Purchase this article with an account.
Julian M. Fernandez, Bart Farell; Near optimal depth cue combination from binocular disparity and motion parallax. Journal of Vision 2006;6(6):344. doi: https://doi.org/10.1167/6.6.344.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The visual system combines information from different depth cues to estimate 3D shape. Previous studies show that humans can optimally combine texture and binocular disparity cues in estimates of surface slant. These estimates agree with predictions of a maximum-likelihood estimation model (MLE), which weights each cue in accordance with its relative reliability. We have tested a MLE model for combining motion parallax and binocular disparity in estimates of surface curvature. The structure-from-motion stimuli consisted of monocularly presented rotating cylinders whose axis of rotation—which coincided with their longitudinal axis—could be inclined away from the frontoparallel plane. The same stimuli, but static and binocular, generated the disparity cue. In a first experiment, reliability for each cue was measured using a 2AFC procedure, in which subjects had to discriminate between two slightly different elliptical cross sections. These reliability estimates were used to predict MLE weights for stimuli containing both cues. In a second experiment we obtained perceived-shape judgments using a method of adjustment on the cross sections when only one cue, and when both cues, were present. When both were present, the cues might signal consistent or inconsistent values of the cross section and of the inclination of the axis of rotation. We compared the observed weights with the MLE predictions obtained from the first experiment. We found that most observers combine disparity and motion parallax cues near optimally in judgments of surface curvature, resulting in more reliable estimates of object's shape than can be obtained from either cue alone.
This PDF is available to Subscribers Only