Dr. David Luebke:
Christopher Lutz:
Rui Wang:
Cliff Woolley:
 
luebke@cs.virginia.edu
ccl9n@cs.virginia.edu
rw2p@cs.virginia.edu
jcw5q@cs.virginia.edu
 
The Department of Computer Science
School of Engineering and Applied Science
University of Virginia
www.cs.virginia.edu
www.cs.virginia.edu/~gfx/

The Scanning Monticello project at the University of Virginia Computer Science Department (a joint project with researchers at the University of North Carolina at Chapel Hill) aims to create an extremely accurate model of Thomas Jefferson's Virginia home. Using recent technology known as a laser-range scanner (see sidebar), millions of data points can be gathered quickly and precisely. This data can then be transformed into a highly accurate three-dimensional model of Monticello, ready to be used for scientific study or simple exploration.

We have developed an interactive "Virtual Monticello" display for the New Orleans Museum of Art, showcased in the major exhibition Jefferson's America & Napoleon's France.  Our work and this exhibit are funded by a grant from the National Science Foundation.

CURRENT RESEARCH TOPICS:

  • Scan Editing:
    • Manual scan repair, allowing a user to explicitly remove anomalies and artifacts introduced during the scanning process
    • Algorithmic data resampling for automatic hole-patching and intelligent "dust particle" filtering
  • Registration and Surfacing:
    • Assembling individual scans into full data-sets (transforming all scans into the same global coordinate system)
    • Novel methods for combining multiple scans to fill in "holes" and create a single coherent 3D model.
  • Representation and Display:
    • 3D Geometry: points, triangles, polygons
    • Image-Based Rendering: splats, layered-depth images
Our partners at the University of North Carolina include Dr. Anselmo Lastra and Dr. Lars Nyland (faculty), Nathaniel Williams, Chad Hantak and Kok-Lim Low (students), Kurtis Keller, and John Thomas (staff).  Additional artwork provided by Ben Cloward.  This work was jointly funded by National Science Foundation awards ACI-0205425 and ACI-0205324, with additional support from Mitsubishi Electric Research Laboratories.
Laser Scanner Diagram DeltaSphere 3000

The Laser Range Scanner:
A laser-range scanner works by emitting a beam and recording the distance the beam traveled before striking a surface (r). By further recording the rotation angles around the y- and z-axes (t and p, respectively), the scanner quickly creates a highly-accurate series of sampled points in 3-space, often taking millions of samples per minute. However, problems can occur when the scanner encounters:
  • Unreflective (matte) surfaces - no range measure
  • Highly reflective surfaces - skewed range value
  • "Dust particle" interference - skewed range value
  • Motor inaccuracies - incorrect (t, p) values
Part of our research focuses on addressing these problems and extending scanning techniques to be more robust, accurate, and automatic.  We acquire our datasets obtained using the DeltaSphere 3000 (shown above), a commercial laser-range scanner built by 3rdTech, Inc. and developed at the University of North Carolina.


A fraction of the samples from a single scan, displayed as individual points to create a "screen door" effect that helps visualize the interior space of the room.

A sub-portion shown at high resolution after triangulation; note the color inaccuracies where the photographs overlap.

A top-down view of the dataset, illustrating a novel viewpoint useful for architectural schematics.

Roughly 1/8 of the dataset displayed using the QSplat program developed at Stanford University, producing a "painterly" effect.

A VRML model of the dataset created by our partners at the University of North Carolina.

A view of a scan with artificial lighting adding; note the minute range details that were not captured in the photographs.



A panoramic view of laser-return intensities (roughly 19.5 million points)
Site created April 1, 2002
All content and images (c) University of Virginia
Relevant links: www.Monticello.org
www.3rdTech.com
www.noma.org || www.nsf.gov
http://graphics.stanford.edu/software/qsplat/