clarke@csri.toronto.edu (Jim Clarke) (03/21/89)
AI SEMINAR - Thursday, April 6, 11 a.m. in Room SF 1105
(SF = Sandford Fleming Building, 10 King's College Road)
Ruzena Bajcsy
University of Pennsylvania
"Perception via Active Exploration
Examples of Disassembly"
It has been accepted by most psychologists (e.g., Gibson) that perception
is an active process, that is, an interaction of the perceptual system with
its environment. In this presentation we shall limit the perceptual system
to only two modalities: the Visual and the Haptic. The goal of our work is
to answer what are the primitive actions and attributes that are measurable
and or computable from Visual and Haptic modality that are necessary for
disassembly of a scene or a two-part object. While a great deal of
attention has been paid to visual information processing both in
psychological and machine perception literature, haptic processing always
took the second seat. Haptic information processing is the identification
of objects by touch (the use of kinesthetic and tactile information). We
shall develop a theory and present two experiments to show an existential
proof of our theory. Another obvious observation is that vision is limited,
especially in discerning two separate objects when touching from the case
of part-whole relationship of one object. Take an example of a cup on a
saucer. From vision alone one cannot establish whether the saucer is glued
to the cup or the cup is just sitting on the saucer. The only way to
disambiguate this situation is to pick up the cup or shake it, in other
words to perform some manipulation operation.
The goal of this research is perception (apprehension) via exploration.
This means to us data driven perception which results in discerning solid
separable objects and describing them in terms of their structural and
geometric properties. Our aim is to explore complex scenes composed of more
than one object in arbitrary positions. Our basic hypothesis is that this
cannot be done by vision alone, that one needs some possibilities of
manipulation and the use of haptic information processing. Much of our
stimulation, especially in the area of haptic information processing, comes
from the studies and discussions of Klatzky and Lederman.
--
Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4
(416) 978-4058
clarke@csri.toronto.edu or clarke@csri.utoronto.ca
or ...!{uunet, pyramid, watmath, ubc-cs}!utai!utcsri!clarke