International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
— picking the point, leading to effects like very small angles between projection ray An
— re-measurement of this point and and terrain surface or hidden terrain regions. These effects, hei;
— confirming the measurement. caused by steep terrain features, occur mainly at the border of DE
b) False points often lay in low textured areas, where the images and therefore also at the border of the stereo models. abo
measurement within the original photographs on analytical Height measurement is often difficult or impossible in such
plotters is much better than measurement within the regions. Hei
digitised images with a digital resolution of 8 bit. As the border regions usually are in addition covered by the pre
¢) Verification and measurement in steep areas, neighbouring stereo model with different insight to the terrain,
especially near ridges, is rather complicated as it is choosing the most suitable stereo model for point measurement m
difficult to find out the correct stereo match of the regular can significantly improve the overall accuracy of DEM points. LA)
distributed points that are often incorrect in this areas. Therefore the preliminary DEM and orientation parameters of 1.0
Recognizing these problems in stereo verification of matched the images are used to calculate the angle between the surface 3.0
points, considerations have been made to improve the accuracy normal vector of the terrain and the projection rays to the aerial :
and the efficiency of the semi-automatic matching procedure. images for each point of the matched DEMs in the areas of
As shown in figure 1, an additional analysing tool was overlap. Thus each DEM point can be allocated to the most
established, using a priori knowledge of the terrain and Suitable model (see figure 2). Tabl
additional information about the matched points.
2.4
Afte
Matched DEM points per poin
stereo model, overlapping at quali
the margins of the model quicl
“goo
Stere(
éecnenenenenencentds = /Ænononoonononocoo) digit:
+ A A he .
# model 1 2 # model 2 5$ Miss.
e (XXXIX) Z/ N BHNODONIODOQOD x
Knowledge based Preliminary DEM, prelir
analysing of the measured breaklines mean
matched points point classes X DEM
ETE em Typic
AE 2 | often
(1) Point deletion near breaklines LN AN 4 chang
(2) Point allocation to suitable stereo model ZZ i, 55 SL 7 Gs are ve
(3) Point ranking based on quality estimation CE ZZ DR Zn ZZ TE of the
n A :
RR EAT erra ; CE a In tc
s " e P o o rA MP a
| I I, v a 2 A | includ
; € -— 1
AN ^. live fi
Overlap of models | live t
and al
Analytical Verification at at digi
measurement of digital plotter for Figure 2: Allocation of terrain points to suitable models in The g
i : Apa | image
2.3 Ranking of matched points based on quality estimation | availal
2 ; ; mount
; i The third step of knowledge based point analysing tries to | image:
Figure 1: Semi-automatic matching strategy for high mountain estimate the quality of the matched DEM points. Useful mr
. x . ~ . ^ . . e .
DEM information for that task can be derived from the preliminary va
DEM and the quality assessment of the matching procedure, Solin
The results of this tool are point to model allocation for points and in the special case of glacier measurement also the area | id
in areas of model overlap and a quality ranking and sorting of covered by glaciers. Initially all points arc ranked in the : |
the matched points for further verification. The strategy and ^ medium quality class 3. Better ranking up to class 2 or 1 and
results are explained in the following chapters. worse ranking down to class 4 or 5 depends on |
— quality assessment provided by the matching |
2.2 Allocation of terrain points to suitable stereo models software, | During
: : ; — height difference between the matched point and the | "ei
First of all matched points near measured breaklines are E Yih eie ur: p | decisiv
: : : gt A : ; corresponding height in the preliminary DEM and | well-ac
deleted, as verification of these points is often complicated like " |
: ; E lr co al — the angle between the surface normal vectors of for ma
mentioned above and not necessary for terrain description. d
: : : x : : Ee matched DEM and preliminary DEM | areas ]
For DEM generation of extensive regions a lot of aerial images : om ; ; : :
: 5 : : Height differences in glacier areas are mainly caused by natural accurac
with partly overlapping model areas are processed. Usually the : ; se rule : : |
. changes. Therefore the additional information about glacier have be
central regions of the models are selected or the DEM of a
U : ML + ; areas can be used to adapt the ranking strategy. The res
overlapping regions are merged together with interpolation : : TR ;
neshôds The parameters for point ranking have been optimised testing for the
methods. : ; He : ps
i : : ; : ; the results of the different classes of point quality at the digital
But, in high mountain areas the orientation of the terrain to the j pp d'ou point quai ee | develop
2 as re à s : stereo workstation to minimize miss-classification and effort for | E =
camera positions often is very different from image to image, verification | uropez
|
754
