Human recovery of the shape and size of a 3D indoor scene

Little is known as to how humans form a 3D representation of a scene from 2D visual data. Previous studies reported that human visual space is systematically distorted. However, these studies cannot be generalized beyond impoverished viewing conditions. The present study investigated how humans perceive a layout of indoor scene under a natural viewing condition. When a subject was asked to draw top view of a scene, he or she reconstructed the scene accurately except for the size, which turned out to be response bias. There was little affine or projective distortion. To test the intrinsic geometry of visual space, Foley\u27s (1972) right isosceles triangle experiment was replicated under a natural viewing condition. Unlike his conclusion in an impoverished viewing condition, the subject adjusted the right isosceles very accurately under a natural viewing condition. The same task was repeated in several viewing conditions varying the lighting condition, level of stimuli, and number of stimuli. The subject formed the right isosceles triangle more accurately in a bright room than in a dark room, with 3 objects than with 2 objects and the subject as one of the vertex. Natural environment provides several effective priors such as gravity, horizontal ground, and known height of the observer. By using these priors, humans can perceive a 3D scene veridically