(A) study of deep neural network-based algorithm for real-time monitoring of physical activity types in children and adolescents

Abstract

The increase in the obese population is closely related to the obese population of children. This is because the probability of obese children becoming obese adults is two times higher than that of normal children. It is very important to manage children's obesity. The devices that measure calorie consumption are often marketed as commercial products, however, they are not applicable to children and there is a need for tools to measure children's physical activity. It is more appropriate to record activity types than to measure calorie expenditure because children's physical activity is not continuous and repeats to act and stop for a short period of time. In this study, a deep neural network based algorithm was developed to improve the time resolution of measuring children's activities, and devised a method to distinguish between continuous motion and intermittent motion at one time. As a result of classifying the physical activity of children, higher accuracy of classification result than the accuracy of the conventional machine learning algorithm was achieved. It could be applied to the real - time activity type classification system in the future.