An assessment of linear wrist motion during the taking of a bite of food

Motivated by the goal of building a device capable of detecting when a user takes a bite of food, we examine whether there is a correlation between linear wrist motion and bite detection. Such a device could be used to manage weight loss by tracking eating trends over time and providing real time feedback to the user regarding eating rate or total consumption. A comprehensive analysis of tracking linear motion using low cost inertial sensors is presented. Based on this theory, we discuss the design and tracking accuracy of a prototype bite counting device. We show that tracking position using low cost, off-the-shelf inertial sensors is impractical over any sustained period of time. However, by detecting peaks in acceleration data we are able to look for a characteristic motion during eating. We recorded data from subjects eating a meal while wearing the device. Their wrist motions were analyzed in an attempt to identify a characteristic linear motion of the wrist that corresponds to the taking of a bite of food. Our results show that there is too much variability in motion during eating to detect bites of food based on acceleration peaks alone.