🚀 go-pugleaf

        ACTIVITIES FOR THE WEEK COMMENCING APRIL 10, 1989
    (SF = Sandford Fleming Building, 10 King's College Road)
         (GB = Galbraith Building, 35 St. George Street)

SUMMARY:

LECTURES ON GRAPH MINORS -- Neil Robertson
   Mon., Apr. 10, 11 a.m., GB119: "An Overview of Graph Minors"

   Wed., Apr. 12, 11 a.m., GB119: "An Outline of a Disjoint Paths Algorithm"

   Thurs., Apr. 13, 10 a.m., GB119: "A Survey of Algorithmic Problems"

COLLOQUIUM - Tuesday, April 11, 11 a.m., SF 1105 -- Gordon Bell
     "New Directions in Supercomputing"

AI SEMINAR - Thursday, April 13, 11 a.m., SF 1105 -- Daniel Huttenlocher
     "Recognizing Solid Objects from a Two-Dimensional Image"

-----------------

THREE LECTURES ON GRAPH MINORS SPONSORED BY ITRC

                         Neil Robertson
                      Ohio State University

            Monday, April 10,  11 a.m. in Room GB 119

                  "An Overview of Graph Minors"

          Wednesday, April 12,  11 a.m. in Room GB 119

           "An Outline of a Disjoint Paths Algorithm"

           Thursday, April 13,  10 a.m. in Room GB 119

       "A Survey of Algorithmic Problems in Graph Minors"

COLLOQUIUM - Tuesday, April 11,  11 a.m. in Room SF 1105

                           Gordon Bell
                         Ardent Computer

               "New Directions in Supercomputing"

                      (Abstract to follow)

AI  SEMINAR - Thursday, April 13,  11 a.m.  in  Room SF 1105

                       Daniel Huttenlocher
                       Cornell University

    "Recognizing Solid Objects from a Two-Dimensional Image"

Model-based recognition, in which stored object models are
matched against an unknown image, is an important paradigm in
machine vision.  Particular difficulties are encountered when
matching solid objects against a two-dimensional image of a
three-dimensional scene.  This talk will describe a method for
computing a transformation from a three-dimensional model
coordinate frame to a two-dimensional image coordinate frame.
What distinguishes the method is that three noncolinear model
points and three corresponding image points always define a
transformation that is unique up to a reflective ambiguity.  The
solution method is closed-form and involves only second order
equations.

We have developed a recognition system that uses this
transformation method to determine possible alignments of a model
with an image.  Pairs of local edge features in the model and the
image are used to hypothesize transformations that may align the
model with the image.  Each hypothesis is then verified by
comparing the entire edge contour of the transformed model
against the edges in the image.  This approach will be contrasted
with two common matching methods: parameter hashing and
interpretation tree search.  The alignment method has been
implemented on both a serial processor and on the Connection
Machine, and these implementations have been successfully tested
on occluded objects in highly cluttered scenes.
-- 
Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4
              (416) 978-4058
clarke@csri.toronto.edu     or    clarke@csri.utoronto.ca
   or ...!{uunet, pyramid, watmath, ubc-cs}!utai!utcsri!clarke