Finger Gesture Recognition Based on Wireless Coupling Resonance

This article proposes a finger state capture and recognition system based on near-field coil coupling and multifrequency resonance. Finger state capture is realized by load coils mounted on the fingers that form coupling resonance at multiple frequencies under sweep frequency excitations. The motions of the finger joints lead to changes of the distance and angle between the load coils, which further affect the amplitude and frequency of the resonance through mutual inductances. These features are extracted from the measurements of reflected signals for finger state recognition. The process of a finger from full flexion to full extension is approximately evenly divided into 20 states. A weighted parallel amplitude and phase (WPAP) classification method is proposed based on the support vector machine (SVM) algorithm. This method processes the amplitude and phase of the complex reflected signal separately in parallel, and then combines their classification results by assigning weights to the output. The experimental results verify the effectiveness of finger state recognition from the reflected signals. The classification result and performance evaluation demonstrate that the proposed method outperforms the two SVM classifiers trained on amplitude data and phase individually. The classification accuracies of the 20 states of the index finger and the thumb are 96.92% and 83.58%, respectively. Moreover, four hand gestures are selected and tested using the measurements of the thumb and the index finger. All gestures are correctly recognized. This work achieves accurate finger state recognition without using additional active sensors on the fingers, which largely preserves the finger flexibility. It has potential applications in finger tracking and manipulator monitoring.