Smartwatch-Based Face-Touch Prediction Using Deep Representational Learning

Hamada Rizk, Tatsuya Amano, Hirozumi Yamaguchi, Moustafa Youssef

Abstract: World Health Organization (WHO) reported that viruses, including COVID-19, can be transmitted by touching the face with contaminated hands and advised people to avoid touching their face, especially the mouth, nose, and eyes. However, according to recent studies, people touch their faces unconsciously in their daily lives, and it is difficult to avoid such activities. Although many activity recognition methods have been proposed over the years, none of them target the prediction of face-touch (rather than detection) with other daily life activities. To address to problem, we propose TouchAlert: a system that automatically predict the occurrence of face-touch activity and warn the user before its occurrence. Specifically, TouchAlert utilizes commodity wearable devices’ sensors to train a deep learning-based model for predicting the variable length face-touching of different users at an early stage of its occurrence. Our experimental results show high accuracy of F1-score of 0.98 and prediction accuracy of 97.9%.