hul 2004
7
A
AS.
J
DD
T
LÀ
alization:
In: Joint
ssing and
A METHOD FOR THE IMPROVEMENT ELEVATION DATA GENERATED FROM
AUTOMATED PHOTOGRAMMETRIC METHODS INTO SIS
A.M. Felicísimo?, A. Cuartero?, F.J. Ariza"
* Dept. of Graphic Expression, Extremadura University, 10071 Cáceres , Spain. acuartero@unex.es amfeli@unex.es
h Dept. of Cartography, Geodesy and Photogrammetry Engineering, University
^
of Jaén, 23071, Jaén. Spain
fjariza(cgujaen.es
Commission VI, WG VI/4
KEY WORDS: DEM, Stereoscopic, Matching, Correlation, Accuracy
ABSTRACT:
DPW generates DEMs with very large storage volumes may require quite large computation loads. Then its integration into a
SIS may lead to conflict between these huge data sizes and the operations of analysis, mapping algebra, simulation, etc. It would
clearly be advisable therefore to avoid a blind inclusion of all of tie data derived from the photogrammetric process, and only to take
those which are both, of good quality and significant in representing the relicf. By eliminating poor quality data, one will improve
the model's accuracy, and by eliminating unnecessary data one will reduce the redundancy, the data volume and the computational
loads. Those effects will improve the efficacy of the SIS or GIS, in which the DEM is just been integrated.
We propose, and analyze, a method based on a single and objective criterion designed to generalize DEMs by the way of data
redundancy removal with no appreciable loss of accuracy. The criterion is to state a correlation threshold for the correlation values
that the DPW assigns to each point of the DEM, so that points which correlation values are below the threshold will be rejected. For
several DEMs we checked the resulting accuracy against a set of more than 7000 ground control points measured using differential
GPS techniques obtaining satisfactory results
1. INTRODUCTION
The digital photogrammetric workstation (DPW) introduced
major changes into the flow of tasks in analytical
photogrammetry. One of these changes was the possibility of
automating part of the steps of the photogrammetric process
itself. For instance, automation allows one to construct DEMs
with an almost arbitrarily large density of points. In the manual
process, however, it had been necessary to make a preliminary
selection of only those points that the operator interpreted as
significant in describing the relief (these were called VIPs, very
important points).
Due to this very high density of points, DPW-generated DEMs
may attain quite massive computational sizes. Then their
integration into a geographic information system (GIS) may
lead to conflict between these huge data sizes and the operations
of analysis, mapping algebra, and simulation. It would clearly
be advisable therefore to avoid a blind inclusion of all of the
data deriving from the photogrammetric process, and only take
those which are both of good quality and significant in
representing the relief. By eliminating poor quality data, one
will improve the models accuracy, and by eliminating
unnecessary data one will reduce the redundancy. Both effects
will improve the efficacy of the characteristic operations of the
GIS, in which the DEM is just one more of the variables to be
considered.
To carry out such a cleaning up process, one needs an objective
criterion for the selection of which data to keep and which to
discard. DPWs generate a correlation value for each point of the
DEM. This value depends on the success of the stereo-matching
operations used to estimate the parallax from which the
elevation is calculated. It is to be expected a priori that high
correlations correspond to more reliable points, whereas
keeping points with low correlations could introduce errors.
N
UA
We here analyse the efficacy of a method of selecting reliable
points on the basis of their correlation values. We evaluate the
error of the DEM as points are eliminated and determine the
threshold separating reliable points from those of high
uncertainty. This process of simplification allows one to
"lighten" the data structure, making it better adapted to
integration in a GIS.
2. BACKGROUND
Constructing a DEM on the basis of correlation methods usually
gives better results than using conventional analytical
techniques. Ackerman (1994) studied the case of aerial
photograms, finding that the precision of the parallax depends
on the stereo-correlation technique: 0.1-0.2 pixels with least
squares matching (LSM) and 0.3-0.4 pixels with feature based
matching (FBM ).
Dependence on analogue aerial images was formally broken in
1980 when the American Society of Photogrammetry and
Remote Sensing (ASPRS) included the possibility of using
digital data from remote sensing in its definition of
photogrammetry (Slama, 1980). The techniques of
photogrammetric restitution have been known for decades, but
the possibility of using satellite images did not arise until 1986
with the launch of the first of the SPOT series satellites. The
work of Priebbenow & Clerici (1988) dealt with the
cartographic utility of panchromatic SPOT images, whose 10 m
pixel size was compatible with a 1:50 000 scale. Mukai et al.
(1989) studied the generation of DEMs from the overlap zones
of contiguous Landsat-TM images. The estimated RMSE with
60 ground control points was 92 m, approximately thrice the
pixel size. An identical study performed with panchromatic
SPOT images of Japan's Central Alps (Mukai et al., 1990) gave
RMSE values of 26 m with 40 ground control points taken from
pre-existing 1:25 000 maps. Sasowski & Petersen (1992)
carried out the same test for a zone of Alaska. They obtained a
