Hand Gesture Recognition Using a Dual Axis Millimeter-Wave Interferometric-Doppler Radar and Convolutional Neural Networks

We demonstrate hand gesture classification based on measurements obtained from a 40 GHz dual axis interferometric-Doppler radar. Using a dual axis, three-element receiver, velocity is measured in three dimensions (two angular, one radial). We compare the results of four millimeter-wave receiver topologies: a system providing two Doppler responses; two Doppler responses plus their corresponding interferometric response; three Doppler responses; and three Doppler responses plus their two corresponding interferometric responses. In almost every case, the system with three micro-Doppler and two interferometric responses performed the best, having a total classification accuracy of 99.7% over all of the test set measurements.