Three-dimensional (3D) cluttered scenes consist of a large number of small surfaces distributed randomly in a 3D view volume. The canonical example is the foliage of a tree or bush. 3D cluttered scenes are challenging for vision tasks such as object recognition and depth perception because most surfaces or objects are only partly visible. This paper examines the probabilities of surface visibility in 3D cluttered scenes. We model how the probabilities of visible gaps, depth discontinuities, and binocular and half-occluded points depend on scene parameters such as the size and density of the surfaces that make up the clutter, as well as on depth and inverse depth. Inverse depth is of particular interest since both binocular disparity and motion parallax depend directly on it. The probability models are verified using data from synthetic 3D cluttered scenes, which are generated using computer graphics.
© 2012 Optical Society of AmericaFull Article | PDF Article
Nasser M. Nasrabadi, Sandra P. Clifford, and Yi Liu
J. Opt. Soc. Am. A 6(6) 900-907 (1989)
Daniel Kersten, Alice J. O'Toole, Margaret E. Sereno, David C. Knill, and James A. Anderson
Appl. Opt. 26(23) 4999-5006 (1987)
G. Sudhir, Subhashis Banerjee, R. Bahl, and K. K. Biswas
J. Opt. Soc. Am. A 12(12) 2564-2572 (1995)