor technologies are suscetible to the effects of lighting environment and have poor body sensations and robustness. This paper proposes an end-to-end Time Series convolutional neural network(TS-CNN) gesture recognition method based on FMCW radar signals. When the FMCW radar captures hand movements, the spectrum of the intermediate frequency signal is first extracted using a two-dimensional fast Fourier transform to obtain a Range Doppler map. Second, the Range Doppler map is used to map the gestures into a set of multi-frame Range Doppler maps over time. Finally, the TS-CNN is used for feature extraction, learning, and classification. Convolutional neural networks are powerful classifiers. To determine the optimal structure, we trained multiple convolutional neural networks by changing the hyperparameters. Experimental results show that compared with traditional machine learning methods, the classification accuracy obtained from the best convolutional neural network structure is improved by about 5%.