Selective Clutter Removal for FMCW-Radar Based Hand Gesture Recognition System

Hand gesture (HG) recognition using radar has been explored not only for human-to-computer interfaces in home appliances but also for device control in automotive applications. In such scenarios, the stationary clutter from the ground (or pavement), vehicles, furniture, human bodies, and other obstacles in the radar’s field of view affects target detection and degrades the accuracy of HG classification. Such stationary clutter is usually removed using a high-pass filtering technique. However, since the relative velocity in radar is calculated from the difference in the range along the radial direction from the radar over multiple chirps, the velocity decreases when the human hand is located near the center position of the radar. When using the conventional filtering method based on simple high-pass filtering, the number of detected points on the HG trajectory decreases. We therefore propose a more intelligent filtering method that removes only stationary clutter while preserving the detected points on the HG trajectory. In this work, we demonstrate the effectiveness of the proposed method through simulations and measurement using 79-GHz band frequency-modulated continuous-wave multiple-input multiple-output radar with three transmit antennas and four receive antennas by classifying 11 classes of HGs. Our findings show that the proposed method performs better than the conventional filtering method, with an accuracy improvement ranging from about 2 to 9 % across different scenarios in both simulation and measurement even when the power of clutter is higher than that of the human hand. The classification performance of the 11 classes of HGs showed an accuracy of 95.6 % when mixture datasets consisting of a small portion of measurement datasets and synthetic datasets were used with the proposed method.