Related Experiment Video
Updated: Jun 25, 2025
07:47
Design and Evaluation of Smart Glasses for Food Intake and Physical Activity Classification
Published on: February 14, 2018
11.2K
A CNN Model for Physical Activity Recognition and Energy Expenditure Estimation from an Eyeglass-Mounted Wearable
Md Billal Hossain1, Samuel R LaMunion2, Scott E Crouter2
1Department of Electrical and Computer Engineering, The University of Alabama, Tuscaloosa, AL 35487, USA.
Sensors (Basel, Switzerland)
|May 25, 2024
Summary
Eyeglass-mounted sensors can accurately track physical activity and estimate energy expenditure, offering a novel approach to managing metabolic syndrome. This technology shows promise for unobtrusive health monitoring and personalized interventions.
Area of Science:
- Biomedical Engineering
- Wearable Technology
- Metabolic Health
Background:
- Metabolic syndrome is a global health concern requiring integrated physical activity and energy expenditure monitoring.
- Wearable sensors are increasingly used for energy intake and expenditure (EE) estimation, traditionally placed on the hip or wrist.
Purpose of the Study:
- To investigate the efficacy of an eyeglass-mounted sensor (AIM-2) for simultaneous physical activity recognition (PAR) and steady-state EE estimation.
- To compare the performance of the eyeglass-mounted sensor against a traditional hip-worn device.
Main Methods:
- Six participants performed six structured activities, with EE measured via indirect calorimetry (COSMED K5) as METs.
- A deep convolutional neural network (Multitasking-CNN) was developed for PAR and EE estimation using a two-step progressive training approach.
- The Multitasking-CNN's performance was evaluated on eyeglass-mounted AIM-2 data and hip-worn ActiGraph GT9X (AG) data.
Main Results:
- The eyeglass-mounted AIM-2 achieved 95% PAR accuracy, 0.59 METs MSE, and 11% MAPE for EE estimation.
- The hip-worn AG achieved 82% PAR accuracy, 0.73 METs MSE, and 13% MAPE for EE estimation.
- The Multitasking-CNN model demonstrated superior performance with the eyeglass-mounted sensor.
Conclusions:
- Eyeglass-mounted sensors are feasible for simultaneous physical activity recognition and energy expenditure estimation.
- This approach offers a potentially more accurate and unobtrusive method for health monitoring compared to traditional wearable sensors.
