Deep Learning Approach for Gesture Recognition on Millimeter- Wave Radar

Gesture recognition has gained world-wide attention in the field of human-computer interaction. Built on millimeter-wave radar, the gesture recognition system has specific advantages of high integration, all-day operation, and robustness in poor lighting. This paper presents a 77 GHz millimeter-wave (MMW) radar to build a gesture recognition algorithm, which uses frequency-domain notch and sliding window search methods to mitigate interference and automatically extract useful signals from gesture echoes. We establish the gesture radar data collection platform on an MMW radar, design common gestures, and collect massive gesture radar data in the time-frequency domain to form a dataset. A convolutional neural network is built for gesture recognition, and the accuracy can reach up to 94.72 % based on the collected gesture radar dataset.