Abstract
Previous studies of figure-ground organization (FGO) used ambiguous 2D images. These stimuli contrast with our everyday experience where we view 3D scenes which are never ambiguous. In order to shed light on 3D FGO we tested the subject's ability to count pieces of furniture in 3D virtual scenes and to reconstruct their spatial arrangement. In the first experiment, the scene contained between 7 and 13 pieces of home and office furniture randomly arranged. In half of the scenes there were strong occlusions of some objects by others. Each of the 100 scenes was shown for 4 seconds. The subject's task was to count the objects. There were 3 experimental factors: monocular vs. binocular viewing, color vs. grey scale, and right side up vs. upside down scenes. The dependent variable was the error in the number of objects reported. The strongest effect was produced by the scene orientation. Binocular viewing produced better performance than monocular although this effect was small. Color had a negligible effect. In the second experiment, the subject's task was to reproduce the spatial positions and orientations of 6 pieces of furniture. The reconstruction room was viewed from a different angle compared to the reference room. Two angles were used, 22.5deg and 45deg. Viewing was monocular or binocular. Performance was very good, with most position errors being within the size of the furniture pieces. There was not much difference between monocular and binocular viewing. The rotation of the scene by 45deg was more difficult than 22.5deg. When the scene was upside down, the errors were twice as large. The presence of shadows did not affect performance. These results show that 3D scene perception is very good and that the effect of visual cues is weaker than the effect of a priori constraints such as the direction of gravity.
Meeting abstract presented at VSS 2016