Date of Award

Spring 2009

Document Type


Degree Name

Doctor of Philosophy (PhD)


Electrical & Computer Engineering

Committee Director

Lee A. Belfore II

Committee Member

Frederic D. McKenzie

Committee Member

Roland R. Mielke

Committee Member

Andreas Tolk


In this dissertation, a strategy and framework for retrieving 3D scenes is proposed. The strategy is to retrieve 3D scenes based on a unified approach for indexing content from disparate information sources and information levels. The TREE-D-SEEK framework implements the proposed strategy for retrieving 3D scenes and is capable of indexing content from a variety of corpora at distinct information levels. A semantic annotation model for indexing 3D scenes in the TREE-D-SEEK framework is also proposed. The semantic annotation model is based on an ontology for rapid prototyping of 3D virtual worlds.

With ongoing improvements in computer hardware and 3D technology, the cost associated with the acquisition, production and deployment of 3D scenes is decreasing. As a consequence, there is a need for efficient 3D retrieval systems for the increasing number of 3D scenes in corpora. An efficient 3D retrieval system provides several benefits such as enhanced sharing and reuse of 3D scenes and 3D content. Existing 3D retrieval systems are closed systems and provide search solutions based on a predefined set of indexing and matching algorithms Existing 3D search systems and search solutions cannot be customized for specific requirements, type of information source and information level.

In this research, TREE-D-SEEK—an open, extensible framework for retrieving 3D scenes—is proposed. The TREE-D-SEEK framework is capable of retrieving 3D scenes based on indexing low level content to high-level semantic metadata. The TREE-D-SEEK framework is discussed from a software architecture perspective. The architecture is based on a common process flow derived from indexing disparate information sources. Several indexing and matching algorithms are implemented. Experiments are conducted to evaluate the usability and performance of the framework. Retrieval performance of the framework is evaluated using benchmarks and manually collected corpora.

A generic, semantic annotation model is proposed for indexing a 3D scene. The primary objective of using the semantic annotation model in the TREE-D-SEEK framework is to improve retrieval relevance and to support richer queries within a 3D scene. The semantic annotation model is driven by an ontology. The ontology is derived from a 3D rapid prototyping framework. The TREE-D-SEEK framework supports querying by example, keyword based and semantic annotation based query types for retrieving 3D scenes.