Intelligent Systems Laboratory
Perspective Projection Expert
Among all the features in the 2D perspective projection image
of a 3D scene, perhaps the most
significant information--carrying feature is the 2D geometry
of the projected object. This
is so because perspective projection preserves, in a definite
mathematical way, much of the geometry of the objects in the 3D scene.
When the scene contains an object whose particular 3D geometry is known,
the mathematical constraints between the observed 2D geometry and the
3D geometry of the object model are so strong that inference about what
in the image constitutes the observed 3D object, where in the 3D world the
object is, and what its orientation is becomes possible without exponential
searches. The conditions under which this is possible constitute an
important part of the knowledge base of the state-space search
to accomplish the image feature to model feature matching.
In this project, we are developing an optimal algorithm which, with
a suitable perspective projection knowledge base, will have an expected
computational complexity which is
linear in the number of
features.
