Visual Motion: Estimation and Understanding

                   Allan Jepson

           Department of Computer Science
              University of Toronto

The motion of the images of objects within a video sequence provide
strong cues to their three dimensional motion, as well as their shape
and relative locations within the scene.
If we have a priori models for the objects we wish to track, then
the estimation problem involves the determination of the position
and pose parameters for these models given video data.  Such a process
is said to track ``things,'' namely the a priori models.   However, it
is often the case that we do not have such detailed prior models.  In
this case we can estimate image motion by tracking ``stuff'', that is, some
locally identifiable image structure.   Recently, techniques involving
the robust estimation of layered representations for image motion
have been demonstrated to be quite successful for tracking both
stuff and things.  In this talk I will motivate these techniques, and 
provide some computational examples.

Finally, we consider developing an understanding for the
motion of various objects within a given video sequence.
For example, on observing a hand lifting a can, we may infer that an
`active' hand is applying an upwards force (by grasping) to lift a
`passive' can.  I will present a system
that derives such dynamic object descriptions directly from tracking data,
without a priori knowledge of the dynamical properties of the
things being tracked.  The approach is based on an analysis of the
Newtonian mechanics of a simplified layered scene model.



In 1976 Allan Jepson received a B.Sc. degree in Mathematics from the
University of British Columbia.  He then went to the California
Institute of Technology, where in 1980 he completed his Ph.D. in 
Applied Mathematics.  He spent two years as a postdoctoral fellow at
Stanford University in the Mathematics Department, and then joined
the faculty of the Department of Computer Science at the University
of Toronto.  From 1989 to 1995 he was named a Scholar of the Canadian
Institute of Advanced Research.  His current research interests include
various aspects of computer vision and perceptual inference (see 
http://www.cs.toronto.edu/~jepson for more information).