Meal-time and duration monitoring using wearable sensors

Abstract

Abstract In many recent studies it has been shown that the frequency and duration of meals taken during a day can have significant impacts on human health. Poor meal habits such as skipping breakfast, nighttime eating, and too many eating episodes can increase the risk of obesity. Researchers have often used self-reported questionnaires for analyzing such meal-time and dietary behaviors. Questionnaire-based data collection, however, often suffers from high errors due to reporting subjectivity. This paper presents a wearable sensor system that can monitor breathing and hand movement for estimating the time and duration of a meal. The system combines swallowing signatures from breathing signal with hand movement signatures from hand acceleration to train a hierarchical Support Vector Machine (SVM) classifier and a Hidden Markov Model (HMM) for mealtime and duration estimation. Algorithms are developed for detecting various types of swallowing events including for solid and liquid in the presence of artifacts such as spontaneous swallows, laughing, coughing, and throat clearance. The experiments were carried out on 14 healthy subjects wearing the proposed system. In each experiment session, the subjects were asked to have lunch, drink water, rest and talk. The subjects were asked to press a button for each swallow, and the whole experiment process was video recorded. The push button and video information were used as a ground truth for verification purposes. Through extensive experimentation in a semi-controlled setting, it was shown that the system is able to detect mealtime with high accuracy.