A Robust mmWave Point-Based Hand Gesture Recognition Algorithm With False Alarm Control

Dynamic hand gesture recognition (HGR) using millimeter wave (mmWave) radar has demonstrated significant potential across various application domains. However, existing methods tend to produce erroneous recognition results when subjected to random motion interferences, such as walking, waving, bending, or other motions beyond predefined gestures. In this paper, we proposed a robust mmWave point-based HGR system with false alarm control. First, a point cloud management algorithm is designed to simultaneously capture both dynamic and static point clouds of the human body, followed by a human pose estimation (HPE) network to generate stable HPE keypoints. Second, a hand-raised posture is served as the trigger signal for gesture recognition. Third, the segmented hand point clouds are processed using a long short-term memory (LSTM) network for gesture recognition. Finally, six volunteers were recruited to acquire seven predefined gesture data. Additionally, personnel interference data over five days in three real scenarios were collected. Experimental results show that the proposed algorithm achieves a recognition accuracy of 98.94% for predefined gestures and a false alarm rate as low as 0.0038% in real scenarios.