非表示:
キーワード:
-
要旨:
INTRODUCTION. Despite impressive gains in computational and graphical power over the
last decade, virtual reality (VR) simulations still often fail to convey a natural impression of
scale, dimensions, or distances. This however is a key requirement for many potential VR applications,
as for example architectural walk-throughs. Recent models of spatial vision suggest
that the core of the problem lies in the way different depth cues are integrated by the human visual
system similar to a mandatory weighted maximum likelihood estimation (e.g., [1]). Since
most VR setups simulate only a few depth cues, strong adversary but in itself consistent depth
information like a xed physical screen distance may therefore substantially compress the perceived
depth dimension. However, cue integration might also bear the key to an applicable
workaround solution: Varying easily adjustable simulation parameters might offer a potential
for compensating for the shortcomings of more intractable depth cues.
OBJECTIVE. The empirical study presented here investigates effects of changing the horizon
height in VR by a frustum shift. Observations on distance estimates in reality, the form
of the human visual eld, and the so-called cognitive framing effect of a restricted simulated
eld of view ([2]) suggested that a raised horizon could particularly affect the perception of
egocentric distances.
METHOD. In a psychophysical experiment eight participants estimated three main dimensions
and two egocentric distances in 20 virtual rectangular rooms from two different scene
sets. In addition, possible side effects of the experimental parameter were roughly evaluated
by a semantic differential rating in eight principal experiential categories in subsets of four
rooms. The stimuli were presented on a calibrated desktop VR system. Radiosity-rendered
interiors were generated using a custom-made tool that allowed to hide the manipulation of the
experimental parameter by a balanced variation of scene features.
RESULTS. Albeit verbal distance estimates are known for showing high variances between
subjects, relative horizon height and perceived egocentric distances were signicantly correlated
(correlation coefcient r=0.16**, p=0.003): The underestimation of distances was reduced
by 29 on average. Whereas the inuence of the experimental parameter on exocentric
dimensions (horizontal r=0.01, vertical r=0.07) and rated experience were below signicance
level and obviously irrelevant.
DISCUSSION. The results are in accordance with both the cognitive framing theory and
the model of mandatory fusion of depth cues. Our ndings suggest that a raised horizon height
is particularly suitable for improving the perceived ego-location in VR without negatively affecting
other perceptual properties. Besides this practical improvement of VR simulations, the
study shows that VR can also be effectively used in basic vision research.
