International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV. Part B2. Istanbul 2004 
2. METHODOLOGY 
The principle of the method is quite simple. Using the DEM 
under investigation, one can produce two orthophotographs 
from two different photographs. Matching the two 
orthophotographs and running the suggested mathematical 
model, a TIN with DEM discrepancies can be calculated. 
Running statistics over this TIN leads to checking of the 
initial DEM while addition of the TIN to the DEM produces 
a more accurate DEM. 
Calculation of the height discrepancy is a two-step problem. 
[t begins with two matched points in left and right 
orthophotographs as input data and should return a height 
correction in a specific position of the DEM. The calculation 
of the height correction is one thing, and the calculation of 
the exact position is another. It is not to forget that if the 
matched points in the two orthophotographs do not coincide 
(that is they do not have exactly the same geodetic co- 
ordinates), neither of them is correct, hence the exact position 
must be calculated. The key point is that the planimetric 
displacement due to height error is always radial to the nadir 
of the corresponding photograph (Kraus, 1992). 
It is critical to calculate the exact height error in each 
planimetric position. The basic quantities can be seen in 
figure 1, and the basic formula for the height discrepancy 
calculation is 
R2" Dh R2'' ; 
—— = —— — & Dh" = — I” 
M'2". 5.41! Acorr 
Hence Dh’’ and similarly Dh’, can be calculated exactly. The 
final Dh on the point can be the average of the two values. 
The complete theoretical model is detailed described by 
Georgopoulos and Skarlatos (2003). 
Acorr (2) 
LQ QR 
Ng -« DEM (erroneous) 
M 
: Mit dr 2 
> + AM 
Oh / .ph" i Ground (truth) 
Dh ne : x 
N Ec 
L me m 
5 
^ corr 
A 
Section 
pue ET Reference Plane 
A Acor A" 
NR Planimetric 
    
Diagram of the basic concept of the proposed 
method. 
Figure 1. 
Un 
Un 
3. TEST AREA. 
In order to test thoroughly the proposed algorithm, a 
manually collected DEM was necessary. It must be noted that 
the manually collected DEM was only to test the integrity of 
the proposed algorithm. The algorithm itself is designed to 
work without reference DEM in any area. 
The test area has been selected from 1:17000 scale colour 
photographs. It is equal to half photogrammetric model and 
can be seen in figure 2. It is 1500x1200 meters and the height 
range is 58 meters, with minimum and maximum of 54 and 
169 meters respectively. The flying height was 2650 meters 
above mean sea level. The original photographs were scanned 
at 21 microns, or 0.364 meters in ground unit. The area 
selected includes many features namely a quarry, agricultural 
and semi urban areas, being ideal for testing. 
The collection of the DEM has been done with 10 meters grid 
spacing in a Z/I SSK stereoplotter. The operator had the 
ability to move up to 7 meters from the predefined grid 
position, meaning that the collected points form a TIN rather 
than a DEM. 
  
Manually collected TIN and orthophotograph of 
the test area. 
Figure 2. 
In order to evaluate the precision of the manually collected 
DEM, 14 repetitive measurements in 16 points of different 
background, approached from accidental addresses have been 
made. The average standard deviation of these measurements 
is a measure of precision and was 0.147 meters. Theoretically 
the expected height accuracy in a single point is equal to the 
pixel size in ground units for a 152 mm lens and a base of 90 
  
Intern 
mm (| 
than h 
the the 
4. T 
N 
4.1 A 
The f 
artific 
calcul: 
from 
orthor 
from | 
collec 
coveri 
size of 
The 1 
orthop 
match 
and el 
(Skarl 
the a 
predef 
grid (I 
additic 
predef 
often 1 
tones : 
The m 
(equal: 
templa 
templa 
adequ: 
correci 
can bi 
plane) 
detecte 
some | 
the ori 
pixels 
undula 
points 
necess; 
along | 
interpa 
Figure 
In orde 
Was ne 
algoritl