Implementing Real-Time Food Intake Detection in a Wearable System Using Accelerometer

The objective of this study was to develop a realtime food intake detection method in a wearable sensor system using an accelerometer sensor. The Automatic Ingestion Monitor v2 (AIM-2) wearable sensor consists of a chewing sensor, accelerometer, camera, microcontroller, battery, and other components. The accelerometer is used to capture signals of chewing and head-movements. Time-domain statistical features were computed from the 3-axis accelerometer signal. A decision tree classifier was trained offline and implemented in the microcontroller to predict food intake events in real-time. The classifier was trained on a dataset collected from 30 free-living volunteers. The classifier achieved a sensitivity of 86.79%, specificity of 96.24%, and F1-score of 57.57% in leave-one-out subject validation. The proposed method was implemented and tested for eight days of free-living data and achieved 97.62% accuracy of eating episode detection. Real-time food intake detection in a wearable sensor system may enable future implementation of just-in-time feedback to the user for improving eating habits and achieving weight-loss.