ays. 
"ical 
the 
] is 
, as 
1g a 
eve 
er, 
| to 
ige 
vas 
ule 
js 
le, 
rd 
Ice 
he 
he 
ng 
Brandt, Steffen 
  
watercourses, analogue map material, and the analytical (P1) or digital (PHODIS-ST10) photogrammetric images all 
ensured an elevation accuracy of 20 — 30 cm. 
Compared with manual photogrammetry, the DTMs generated completely automatically using MATCH-T (feature- 
based matching) were distinguished by a very dense DTM grid with a very small mesh size. Approx. 1 million points 
(pixel size 30um or 2.5 million points for pixel size 15um) per stereo model were matched, of which as a rule some 
80,000 mesh points represented the DTM, according to the finite elements interpolation (pixel size 15pm). 
According to the manufacturer, in favourable flat areas a DTM accuracy of 0.05 per thousand of the flying altitude can 
be attained or in hilly terrain 0.2 per thousand of the flying altitude. The generation was based on the classification of 
homologous points and the reconstruction of DTM bases. Corresponding with each level of the image pyramid a DTM 
was calculated. The result was a DTM pyramid illustrating the terrain surface more densely and precisely from level to 
level. The criteria according to which MATCH-T made the provisional classification were the epipolargeometry, the 
parallel axis value, the gradient signs, the interest value and the correlation coefficients. First, using an intersection of 
lines, calculation points were provisionally allocated for each model. The use of robust methods allowed the 
identification and elimination of interfering surface objects such as individual houses or trees. Additional, previously 
recorded data can also be added to the process of the DTM calculation, i.e. breaklines, recess areas and outlines. 
Within the context of research and development, the correlation approach for the DTM calculation was analysed from 
an economic and qualitative aspect. Parallel, analytical measurements on the C100 were taken for several selected areas 
as reference measurements for comparison of the correlation results at a later date. Each test area was selected 
according to prominent object structures. So that a separate assessment of the correlation accuracy in dependency on the 
object structure could be undertaken later. 
For a detailed illustration of the 
errors in elevation between 
manual measurement and 
automatically calculated digital 
terrain models, the test areas 1 
and 5 of the stereo pair 680/681 
were selected as examples (for 
further information concerning 
the test set-up, please contact 
DSK-DG). An analytical 
measurement twice is available 
for both test areas. The square 
mean of differences for Area 1 
was + 0.053 m and for Area 5 € 
0.057 m. This represents an 
accuracy 0.06 per thousand of the 
flying altitude. 
  
Figure 2. Illustration of the elevation differences of analytical / digital 
photogrammetry 
Both examples were calculated with 19 different parameter combinations in two different scan resolutions. They 
contrast with the other test areas in that they were not disturbed by the shadows cast by trees, bushes or urban structures 
such as houses or bridges. They were simply examples of agricultural land (farmland). The error vectors in the 
orthophotos showed that the accuracy of the correlated DTM sharply decreased for the ditch systems, a particularly 
sensitive segment of the watercourse network between farmland and meadows. 
The square mean of differences for the whole of Area 1 was 17 cm (for 1000 and 2000 dpi) and for the whole of Area 5 
it was 24 cm for the 2000 dpi images and 47 cm for the 1000 dpi images. However in both Area 1 and Area 5, 
especially in the field borders, errors in the order of 0.80 — 1.20 m were not uncommon. This can probably be accounted 
for by the thick vegetation of the ditch. Figure 2 shows an example illustrating the uncertainty regarding elevation 
accuracy which is inherent to the system of a correlated DTM. Depending on the topography, the original data record 
can attain accuracy levels of 20 cm to 3.50 m (cf. Figure 3). 
In many places in recent years, intelligent filter techniques have been developed which reduce and correct either 
incorrect correlations or a surface distorted by the actual topography. Scattered trees or groups of trees are smoothed out 
and recess areas are interpolated. Depending on the required accuracy and the topography, re-working of this kind can 
certainly be effective. However, the DSK-DG has established that the systematic error rate of manual or semi-manual 
re-working techniques, particularly for breaklines, does not represent a realistic alternative to manual stereo 
photogrammetry for cost and quality reasons. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 215