Rule-based roof plane detection and segmentation from laser point clouds

This paper presents a combined bottom-up and top-down approach to 3D roof plane detection and segmentation from laser scanning point clouds. Laser scanning data of city scenes often shows noise and incompleteness because of, e.g., the clutter by trees and the reflection of windows/waterlogged depressions on the roof. Results of the bottom-up plane reconstruction may thus be limited to a number of incomplete and/or irregular regions. We proposed a joint multiple-plane detection scheme to improve the performance of the 3D Hough transform. A model-driven segmentation, which works with the constraint-rules derived from the basic roof model, is conducted to overcome the clutter and flaws in the point cloud ensuring a plausible reconstruction.