Martin A. Giese
CBCL, MIT
will speak on
Representation of Biological Motion Based on Learned Prototypical
Example Patterns
Abstract:
Learning-based methods have been applied successfully for the
representation of the shape of stationary objects, like faces. The
interpolation between learned two-dimensional prototypical
views has provided a powerful techniques for applications in
computer vision and computer graphics, like face recognition and the
synthesis of artificial views of a face. Learning-based representations
have also been shown to provide a consistent account for psychophysical
and neurophysiological results on stationary object recognition. The
work presented in this talk explores if similar concepts are also feasible
for the representation of complex movement patterns and actions. The
usefulness of learning-based representations of biological motion is
demonstrated for two different fields of applications.
A new computer vision method is presented that permits to approximate
patterns of biological motion by interpolation between a small number of
learned prototypical actions. It is shown that this method can be applied
for the recognition of complex movement patterns and the style with which
actions are performed. It is shown that the same theoretical approach
permits also to synthesize artificial naturally-looking movements
by "motion morphing".
The idea of a representation of biological movements based on learned
prototypes seems also to provide a reasonable theoretical
approach to account for action recognition in the visual system.
A neural model is presented that is compatible with the known
facts from neurophysiology, and which achieves a recognition of biological
movement patterns, The model reproduces in particular the spatial and temporal
invariance properties of biological motion recognition that have been
demonstrated in psychophysical experiments. The model makes a number
of specific predictions that can be tested in psychophysical and
neurophysiological experiments.
The lecture will take place in the
Lecture Hall, Room 203, 44 Cummington St.
on Monday, February 28, 2000
at 1:00 pm
Hosted by the
Brain and Vision Research Laboratory