Hand Gesture Recognition using MIMO Radar and Lightweight Convolutional Neural Network.

Frequency-Modulated continuous-wave (FMCW) radar is a promising modality for gesture recognition, as it has the potential to overcome the limitations of cameras, which are commonly used for this purpose. Ambient light conditions and obstructions pose significant challenges to optical sensors. This paper introduces a method for hand gesture recognition with ten different gestures by processing the data obtained from the continuous waves captured by the radar system of Texas Instruments. Raw radar data is performed FFT (Fast Fourier Transform) to generate the Range-Doppler Heatmap (RDH), and then micro-Doppler (time-Doppler) is analyzed to extract velocity hand gesture data over time from RDH. Subsequently, we propose a lightweight Convolutional Neural Network (CNN) model designed to extract complex structures from the processed data, comprising velocity, angles, and phase features. The CNN model we developed was trained on a dataset consisting of eight individuals and evaluated on two additional individuals. The average accuracy across different gestures achieved by the model is 98.35%.