dc.contributor.author	Foudeh, Pouya
dc.date.accessioned	2024-02-21T00:27:29Z
dc.date.available	2024-02-21T00:27:29Z
dc.date.issued	2020
dc.description	Thesis (PhD. (Computer Science))
dc.description.abstract	This research proposes an ontological model for sensor-based human activity recognition (HAR). Ontology is one of the most effective models for knowledge representation, reasoning and reuse. Even though they are widely used, ontologies as flat text files are not efficient enough for query processing over large knowledge bases as compared to relational databases. Handling uncertainty is another ongoing and challenging research topic in ontologies area. Previous researches address these problems independently, and not both simultaneously. A model for storing ontologies in relational databases is proposed in the form of tables that contain ontology’s semantic material with accompanying probability values. Subsequently, SQL functions and triggers for keeping probabilistic database constraints are defined, in which it later performs probabilistic reasoning to answer queries. To assess these approaches, a European Union research dataset, “OPPORTUNITY” that provides data from the body and environment sensors with the aim of identifying high-level activities is utilized. Firstly, signal processing methods is proposed to convert raw signals from different types of sensors into probabilistic information about low-level subjects’ activities and location, in which information was stored as probabilistic semantic triples in a relational database. Secondly, probabilistic implication between subjects’ location, low-level activities and high-level activities are determined from labelled instances and can be developed or edited by the user according to his own background knowledge. Coarse-grained activities are obtained from a reasoning process that uses fine-grained instances and assertion axioms that are both probabilistic in nature. The results of the first stage consisting of the most probable candidates of low-level activities are compared with a real challenge participant, organized by developers of the dataset. The proposed method obtained very close results in terms of accuracy while it is more optimal in terms of the number of features and required time. In the knowledge driven stage, in addition to semantic advances of the proposed model, the results indicate improvement in terms of accuracy and significant performance in terms of time. The qualitative assessment of the proposed model reveals its advantages against existing models. The results provided support for the hypotheses that utilization of probabilistic ontological modelling and relational database management systems in a HAR system can improve the performance and efficiency of the knowledge-based system. Unlike most efforts on sensor-based HAR which focus on real-time systems, this research has yielded many contributions: scalability; processing over a considerable amount of sensor data in a reasonable time; beneficial in different applications including employee monitoring, under-parole criminal monitoring; and medical or praxeological studies on people behavior.
dc.description.sponsorship	Faculty of Engineering - School of Computing
dc.identifier.uri	http://openscience.utm.my/handle/123456789/1007
dc.language.iso	en
dc.publisher	Universiti Teknologi Malaysia
dc.subject	Ontologies (Information retrieval)
dc.subject	Human activity recognition
dc.subject	Human behavior—Research
dc.title	Scalable probabilistic ontology-based method for human activity recognition
dc.type	Thesis
dc.type	Dataset

Scalable probabilistic ontology-based method for human activity recognition

Files

Original bundle

License bundle

Collections