STEREO SLANT PERCEPTION IS MORE PRECISE IN AN OCULOCENTRIC TASK THAN IN A HEADCENTRIC TASK

((J.M. Hillis, B.T. Backus, R. van Ee, & M.S. Banks))

School of Optometry & Vision Science Program, University of California, Berkeley

Purpose. To describe the slant of a plane, a coordinate system is required. Some computational models assume that stereoscopic slant is computed with respect to the line of sight (oculocentric coordinates; Koenderink & van Doorn, 1976) while others assume that it is computed directly in headcentric coordinates (Gillam & Lawergren, 1983; Erkelens & van Ee, 1997). The variability of a computation cannot be less than the variability of the measurements on which it is based. We used this fact to examine whether slant is estimated primarily in oculocentric or headcentric coordinates. Specifically, we compared the variability of slant estimates in oculocentric and headcentric tasks. Methods. Stereoscopic planes were presented in a haploscope. The stimuli were back-projected, random-dot planes, so no perspective cues to slant were available. Observers were asked to discriminate departures in slant from internal standards of gaze-normal (oculocentric task) or frontoparallel (headcentric task). Surfaces were presented randomly at many gaze angles. A fixation point appeared 500ms prior to the stimulus to allow observers to fixate the surface location. Slant was varied according to interleaved 1-up/1-down staircases. Psychometric slopes were the index of variability. Results. Variability was significantly lower in the oculocentric than in the headcentric task. Thus, slant discrimination is more accurate when the task is specified oculocentrically. Conclusions. Greater precision in oculocentric slant estimations suggests this is a first and necessary step to computing slant with respect to the head.

Supported by AFOSR (93NL366), NSF (DBS-9309820) & HF here None