Effect of 3D Structure on Motion Segmentation

A smooth surface imaged on the retina produces a smooth flow field. Thus, the visual system may group regions of smoothly varying flow to segment surfaces. We tested this idea by having observers perform a segmentation task on several stimuli that differed in their 3D interpretations but were all matched in the smoothness of their 2D flow fields. Performance varied across conditions with the best performance occurring when the stimulus simulated a rigid plane. This result suggests that while observers may use deviations from smoothness to segment a broad class of motion stimuli, they use a more precise strategy to segment stimuli with a familiar 3D interpretation.