McGlone - 1
»
SOME CONSIDERATIONS IN 3D
OBJECT MODELING
Chris McGlone
Digital Mapping Laboratory
School of Computer Science
Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh, PA, 15213-3891
jcm@cs.cmu.edu
1. Introduction
Our goal is to use images to describe the world; object models are how we ’’talk
about” objects in the world. The language we define by our models determines
what we can say about objects, how easily we can say it, and how well we can be
understood by others.
There are a large number of competing model types—languages—currently in use.
Rather than a detailed technical discussion of model types, this paper will focus
on model selection criteria. The selection of a model type must be based on more
than purely computational or algorithmic considerations. While it is important for
the chosen model to be usable efficiently by the vision algorithms, for robust and
meaningful results other considerations must be taken into account.
This paper briefly discusses selection criteria in the context of model types, pointing
out comparative strengths, weaknesses, and tradeoffs. The optimal model type can
change at each phase of the computer vision process, from the initial processing to
delivery of models to an end user; pros and cons of models for each phase will be
discussed. The design decisions made in some existing systems will be discussed.
Finally, some conclusions will be outlined.
2. Model types
A few main types of models and their variations have become standard within com
puter vision research. Some have been borrowed from computer aided design, such
as constructive solid geometry (CSG) models, while others have been developed by
the vision community. This section gives an overview of some general types of object
models, in preparation for discussion of their specific applicabilities and properties
later in the paper.