tanbul 2004 
location of 
utilized by 
in parallel 
the digital 
tically. It is 
hree sensor 
amera lens. 
ensor line b 
surface. By 
n along the 
e, a strip of 
surface in 
he fact that 
or line, an 
figure 1, is 
true ground 
ndependent 
the mapped 
As already 
in parallel 
linates with 
F1 
jmage by 
approach. 
represents 
/ flight F1. 
is imaged 
By use of 
can direct- 
0. 
3 to locate 
ct them to 
ved in two 
matching 
hotogram- 
he second 
'mentation 
ments into 
nts can be 
tion coor- 
International Archives of the Photogrammetry; Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
4. EXPERIMENTAL TESTS 
WITH SYNTHETIC IMAGES 
In order to verify the new approach, airborne pushbroom scan- 
ner image data were simulated. For this purpose an image se- 
quence was acquired, and the medial rows of each image were 
cut out and finally merged to one dataset. Due to the fact that 
the swath angle of one row of a central perspective image is so 
small, parallel projection can be assumed in the final dataset. A 
parallelepiped solid was taken as the test object, because in this 
case the expected geometrical impacts are easy to understand, 
and can also readily be visualized. To provide photogrammetric 
analysis, a reference object was imaged together with the test 
object at the same time. Figure 3 shows the two objects from a 
perspective view. The reference object defines the object space 
coordinate system. In addition, two Cartesian coordinate axes 
were drawn in for a better visualization. The longer one indi- 
cates the x-axis and the shorter one the y-axis. 
  
Perspective view of the parallelepiped solid together 
with the reference object, at the top. Below, the 
simulated pushbroom scanner data in x-direction 
(left) and in y-direction (right). The ground track of 
the two flight lines are marked by the arrows. 
Figure 3. 
First of all, the objects were "overflown" in x-direction and the 
dataset was produced, as described before. In the lower left pic- 
ture of figure 3, the parallel projection in x-direction is well 
visible, because there are no displacements left in flight direc- 
tion, whereas the upper surface of the parallelepiped solid is 
dislocated in y-direction, as expected. The inverted distortions 
appear by the acquisition in y-direction, the displacements are 
solely x-directed in this case (lower right). The corner coor- 
dinates of the upper surface of the parallelepiped solid were 
measured in the next step and transformed from the image to 
the object space coordinate system. As explained earlier, there 
should be truely located x-coordinates in the simulated airborne 
pushbroom scanner image available in x-direction, respectively 
truely located y-coordinates in the image in y-direction. The 
combination of both information will result proper ground coor- 
dinates in x and y direction. 
In order to establish geometric control of the results and to 
check this hypothesis, the measured coordinates were compared 
with coordinates calculated with traditional photogrammetric 
695 
methods. Therefore, a bundle block adjustment for the imaged 
reference object was accomplished with data from an array 
camera. Truely located three-dimensional coordinates were de- 
termined by a spatial intersection as a result of the bundle block 
adjustment. Thus, there exist no object displacements in these 
calculated coordinates. The result of the comparison is shown 
in figure 4. The round dots mark the points representing the 
measured coordinates derived by the new approach, and the 
squares show the points representing the coordinates as the 
result of the bundle block adjustment. The displacements in the 
two simulated scanner datasets are also shown in figure 4 as 
black triangles for the y-directed dataset and white triangles for 
the x-directed dataset. For example the difference between the 
filled triangles and the square points demonstrate the impact of 
the displacements in x-direction by flying towards y-direction. 
The four points around the reference object are control points 
for checking and for the calculation of the parameters like 
translation and rotation between the coordinate systems. The 
arrows in the right small picture in figure 4 illustrate the 
attitude of the camera relative to the observed objects. 
^ simulated image in x direction 
à simulated image in y-direction 
| = from bundle bock adjustment 
|o combination of the correct XY. 
  
  
  
CP 
| » ground track of the 
CP | two flight lines 
CP control point 
Figure 4. Result of the comparison between the coordinates 
obtained by means of the new approach and by 
bundle block adjustment. 
The deviations between the measured coordinates and the ones 
calculated by bundle block adjustment are smaller than one 
pixel, and therefore in the accuracy of the measurement as 
expected. 
5. GENERATING TRUE ORTHOIMAGES 
The experiment has shown the functionality of the new 
approach in principle. However, in order to meet the require- 
ments of orthoimages, all points in the entire dataset have to be 
displayed in their correct position and not only the exemplarily 
shown outstanding points in figure 4. In order to achieve this 
area-wide, two methods can be applied which were already 
mentioned in section 3. The first possibility is finding corres- 
ponding points in both images by means of image matching 
algorithms. Many useful general algorithms for image matching 
have been developed so far, but due to the complexities of the 
imaged real world any method has also its shortcomings and 
yields insufficient results in particular situations. Therefore 
three basic reasons for such problems should shortly be ex- 
plained. The first reason is, finding corresponding points is im- 
possible, if an object point is located in a hidden area in one