Caloric expenditure estimation for health games

Abstract

With the decline in physical activities among young people, it is essential to monitor their physical activity and ensure their calorie expenditure is within the range necessary to lead a healthy life style. For many children and young adults, video gaming is one favorable venue for physical activities. A new flavor of video games on popular platforms, such as Wii and Xbox, aim to improve the health of young adults through competing in games which require players to perform various physical activities. These popular platforms detect the user movements, and through an avatar in the game, players can be part of the game activities, such as boxing, playing tennis, dancing, avoiding obstacles, etc. Early studies used self-administered questionnaires or interviews to collect data about the patient's activities. These self-reporting methods ask participants to report their activities on hourly, daily, or weekly basis. But self-reporting techniques suffer from a number of limitations, such as inconvenience in entering data, poor compliance, and inaccuracy due to bias or poor memory. Reporting activities is a sensitive task for overweight/obese individuals with research evidence showing that they tend overestimate the calories they burn. Having a tool to help estimate calories consumption is therefore becoming essential to manage obesity and over-weight issues. We propose providing a calories expenditure estimation service. This service would augment the treatment provided by an obesity clinic or a personal trainer for obese children. Our energy expenditure estimation system consists of two main components: activity recognition and calories estimation. Activity recognition systems have three main components: low-level sensing module to gather sensor data, feature selection module to process raw sensor data and select the features necessary to recognize activities, and a classification module to infer the activity from the captured features. Using the activity type, we can estimate the calories consumption using existing models on energy expenditure developed on the gold standard of respiratory gases measurements. We choose Kinetic as our test platform. The natural user interface in Kinect is the low-level sensing module providing skeleton tracking data. The skeleton positions will be the raw input to our activity recognition module. From this raw data, we define the features that help the classifier, such as the speed of the hands and legs, body orientation, and rate of change in the vertical and horizontal position. These are some features that can be quantified and passed periodically (e.g., every 5 seconds) to the classifier to distinguish between different activities. Other important features might need more processing, such as the standard deviation, the difference between peaks (in periodic activities), and the distribution of skeleton positions. We also plan to build an index for the calories expenditure for game activities using the medical gold standard of oxygen consumption, CO2. Game activities, such as playing tennis, running, and boxing are different from the same real world activities in terms of enegy concumption and it would be useful to quantify the difference in order to answer the question of whether these "health" games are useful for weight loss.