15 pm ] 
10.9 
120 
1034 
5.4 
6.5 
  
0.43 
2.1 
  
68 
0.9 (0.4) 
1.0 (0.5) 
2.9(1.4) | 
.4 (1.15) | 
2.3 (1.1) 
5.0 (2.4) 
5.3 
5.1 
2.6 
3.2 (1.5) 
3.3 (1.6) 
1.9 (0.9) 
n 
3300) 
  
  
dd 
^M e—HÁáÓ X 
  
he power of 
orocedure in 
e computer 
rkably short 
quite easily. 
oerformance 
de, but it is 
ff automatic 
| in MATCH- 
3.3 DEM generation 
Finally, we want to mention preliminary results which 
were attained with MATCH-AT in DEM generation mode. 
We selected the OEEPE block FORSSA as a test block 
(photo scale 1:4000; 28 images at 30 pm pixel size; 60 
% forward lap; 20 - 40 % side lap). A DEM area was 
defined in the center of the block covering approximately 
the area of 2 neighboring stereo models (Figure 14). The 
block adjustment ran through the entire image pyramid 
creating in each level a DEM with a post spacing being 
equivalent to about 30 pixel. Figure 15 shows the block 
DEM obtained for 120 ym pixel size. The block 
adjustment gave the same results for co and the 
theoretical z-coordinate accuracy oz as recently reported 
by Heuchel et al. (1996). Table 4 quotes also the 
theoretical DEM accuracy cpem Which was derived by the 
surface reconstruction part of MATCH-AT by using the 
inverse of the normal equation system and the residuals 
of the automatically matched 3D points (see details 
Krzystek, Wild, 1992). Those values are better almost by 
a factor of 2 than the standard deviation oz for the z- 
coordinate of the block adjustment. In detail, a standard 
deviation opgw Of 8 cm = 0.13 %o h was obtained for 60 
um pixel size. For the 30 pm case the corresponding 
value amounts to 5.2 cm = 0.08 %o h. The improvement 
against the oz of the block adjustment is influenced by 
the redundancy effect, since a large number of points per 
grid mesh are filtered in the surface reconstruction 
process. Those results are very promising in every 
aspect. 
  
  
Figure 14: Block DEM of OEEPE Block "FORSSA" 
In general, it is to be expected that DEMs derived from 
more than two images by multiple-image matching 
techniques will be more accurate and more reliable than 
DEMs  conventionally derived from two images. 
Especially, in the case of a 60 % side lap this benefit 
would be fully visible (Thorpe, Schickler, 1996). Future 
work will be focused on optimizing that special DEM 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
generation approach. Also, ground truths will be used for 
empirical accuracy checks to independently confirm the 
theoretical expectations. 
4. Conclusions 
We have presented an integral approach to automatic 
aerial triangulation which incorporates a DEM generation 
process. This strategy helps the system to initialize 
accurately enough homologous image patches in the 
presence of large height undulations. Also, the matching 
of image patches becomes more effective and an entire 
block DEM can be determined using consequently the 
multiple image overlap. The practical results of the 
automatic aerial triangulation indicate that even with 30 
ym pixel size excellent results can be achieved which are 
very close to those with a high precision triangulation. 
Furthermore it is to mentioned that the system was 
successfully applied to a block of complex form with 
many side strips The preliminary results of the DEM 
generation for a block DEM are promising and indicated 
an expected height accuracy of 0.1 %o h. Future work 
will be focused on the system development and, 
especially, on optimized matching strategies to take 
advantage of the accuracy potential of the 15 pm pixel 
size. 
  
Figure 15: Block DEM 
  
  
  
  
  
  
  
pel | co[um] | Oz | ODEM ODEM 
[um] ([pel])([em]) [cm] [em] _| [%o h] 
60 ]19.1(0.31)(7.6)]| 136 | 80 | 0.13 
30 | 13.5(0.35)(4.1)|| 9.1 | 52 | 0.08 
  
  
Table 4: Theoretical DEM accuracy of block DEM 
5. Literature 
Ackermann, F., 1996a. Experimental Tests on Fast 
Ambiguity Solutions for Airborne Kinematic GPS 
Positioning. XVIII ISPRS Congress Vienna, 9 - 19 July. 
Ackermann, F., 1996b. Some Considerations about 
Automatic Digital Aerial Triangulation. In Proceedings, 
OEEPE Workshop "On the Application of Digital Image 
Workstations", Lausanne, 4 - 6 March. 
Ackermann, F., Tsingas, V., 1994. Automatic Digital 
Aerial Triangulation. ASPRS/ACSM Annual Convention, 
pp. 1 -12, Reno. 
413