The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008 
(2002) separate ground and above-ground objects by a method 
that recursively fragments the entire LIDAR data into a set of 
piecewise planar surface models in order to make underlying 
terrain slope variations regularized into homogeneous plane 
terrain. Its performance is good for large buildings and all types 
of vegetation but poor for buildings located on slopes and low 
vegetation (Sithole and Vosselman, 2003). Roggero (2002) 
first separates ground from above-ground objects, and then 
from the latter dataset, buildings and trees are extracted. The 
method has been tested at various test sites but it is very 
sensitive to the input parameter values. Brovelli et al. (2002) 
applies edge detection before region growing, and then the 
terrain model is computed. Buildings and trees are extracted 
after a region growing process. Wack and Wimmer (2002) 
interpolate the original data to a regular grid and use a 
hierarchical approach in combination with a weighting function 
for the detection of raster elements that contain no ground data. 
The weighting function considers the terrain shape, as well as 
the distribution of the data points within a raster element. 
Many researchers use 2D maps as prior information for building 
extraction in Lidar data. In Haala and Brenner (1997), 
geometric primitives were estimated based on histogram 
analysis. In general, in order to overcome the limitations of 
image-based and Lidar-based techniques, it is of advantage to 
use a combination of these techniques (Ackermann, 1999). 
Building reconstruction fusing Lidar data and aerial images was 
presented in Rottensteiner and Briese (2003). Firstly, they 
detected building regions in raw data, then, roofs were detected 
using a curvature-based segmentation technique. Additional 
planar faces were estimated with aerial images. Sohn and 
Dowman (2007) used IKONOS images to find building regions 
before extracting them from Lidar data. Straub (2004) combines 
information from infrared imagery and Lidar data to extract 
trees. 
Few commercial software allow automatic object extraction 
from Lidar data. In TerraSCAN, a TIN is generated and 
progressively densified, using as parameters the angle points to 
make the TIN facets and distance to nearby facet nodes 
(Axelsson, 2001). In SCOP++, robust methods operate on the 
original data points and allow the simultaneous elimination of 
off-terrain points and terrain surface modelling (Kraus and 
Pfeifer, 1998). 
In summary, most approaches try to find objects using single 
methods. In our strategy, we use different methods using all 
data with the main aim to improve the results of one method by 
exploiting the results from the remaining ones. 
3. INPUT DATA AND PREPROCESSING 
The methods presented in this paper have been tested on a 
dataset located in the Zurich airport. The area has an extension 
of 1.6km x 1.2km (Figure 1). The available data for this region 
are: 
3D vector data of airport objects (buildings) 
Colour and CIR (Colour InfraRed) images 
Lidar DSM/DTM data (raw and grid interpolated) 
3.1 Vector data 
The 3D vector data describe buildings with 20 cm vertical 
accuracy. It has been produced from stereo aerial images 
(Section 3.2) using a semi-automatic approach using the CC- 
Modeler software (Gruen and Wang, 1998, 2001). 
Figure 1. Aerial image of Zurich Airport. 
3.2 Image data 
RGB and CIR images were acquired with the characteristics 
given in Table 1. 
RGB | CIR 
Camera Type 
Analogue 
Focal Length 
303 mm 
Scale 
1:10150 
1: 6000 
Forward overlap 
70% 
70% 
Side overlap 
26% 
26% 
Scan resolution 
20 micron 
GSD (calculated) 
12.50cm 
7.25cm 
Date of acquisition 
July 2002 
July 2002 
Table 1. Main characteristics of RGB and CIR flights. 
The images have been first radiometrically preprocessed (noise 
reduction and contrast enhancement), then the DSM was 
generated with the software package SAT-PP, developed at the 
Institute of Geodesy and Photogrammetry, ETH Zurich (Gruen 
and Zhang, 2003; Zhang, 2005). The matching uses multiple 
matching primitives and multi-image matching methods and 
automatic quality control. A detailed description of the 
matching method is given in Zhang (2005). For the selection of 
the optimum band for matching, we considered the image 
ground resolution, and the quality of each channel based on 
visual checking and histogram statistics. Finally, the NIR band 
was selected for DSM generation. The final DSM was 
generated with 50cm grid spacing. 
Figure 2 shows a part of the generated DSM. 
Figure 2. Matching DSM at Zurich Airport. 
Using this DSM, CIR orthoimages were produced with 12.5cm 
ground sampling distance. 
314