|
|
|
| RI | Thesis Oral | 23 Jul 2002 | |
Robotics Institute Thesis Oral 23 Jul 2002
Place and Time |
Abstract |
Further Details |
Thesis Committee
Automatic Three-Dimensional Modeling from Reality
Daniel Huber
Robotics Institute
Carnegie Mellon University
| Place and Time |
WEH 4625
1:30 PM
| Abstract |
Three-dimensional (3D) modeling from reality is the process of creating 3D
digital models of real-world objects and environments. Modeling from
reality finds use in a variety of domains ranging from archaeology to
entertainment. Research has shown that range sensors are well-suited for
creating geometrically accurate models of complex scenes. However, range
sensors only capture the 3D scene structure as seen from a single
viewpoint. Multiple views must be combined in order to create a complete
model. Current methods for modeling from reality are limited in that they
require careful measurement of the sensor viewpoints or manual registration of each view. We have developed an alternative approach that fully automates the modeling process without any knowledge the sensor viewpoints and without manual intervention.
Given a set of 3D views of a scene, the challenge is to align all the views
in a common coordinate system without knowing the original sensor
viewpoints or even which views overlap one another. We call this the
multi-view surface matching problem. It can be seen as an extension of the
pair-wise surface matching problem, in which pairs of views are aligned
from arbitrary, unknown starting positions.
To solve the multi-view surface matching problem, we use pair-wise surface matching to align pairs of views, collecting the resulting matches in an attributed graph called the model graph. Unfortunately, these pair-wise
matches are frequently incorrect, and it is sometimes impossible to
distinguish the correct matches from incorrect ones at the local (pair-wise) level. However, by considering the global consistency of a network of views, we can eliminate incorrect, but locally consistent matches. We formulate the problem as a mixed discrete/continuous optimization over sub-graphs of the model graph, and we have designed several algorithms for performing this optimization. Once the views are correctly registered using one of our multi-view surface matching algorithms, we merge them and texture-map the resulting object.
Our multi-view surface matching framework 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 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.
| Further Details |
A copy of the thesis oral document can be found at http://www-2.cs.cmu.edu/~dhuber/files/huber_thesis.pdf.
| Thesis Committee |
This page maintained by robotwebmaster@ri.cmu.edu.