The prevalence and incidence of cost-intensive, cardiovascular diseases such as heart arrhythmia have increased con-tinuously in the past few years. According to the demographic change and the medical progress it must be assumed that further augmentations will follow. For this reason, the development of telemedical devices and systems has been pushed in order to combine a reliable diagnosis with the benefits of automation for the early detection of emergencies to prevent serious consequences by the application of anticoagulants for example. However, usual body attached sensor applications of the telemedicine have not been approved by the patients because they restrict them in their autonomy and often convey an impression of surveillance. This paper propounds a human-centered assistance system as an alternative and comple-ment to well established telemedical solutions. The architecture of the assistance system unifies probabilistic methods with techniques of description logic in order to gain knowledge of a higher level from the physical environmental data which is essential for the predictive situation recognition with coherent control procedures. With the focus on AF as one predisposing factor of cerebral insults in particular it is essential that most of the information is acquired through non-contact techniques in order to realize a non-stigmatizing assistance system which implements functionalities for the com-fort, the individual security and the safety of its user in consideration of their individual preferences.