Automated dietary monitoring (ADM) is increasingly recognized as a critical component to advance health tracking and inform behavioral interventions. Although prior research has investigated ADM through various forms of wearable devices and sensing modalities, e.g. wrist-worn motion sensors, acoustic sensing, and vision-based approaches, the potential of smart rings, an emerging and increasingly popular wearable form factor, remains largely underexplored. In this work, we focus on detecting eating events, a key component of ADM, leveraging inertial data acquired from a custom-designed wireless ring. We developed a prototype ring device accompanied by a smartphone application to facilitate data collection. To evaluate the feasibility and effectiveness of ring-based ADM, we conducted studies with 42 participants in three settings: a controlled laboratory environment, a semi-controlled environment, and a deployment in the wild. Our analysis addresses key research questions regarding the accuracy of eating event detection in various contexts, the influence of ring placement on performance, the impact of personalization, and the potential benefits of integrating inertial data from both the ring and a commodity smartwatch. Our ring wearable achieved F1-scores of 82.6% in the controlled laboratory setting and 65.9% in the semi-controlled condition. We showed that while a ring wearable alone does not perform well in naturalistic settings, it can complement a smartwatch sensor at this task, and performance can be further improved through personalization. To foster continued advancement in ring-based ADM research, we will publicly release a paired ring and smartwatch dataset comprising 80 hours of annotated eating and non-eating activities. To our knowledge, this will constitute the largest publicly available dataset focused on eating event detection using ring-based inertial sensing.
Liang et al. (Mon,) studied this question.