Finger Bending Biometrics — Smart Glove
Signal segmentation - Continuous flex-sensor streams segmented by typing episode
- PIN entry → fixed-length numeric sequence
- Password entry → fixed-length alphanumeric sequence
- Segmentation aligned to task boundaries, not individual keystrokes
- Preserves inter-key motion dynamics
Preprocessing decisions - Per-sensor normalization to account for:
- Manufacturing variance across flex sensors
- Baseline resistance differences across users
- Temporal alignment across fingers to maintain coupled motion structure
- Minimal filtering to avoid smoothing out micro-bend dynamics
Feature families used - Time-domain statistics
- Mean, variance, range, RMS
- Captures individual bend amplitude and stability
- Frequency-domain features
- Dominant frequencies, spectral energy
- Reflects rhythmic finger motion during typing
- Entropy-based features
- Measures predictability vs irregularity of finger motion
- Cross-finger correlation features
- Quantifies coordination between fingers
- Typing is not independent per finger.
- Even unused fingers exhibit involuntary motion driven by neuromuscular coupling.
- Aggregating statistics across multiple fingers and their correlations captures biomechanical signatures that keystroke timing alone cannot.