Wednesday 30th January 2008, 4pm-5pm
Jean Ponce (ENS, INRIA Rocquencourt, France)
High-Fidelity Image-Based Modeling
Slides: not yet available
Abstract
I will discuss some of our recent work on image-based
modeling---that is, the process of acquiring geometric models of
complex 3D objects and scenes from multiple images. I will first
present a method for computing the visual hull of a solid bounded by a
smooth surface, then carving it to recover concavities and optimize a
global measure of photometric consistency. I will then discuss a new
approach to multi-view stereopsis that outputs a (quasi) dense set of
rectangular patches covering the surfaces visible in the input
images. This algorithm does not require any initialization in the form
of a bounding volume, and it detects and discards automatically
outliers and obstacles. It does not perform any smoothing across
nearby features, yet is currently the top performer in terms of both
coverage and accuracy for four of the six Middlebury benchmark
datasets. A simple but effective method for turning the resulting
patch model into a mesh appropriate for image-based modeling is also
presented. I will also discuss applications, and ongoing work on
accurate camera calibration from stereo and motion capture from
multiple, synchronized video streams. Finally, I will conclude with a
brief overview of the Willow research activities.
Joint work with Yasutaka Furukawa at the University of Illinois at
Urbana-Champaign.