Christopher S. Gourley
Electrical Engineering
Ph. D. 1998

Pattern Vector Based Reduction of Large Multi-Modal Data Sets for Fixed Rate Interactivity During Visualization Of Multi-Resolution Models

PART 1: Virtual Reality data visualization and manipulation

Research Objectives:
The research objective is to build hardware and software for use in virtual reality visualization of large-scale models. This task consists of building a virtual reality hardware environment allowing data display and manipulation. It also involves dealing with virtual reality input devices such as a space ball, to ensure human interaction with the data being displayed. A complete virtual reality environment is to be built for efficient display and manipulation of data sets. This issue of building a virtual reality environment arises because a reconstructed scene in virtual reality typically consists of millions of triangles. Data sets are taken from various sensors such as range, color, thermal, and radiation sensors. Data sets are heterogeneous and consist not only of geometric coordinates, but also of multi-modal data. Therefore, a large panel display setup is to be built to allow efficient display of large-scale, multi-modal models.

Methodology and Results:
A virtual reality system, known as CAVE, has been built to display large-scale models. The CAVE setup is built to immerse users with the data. The system is composed of a virtual reality input device (space ball), a SGI machine (indigo2), three VGA projectors, and three large screens for display. This setup has been built with off-the-shelf hardware to optimize the trade-off cost performance. The CAVE set-up allows very efficient data display of large-scale models when multi-resolution display capabilities are used.

This work was conducted by Christopher S. Gourley under the supervision of M. A. Abidi. This work was supported by DOE's University Research Program in Robotics under grant DOE-DE-FG02-86NE37968.

Pattern Vector Based Reduction of Large Multi-Modal Data Sets for Fixed Rate Interactivity During Visualization Of Multi-Resolution Models

PART 2: Data Reduction and Constant Interactivity During Display

Research Objectives:
This research addresses the issue of maintaining constant interactivity during data display and manipulation of 3D models. Techniques are investigated to facilitate display of large data sets in real-time while maintaining both high and low resolution for easy manipulation. Efficient manipulation of models necessitates both displaying details and obtaining an overview. 

The details of the objects in the scene may reveal one very small radioactive barrel in the corner of a large room. The overview allows the fast selection of areas of interest and decision-making on orientation of the model. Therefore, this research addresses the issue of enhancing data sets with multi-resolution capabilities. This project involves defining a strategy for reducing data and building a progressive multi-resolution data structure for maintaining a constant interactivity during data display.

Methodology and Results:
A multi-resolution data reduction process has been developed to build models with multi-resolution capabilities for use during data display. The data reduction method based on pattern vector allows information selection. The algorithm projects data into a space of patterns; a pattern vector is computed for each vertex composing the initial model and the reduction process reduces one vertex at a time. Patterns are geometric and indicate the color and the type associated with each vertex. This method allows data reduction of multi-modal data sets and the building of a new data structure with multi-resolution capabilities. The constant interactivity is maintained by selecting the optimal resolution that allows users to reach a desired frame rate during data display.

This work was conducted by Christopher S. Gourley under the supervision of C. Dumont and M. A. Abidi (Thesis Chair). This work was supported by DOE's University Research Program in Robotics under grant DOE-DE-FG02-86NE37968.

Papers

C. S. Gourley, C. Dumont, and M. A. Abidi, Model building for fixed rate interactivity during display of large range data sets, Submitted to Robotics today Journal of SME.

C. S. Gourley, C. Dumont, and M. A. Abidi, Enhancement of 3D models for fixed rate interactivity during display, To be presented to Robotics and Applications IASTED Conference in Santa Barbara, CA USA, October 28-30, 1999

C. S. Gourley, C. Dumont, and M. A. Abidi, Pattern Vector Based Reduction of 3D Meshes Created from  Multimodal Data Sets, In proceedings of SPIE AeroSense (Aerospace/Defense Sensing, Simulation and Controls), Orlando USA, April 1999, 3693, pp. 112-123.

C. S. Gourley, C. Dumont, and M. A. Abidi, Fixed-Rate Interactivity for Visualization of Photo-Realistic Multiresolution Models, American Nuclear Society, 8th Topical Meeting on Robotics and Remote Systems, CD ROM publication (ISBN 0-89448-647-0, ANS order 700271), Pittsburgh, 25-29 April 1999

C. Gouley,  C. Dumont, R. Barry, and M. A. Abidi, Model building for large scale inspections,  In Proceedings of the 5th conference Quality Control by Artificial Vision (QCAV99), Les trois-rivieres, Quebec (Canada), May 18-21, 1999, pp. 235-240.