Applying Ontology and Probabilistic Model to Human Activity Recognition from Surrounding Things

Abstract

This paper proposes human activity recognition based on the actual semantics of the human's current location. Since no predefined semantics of location can adequately identify human activity, we automatically identify the semantics from things by focusing on the association between things and human activities with the things. Ontology is used to deal with the various possible representations (terms) of each thing, identified by a RFID tag, and a multi-class Naive Bayesian approach is applied to detect multiple actual semantics from the terms. Our approach is suitable for automatically detecting possible activities even given a variety of object characteristics including multiple representations and variability. Simulations with actual thing datasets and experiments in an actual environment demonstrate its noise tolerance and ability to rapidly detect multiple actual semantics from existing things.