Volume III-3
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., III-3, 407-414, 2016
https://doi.org/10.5194/isprs-annals-III-3-407-2016
© Author(s) 2016. This work is distributed under
the Creative Commons Attribution 3.0 License.
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., III-3, 407-414, 2016
https://doi.org/10.5194/isprs-annals-III-3-407-2016
© Author(s) 2016. This work is distributed under
the Creative Commons Attribution 3.0 License.

  06 Jun 2016

06 Jun 2016

ROOF RECONSTRUCTION FROM AIRBORNE LASER SCANNING DATA BASED ON IMAGE PROCESSING METHODS

S. Goebbels and R. Pohle-Fröhlich S. Goebbels and R. Pohle-Fröhlich
  • Niederrhein University of Applied Sciences, Reinarzstr. 49, 47805 Krefeld, Germany

Keywords: building modeling, airborne LiDAR, planar faces, CityGML, complex roof structure

Abstract. The paper presents a new data-driven approach to generate CityGML building models from airborne laser scanning data. The approach is based on image processing methods applied to an interpolated height map and avoids shortcomings of established methods for plane detection like Hough transform or RANSAC algorithms on point clouds. The improvement originates in an interpolation algorithm that generates a height map from sparse point cloud data by preserving ridge lines and step edges of roofs. Roof planes then are detected by clustering the height map’s gradient angles, parameterizations of planes are estimated and used to filter out noise around ridge lines. On that basis, a raster representation of roof facets is generated. Then roof polygons are determined from region outlines, connected to a roof boundary graph, and simplified. Whereas the method is not limited to churches, the method’s performance is primarily tested for church roofs of the German city of Krefeld because of their complexity. To eliminate inaccuracies of spires, contours of towers are detected additionally, and spires are rendered as solids of revolution. In our experiments, the new data-driven method lead to significantly better building models than the previously applied model-driven approach.