GlasSense I

GlasSense는 섭식 활동을 비롯한 다양한 얼굴의 활동을 안경을 통해 감지하고 머신 러닝을 통해 분류하는 연구이다. 안경 힌지에 로드 셀을 내장하면 측두근의 미세한 수축과 이완을 지렛대의 원리를 통해 기구학적으로 증폭할 수 있다. 이렇게 측정된 데이터로부터 시간과 주파수 특징 벡터를 추출하고, 이를 학습한 SVM 분류기 모델은 좌/우 저작, 좌/우 윙크, 말하기, 머리 움직임을 평균 94.0%의 F₁ 점수로 분류할 수 있다.

Design was conducted under following requirements.

1. The glasses has to be in ergonomic shape to be wearable.
2. Both hinges has to be functional, since the temples work as levers on load cells.
3. Every component should be 3D-printable.
4. Load cell should be detachable and be assembled in two positions, hence enabling the comparative experiment between the following two cases.
  – load cell being sensing the pressure directly in contact with temporalis epidermis.
  – load cell sensing the pressure amplified by the leverage.

[Abstract]
Here we present a new method for automatic and objective monitoring of ingestive behaviors in comparison with other facial activities through load cells embedded in a pair of glasses, named GlasSense. Typically, activated by subtle contraction and relaxation of a temporalis muscle, there is a cyclic movement of the temporomandibular joint during mastication. However, such muscular signals are, in general, too weak to sense without amplification or an electromyographic analysis. To detect these oscillatory facial signals without any use of obtrusive device, we incorporated a load cell into each hinge which was used as a lever mechanism on both sides of the glasses. Thus, the signal measured at the load cells can detect the force amplified mechanically by the hinge. We demonstrated a proof-ofconcept validation of the amplification by differentiating the force signals between the hinge and the temple. A pattern recognition was applied to extract statistical features and classify featured behavioral patterns, such as natural head movement, chewing, talking, and wink. The overall results showed that the average F1 score of the classification was about 94.0% and the accuracy above 89%. We believe this approach will be helpful for designing a non-intrusive and un-obtrusive eyewear-based ingestive behavior monitoring system