Document Type

Theses, Ph.D

Rights

Available under a Creative Commons Attribution Non-Commercial Share Alike 4.0 International Licence

Disciplines

Computer Sciences, Information Science

Publication Details

Phd Thesis; University College Dublin, 2011

Abstract

Weiser’s [111] vision of pervasive computing describes a world where technology seamlessly integrates into the environment, automatically responding to peoples’ needs. Underpinning this vision is the ability of systems to automatically track the situation of a person. The task of situation recognition is critical and complex: noisy and unreliable sensor data, dynamic situations, unpredictable human behaviour and changes in the environment all contribute to the complexity. No single recognition technique is suitable in all environments. Factors such as availability of training data, ability to deal with uncertain information and transparency to the user will determine which technique to use in any particular environment. In this thesis, we propose the use of Dempster-Shafer theory as a theoretically sound basis for situation recognition - an approach that can reason with uncertainty, but which does not rely on training data. We use existing operations from Dempster-Shafer theory and create new operations to establish an evidence decision network. The network is used to generate and assess situation beliefs based on processed sensor data for an environment. We also define two specific extensions to Dempster-Shafer theory to enhance the knowledge that can be used for reasoning: 1) temporal knowledge about situation time patterns 2) quality of evidence sources (sensors) into the reasoning process. To validate the feasibility of our approach, this thesis creates evidence decision networks for two real-world data sets: a smart home data set and an officebased data set. We analyse situation recognition accuracy for each of the data sets, using the evidence decision networks with temporal/quality extensions. We also compare the evidence decision networks against two learning techniques: Naïve Bayes and J48 Decision Tree.

Download


Share

COinS