Self-organising object networks using context zones for distributed activity recognition

Abstract

Activity recognition has a high applicability scope in patient monitoring since it has the potential to observe patients' actions and recognise erratic behaviour. Our activity recognition architecture described in this paper is particularly suited for this task due to the fact that collaboration of constituent components, namely Object Networks, Activity Map and Activity Inference Engine create a flexible and scalable platform taking into consideration needs of individual users. We utilise information generated from sensors that observe user interaction with the objects in the environment and also information from body-worn sensors. This information is processed in a distributed manner through the object network hierarchy which we formally define. The object network has the effect of increasing the level of abstraction of information such that this high-level information is utilised by the Activity Inference Engine. This engine also takes into consideration information from the user's profiles in order to deduce the most probable activity and at the same time observe any erratic or potentially unsafe behaviour. We also present a scenario and show the results of our study.