Warning:
This page is
provided for historical and archival purposes
only. While the seminar dates are correct, we offer no
guarantee of informational accuracy or link
validity. Contact information for the speakers, hosts and
seminar committee are certainly out of date.
GE Corporate Research and Development
One of the most difficult problems in machine vision is to find attributes of an object's shape which are invariant to viewpoint. For example, a circle appears as an ellipse when viewed off center. Extensive work over the last five years, indicates that a field of mathematics, called projective invariant theory, provides a significant advance in the problem of geometric shape description.
The theory of projective invariants was developed almost entirely in the 19th century as an important mathematical problem. Interest in the theory of invariants decreased considerably when Hilbert proved that the invariants for algebraic forms have a finite number of independent invariants, which was the central mathematical issue. However a rich body of results are available for application to computer vision.
Initial work carried out jointly between GE and Oxford University has demonstrated that projective invariant theory is a rich source of ideas in machine perception. We have constructed a successful model-based vision system using projective invariants of planar shapes, called LEWIS, which uses invariant values to index into an object database.
More recently, these ideas have been extended to the case of classes of 3D objects which provide constraints so that 3D invariant descriptions can be extracted. Examples which have been implemented so far are: rotationally symmetric objects, polyhedra, extruded surfaces and objects with repeated structure, such as bilateral symmetry.
The talk will also discuss a new system, MORSE, which is being developed to exploit the invariant approach to recognize curved 3D objects from a single intensity image. The major issues involved in the design and implementation of a recognition system will be discussed, such as grouping and global scene consistency.
Host: Yangsheng Xu (xu@cs.cmu.edu) Appointment: Lalit Katragadda (lalit@cs.cmu.edu)