MealWatcher: A New Toolset to Record Meals for the Development of Automated Energy Intake Estimation

Faria Armin, James Jolly, James Nguyen, Lakshmi Rangaraju, Daniel Regan, Zoe Brown, Sharon O’Toole, Hollie Raynor,, Leslie Brick, Elissa Jelalian, J. Graham Thomas, Stephanie P. Goldstein, Adam Hoover

Keywords: smart ring, motion sensor, wearable device, gesture recognition, energy intake measurement

TL;DR: This paper introduces a new toolset for collecting motion data simultaneously from smartwatch and smart ring, and an ongoing dataset collected with these tools.

Abstract: Classic tools for measuring energy intake, such as food diaries and 24 hour recalls, are burdensome to use and have significant measurement error. This hinders research and interventions in obesity treatment and comorbidities such as diabetes and heart disease. New tools are being developed to automate the measurement of energy intake, such as wearable devices like smartwatches. Towards this goal, several datasets have been collected and made publicly available that include hand motion. However, these datasets have been limited to the wrist part of hand motion, and have only been collected in controlled environments such as labs or cafeterias. In this work we describe a new toolset that supports data collection from a smartwatch and smart ring simultaneously, to be used in home, cafeteria, and free-living environments, with real-time user feedback to assist with energy intake estimation. This data collection is ongoing and will eventually encompass 600 subjects. This paper describes the toolset and preliminary results for 16 subjects, including a comparison of smart ring versus smartwatch intake gesture detection. The smart ring achieved an F1 score of 0.74 compared to an F1 score of 0.80 from the smartwatch. Finally, we describe the full set of experiments we intend to perform with the complete dataset.

Track: 11. General Track

Registration Id: K6NZDVKKZ97

Submission Number: 378