Face touch monitoring using an instrumented wristband using dynamic time warping and k-nearest neighbours

Abstract

One of the main factors in controlling infectious diseases such as COVID-19 is to prevent touching preoral and prenasal regions. Face touching is a habitual behaviour that occurs frequently. Studies showed that people touch their faces 23 times per hour on average. A contaminated hand could transmit the infection to the body by a facial touch. Since controlling this spontaneous habit is not easy, this study aimed to develop and validate a technology to detect and monitor face touch using dynamic time warping (DTW) and KNN (k-nearest neighbours) based on a wrist-mounted inertial measurement unit (IMU) in a controlled environment and natural environment trials. For this purpose, eleven volunteers were recruited and their hand motions were recorded in controlled and natural environment trials using a wrist-mounted IMU. Then the sensitivity, precision, and accuracy of our developed technology in detecting the face touch were evaluated. It was observed that the sensitivity, precision, and accuracy of the DTW-KNN classifier were 91%, 97%, and 85% in controlled environment trials and 79%, 92%, and 79% in natural environment trials (daily life). In conclusion, a wrist-mounted IMU, widely available in smartwatches, could detect the face touch with high sensitivity, precision, and accuracy and can be used as an ambulatory system to detect and monitor face touching as a high-risk habit in daily life.

Applying Human Factors to Reduce Healthcare-Associated Infections Caused by Face Touching

“…The most prevalent potential solutions are awareness enhancement devices, specifically artificial intelligence-based wearable detection and warning systems (e.g., D'Aurizio et al, 2020;Patel et al, 2021;Rizk et al, 2021). These systems use wristbands, smart watches, or headphones with built-in (ultrasonic) sensors, integrate algorithms to detect and differentiate between FT and non-FT behaviours based on hand or wrist orientations, and alert the user with feedback (e.g., haptic or auditory) to prevent the execution of FT actions (Bai et al, 2021;Fathian et al, 2022;Mustaqim et al, 2021;Roy et al, 2022). However, watchbased solutions can only track one hand's movements, and HCWs are unlikely to wear headphones when working (Spille et al, 2022a).…”