ROBUST PROCEDURES FOR GIS DATA HOMOGENIZATION
D. Fritsch, F. Crosilla
Chair for Photogrammetry and Remote Sensing/Istituto di Urbanistica et Pianificatione
Technical University Munich/Universita degli Studi di Udine
Arcisstr. 21, D-8000 Munich 2, Germany/Via Larga, 42, 1-33100 Udine, Italy
Tel.:+49-89-2105 2671; Fax +49-89-2809573; Telex: 522854 tumue d
email:dieter@photo.verm.tu-muenchen.de
Commission III
ABSTRACT
Vectorial data acquisition for Geographic Information Sys-
tems (GIS) is a real bottleneck which is to be overcomed by
a combination of results of surveying, photogrammetry, and
map data digitization. The homogenization problem consists
of consistency checks in first part with the more accurate
data set, therefore math models must be developed to de-
cide on data acceptance and rejection respectively.
The paper introduces with overall accuracy measures for
the three data acquisition methods. Its main part solves the
mathematical problem when all three data sources are joined
together. The corresponding linear models and hypothesis
tests are shown. It concludes with pros and cons if different
objective functions (L1, La, Lo;) are used for parameter esti-
mation.
Key words: GIS data acquisition, data homogenization,
math models, hypothesis tests, objective functions.
1 Introduction
Geometric data acquisition for Geographic Information Sys-
tems (GIS) can be done by different methods of surveying,
photogrammetry and cartography. This process is driven by
two main parameters: costs and accuracy which are depend-
ing on each other. In order to fill the databases of a GIS very
fast maps are digitized and preprocessed to fit into a refer-
ence frame of control points, to overcome isolated mapping
regions, and to realize constraints such as straight lines, per-
pendicularity and others. Map digitization is cheap in terms
of acquisition time but bad in accuracy. It can considerably
be improved when photogrammetry and surveying deliver a
set of control points by means of photogrammetric restitu-
tion, tacheometry and GPS, as it is well-known.
In this context the homogenization process consists of
similarity transforms between mass points obtained during
map digitization and control and additional check points ob-
tained by photogrammetry and surveying. Moreover, also
photogrammetric models can be transformed to fit into
the frame given by more precise reference points, e.g. GPS
points.
The math model dealing with such transforms can be
block adjustment with independent models (K. Schwidef-
sky/F. Ackermann, 1976). This model is capable for a rig-
orous handling of observations not only of photogrammetry
but also of digitized maps and surveying. A realization of
this approach can be found in H. Wiens (1986) and W. Ben-
ning/Th. Scholz (1990) — in the following not only the trans-
form itself will be treated but also a comprehensive hypothe-
sis test procedure. The combination of parameter estimation
and hypothesis testing leads to a chaining procedure which
is a feedback loop: after testing the residuals on Gaussian
distribution the data snooping starts to detect points which
do not fit into the given reference frame. After some itera-
tions the overall accuracy of digital cartography is estimated
which should be improved considerably compared with a pri-
ori values given in table 1. In this table accuracy measures
are given according to different map scales.
Table 1: The ground tracking speed and accuracy of manual
digitizing.
Scale Ground Speed (km/hr) Ground Accuracy (m)
1:10 000 54 2
1:20 000 108 4
1:25 000 135 5
1:50 000 270 10
1:100 000 540 20
Regarding the tracking speed during map digitizing a
good operator captures data in a rate of about 1.5 mm per
second. This is to maintain a tracking accuracy of about 0.2
mm. These figures indicate that there is no room left for
more accurate data acquisition but the final data processing
should result into much more accurate values in particular if
large scale maps are digitized.
In order to complete the overall measures of accuracy for
photogrammetry o, and surveying c, we can state the fol-
lowing values:
