🚀 go-pugleaf

           Department of Computer Science, University of Toronto
              (GB = Galbraith Building, 35 St. George Street)

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

                                    AI
                   GB221, at 11:00 a.m., 8 October 1992

                             Christopher Brown
          Department of Computer Science, University of Rochester

                      "A Goal-Oriented Vision System"

A brief historical review of the place of selective (or attentional)
mechanisms in machine vision is followed by illustrations of real-time
sensori-motor capabilities that current technology makes available fairly
easily.  A flexible  task-oriented vision system is described for use in
domains with predictable visual structure (e.g. anatomy, airports, table
settings, traffic).  in which a visual task is formalized as a question
about the scene.  Bayes nets represent a priori and dynamically-acquired
knowledge.  Decision theory, using benefit/cost ratios for visual actions,
provides a basis for choosing which visual action to perform next.  The
goal of the  system is to choose the sequence of visual actions that has
the maximum expected utility for answering the visual question.
Implemented on a parallel processor (eight-node SGI computer), the system
runs in simulation with dynamic scenes and with real data for  static
scenes.