Universität Bielefeld - Sonderforschungsbereich 360
Towards Generic 3-D Object Recognition
Sven Dickinson
Department of Computer Science and Center for Cognitive Science
Rutgers University
Wednesday, May 20th, 1998
10.15 h
Room: D6-135
Although generic shape models typically denote representations which
capture the coarse, prototypical shape of an object, we can also ask
if a given shape representation is generic in terms of supporting
different recognition tasks. For a vision-equipped agent, 3-D object
recognition is not a single isolated task, but rather a complex
interaction of a set of closely-related behaviors, including both
bottom-up and top-down recognition, localization, subclass
recognition, active recognition, and object tracking. Not only must
these individual problems be solved, they must be integrated within a
single framework, since a complex recognition task may involve rapid
movement from one behavior to another. In previous work, we have
introduced a novel object representation integrating object-centered
and viewer-centered models. This representation, a probabilistic
part-based aspect hierarchy, has formed the basis for novel approaches
to each of the above recognition problems. In this talk, I will begin
by motivating the representation, and will proceed to describe our
current progress towards addressing these various recognition problems
within this representational framework.
Anke Weinberger, 1998-03-02, 1998-04-02