A Bayesian approach to the aperture problem of 3D motion perception

Abstract

We suggest a geometric-statistical approach that can be applied to the 3D aperture problem of motion perception. In simulations and psychophysical experiments we study perceived 3D motion direction in a binocular viewing geometry by systematically varying 3D orientation of a line stimulus moving behind a circular aperture. Although motion direction is inherently ambiguous perceived directions show systematic trends and a Bayesian model with a prior for small depth followed by slow motion in 3D gives reasonable fits to individual data. We conclude that the visual system tries to minimize velocity in 3D but that earlier disparity processing strongly influences perceived 3D motion direction. We discuss implications for the integration of disparity and motion cues in the human visual system.