Abstract
Sensing performance of capacitive touch sensor is significantly degraded in electronically harsh environments, for example, underwater. In particular, a capacitive touch sensor used in a general mobile phone cannot recognize a touch in the underwater. Based on the observation that contact between two physical bodies (e.g., fingertip and display screen) induces object motion, although tiny, we propose a novel touch interface system that learns multivariate sequential signals to recognize the touched position while underwater. To that end, we first collected multivariate sensor data utilizing a commercial robot arm system to obtain sufficient amount of touch data in the underwater condition. Then, we trained deep neural network models using the collected data along with predefined touch regions in a supervised fashion. The experimental results obtained demonstrated higher recognition performances with overall accuracy of 96.74%. We conclude this paper by discussing the issues and highlighting future research directions.
Original language | English |
---|---|
Pages (from-to) | 8924-8932 |
Number of pages | 9 |
Journal | IEEE Sensors Journal |
Volume | 22 |
Issue number | 9 |
DOIs | |
State | Published - 1 May 2022 |
Keywords
- Deep neural network
- convolutional neural network
- recurrent neural network
- sequence learning
- touch-induced motion
- virtual sensing