Return to search

A hybrid knowledge-driven approach for composite activity modelling and recognition in smart environments

The rising ageing population worldwide impacts social and economic facets of modern society, and eventually the quality of human life. There is an increasing demand on a new paradigm of healthcare provisioning that can help address the growing needs of the elderly who are more likely to experience age-related physical frailties and cognitive decline, and also the shortage of public healthcare resources. Ambient assisted living (AAL) with smart homes as a specific realisation of the metaphor has emerged as a realistic technology-driven approach to supporting independent living and delaying institutionalisation. Activity recognition plays a pivotal role in the identification of users' behavioural needs, thus allowing AAL applications to provide context-aware assistive services. This Thesis conceives, designs and develops a hybrid knowledge-driven ontology-based approach for activity modelling and recognition and underlying technologies and tools, which provides a viable technological solution for AAL and advances research frontiers of associated research areas. The Thesis has been conducted through four complementary studies, each addressing a core aspect of the proposed approach. The first study developed and evaluated a hybrid approach to activity modelling that uses ontologies to specify activity models for activities of daily living (ADL), and temporal logic to represent inter-activity relationships for composite activities. The second study developed a dynamic sliding time window-based mechanism for segmenting streaming sensor data to support real-time activity recognition. The mechanism includes a fonnal time window model and its parameters together with algorithms that dynamically manipulate the parameters at runtime to vary the length of the time window. The third study developed a unified approach to simple and composite activity recognition. The approach provided a modular architecture that was realized as a multi-agent system, with agents playing various roles and cooperating to identify simple and composite activities. The final study developed techniques that analyze logged activity data to learn new activities and preferences to adapt initial activity models to make them more complete and responsive. This Thesis implemented ontologies, a software prototype for the proposed approach, and supportive utility tools, e.g. simulator, and synthetic ADL data generator, for experimentation. In addition, all methods and algorithms have been tested and evaluated using various synthetic and real ADL datasets. Experiment results have demonstrated the feasibility of the approach to support real -time activity recognition for both simple and composite activity recognition.

Identiferoai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:591053
Date January 2013
CreatorsOkeyo, George Onyango
PublisherUlster University
Source SetsEthos UK
Detected LanguageEnglish
TypeElectronic Thesis or Dissertation

Page generated in 0.0021 seconds