Tilt perception from optic flow in two-view stimuli

Abstract

We studied the visual perception of plane orientation from motion parallax. Tilt is the direction of a surface normal, as projected in the frontoparallel plane. Slant is the angle between the surface normal and the line of sight. For a rotation in depth, the component of frontal translation Tf is orthogonal to the frontoparallel rotation axis, and W (winding angle) is the angle between tilt and Tf. We measured the ability to report the tilt of a moving plane, investigating the role of the field of view (FOV), W, slant and the dot speed. In theory, the assumption of a zero translation in depth yields a unique solution for tilt in a large field (LF) and two solutions (tilt and Tf) in a small field (SF). The stimuli were sequences of 2 views of a dotted plane rotating in depth under perspective projection (PP) presented monocularly with a FOV of 8 (SF) or 60 deg (LF). W was equal to 0, 22.5, 45, 67.5, or 90í Slant ranged from 17.5 to 35í 3D rotation angle was chosen such that dot speed did not depend on slant in SF. Perceived tilt was indicated by adjusting a probe superimposed on the stimulus. We found that a large FOV decreases the rate of depth reversals (35% in SF vs 0.1% in LF). Considering the errors in tilt direction, independently of the tilt sign, we measured the absolute tilt error (from 0 to 90í . This error is significantly smaller in LF than in SF, and significantly increases with W in SF. In SF, the perceived tilt lies in-between tilt and Tf for W=0í to W=75í, and at W=90í the tilt responses present 2 shallow peaks centered at tilt and Tf. Tilt judgments are little affected by slant, but improve significantly as dot speed increases in SF. In conclusion, the critical variables for tilt estimations are the FOV, W and dot speed. The effect of W and dot speed is stronger in SF. The shift of the perceived tilt toward Tf in SF is not explained theoretically so far.