E-Locus - Institutional Repository of the University of Crete - A real-time semantics-aware activity recognition system

Home A real-time semantics-aware activity recognition system

Results - Details

[Add to Basket]

Identifier

000372833

Title

A real-time semantics-aware activity recognition system

Alternative Title

Ενα σημασιολογικά-ενήμερο σύστημα αναγνώρισης δραστηριότητας σε πραγματικό χρόνο

Author

Ευθυμίου, Βασίλειος Παναγιώτης

Thesis advisor

Αντωνίου, Γρηγόρης

Abstract

This paper introduces a system that exploits the semantics of ontologies, to improve the accuracy of machine-learning methods. This system is intended to perform real-time activity recognition that could assist users in an Ambient Intelligent (AmI) environment. Ontologies can offer quite rich semantics that, to our knowledge, have not been utilized yet in a system that performs real-time activity recognition, mainly due to the time-cost of loading/querying ontologies. For the utilization of the information that ontologies offer (such as hierarchies), we introduce novel methods to record them in machine learning datasets. We also present ways that make real-time activity recognition viable, by keeping execution-time and memory requirements low. Post-processing of the suggested activities is also involved to make sure that the returned results will contain as few errors as possible. Results on the experiments that we held, in order to present the performance of our system, are also provided.

Language

English

Subject

Activity Recognition

Ambient Intelligence

Semantics-Aware System

Αναγνώριση δραστηριότητας

Διάχυτη νοημοσύνη

Σημασιολογικά-ενήμερο σύστημα

Issue date

2012

Collection