Finding Planar Regions in Noisy 3D Grid Point Data 1996

Abstract:  This paper describes an algorithm for finding regions of planar surfaces in data given as three-dimensional points located in a (possibly incomplete) grid. It was motivated by noisy range data obtained from IFSAR or stereo correlation. The procedure basically consists of the following steps: (a) fitting of robust planar patches, (b) region merging, (c) region refinement, (d) region splitting, and (e) boundary finding. The steps (b)--(d) are iterated until a convergence criterion is met. A major design principle was the lack of assumptions about specific characteristics of the input data. Hence, heuristic parameters common in most of today 's range image segmenters were avoided where possible. Rather, the same fit criteria are applied at all steps of the procedure. The algorithm can incorporate information from optical imagery where available, and cooperate with higher-level site reconstruction tasks. However, a major drawback is the resultant computational complexity of the procedure,...