The comparison is always made on two levels of pixel Project 3 (1:22600, h = 3200 m) considered 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
: : : d certainl 
resolution, for 15 pm and for 30 pm pixel size. The digital PACS p cru 
images of the projects 1, 2 and 3 with 30 pm pixel size were isum DEM DIAG! ] In relati 
i i : 0.10 0.10 0.10 in ielauon 
derived from the 15 pm-imagery by applying a Gauss-filter, 9 int 
and processed separately. The respective results of the projects r.m.s.d. 0.22 0.20 021 
4 and 5 are related to the last and next to last level of the DEM 9 emp 0.10 0.09 0.09 3. Time P 
pyramid. Thus, they are not entirely independent. 0.05 %oh | 0.05 %oh | 0.05 %o h : 
os sa. 30um DEM DIAG! | DIAG2 The test m 
The accuracy investigation presented here is restricted to : 
à 3 ; E a G int 0.18 0.19 0.18 processing 
DEMS without consideration of break lines. They will have to 
rods Ts r.m.s.d. 0.24 0.25 0.24 from read 
be studied in more detail in future. 
9 emp 0.17 0.18 0.17 processed 
A : workstati 
The results of the empirical accuracy tests are listed below, for 0.09 %oh | 0.10 %oh | 0.08 ooh «m seno 
each project separately. The tables include several accuracy # points 4119 122 134 ue > al 
parameters and distinguish 15 pm and 30 pm pixel size. ben 
operating 
: a ; computatio 
S im the internal accuracy indicator, as obtained Project 4 (1:22600, h = 3200 m) abs N late Or 
from the DEM generation, in m were taken 
15pm DEMI DEM2 DIAGI DIAG2 
rm.s.d.:  r.m.s. difference between DEM heights and Sint 0.28 0.28 0.28 0.28 Pro 
check heights, in m r.m.s.d. 0.56 0.65 0.55 0.57 
= G emp 0.37 0.43 0.37 0.38 
© emp‘ empirical height accuracy of the DEM, as 0119.h | 0.139.h | 0.11%.h [| 0.12 9»h 
Supe from the r.m.s.d.. in m and in relative 30pm DEMI DEM2 DIAGI DIAG2 
Sight accuracy S int 0.37 0.37 0.37 0.37 
4 points: the number of check points involved in the Em d. 0.65 0.67 0.58 058 y part of 
comparison 9 emp 0.50 0.52 0.50 0.50 
0.16 %oh | 0.16 %oh | 0.14 %oh | 0.14 Hoh 
Project 1 (1:7000, h = 1000 m) # points 2200 1092 200 152 Table 2: T: 
ge 
15pm DEM1 DEM2 DIAGI DIAG2 
Sint 0.05 0.06 0.05 0.05 The accuracy analysis of project 5 has unfortunately not been 
r.m.s.d. 0.10 0.13 0.10 0.11 completed at the time of writing. (It is nevertheless included in The table st 
Samp 0.04 0.06 0.05 0.05 this paper, as it does contribute to the computing time statistic six hours of 
0.04 %oh | 0.06 %oh I 0.05 %oh | 0.05 %o h below. It also is worth mentioning that the automatic DEM With 30 pr 
. C : It 
30pm DEMI DEMO DIAGI DIAG2 generation run through, although it was extremely difficu 
: terrain). 
© int 0.06 0.09 0.06 0.06 | 
r.m.s.d. 0.13 0.14 0.10 0.13 The above results speak for themselves. They can be 
9 emp 0.07 0.10 0.05 0.07 summarized in some statements: 
0.07 %oh | 0.10 %oh | 0.05 %oh | 0.07 %oh 
# points 2200 1092 200 152 - In smooth terrain the accuracy of the automatically 
  
  
  
  
  
  
  
generated DEM is extremely good, well below the threshold 
of 0.1 %o of flying height. In steeper terrain and with 
Project 2 (1:14000, h = 2100m) larger grid units the accuracy deteriorates somewhat, but is 
still considerably better than conventional DEMs. 
  
  
  
  
  
  
  
  
  
Lm DENI ME DL DIAC? - The pixel size of 30 pm gives, in smooth terrain, practically 
2 int en 0.10 odi oil the same accuracy as 15 pm pixel size. 
r.m.s.d. 0.15 0.11 0.18 0.20 
9 emp oi 0.08 0.13 0.14 - The internal accuracy indicator 6. Seems to represent the 
0.06 %oh | 0.04 %oh | 0.06 %oh | 0.07 %o h DEM accuracy quite well. The discrepancy appearing in 
30pm DEMI DEM2 DIAG1 DIAG2 project 4 will have to be investigated. 
€ int 0.17 0.18 0.18 0.18 
rmsd. 0.20 0.14 0.18 0.22 - Although not shown here in detail, it can be reported that 
G emp 0.17 0.12 0.15 0.19 the DEM generation succeeded very well in eliminating 3D 
0.08 %oh | 0.06 %oh | 0.07 %oh | 0.09 %oh obstacles like trees or houses, and was even acceptable for 
Pom 1428 198 12 12 most parts of the built-up areas. 
  
  
  
  
  
  
  
The test results presented here demonstrate a surprisingly high 
DEM accuracy which the automatic DEM generation is 
capable of providing, and which is not reached by conventional 
photogrammetric DEM generation. This result is almost 
exclusively based on the high density of point measurements 
in the automated system. How far the test projects can be 
988