A Terahertz Radar Feature Set for Device-Free Gesture Recognition

This paper proposes a set of simple but effective features using a terahertz radar, specifically for device-free gesture recognition based on high resolution range profiles. Three types with seven features are extracted, including the tracking features, directional features, and behavioural features. The proposed method is evaluated on a dataset based on 0.34 THz radar, which contains 10 kinds of 5 pairs of frequently-used gestures. These features are demonstrated to be effective to encode the morphological differences among various gestures and be sensitive to the moving direction in a short period of time. The results show that the proposed method achieves 95.5% accuracy on frame-level gesture recognition.