SEAbIRD: Adaptable Daily Living Activity Identification from Sensor Data Streams

Abstract

One of the biggest concerns in Ambient Assisted Living (AAL) proposals is helping the population of elderly people in order to maintain their independence and autonomy. A relevant task done by AAL is the automatic inference of a person's activities of daily life (ADL) from data streams recorded by sensors deployed on an active environment. This work proposes an ADL discovering system which consider factors as personal behavior changes and respect for privacy. The proposed system is tested and validated under a dataset from a real user. The results show that our system can operate adequately on a real scenario with the respective constraints. The main contribution of this work is a system for ADL detection that can adapt to user's behaviors changes without retraining the model, considering sensor failures and preserving the user's privacy.