VASC Seminar Announcement
       =========================

Date:  Monday, March 18
Time: 3:30-4:40
Place: NSH 1507

Title:  Automatic 3D modeling from reality

Speaker: Daniel Huber

Abstract:

Modeling from reality is the process of creating realistic digital 3D models
of real-world scenes.  Recent high-profile modeling from reality projects,
such as the Digital Michelangelo Project, have shown us that range sensors
can be used for accurately modeling complex objects.  They have also taught
us that the modeling process is immensely labor intensive and demands
significant expertise.  We believe that modeling from reality should be
simple enough that an average person could snap a few 3D pictures of a
statue, their house, or any scene, and quickly and effortlessly create a
digital model.  This talk present an automatic modeling framework that takes
the first steps in achieving this goal.

Our algorithm fully automates the 3D modeling process.  Given an unordered
set of 3D views obtained from unknown viewpoints, our task is to align all
the views in a common coordinate system and then merge them into a single
representation.  This problem is difficult because we do not have any
knowledge of the sensor viewpoints or even which data sets overlap.  We
formulate this multi-view registration problem as a combinatorial
optimization over a graph of pair-wise registration results.  By considering
the global consistency of a network of views, we can eliminate incorrect,
but locally consistent pair-wise matches.  Once the views are correctly
registered, we merge them and texture-map the resulting object.

Our modeling algorithm is sensor independent and can be applied to 3D data
at any scale.  We show results using two different range sensors to
automatically model scenes varying in size from from small, "desktop"
objects to large-scale terrain.  For small objects, we have developed an
application called hand-held modeling, in which the user holds an object and
scans it from various viewpoints.  This is an easy modeling method,
requiring no specialized hardware, minimal training, and only a few minutes
to model an average object.