!!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!!
-------------------------------------------------------------------------------
                                       
                                       
                   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
                                       
         Recognition of multi-agent action from perceptual features

			   Stephen Intille

                      Vision and Modeling Group
                         The Media Laboratory
                Massachusetts Institute of Technology


			 Wednesday, March 17th
                                       
                                11:00am
                       (Coffee served at 10:45am)

                                       
                         Seminar Room / MCS 135
                                       

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

    The goal of my research is to create computational systems that can
    model, recognize, and respond to human activity using non-obtrusive
    sensing.  Algorithms that automatically assign semantic labels to
    observed activity can be used to simplify and improve manually
    intensive and tedious tasks like video annotation and video
    surveillance and to create entirely new types of interactive
    interfaces and environments.  Existing computer vision systems use
    representations of action that are adequate for recognition of some
    single-person action but that are not suitable for recognizing the
    coordinated activity of multiple people and objects from noisy
    perceptual data in real scenes.

    In this talk, I will present a framework for representing and visually
    recognizing coordinated multi-agent action from perceptual data in
    scenes with many agents.  The representation is motivated by work in
    model-based object recognition and has three components: (1) temporal
    structure descriptions representing the temporal relationships between
    agent goals, (2) belief networks for probabilistically representing
    and recognizing individual agent goals from visual evidence, and (3)
    belief networks automatically generated from the temporal structure
    descriptions that support the recognition of the complex action.  I
    will discuss a system that uses the representation to recognize
    football plays from noisy trajectory data.


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

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