Sander Oude Elberink 
  
THE USE OF ANISOTROPIC HEIGHT TEXTURE MEASURES FOR THE SEGMENTATION gp 
AIRBORNE LASER SCANNER DATA 
Sander Oude Elberink* and Hans-Gerd Maas** 
*Faculty of Civil Engineering and Geosciences 
Department of Geodetic Engineering, Section of Photogrammetry and Remote Sensing 
Delft University of Technology 
S.J.OudeElberink G student.tudelft.nl, 
**Department of Forestry- Geo- and Hydro Sciences, Institute for Photogrammetry and Remote Sensing 
Dresden University of Technology 
hmaas @rcs| .urz.tu-dresden.de 
Working Group III/3 
KEY WORDS: Airborne laser scanner, image processing, segmentation, classification 
ABSTRACT 
Airborne laser scanning data has proven to be a very suitable technique for the determination of digital surface model 
and is more and more being used for mapping and GIS data acquisition purposes, including the detection and modeling 
of man-made objects or vegetation. The aim of the work presented here is to segment raw laser scanner data inq 
unsupervised classification using anisotropic height texture measures. Anisotropic operations have the potential t 
discriminate between orientated and non-orientated objects. The techniques have been applied to data sets from 
different laser scanning systems and from different regions, mainly focussing on high-density laser scanner data, The 
results achieved in these pilot studies show the large potential of airborne laser scanning in the field of 3-D GIS di 
acquisition. | 
1 INTRODUCTION 
In the last few years laser altimetry has become a very attractive and reliable technique for the acquisition of 3) 
information. Beyond pure elevation model oriented applications, users began to examin the suitability of laser scanner 
data for the generation of 3-D city or landscape models. A crucial pre-requisite for the generation of object models fron 
laser scanner data is the segmentation of data sets. The segmentation of laser scanning data has often been performed 
using an external data source like available 2-D GIS data or multispectral image data, acquired independently from th: 
laser scanner data. Haala et al. (1998) describes the use of ground plan information to improve the reconstruction 
buildings. Lemmens et al. (1997) shows the fusion of laser-altimeter data with a topographical database to deriv 
heights for roof-less cube type building primitives. However, in some cases suitable external data will not be available, 
so that the segmentation has to be performed based purely on the laser scanner data itself without any additional source 
of information. Maas and Vosselman (1999) show two approaches for the automatic derivation of building models from 
raw laser altimetry data, based on the analysis of invariant moments of point clouds and the other approach based on the 
intersection of planar faces in triangulated points. 
The aim of the work presented in the following is to segment raw laser scanner data in a classification using anisotropi 
height texture measures. Texture is qualitatively and quantitatively defined by height, variation of height in loci 
windows and measures such as homogeneity and contrast. Some of the measures are obtained by anisotropic operation 
to avoid false segmentations along object borders. Objects to be segmented are buildings, trees, infrastructural objects 
and several sorts of agriculture ground use. While the digital surface model given by first pulse laser scanner data wil 
for example show large local height variations in regions with vegetation, it will show much lower variations and 
systematic behaviour on man-made objects such as roads or roofs of buildings. An important aspect of the work is the 
analysis of the benefit of reflectance measurements and simultaneous first and last pulse registration data. The 
reflectance measurements will be a crucial band to discriminate between several sorts of agricultural fields and also 
between roads and agricultural fields. If first and last pulse registration is available, the completeness of tit 
classification will increase, especially the discrimination between buildings and trees. 
The techniques have been applied to data sets from different high-density laser scanning systems. In this paper t 
kinds of data will be described. Figure 1 shows a part of a data set acquired with the FLI-MAP system over an urban 
area in the Netherlands. The FLI-MAP system was installed in a helicopter; the average point density within the last 
scanner strips is in the order of 5 points / m”. Figure 2 shows a part of a data set acquired with an Optech ALTM 1210 
  
678 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
ta