Special AI Seminar - Nov 10
Thomas Bartold
Friday, Nov. 10, 2000
12:00 noon
175 ATL (Large Conference Room)
"Structure from Motion without Correspondence"
Frank Dellaert, Ph.D. Student
Carnegie Mellon University
Robot Learning Laboratory (http://www.cs.cmu.edu/~rll/)
ABSTRACT:
Structure from motion (SFM) is concerned with recovering the 3D structure
of a scene from images taken at different viewpoints. This has many
applications in robotics as in other areas. Most if not all SFM methods
that work with sparse features assume that the correspondence between
features in different images is given, or has been obtained in a separate
pre-processing step.
In my talk I will present a method to recover 3D scene structure and
camera motion from multiple images *without* the need for correspondence
information. Instead of a separate preprocessing step that looks at pairs
or triples of images, our method works with all the images at the same
time, and solves for structure, motion and correspondence simultaneously.
To this end, we frame the problem as finding the maximum likelihood
structure and motion given only the 2D measurements, integrating over all
possible assignments of 3D features to 2D measurements. The approach is
cast within the framework of Expectation-Maximization, which leads to an
intuitive iterative algorithm: at each iteration a new structure from
motion problem is solved, using as input a set of 'virtual measurements'
obtained from a distribution over feature assignments. This distribution
can be efficiently obtained by Monte Carlo Markov Chain sampling. The
algorithm works well in practice, as will be demonstrated using results on
several real image sequences.
BIO:
I graduated from the University of Leuven, Belgium, with an M.Sc. in EE,
in 1989. After a brief (obligatory) stint in the Belgian Air Force, I
worked in industry for two years before returning to graduate school in
the US. I obtained an M.Sc. in Computer Science from Case Western Reserve
University, Cleveland, in 1995. Since August 1996, I am a Ph.D. candidate
at the department of Computer Science at Carnegie Mellon University,
Pittsburgh. My main research focus is on probabilistic methods for
computer vision and robotics, in particular sampling based methods.
Altogether, I have authored over 20 publications on these topics. I work
mostly within the setting of the Robotics Institute at CMU, where I am a
member of the VASC group (Vision and Autonomous Systems Center).
Please peruse my homepage at www.ri.cs.cmu/~dellaert