A Novel Approach of Chewing Detection based on Temporalis Muscle Movement using Proximity Sensor for Diet Monitoring

This paper presents a novel chewing detection system by using a proximity sensor in capturing the temporalis muscle movement during intake for diet monitoring application. The sensor is attached to eyeglasses by using 3D printed housing. A single-subject with multiple data collection was used in forming a dataset. The main activity of eating and resting was taken into consideration with a total time of 240s for a set of data. Three test food (carrots, bananas, and apples) with a portion of one spoonful for each intake of food. The signal was pre-processed using z-score normalization and a bandpass filter. Analysis based on several parameters of signal processing such as k-fold of cross-validation value and bandpass filter value was done. Results show that the proposed proximity sensor could achieve an accuracy of 97.3% with 97.3% F1-score (quadratic SVM) using a 10-fold and bandpass filter of 0.5Hz to 5Hz.