Aerial image based geometric refinement of building models derived from airborne LIDAR data

Airborne laser scanning has proven to be a strong basis for the automatic generation of 3D building models. A drawback, however, is
often in the point spacing of typical datasets. As a consequence, the precision of roof plane and ridge line parameters is usually
significantly better than the precision of gutter lines. To cope with problem the paper presents a novel approach for geometric
refinement of building models reconstructed from ALS point clouds using single aerial imagery. The basis idea of our modeling
approach it to obtain refined roof corners by direct intersection of 3D roof planes previously extracted from ALS data with viewing
planes assigned with the edges detected in high resolution digital photographs. The synergy between LiDAR and optical imagery
allows for obtaining building models with high vertical and plannimetric accuracy. In order to evaluate performance of our
refinement algorithm, we compare the results of 3D reconstruction executed using only laser scanning data and enhanced by image
information. Furthermore, quality assessment of both modeling outputs is performed based on a reference data provided by the
ISPRS benchmark for 3D building reconstruction. Integration of linear cues retrieved from single imagery allows for average
improvement of planar accuracy of 9 cm (RMS error for roof plane outlines). The overall quality of final building models calculated
on a per-area level reaches nearly 90%.