The Dominance of Static Depth Cues over Motion Parallax in the Perception of Surface Orientation

Under polar projection (the natural projection for visual scenes) motion parallax is a powerful cue specifying relative depth. For small-field stimuli, it is ambiguous in the sense that a concave surface can be perceived as convex and deforming. By contrast, concavity/convexity of wide-field surfaces is unambiguously perceived. This led us to hypothesise a critical role of the 3-D rigidity constraint for large visual scenes in motion (Dijkstra et al, 1995 Vision Research 35 453 – 462).

To examine this hypothesis, we exposed subjects to planes inclined in space, and asked them to report the tilt (direction of inclination). Depth was specified either by motion parallax (MP, the surface oscillated around a frontoparallel axis) or by static perspective cues (SP, orthogonal square grids drawn on the plane).

At ECVP95, we had reported a predominance of SP over MP when the tilts specified by these two cues (t_MP and t^SP respectively) differed (1995 Perception 24 Supplement, 137). Since these results were obtained for fast movements (oscillation frequency for MP: 3.6 Hz), we extended our investigation to a slower frequency (0.5 Hz) which is more likely to be involved during natural head-movements. We found that: (i) errors in tilt reports were larger for MP than for SP, and decreased with increasing field-size; (ii) in the case of conflict (t_MP=t_SP±90°), the reported tilt was either t_MP or t_SP, rather than an average of these two values; (iii) in this case, tilt was most often reported according to SP, rather than to MP cues; this effect occurred even when the accuracies for the two individual cues were similar.

Therefore, in a conflict situation between MP and SP, surface orientation is reported according to a winner-take-all rule, which is largely in favour of static grid-cues. Hence, even for wide-field movements, the image contrast distribution can lead the visual system to prefer an unrigid, rather than rigid, solution to the 3-D shape-from-motion problem.