FaceSense: Sensing Face Touch with an Ear-worn System
summary
An ear-worn prototype combines 8x8 thermal cameras and a bio-amplifier skin impedance subsystem in flexible TPU earpieces to detect actual hand-to-face contact and localize touches to mucosal versus non-mucosal zones. A multimodal deep model, TouchNet with a user discriminator, fuses low-res thermal frames and bilateral impedance signals to suppress proximity false positives and user-specific noise. Evaluation with 14 participants shows binary touch detection at ~83.4% macro‑F1 (leave‑one‑user‑out) and personalized zone ID at ~90% macro‑F1 after a ~5‑minute calibration.
AI-written summary · about