FMCW Radar-Based Hand Gesture Recognition Using Dual-Stream CNN-GRU Model

Data derived from mmWave radars, such as frequency modulated continuous wave (FMCW) radars, contains distinctive movement-based features that characterize each gesture uniquely and facilitate contactless human hand gesture recognition. This paper aims to use an advanced 77-GHz multiple-input-multiple-output (MIMO) FMCW radar with a deep-learning model to automate the extraction of these unique features. We forward this radar’s pre-processed range-Doppler and range-angle images (RDIs and RAIs) into a dual-stream artificial neural network to classify human hand gestures. The proposed multiple-input, single-output architecture comprises 2D convolutional neural network-gated recurrent units (2D CNNGRU). According to the conducted experiments, the average accuracy of the proposed classification model with 8-fold cross-validation achieves 92.50%.