Serendipity: Finger Gesture Recognition using an Off-the-Shelf Smartwatch
summary
System uses only accelerometer and gyroscope on an off-the-shelf Samsung Gear watch sampled at 50 Hz with 1 s sliding windows, FFT (keep lower 10 bands), DTW-based gating plus K-NN and SVM classification. A 10-participant study with two sessions and two-minute non-gesture baselines reports an average f1-score of 87% for 5 fine-motor gestures (pinch, tap, rub, squeeze, wave); rub is easiest, pinch/tap are often confused, and a three-rub activation yields ~0.38 false positive and 5% false negative.
AI-written summary · about