!!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!!
-------------------------------------------------------------------------------
                                       
                                       
                   B O S T O N   U N I V E R S I T Y
                                       
                      Computer Science Department
                                       
                          C O L L O Q U I U M
                                       

                           Vision as Feedback:
      Toward Desktop Software Systems for Vision-Based Interaction

			      Greg Hager 

                    Department of Computer Science
		            Yale University


 			 Wednesday, March 31st
                                       
                                11:00am
                       (Coffee served at 10:45am)

                                       
                         Seminar Room / MCS 135
                                       

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


       Real-time vision is a rich source of feedback for dynamic
    interactions with the physical world.  In particular, the ability to
    visually track people and the objects they manipulate can be expected
    to play a central role in future systems for human-computer
    interaction.  Our goal is to develop effective, general-purpose
    algorithms and programming abstractions for real-time vision upon
    which such systems could be based.

      In this talk, I will first present some of the ideas behind XVision,
    a modular, portable software system for visual tracking.  In
    particular, one of the central abstractions in XVision is to model
    visual tracking as a problem of stabilization over variations in the
    appearance of a target.  To illustrate this idea, I will describe an
    algorithm that can track a general 3-D object while accounting for
    changes in pose and illumination as well as partial occlusion.  We
    have shown that, for any n linear deformations of an m pixel image,
    this algorithm can be computed using as little as O(n(n+m)) operations
    per frame.  Experiments confirm that this amount of computation can
    easily be performed in real time by off-the-shelf, commodity
    processors.

       In the second half of the talk, I will briefly describe some of the
    applications we have developed using XVision.  Our experiences with
    these applications have led to consider how one might construct the
    "next generation" programming environment for real-time vision
    systems.  In particular, we have begun to use a set of techniques,
    referred to collectively as "functional reactive programming," to
    create just such an environment.  At the end of the talk I describe
    some of the directions this work is taking us.


    Host: Stan Sclaroff (sclaroff@cs.bu.edu)

-------------------------------------------------------------------------------
For colloquium info, including directions, see http://cs-www.bu.edu/colloquium
-------------------------------------------------------------------------------