Recognition of Human Eating Behavior Using Kinect

Abstract

The broad range of data collection methods has introduced to the rest of the world, Kinect is one of the common tool in world. In the world of motion capturing, Microsoft Kinect has a low-cost depth mapping sensor which can be applied in many sorts of applications. Kinect sensor V2 is the chosen in this project. Recognition of eating is a great challenge because of it involved multiple objects and huge variety of food. But this is a good investigation for eating actions of people. My research focus on doing skeleton tracking and RGB-D data collection by using Kinect sensor V2. The dataset would be built up by capturing such data from multiple subjects who will eat different kinds of food and with various types of utensils. Then, analyze the captured data by exploring machine learning techniques in order to recognize the subjects and movement so that the eating behavior of individuals can be characterized. It also include to find higher accuracy of prediction rate. Meanwhile, the improvement is to find out the way to reduce recognition variation and compare different kind of classification result and accuracy. My research finding indicated that Kinect sensor V2 has the ability to record movement changes such as people are eating or drinking clearly. In this paper, I proposed recognition system for 4 types of targeted food by few kinds of features extracted from skeletons tracking and RGB-D including Support Vector Machine (SVM).