Building Reconstruction and Visualization from LIDAR Data

A strategy for building extraction and building reconstruction from dense LIDAR data is presented. Roofs are modeled as plane surfaces, connected along ridges and bordered by the eaves lines. Initial segmentation of terrain and non-terrain features is performed on grid data; after segmentation by region growing and region topology, classification is achieved based on a hierarchical set of aggregation rules. Areas labelled as potential buildings are further segmented in plane surfaces (roofs slopes) based on gradient orientation and plane fitting by RANSAC; a similar procedure applies to edge pixels to extract eaves lines. The topology of the roof slopes and walls is reconstructed, deriving also roof ridges and roof corners. A 3D model of the building is thereafter obtained which can be imported in CAD and visualization environments, such as those provided by the VTP group. Results from a laser scanning survey with a ground resolution of 1 m are presented for a suburban area in Pavia.