An algorithm to extract 3-D building roof points from airborne LIDAR data

Abstract

Building roof points in airborne LIDAR data are very important for 3-D building reconstruction. This paper will present an algorithm to automatically acquire the 3-D building roof points from airborne LIDAR data. Firstly, for roughly locating the area outlines of the roof(s) from pure LIDAR data, the aerial images with known orientation and original LIDAR data are used to generate the orthoimages. Then, the user indicates the area outlines of the building(s) on the orthoimages by mouse device. Afterwards, on the assumption that roofs are composed of either horizontal or slope planes, some better plane information, called GRID planes, are extracted by means of least squares fitting based on quad-tree segmentation from the LIDAR data in the area outlines. These GRID planes will be further merged according to the height constraints for providing SEED regions for plane growing by employing forward selection data snooping approach to merging the neighboring roof LIDAR points in order to extract the whole roof points. From the experiments, the efficiency of the proposed algorithm will be shown.