Rfid Based Activities Of Daily Living Recognition

In this paper, we address the issue of indoor activities of daily living recognition with a novel Activity Recognition System (ARS). This system only uses interactions between objects. Their locations is provided by a tracking system based on passive RFID tags to compute activity probabilities. Classification within the tracking system is done through a random forest that gives over 97% in accuracy. Activities are represented by the use of a custom behaviour tree, which makes it possible to compose interactions to form activities of any complexity. Interactions and activities are defined in a human readable form to allow everyone to expand them with minimal prior knowledge.