3. DESCRIPTION OF THE TEST 
3.1 Gray Level Data 
3.1.1 Synthetic Data 
x2+y2 
A discrete 2D-Gaussian exponential function of the form e^ CT (CT=constant) was used for 
simulating gray level patches. The correct match point was taken at the peak of the surface. The 
gray levels, after scaling and addition of random noise, had a maximum range of 1 - 256. The 
peak gray level, the standard deviation of the noise and the Slope of the surface can be 
interactively varied. 
The geometric configuration simulated included two images with variable overlap, wide angle 
camera, 0 = ¢ = x = 0 and same flying height. The image scale and the image position of the 
points are variable. 
The synthetic data was used to test the correctness and functionality of the program and provide 
some first results and experience for the behaviour of the algorithm. After this stage real data was 
used. 
3.1.2 Real Data 
This data was taken from digitized images of the testblock Echallens ( Koelbl 1984). A3x3 
block was chosen and for each photo, squares of dimensions 1.1? or 1.4? cm? at the nine 
standard positions were digitized. Each square included at least one signalized control point, 
good and bad natural points. The data was digitized with a Perkin Elmer PDS 2020 G 
microdensitometer at the "Laboratory for Signal Processing” of the Federal Institute of 
Technology, Lausanne. The quantization was 12 - bit, the pixel size 202 um? .The data was 
transformed to intensities and reduced to 8 - bit. 
The testruns presented in this paper refer to one model only. From the six overlapping squares 
only two were used, since the remaining ones have similar geometry for symmetry reasons. The 
image scale is approximately 1:5300, the average flying height 815 m, the base to height ratio 
0.48 and the overlap 68%. From the two Squares 18 points of different geometric and 
radiometric quality were chosen and measured manually at the AC1. Their image coordinates 
were input into a bundle program without additional parameters to determine the X,Y,Z ground 
coordinates and the exterior orientation elements. 
A major problem was the establishment of a relationship between digital image coordinates and 
AC1 image coordinates. First the measured points had to be identified in the digital images with 
high ( subpixel ) accuracy. This was done using large format contour plots of the digitized 
intensities from which the measured points’ row and column were read. Second, a rotational 
relationship between scanning direction and the image coordinate system had to be established. 
The microdensitometer has no device that permits to measure the position of the film on the 
carriage with respect to the scanning direction, so the films were placed by hand on the carriage 
trying as much as possible to align the image coordinate system with the scanning direction. The 
errors introduced in the two processes described above cause a parallax and height error and a 
deterioration of the iterations" convergence. From a comparison of the results with the heights 
from the bundle adjustment, we can conclude though, that no significant errors were introduced. 
3.2 Brief Description of the Program 
Different aspects and modifications of the algorithm were tested with FORTRAN 77 software, 
developed at first stage on a VAX 11/750 and transfered to the Digital Photogrammetric Station 
(DIPS) of the Institute of Geodesy and Photogrammetry, ETH Zurich (Gruen 1986). 
The main characteristics of the program are: 
* The gray level derivatives are computed as discrete first order partials using the two 
neighbouring pixels with the exception of the edge rows and columns, where the difference 
to the only neighbouring pixel is used 
- 288 -