This paper describes a Bayesian approach for modeling 3D scenes as a collec
tion of approximately planar layers that are arbitrarily positioned and ori
ented in the scene. in contrast to much of the previous work on layer-based
motion modeling, which computes layered descriptions of 2D image motion, o
ur work leads to a 3D description of the scene. There are two contributions
within the paper. The first is to formulate the prior assumptions about th
e layers and scene within a Bayesian decision making framework which is use
d to automatically determine the number of layers and the assignment of ind
ividual pixels to layers. The second is algorithmic. In order to achieve th
e optimization, a Bayesian version of RANSAC is developed with which to ini
tialize the segmentation. Then, a generalized expectation maximization meth
od is used to find the MAP solution.