Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Estimation of Food Intake Quantity Using Inertial Signals from Smartwatches

Ioannis Levi, Konstantinos Kyritsis, Vasileios Papapanagiotou, Georgios Tsakiridis, Anastasios Delopoulos

TL;DR

The paper demonstrates that bite-level food intake can be quantitatively estimated using only wrist-worn inertial data by extracting six engineered features (two behavioral, four statistical) and applying a Support Vector Regression model, achieving a mean absolute error of 3.99 g per bite across 342 bites from 10 participants under LOSO CV. It compares against an adapted multimodal baseline and an early-fusion deep architecture, with the proposed SVR method delivering the strongest relative improvement over a simple baseline. The study leverages a detailed preprocessing pipeline and a carefully defined bite annotation protocol, contributing a valuable dataset and a practical pathway toward single-device, non-invasive dietary monitoring. The findings highlight the viability of wearable sensors for quantitative dietary metrics and set the stage for automated bite detection and deployment in real-world settings.

Abstract

Accurate monitoring of eating behavior is crucial for managing obesity and eating disorders such as bulimia nervosa. At the same time, existing methods rely on multiple and/or specialized sensors, greatly harming adherence and ultimately, the quality and continuity of data. This paper introduces a novel approach for estimating the weight of a bite, from a commercial smartwatch. Our publicly-available dataset contains smartwatch inertial data from ten participants, with manually annotated start and end times of each bite along with their corresponding weights from a smart scale, under semi-controlled conditions. The proposed method combines extracted behavioral features such as the time required to load the utensil with food, with statistical features of inertial signals, that serve as input to a Support Vector Regression model to estimate bite weights. Under a leave-one-subject-out cross-validation scheme, our approach achieves a mean absolute error (MAE) of 3.99 grams per bite. To contextualize this performance, we introduce the improvement metric, that measures the relative MAE difference compared to a baseline model. Our method demonstrates a 17.41% improvement, while the adapted state-of-the art method shows a -28.89% performance against that same baseline. The results presented in this work establish the feasibility of extracting meaningful bite weight estimates from commercial smartwatch inertial sensors alone, laying the groundwork for future accessible, non-invasive dietary monitoring systems.

Estimation of Food Intake Quantity Using Inertial Signals from Smartwatches

TL;DR

The paper demonstrates that bite-level food intake can be quantitatively estimated using only wrist-worn inertial data by extracting six engineered features (two behavioral, four statistical) and applying a Support Vector Regression model, achieving a mean absolute error of 3.99 g per bite across 342 bites from 10 participants under LOSO CV. It compares against an adapted multimodal baseline and an early-fusion deep architecture, with the proposed SVR method delivering the strongest relative improvement over a simple baseline. The study leverages a detailed preprocessing pipeline and a carefully defined bite annotation protocol, contributing a valuable dataset and a practical pathway toward single-device, non-invasive dietary monitoring. The findings highlight the viability of wearable sensors for quantitative dietary metrics and set the stage for automated bite detection and deployment in real-world settings.

Abstract

Accurate monitoring of eating behavior is crucial for managing obesity and eating disorders such as bulimia nervosa. At the same time, existing methods rely on multiple and/or specialized sensors, greatly harming adherence and ultimately, the quality and continuity of data. This paper introduces a novel approach for estimating the weight of a bite, from a commercial smartwatch. Our publicly-available dataset contains smartwatch inertial data from ten participants, with manually annotated start and end times of each bite along with their corresponding weights from a smart scale, under semi-controlled conditions. The proposed method combines extracted behavioral features such as the time required to load the utensil with food, with statistical features of inertial signals, that serve as input to a Support Vector Regression model to estimate bite weights. Under a leave-one-subject-out cross-validation scheme, our approach achieves a mean absolute error (MAE) of 3.99 grams per bite. To contextualize this performance, we introduce the improvement metric, that measures the relative MAE difference compared to a baseline model. Our method demonstrates a 17.41% improvement, while the adapted state-of-the art method shows a -28.89% performance against that same baseline. The results presented in this work establish the feasibility of extracting meaningful bite weight estimates from commercial smartwatch inertial sensors alone, laying the groundwork for future accessible, non-invasive dietary monitoring systems.

Paper Structure

This paper contains 13 sections, 2 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. METHODS
  3. Preprocessing
  4. Feature engineering
  5. Behavioral features
  6. Statistical Features
  7. Bite weight estimation model
  8. Comparative architectures
  9. EVALUATION
  10. Dataset
  11. Framework
  12. RESULTS
  13. CONCLUSIONS

Figures (2)

  • Figure 1: Overall pipeline of the proposed bite weight estimation method. From left to right: Raw IMU streams ($N = t \cdot f_s$) undergo preprocessing for signal quality optimization. Bite intervals are then identified through manual video annotation (dashed box). These intervals serve as input for automatic feature extraction, where behavioral characteristics are derived from micromovement analysis and statistical features from IMU signal processing. The resulting feature vector feeds an SVR model that generates weight estimations for each bite in the meal.
  • Figure 2: Figure illustrating the absolute error distribution across evaluated methods. The dotted vertical line represents the MAE.