ONR MURI Presentation Computer Vision Research

1
ONR / MURI PresentationComputer Vision Research

• MIT Seth Teller et al.
• UCB Jitendra Malik, Paul Debevec et al.

2
The Capture Problem

• Urban geometry
• Appearance data
• Reillumination info

Develop effectivesensors, automatedand
semi-automatedsoftware tools for rapid
environment capture
How can we import 3D scene data quickly and
automatically? Starting point for
visualization, design, simulation, teaching.
3
Toward Automated Site Modeling

• Synergistic MIT/UCB efforts

3) Geometry, reflectance estimated
combinatorially, and with robust statistics
1) Many geo-located images acquired
2) Images are spatially indexed
4
Scientific Issues

• Capture urban exteriors (built structure
  foliage)
• Acquire images near-ground, inside scene
• Generate accurate camera pose estimates
• Reduce/eliminate load on human in the loop
• Develop (semi)-automated optimization schemes

5
Challenges (Research/Engineering/Systems)

• Robust instrumentation for absolute geolocation
• Effective optimization schemes for pose,
  structure
• Sparse/dense correspondence algorithms
• Incremental/multiresolution reconstruction
• Scaleability in of images, output features
• Estimation of surface appearance (texture, BRDF)
• System assessment (speed, error, cost)

6
Technical Approach

• MIT
• Geo-located camera (earth coordinates)
• Acquire thousands of digital images
• Structure, texture estimated automatically
• UCB
• Paul ...
• Synergies/Complementarities
• Appearance estimation (BRDFs, sky models)
• Cost, throughput, error assessments

7
Preliminary results
MIT
Prototype sensor 4,000 images 1
CPU-day Textured CAD model
UCB
8
Next steps / success
Capture MIT Campus (200 structures) from 1 Tb
of ground, 1 Tb of aerial imagery