ion 
For this test a 
Jbmitted to 20 
). One goal of 
is to verify the 
N thousands of 
e) analysis. 
pse operators, 
eved by edge- 
an accuracy in 
from test field 
1:140.000 with 
stungsfähigkeit 
Ider erzeugt, in 
schaftliche und 
unkte mit ihren 
tmeBverfahren 
Aef5genauigkeit 
rt wurde. Der 
alyse. 
Schwerpunkt- 
| den mit einer 
iten von 2/100 
g erreicht. Sie 
elegt wird. 
> world of CMM 
oped and they 
rammetry there 
ptance test, or 
ge in terms of 
tigation of sub- 
ment methods. 
damental steps 
basically affect 
ormation lost in 
rwards. 
on test for the 
measurement 
athetic and real 
generated and 
6 
  
submitted to different institutions. One goal of this test is 
to compare the suitability of different point operators 
under equal conditions. A second goal is to verify the 
current limits of sub-pixel point positioning which, by 
numerous authors, have been reported down to a few 
thousands of a pixel. 
Up to now the comparison study reports of the results of 
the following point operators: structural edge-based 
ellipse operators, center-of-gravity operators, template 
matching operators and least-squares matching 
operators. After a first project stage it is planned to deliver 
the test material to a wider group of experts on an 
international basis. The test is strictly non-commercial 
and results shall be handled confidentially if required. 
The investigation of the image test material is based on a 
2-D (image space) and 3-D (object space) analysis. In 2- 
D analysis the measured image coordinates of point 
centers will be compared to their (assumed) nominal 
values given by a program for the generation of synthetic 
point patterns. 
The 3-D investigation is based on a set of real test field 
images taken with a fully-digital CCD-camera. The test 
field was prepared with a number of retro-reflective 
targets which are widely used in practice. The accuracy 
analysis includes a comparison of RMS values of image 
coordinates as well as the RMS values of adjusted object 
coordinates given by a bundle adjustment with self- 
calibration. 
The test does not cover an independent comparison with 
given object coordinates measured by a system of higher 
accuracy, e.g. a theodolite measuring device. This will be 
done at a later stage of the investigation. 
2 PARTICIPANTS 
Approx. 20 scientific and commercial institutions 
throughout Germany, Switzerland and Austria received 
the test material. Up to now (March 1996) a total of 10 
participants have returned their results. 
Synthetic Point Pattern 
  
Submitted Returned 
Scientific institutes 10 7 
S 7 3 
Industrial users 2 1 
Total 20 11 
All participants used their own implementation of different 
point operators. Some members of the test returned more 
than one result, for instance, if they wanted to compare 
different algorithms. Although all results have been 
handled confidentially and published without names and 
affiliations, the number of system developers in the test is 
rather poor. 
3 IMAGE MATERIAL 
3.1 Synthetic Images 
In order to test the performance of point operators under 
controlled conditions a number of images with artificial 
targets have been created. These targets can be 
overlayed in any type of background image. Each pattern 
can be varied in terms of sub-pixel center coordinates, 
size, orientation angle, contrast and noise. In addition it is 
possible to simulate occlusions or other point 
disturbances. 
The mathematical model of point generation is not 
claimed to be totally error-free. The program is able to 
calculate the greylevel function of circular targets based 
on the rigorous ellipse model and a bilinear greylevel 
interpolation. Figure 1 shows an example of a synthetic 
ellipse pattern compared with a patch from a real retro- 
reflective target (Broers & Hemken 1996). Slight 
differences occur at the edge of the point. 
Using this program a set of five test images (500 x 500 
Pixel, 8 bits per pixel ) has been created (Table 1, Figure 
2). Each image contains a total of 25 targets with different 
properties. 
Real Point Pattern 
  
Figure 1a: Synthetic and real point pattems 
325 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996