Building detection and roof extraction in laser scanning data

A strategy for building detection and roofs extraction, relying on airborne laser scanning data only is presented, which ultimately aims at object reconstruction. Roofs are modeled as plane surfaces, connected along ridges and bordered by the eaves lines. Detection is performed by first removing the terrain pixels by smooth interpolation and then applying region growing to group elevated regions, later filtered out to select those likely to represent buildings. Roof extraction is obtained by classifying pixels in each region as roof slopes, ridges, building outlines. Roofs slopes are identified by finding plane surfaces which are segmented based on gradient orientation. The same procedure is used to extract roof outlines, further segmented in eaves lines. The topology of the roof slopes and walls is finally reconstructed, which allows to compute roof ridges and roof corners. Examples from a laser scanning survey with a ground sampling of 1 m are presented.