Abstract
Optic flow has traditionally been assumed to be the input to structure-from-motion (SFM). To test this assumption we compared the perception of 3D structure from optic flow, actively generated by observer head translation around a stationary object, to the same optic flow, but as experienced by a non-moving observer. For stimuli composed of an Ames window-like object, where motion cues to depth conflicted with perspective, we found that active and passive observers perceived 3D structure differently: active observers relied on motion cues much more than passive observers. If, on the other hand, the virtual used to generate the stimulus was itself in motion (with respect to an earth-fixed reference frame) during observer movement, the active-passive difference disappeared. From these two results we conclude that (1) optic flow is not the sole input to SFM, but that extra-retinal information is also integrated, and (2) extra-retinal information is used in SFM to pick out objects that are stationary in an allocentric, earth-fixed reference frame. In other words, the rigidity assumption carries greater weight for a subclass of rigid objects, ones that are stationary. We tested whether stationarity was an independent criterion for 3D structure and motion in another set of experiments, where we created displays that could either be perceived as a rigid, non-stationary object, or an almost stationary, non-rigid one. When non-rigidity was not too great, subjects preferred the more stationary interpretation. We conclude that extra-retinal, self-motion information plays an important role in spatial vision, and is used to create representations of 3D structure in an allocentric reference frame that is independent of observer movement.