Enhancement of Human-Preference Assisted Activity Recognition Using a Cooperative ADL Infrastructure

Abstract

In ubiquitous computing it is important to recognize a user's activities and provide corresponding services based on the user's preferences. However, learning human preferences and recognizing their activities has traditionally been dealt with separately. Furthermore, most currently available approaches first learn offline and generate static activity models which are later used to recognize activities of interest on a real time basis. Unfortunately, these static activity models become outdated when human behavior, sensor deployment, or even system configuration have changed in a dynamic environment. To address these issues, this paper demonstrates an activity-aware system built on a cooperative ADL (Activity of Daily Living) infrastructure to facilitate human-preference assisted activity recognition. In addition, the system can improve accuracy and adaptability in activity recognition via model cooperation and also can enhance the reusability of all underlying components via the proposed ADL infrastructure. We demonstrate that the activity-aware system improves the overall performance in a dynamic environment and therefore has the potential to achieve a better user experience.