Development of a personalized activity monitoring system for everyday life
Being physically active is – apart from not smoking – the most powerful lifestyle choice individuals can make to improve their health. With recent progress in wearable sensing and computing it becomes reasonable for individuals to wear different sensors all day, thus global activity monitoring is establishing: type, intensity and duration of performed physical activities can be detected. However, available systems still have many restrictions, allowing an accurate monitoring only in very limited scenarios. The project ActivityPlus addresses these restrictions by developing a personalized activity monitoring system for everyday life.
One of the main challenges is to extend the number of activities to be recognized, and to deal with all other activities occurring in daily life. This would allow to use the developed system in everyday situations, and not restrict it to scenarios where only the few traditionally recognized activities (e.g. walking, running, cycling) are allowed.
Therefore, activity recognition is regarded as a complex classification problem in ActivityPlus. New algorithms are developed to approach the challenges, focusing on the research of meta-level classifiers. The other main goal of the project is the personalization of activity monitoring. The basic idea hereby is that systems trained generally will always have a certain inaccuracy when used by a new individual. Therefore, personal information should be taken into account when developing and training activity monitoring methods.
In ActivityPlus personal data (such as age, weight or resting heart rate) is considered to create new features for classifiers, and individual movement patterns of various activities are integrated into the activity monitoring algorithms.
Funding by: Stiftung RLP Innovation