ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002
APPLICATIONS OF THE ROBUST INTERPOLATION FOR DTM DETERMINATION
Ch. Briese*, N. Pfeifer, P. Dorninger
Institute of Photogrammetry and Remote Sensing, Vienna University of Technology, Gusshausstrafe 27-29,
A-1040 Vienna, Austria — (cb, np, pdo)@ipf.tuwien.ac.at
Commission III, WG III/3
KEY WORDS: DEM/DTM, Hierarchical, Scanner, Laser Scanning, Robust Method
ABSTRACT:
Different data sources for the determination of a digital terrain model (DTM) are available. New measurement techniques like laser
scanning (LS) and interferometric synthetic aperture radar (InSAR) allow a high degree of automation and dense sensing. They
pushed the development of new methods for data modelling. This paper presents robust methods for the automatic determination of
a DTM from point cloud data. These robust methods operate on the original data points and allow the elimination of off-terrain
points and the modelling of the terrain surface within one process. After a short description of the algorithms the paper focuses on
the results of this technique applied to datasets from different sensors (airborne LS, terrestrial LS, satellite LS, tacheometry and
photogrammetry).
1. INTRODUCTION
For the determination of a digital terrain model (DTM) different
methods for data capturing do exist. The choice of the most
appropriate measurement system depends on the specific
application. Apart from the "classical sensors" like
photogrammetry and tacheometry, new measurement systems
like laser scanning and interferometric synthetic aperture radar
(InSAR) have been developed in the last years and offer new
possibilities such as increasing measurement density and higher
degree of automation. These systems pushed the development of
new methods for DTM determination.
This article presents algorithms developed at our institute for
the generation of DTMs from irregular distributed point cloud
data (sec. 3). Robust methods for error reduction and
elimination are integrated into the modelling process and are
used for the classification into terrain and off-terrain points.
Finally, the results of these techniques applied to datasets
obtained by different sensors (airborne LS, terrestrial LS,
satellite LS, tacheometry and photogrammetry) are presented.
This demonstrates the generality of the proposed method (sec.
4). However, we expect that other algorithms developed for the
elimination of off-terrain points in airborne laser scanner data
(sec. 4.2) can be used in a more general context, too. In this
section we also present different strategies for data densification
and homogenisation in order to close data holes. Although it is
an extrapolation of the measured points, it is necessary for many
DTM applications.
It shall be made clear, that the algorithm for robust DTM
generation has already been presented elsewhere (e.g. Kraus and
Pfeifer, 1998), nevertheless the formulas are given in the
appendix. Its versatile application possibilities have not been
published so far and we expect that the general concept is very
useful for all who have data with gross errors and want to
interpolate a surface (not only a DTM) from it.
* Corresponding author.
2. ERRORS WITH RESPECT TO THE DTM
In the context of this paper the term "error" is used for the
disagreement between the z-value of the measured point and the
“true” terrain surface. We speak of residuals if we mean the
“true” height (t) minus the observed height (z). Under filter
value (f) we understand the negative residual, thus: t + f = z.
Because we restrict our applications to 2.5d-problems, which
means that the surface is the graph of a bivariate function, the
errors are measured in z-direction only.
As mentioned above, the data used is a cloud of points. This
point of view can be characterized as data driven and has the
advantage that it is very general and independent from the
sensor type. However, the parameters for the model derivation
as well as those of the derived models depend on the
characteristics of the sensor, which will be seen in the examples
section. On the other hand, the consideration of measurement
errors with respect to the individual sensor components would
allow a closer look on the realisation of all the measurements
for one point observation. Modelling and filtering all sorts of
errors (see below) in such a model would yield the most precise
solution (approaches for the consideration of systematic errors
in airborne laser scanner data can be found in Burmann, 2000;
Filin, 2001 and Schenk, 2001).
We aspire a surface determination technique considering errors
in the measured points in the modelling process (i.e. generating
the DTM). Apart from random errors we have to consider gross
and systematic errors in the z-values of our data points.
Random errors Depending on the measurement system, the
points measured on the terrain surface have a more or less
random distribution with respect to the “true” terrain surface. In
general this random error distribution is characterized by the
standard deviation of the measurement system and should be
considered in the DTM generation process.
Gross errors Gross errors can occur in all datasets. Especially
automatic systems like laser scanners produce a high number of
gross errors with respect to the terrain surface. In the case of
