| |
Abstract:
Previous studies suggest that observers can recognize novel
2D views of moving objects more readily than corresponding views of
static objects (Kourtzi & Shiffrar, 1997). To understand how
observers recognize novel views of highly articulated objects
moving non-rigidly, we investigated the perception of human bodies
in motion. Subjects viewed two views of a human model moving one
limb. By varying the spatio-temporal separation between the views,
the human model either did (apparent motion condition) or did not
(static condition) appear to rotate smoothly in depth while moving
one limb. After this priming sequence, a pair of targets was
presented and subjects judged if these two targets matched each
other. Target models had either the same or a different pose from
the prime model. The target orientation could be the same as the
first or second prime view or a novel orientation falling either in
between or beyond these views. Faster, correct judgments of target
models in the same pose as the prime models indicate priming. Our
results showed priming for novel views in the path of the human
model's rotation. Interestingly, these novel views were not primed
when constraints of human movement were violated; that is, when one
of the model's limbs was presented in a biomechanically impossible
position. However, these same views were primed for mannequin
figures that had a similar hierarchical structure to the human body
but were not biomechanically constrained. These results suggest
that human body movements can be readily represented in a
viewpoint-invariant manner based on biological constraints of the
human body.
|
