Multi-Hand Gesture Separation and Recognition using Millimeter-wave Radar

Abstract

This paper presents a multi-hand gesture separation and recognition system based on frequency modulated continuous wave (FMCW) radar. Specifically, we separately construct the parameter maps containing range-Doppler and range-angle. Then the multi-gesture targets separation algorithm based on spatiotemporal path selection is proposed. And then, we design a dualstream 3D convolutional neural feature fusion network to extract gesture features and classify gestures. Finally, experimental results prove that the average classification precision of the method we proposed is 93.12%.