illumination of the target at exposure. The regular 
difference in position in case of a double detection 
is about 15 um, which is an indication for the 
quality of these positions as start-estimates for the 
subsequent position estimation procedure. 
Although it has not been tested, this astronomical 
target detection procedure is expected to give even 
better results if it is adapted for close-range 
photogrammetry. Especially the background 
estimation needs further investigation. 
So far we only discussed the detection of targets. 
For the automatic position estimation of reseau 
crosses approximate locations are needed as well. 
With a sign reversal (densities lower than a chosen 
threshold below the "background" density) the 
procedure applied for target detection can be used 
for the reseaucross detection also. The main 
characteristics of reseaucrosses to be used for their 
recognition are: 
- the known size (number of pixels) of a 
reseaucross image; 
- the density of the reseaucross pixels should be 
among the lowest (and most likely a known fixed 
value) present in the image; 
- the location of the reseaucrosses in a regular grid. 
Although the procedure for searching reseaucrosses 
has not been implemented for this experiment, the 
detection of the reseaucrosses in the image is not 
expected to present any difficulty, apart for those in 
heavily underexposed areas. 
In our experiment a wide range of photoscale and 
illumination of targets was used. Under regular 
circumstances and with a for close-range imagery 
optimized search methode it will be possible to 
detect almost all complete target images. Images 
having defects f.i. due to partial obstruction will not 
only be difficult to detect, but usually it is very 
difficult to generalize algorithms to model their 
position estimation with sufficient succes. 
5. Measurement 
The search method described in the previous 
paragraph can be seen as the first step in the 
position estimation process. The final improvement 
of the position estimate is made with the "least 
squares template matching" method. This matching 
N 
method requires startpositions with a precision of 
about 2 pixels (20 um), in order to assure 
convergence of the iterative procedure. In those 
cases where the search method fails to produce a 
startposition with this precision, an intermediate 
step is necessary. In all three steps the procedures 
are different for reseaucrosses and targets. 
5.1 Reseaucrosses 
The startposition for a reseaucross is improved in a 
straight forward manner by detecting the four legs 
of the cross using a locally determined threshold 
density level. Then the positions of the pixels 
determined to belong to the cross legs are averaged 
using the measured densities for weighting. This 
yields a centre of gravity in density. The average of 
two opposite legs gives the improved position in 
one of the two directions.The procedure is code 
with a (weak) constraint on the alignment of the 
cross legs and the scanning directions. The 
improved position has a precision better than 1 
pixel (10 pm). 
  
figure 3: template and reseau cross image (negative 
converted to positive)