Peter Lohmann
APPROACHES TO THE FILTERING OF LASER SCANNER DATA
Peter Lohmann, Andreas KOCH, Michael SCHAEFFER
University of Hanover, Germany
Institute of Photogrammetry and Engineering Surveys (IPI)
Lohmann@ipi.uni-hannover.de
koch@ipi.uni-hannover.de
mschaeff(@)gmx.net
Working Group III-2
KEY WORDS: Laser scanning, filtering, DEM/DTM generation, surface modelling
ABSTRACT
Recently several papers have been published on the precision of Digital Terrain Models (DTM) which were produced
by airborne laser-scanning. This precision is influenced by several factors, like navigation accuracy, quality of reference
data but also by the way the "raw" laser scanner data is filtered. For the latter task mathematical models as well as
algorithms can be used.
A specific problem in quantifying the accuracy is the typical asymmetric error distribution of laser measurements when
compared to reference data. The data show only small negative deviations to the terrain-surface (below the terrain-
surface), however, relatively large positive deviations due to vegetation or buildings (above the terrain-surface). This
general problem must be taken into account.
Two different approaches for the filtering of laser-scanner data are presented in this paper, namely the use of linear
prediction and the use of dual rank filters. Both methods are presented and their results are compared. If linear
prediction is used as filtering method, it must be applied iteratively, because otherwise the results are strongly
influenced by height points lying far above the mid-terrain-level. The use of linear prediction shows satisfactory results
in forest areas, whereas other areas with steep terrain show problems which makes it necessary to locally adapt the
method to the shape of the terrain.
Using the above approach of dual rank filtering some pre-knowledge of the area of concern is required and very often
the process of filtering is influenced by interactive procedures like restricting the filter process to special areas of
interest, to avoid a smoothing of the "natural" terrain. Dual rank filters show very promising results especially when
filtering artificial objects like buildings but require interactive control and some pre-knowledge in order to properly set
the necessary parameters.
1 INTRODUCTION
Since more than a decade the use of airborne laser scanner systems leads to height-models (Digital Surface Models,
DSM), which describe the surface (vegetation-horizon, roof-heights of the houses, etc) very precisely . While there are
application areas, in which the height of surface itself plays a major role, (urban planning, radio or mobile phone
network-planning, etc), in many cases a description of the true topography (the terrain) is required (forest-stands, traffic
line-planning, terrain-supervision, Geographic Information systems, for example ATKIS, the Authorative Topographic
Cartographic Information System of the German surveying administration).
In the course of processing, higher vegetation-horizons and also buildings are removed, in order to generate a digital
terrain-model (DTM). Removing of the height-values is called filtering. There are diverse methods and procedures for
generating the DTM (Lohmann et al., 1997), like:
= Morphological Filters (Eckstein et al., 1995)
= Linear Prediction (Kraus et al., 1997)
= Spline-Approximation (Axelsson, 1998)
= General Digital Image Processing (Von Hansen et al., 1999)
Obviously, a fully automatic approach would be of advantage for reducing the DSM to a DTM. Different filter-
algorithms have been reported upon (Fritsch et al. 1994, Kilian et al. 1996, Kraus et al.1998, Hug et al. 1997, Huising et
al. 1998, Axelsson 1998 etc.), however nearly all of these methods still need improvement. In some cases the additional
540 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
use 0
last-[
This
infor
(Krai
(Eck
Data
Syste
is lox
is an
Steir
worl
scan
a cru
70-8
The
terra
deta:
the f
Surv
The
DTN
The
othe
fror
defi
