Abstract
Humans can recover structure from motion (SFM) based on the projected 2D motion field of a rotating or translating object. Recovery of SFM from rotation has been studied almost exclusively in the condition where the axis of rotation lies in the frontoparallel plane. Here we assess the ability to recover SFM in the general case, where the axis of rotation may be tilted out of the frontoparallel plane.
Subjects observed simulated elliptical cylinders whose cross-section was constant along the axis of rotation. The cylinders were defined by short-lived random-dots to minimize depth cues from changing dot density; occluders masked the cylinders' borders to eliminate cues from boundary shape. Subjects adjusted a comparison cross-section so it matched the perceived cross-sectional shape of the test cylinder. We found that subjects accurately matched the simulated shape of the cylinder regardless of the inclination angle of the axis of rotation. This result is surprising: the cylinder's shape should depend on the perceived axis of rotation, but using an independent measure we show that subjects do not perceive the axis of rotation veridically.
The results suggests that SFM is computed in two stages. The first stage uses only velocity components perpendicular to the projected axis of rotation. It computes relative depth as if the axis were in the frontoparallel plane. The second stage transforms this depth representation to take into account the angle of inclination. The model predicts that depth order violations can occur when shape changes along the axis of rotation. We tested this prediction in a second experiment using circular cylinders whose radius changed along the axis of rotation. Depth order violations were found. We propose a physiologically plausible implementation of the two-stage model for computing SFM for the general case of arbitrary axis of rotation.
This research was supported by NEI Grants EY12286 to B.F. and F32 EY015637 to J.F.