The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1178 
orientation. The methodology is essentially based on extracting 
a relative DEM from a stereopair, that only went through a 
relative orientation, and then bring it to a correctly geo- 
referenced position by matching the SRTM-DEM. This 
procedure is known to have a planimetric accuracy as good as 
l/20 th of the SRTM pixel size, i. e. approximately 5 m. This 
methodology was successfully applied with SPOT and ASTER 
stereo images, completely avoiding the use of GCPs (Gonfalves, 
2006, Gonsalves and Mar?al, 2007). In the case of ALOS, 
image orientation requires the use of GCPs since exterior 
orientation provided by navigation equipment on board of the 
satellite have a geo-location uncertainty of some 50 meters 
(Fraser, 2007). The main interest in DEM extraction from 
PRISM images is for the remote and poorly mapped areas, in 
undeveloped countries, where ground control surveys are 
difficult to carry. The methodology proposed in this paper may 
overcome this limitation, avoiding any field collection of GCPs. 
The second synergistic use of SRTM is in the DEM extraction. 
The previous knowledge of the terrain form provided by SRTM, 
in terms of elevation and slope orientation, facilitates the 
matching procedures, allowing for a faster and more effective 
automatic DEM extraction. 
The test area is a mountainous area in North Portugal, near the 
city of Porto, with heights ranging from 20 to 1400 meters 
above sea level. 
where (xo,yo) are the predicted image coordinates, (A\A 6 ) are the 
given parameters and (E,N) are the UTM coordinates in Km (more 
precisely the E coordinate does not include the false easting of 500 Km). 
This relation is not exact due to the relief displacement and due 
to the inaccuracy of the initial exterior orientation. Figure 1 
represents the displacement suffered by an object of height h. 
This effect can be estimated from equation (2), using the local 
incidence angle. 
Figure 1. Relief displacement of an object with height h. 
The corrected image coordinate, x 0 , 
x o = x o + to = x o + 
h ■ tan /? 
2.5 
(2) 
2. APPROXIMATE SENSOR MODEL 
The PRISM images available for this project were of mode 
1B2-R, in which images are resampled to a UTM projection. 
Using approximate sensor position and attitude, images are 
projected onto the ellipsoid level (Saunier et al., 2007), in the 
same manner as with other high-resolution imagery, and then 
projected to UTM. However, in image mode 1B2-R, the image 
still needs to be rotated to be aligned with UTM axes, so image 
lines approximately coincide with image lines in the original 
sensor geometry. An approximate sensor model was developed 
to deal with these images, not making use of co-linearity 
equations. 
Several aspects of PRISM images, such as the very narrow field 
of view (2.9°), the fact that a very good initial orientation is 
known and that terrain height is much smaller than the satellite 
altitude allow for important approximations by linear 
relationships. 
2.1 Object to image projection for the Nadir image 
Sensor model equations establish a projection from object to 
image space. That will be done stepwise for the 1B2-R images, 
considering an estimation of relief displacement and additional 
positional corrections. The sensor model is established for the 
Nadir image. 
Within the image ancillary data, an affine formula is given to 
convert from image coordinates to UTM and vice-versa 
(equation 1). 
A A 2 
A 3 A 4 
(1) 
where h is the terrain elevation, P is the local incidence angle 
and 2.5 is the pixel size in meters. The tangent of P can be 
obtained dividing the distance from the point to the ground 
track by the satellite altitude. Figure 2 represents the image and 
the ground track in UTM map space. Satellite position and 
approximate image location given in the ancillary data are 
enoughly accurate to calculate the value of P, which is not 
larger than 3° (in that extreme case the relief displacement 
would be of 1 pixel for a height of 50 meters). 
Figure 2. Nadir image and ground track drawn from the 
ancillary data. 
The image coordinates corrected for the height effect will not 
coincide with the actual position of the point due to the 
uncertainty in the exterior image orientation, which was used in 
the processing of 1B2-R images. This effect is essentially a 
systematic shift, which can be corrected in image space, in the 
same way as is usually done for Ikonos images (Grodecki and 
Dial, 2003) and other sensors, by an affine transformation, 
according to equation (3):