Spatial and temporal surface interpolation in structure from motion

The built-up of perceptual performance with time in structure-from-motion (SFM) tasks has been much debated (e.g., Treue, Husain & Andersen, 1991; Ullman, 1984), even though most of the current models trying to account for human SFM are based on an instantaneous analysis of the optic flow (e.g., Domini & Caudek, 1999). In this study we investigated this problem with a novel paradigm. The ability of human observers to discriminate between SFM displays specifying two different surface orientations was studied in cases in which no single frame pair of the motion sequence provided sufficient information to perform the task. Specifically, two planar surfaces having same slant but different tilt (+45 or −45í) were simulated by using constant optic flow fields. Each frame of the stimulus sequence displayed two dots only. The number of frames of the motion sequence (40, 80, 160, 240) and dot-lifetime (10 or 60 frames) were manipulated. Observers were asked to discriminate between the two simulated tilts. The results reveal that, with the shortest sequence length, performance was at chance with the longest lifetime, but above chance with the shortest lifetime. Moreover, performance improved with the length of the stimulus sequence, but was always better with the shortest lifetime. These findings indicate that, within a limited temporal window, human analysis of the optic flow benefits from the motion signals presented in previous moments in time, thus suggesting that human SFM may be based on a process of spatial and temporal surface interpolation.