Hand Gesture Recognition Using a Radar Echo I–Q Plot and a Convolutional Neural Network

Abstract

We propose a hand gesture recognition technique using a convolutional neural network applied to radar echo in-phase/quadrature (I/Q) plot trajectories. The proposed technique is demonstrated to accurately recognize six types of hand gestures for ten participants. The system consists of a low-cost 2.4-GHz continuous-wave monostatic radar with a single antenna. The radar echo trajectories are converted to low-resolution images and are used for the training and evaluation of the proposed technique. Results indicate that the proposed technique can recognize hand gestures with average accuracy exceeding 90%.