Eating Event Recognition Using Accelerometer, Gyroscope, Piezoelectric, and Lung Volume Sensors

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Health Informatics

Background:

• Overweight and obesity are significant global health challenges.
• Accurate dietary monitoring is crucial for weight management but often hindered by self-reporting inaccuracies.
• Automatic Dietary Monitoring (ADM) offers a potential solution for objective and continuous food intake assessment.

Purpose of the Study:

• To develop and evaluate a multi-sensor system for automatic detection of eating events.
• To assess the effectiveness of combining different sensor modalities for improved eating event classification.
• To support individuals in weight management through objective dietary tracking.

Main Methods:

• A sensor system was designed integrating a smartwatch (accelerometer, gyroscope) for gesture detection, a jaw-worn piezoelectric sensor for chewing detection, and respiratory inductance plethysmography for swallowing detection.
• Features were extracted from sensor data to train a support vector machine (SVM) model.
• Experiments were conducted with six subjects in a controlled setting involving eating and non-eating events.

Main Results:

• The SVM model achieved an F1-score of 0.82 for eating gesture detection.
• Chewing detection yielded a high F1-score of 0.94.
• Swallowing detection achieved an F1-score of 0.58.

Conclusions:

• The integrated multi-sensor system demonstrates significant potential for automatic dietary monitoring.
• Combining sensors for different stages of the dietary cycle improves eating event classification accuracy, particularly for gestures and chewing.
• Further refinement is needed to enhance the accuracy of swallowing detection for comprehensive ADM.