The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008 
1180 
Step 1 - The first step is to identify tie points (TP) on the 3 
images. These points should be in areas with good contrast to 
facilitate the image matching to be done around the TP. They 
should also be in areas with moderate relief because DEM 
matching does not work in flat areas. 
Step 2 - From the nadir image coordinates it is possible to 
estimate terrain coordinates, by doing the image to object 
projection (intersection with the SRTM-DEM). Since the 
planimetrie adjustment required is relatively small (of the order 
of 1 SRTM pixel), and the relief displacement in the nadir 
image is small, terrain coordinates only have a systematic shift. 
Step 3 - Displacements in y image coordinates between F and 
B images (the parallax) are calculated. Using the heights 
estimated in step 2 the linear relationship can be determined by 
linear regression. 
Step 4 - A large number of conjugate points are extracted by 
stereo-matching around each tie-point (in a rectangle of a few 
square km). Only very good matches, with high correlation are 
maintained. Coordinates and heights can now be determined for 
these conjugate points. A small DEM around the tie-point can 
be obtained, which is not properly geo-referenced, since images 
only have a relative orientation. 
Step 5 - The fifth step is to determine the displacement 
required to match these small DEMs to the SRTM-DEM. Once 
this displacement is determined the tie point will act as GCP, 
since the local displacement is known. 
Step 6 - Displacements in map coordinates are converted to 
the equivalent in image coordinates. For the set of tie points it is 
possible to determine the required planimetrie adjustment 
(equation 3). 
3.2 Application of the method 
A total of 16 TPs, regularly distributed on the N image, were 
identified and measured on the 3 images. Using the projection 
from image to object (the SRTM-DEM), approximate 
coordinates and heights were obtained. The linear relation 
between parallax (p y ) and height (h) was obtained: 
h = 2.5130- p y -5.9 (4) 
with an RMS of residuals of 7.2 m. At this stage heights are 
determined with some error, however in a large number of TPs, 
positive and negative errors will tend to be balanced, yielding a 
reasonable quality calibration. 
A regular grid of 20 pixel spacing, centred on a TP, covering 
5x5 km 2 was defined on the N image. Conjugate points were 
obtained on images F and B by stereo-matching. Only those 
with higher correlations (p>0.8) were kept. Heights were 
determined by equation (4). Figure 4 shows in 3D the points 
extracted around one of the TPs. 
Figure 4. 3D view of points extracted in a rectangle of 25 km 2 . 
Figure 5 shows a profile of the SRTM-DEM (line) and the 
extracted points (dots), along that line. The slight displacement 
required to make the points fit the line can be detected. 
0 100Q 2000 3000 4000 5000 
Distance (m) 
Figure 5. DEM profile and extracted points along that line. 
The displacement, with components in easting and northing 
directions, is determined by matching the SRTM-DEM. 
Successive shifts (AE, А/V) are applied to the points in order to 
determine the one that maximises the correlation coefficient 
between point elevations and SRTM elevations. 
The same was done for areas around all the 16 tie points. 
Displacements in map coordinates were converted to the 
equivalent in image space. Finally the affine transformation 
(equation 3) was determined by least squares fit. The formula 
determined was: 
(AxW- 0.000075 -О.ООООЗОУхсЛ (-8.2^ (5) 
{AyJ ~ {- 0.000677 - 0.000079J[y 0 J + 24 -9J 
with RMS of residuals smaller than 1 pixel in both directions. 
The precision obtained in the image orientation process is 
nearly as good as it was when the sensor mode was applied with 
standard GCPs. 
3.3 Assessment with Independent Check Points 
This orientation procedure used only SRTM as ground control. 
The 56 GCP previously used (section 2) will now be taken as 
independent check points (ICP) in order to assess the quality of 
the alternative orientation method. 
These 56 ICPs were projected from map space to image space, 
using equation (5) and the corresponding errors in image space 
were obtained. Using relation (4) parallaxes (F-B) of 45 ICPs 
were converted to heights and compared to actual point heights. 
The RMS errors were the following (planimetric error 
converted to meters):