McGlone - 7
• An interchange format should pass along any inexpressible information, even if
only as unformatted general attributes, for possible later use by other systems
which could possibly utilize the information, or for later retrieval by the initial
system.
• The interchange file must contain information about the provenance of the
data (“metadata”), such as how it was derived, what the input data sources
were, and the processing performed. Such information is crucial in correctly
interpreting and evaluating the data for import into another system.
Nodal models without point connectivity information, such as gridded models, can
be relatively simple to interchange, as long as the underlying grid does not need to
change due to a datum shift or coordinate system change. However, the addition
of point connectivity information complicates the situation. It is difficult to convert
this type of information into other types of models since there are no defined surface
boundaries for use in boundary representations, and no defined shapes for parametric
models. Instead, the data must be processed to either fit parametric shapes to the
surface, or discontinuities must be extracted to define region boundaries.
To interchange boundary representations with parametric models first requires that
they represent objects for which both systems have elementary representations.
Some semantic decisions may be necessary; consider a rectangular prism given in
both a boundary representation and as a parametric model. The points or edges
in the boundary representation may not form a mathematically perfect prism. In
translating this into the parametric model, some sort of fitting operation will then
be necessary.
3.5. Model utilization
The end use of a 3D model produced by a computer vision system may involve
presentation for human interaction, such as editing or simulation, incorporation
into a database, such as a GIS, or input to another system for further processing or
exploitation [Polis et al ., 1995]. Each application has different requirements which
must be kept in mind. For instance, if the user is doing rendering of the models,
boundary representations with their explicit representations of surfaces are optimal.
If the application is cartographic, then models which allow maximum accuracy are
preferred.
4. 3D Modeling in existing systems
This section discusses some computer vision systems which utilize 3D object models,
emphasizing the type(s) of models they use and the tradeoffs involved.
4.1. ACRONYM—generalized cones
ACRONYM, one of the earliest model-based vision systems, is based on generalized
cone models [Brooks, 1981, Brooks, 1983]. The object models exist in two hier-