nates 
each 
ie re- 
sults 
lized 
algo- 
eria. 
> the 
rs of 
or all 
SIC 
mote 
bhani 
only 
rer 
mpa- 
e the 
y the 
n are 
 car- 
The 
1 the 
rinci- 
d the 
) are 
f the 
nclu- 
enta- 
three 
e de- 
ffers 
1 the 
naxi- 
time. 
ut in 
  
  
  
  
  
  
  
  
  
  
  
  
  
epoch 1 epoch 2 epoch 3 
value ag value g value g 
Puncpal |. 0zmm Sum 16.067 mm 5um 16.064 mm 5um 
distance 
Xo 21 um 6 um 19 um 6 um 19 um 6 um 
Vo 119 um 6 um 119 um 6 um 126 um 6um 
dXmax 59 um 2um 59 um 1um 59 um 2um 
dun 47 um 1um 47 um 1um 47 um 2um 
Jo 0.39 um 0.39 um 0.37 um 
  
  
  
Table 1: Results of calibration 
- O,,the estimated accuracy of the image coordinates, 
lies at about 0.4 um or 0.15 pixels. This confirms that 
the measurement of well signalized points can be 
performed with very high subpixel accuracy. 
However, o, is significantly higher than the root mean 
square error of the comparison between the two sets of 
image coordinates. Thus, a number of error sources has 
not yet been properly modelled. Some can be named: the 
accuracy of the sensor movement during image acquisi- 
tion, and the time stability of the testfield. Less likely are 
systematic errors in the observations (they should show 
up in an analysis of the residuals, and didn't), in the 
applied distortion model (the introduction of higher 
order polynomials did not improve the results), and in 
the control points (a free net adjustment resulted in a 
0, 0f 0.33 um). Further investigations have to be conduc- 
ted in order to identify and model the relevant error 
sources. 
3.2 Photogrammetric application example 
  
As an example for the attainable accuracy in object 
space two conventional photogrammetric evaluations 
were carried out. Images of the testfield at a scale of 
approximately 1:100 were acquired using 
- two images with parallel optical axes and a distance 
between the projection centres (baselength) of0.4 m 
and 
- two convergent images with a baselength of 3 m. 
Image coordinates of the signalized points were measu- 
red using the chain code matching algorithm, and each 
pair of images was processed twice, once including the 
calibrated interior orientation parameters, and once 
without them. Table 2 shows the results. Empirical 
standard deviations derived from a comparison with the 
known testfield are given for each coordinate in object 
space. The following can be concluded: 
- The attainable accuracy in object space lies at 
0.02 mm - 0.04 mm in each coordinate for the con- 
vergent case. This confirms that digital photogram- 
metry can compete very successfully with other tech- 
niques in high precision measurement tasks. 
- The accuracy for the parallel case is worse, especi- 
ally in the direction of the optical axis (Z coordina- 
te). This is due to the small field of view of the 
camera. À parallel setup, however, is only necessary, 
if stereo viewing of the images is of interest. À 
convergent setup is superior in terms of accuracy. 
- The comparison between the results with and with- 
out calibrated parameters of interior orientation 
clearly demonstrates the effectiveness of calibration. 
  
2 images, parallel | 2 images, conver- 
setup gent setup 
sx Sy Sz Sx Sy Sz 
[mm] | [mm] | [mm] | [mm] | [mm] | [mm] 
  
  
  
  
  
  
  
  
  
  
meiibised] opos 353] 32 19 | a5 
version 
calibrated | 0.05 | 0.11 | 0.52 | 0.02 | 0.02 | 0.04 
Version 
  
Table 2: Results of photogrammetric evaluation 
4. CONCLUSIONS AND OUTLOOK 
This investigation covered a geometric calibration of the 
ProgRes 3000 camera and an example for a photogram- 
metric evaluation. It could be shown that the parameters 
of interior orientation and lens distortion can be deter- 
mined precisely by the calibration procedure and they 
were found to be stable over time. Accurate object point