tography 
able for 
ibes first 
t camera 
on-board 
el sensor 
Che pixel 
conds an 
lor LCD 
vable IR 
ngle shot 
sulting in 
-22 MB. 
  
3. GEOMETRIC QUALITY 
3.1 Testfield camera calibration 
Usually a testfield calibration is applied as the standard method 
for a photogrammetric accuracy investigation. The IAPG 
laboratory is equipped with a number of different testfields 
consisting of retro-reflective and diffuse-reflective targets. 
Plane 2-D testfields are available as well as 3-D testfields with 
spatially distributed points. In order to get a first impression of 
the photogrammetric accuracy potential of the camera a 3-D 
testfield has been imaged by 76 photos according to well- 
established imaging configurations. The camera was equipped 
with a 35mm wide-angle lens (a 24mm lens is also available 
but not yet tested). The testfield was illuminated by a ring flash 
that could not perfectly be mounted with respect to the viewing 
direction of the camera due to mechanical limitations. 
The test set-up (FIG) consists of two independent 3-D 
testfields. While the outer frame (2m x 2m x 2m) serves 
primarily as a mounting device for reference scale bars and tie 
points, the inner frame (0.8m x 0.8m x 0.8m) consists of 
calibrated 3-D target points measured on a high-precision 
CMM. The overall dimension was designed according to the 
German guideline for acceptance and reverification tests of 
optical 3-D imaging systems VDI 2634 that has been developed 
over the past years (Luhmann & Wendt, 2000). 
Image point measurement has been carried out by two different 
programs. Firstly, all points have been measured automatically 
by the AICON system DPA-Win. Although the measured 
points provide sub-pixel accuracy down to 1/20” of a pixel, all 
points are remeasured using the AXIOS Ax.Ima package that 
allows for the measurement of problematic points based on a 
sophisticated ellipse operator. 
Initial bundle adjustment with self-calibration has been 
performed by DPA-Win. Due to significant changes in 
geometry of the camera over the series of images the program 
(as well as other standard bundle adjustments) could not 
process the data successfully. It has been investigated that the 
sensor back was subject to a severe geometric shift in x'- 
direction that could not be modelled by a standard set of 
parameters of interior orientation. Although this is a rather bad 
result it should be pointed out that the camera has been used as 
delivered with no additional fixtures of the camera back, hence 
the x'-shift is probably caused by uncareful camera handling. It 
is recommended to add suitable mechanical aids to fix the 
sensor back with respect to the camera body if no extended 
camera model is available in the bundle adjustment. 
3.2 Image-variant camera modelling 
An extended model for camera calibration has been developed 
at IAPG (Tecklenburg et al 2001, Hastedt et al. 2002). Briefly 
described, the model does not assume a bundle-invariant 
interior orientation, but uses additional parameters for image- 
variant shifts of the perspective centre with respect to the 
image. In addition, a finite element correction grid is estimated 
within the bundle adjustment in order to compensate for 
remaining imaging errors and sensor unflatness as well. The 
effect of lens distortion is kept fixed for the whole set of images 
since it is assumed that distortion is mainly caused by optical 
refraction inside the lens, hence should be independent of the 
sensor related parameters. Due to the grid correction the 
distortion parameters B1,B2 (decentring) and C1,C2 (affinity, 
sheering) must not be determined. 
  
principal distance 
  
  
i 
  
e 
e 
e 
e 
d 
  
  
  
il 
  
  
  
  
  
  
  
" 
  
  
= 
deviation in mm 
o 
= 
62 j 
66 
70 
74 5“ 
  
EEE 
r1 
-0,005 
-0,01 
  
  
  
  
  
  
  
  
-0,015 
-0,02 : 
images 
  
a) variation of principal distance 
  
principal point x 
0,05 
0,04 
0,03 
0,02 
0,01 
deviation in mm 
e 
-0,01 
-0,02 
-0,03 
-0,04 
-0,05 
images 
  
  
b) variation of principal point in x' 
  
principal point y 
0,02 
0,015 
0,01 
0,005 
deviation in mm 
e 
-0,005 
-0,01 
-0,015 
-0,02 
  
images 
  
  
c) variation of principal point in y' 
Fig. 3: Image-variant shift of perspective center 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
number of images 76 
image observations 2984 
object points 209 
additional observations 1 
finite element grid spacing 2mm 
grid stability a priori 2um 
sigma 0 0.33pm 
principal distance c * 35.4377mm 
principal point x'o * -0.0256mm 
principal point y'o * 0.2942mm 
distortion parameter Al -9,80E-05 
distortion parameter A2 7,01E-08 
distortion parameter A3 -1,18E-11 
RMS X 0.008mm 
RMS Y 0.008mm 
RMS Z 0.011mm 
  
  
  
  
  
  
  
Table 1: Result of bundle adjustment with image-variant 
interior orientation and finite element sensor grid 
* variant values, here given for image 1 
43