1e possible 
the image 
, extended 
distortion 
is Can no 
ates rather 
cal shift of 
n for each 
@) 
(3) 
d image 
osition of 
and the 
ıtify the 
on results 
tions are 
sensor 
the lens, 
ervations 
(4) 
grid in 
croup of 
ons are 
mated a 
flatness 
mber of 
images. 
rities of 
without 
camera 
nt data 
ication 
have a 
).. The 
ls to be 
ed in a 
specific manner (Luhmann & Wendt 20002). The quality of the 
photogrammetric system is reflected by the error of length 
measurement with respect to the expected accuracy. 
For evaluation and comparing purposes a 3-D testfield (Fig. 3) 
with a dimension of app. 2m x 2m x 2m (exterior testfield) has 
been designed for camera calibration. This test range covers 
app. 200 object points. In the centre a second 3-D testfield of 
app. 0.8m x 0.8m x 0.8m is placed forming the reference 
testfield. The reference field consists of 16 reference points 
(Fig. 3) whereof 120 different distances are computable. The 
reference points are measured by a coordinate measuring 
machine (CMM) by two independent measuring passes. An 
absolute deviation of the reference coordinates between the two 
passes of one CMM results with maximal 15um. The 
eccentricities of the retro-reflective targets to the CMM 
measured target axis have been verified. The values do not 
effect the photogrammetric adjustment. The systems scale is 
based on one scale bar in y-direction. The scale bar has a 
verified standard deviation a priori of Sum. 
  
   
  
  
reference testfield 
  
  
  
testfield based on guideline VDI/VDE - 
2634 with interior reference testfield 
principle of 
reference points 
  
  
  
  
  
  
o [um] | RMSxyz | Relative 
CLC | camera | lens Eq. (5) | precision 
Fuji 28 
1 S1 mm 0.406 0.0543 1:49000 | # 
Pro (1) 
Fuji 28 
2 S1 mm 0.346 0.0389 1:69000 | # 
Pro (1) 
28 
3 DCS | mm 0.352 0.0491 1:55000 | * 
460 (1) 
DCS 35 
4 Pro mm 0.333 0.0337 1:80000 | - 
Back 
645M 
  
  
  
  
  
  
(Hubbs-targets) 
Figure 3: Testfields and reference points 
Four data sets of different configurations, varying camera 
handling and camera-lens-combinations (CLC) (Table 1) are 
taken as hand-held shots. With an Fuji S1 Pro and a 28mm lens 
two sets of images (CLC1,2) are taken whereby as few as 
possible imaging directions (Table 2) are mixed among one 
other. Therefore the different camera stations are visited several 
times for taking images. These sets of images are taken by 
careful camera handling. In contrast a data set with a Kodak 
DCS 460 (not stabilised CCD-sensor) and the same 28mm lens 
(CLC3) was taken. All necessary images in all imaging 
directions are taken at once by careless handling. For comparing 
aspects the calibration results of the DCS Pro Back 645M are 
added to the analysis (Table 1). 
  
  
  
Table 1: Data set overview and adjustment results 
# rolling in four directions, care handling, 5mm targets 
* rolling in four directions, careless handling, 5mm targets 
all investigations on this are legible at Jantos (2002) 
  
  
0? -90° +90° 180° 
Ed 
  
  
  
  
  
Table 2: Principle of imaging direction 
Exposures are taken in four imaging directions with a 
proportion of 40:12:12:16 for the directions plotted in Table 2. 
The camera stations are chosen around the 3-D testfield with 0.5 
to 2m object distance in 3 main height levels (bottom — mid 
level — above the testfield). 
All image points are measured by ellipse fitting. The accuracy 
of the ellipse fitting method yield to 0.39um for CLCI, 0.33pm 
for CLC2 and 0.44pm for CLC3 a priori. The average of ellipse 
sizes for all data sets results in less than 6 pixel. 
3.1 Interior accuracy 
The image measurement precision a posteriori lies between 0.3 
and 0.4um. The object point accuracy, in this case RMS values, 
for X, Y and Z direction for all object points within the 
photogrammetric projects (CLCI1-3) yields a mean value of 
30um. An average of 20pm could be obtained with the DCS Pro 
Back 645M (CLC4). This leads to an object precision RMS xy7, 
(5) of 54um up to 34um, the corresponding relative precision 
results in 1:49000 — 1:80000 (Table 1). 
  
RMS, = V[RMS(X)F + [RMS(Y)F + [RMS(Z)P (5) 
The required relative precision for standard industrial 
applications is obtained with these data sets, though they are 
borderline cases because of image configuration with almost 
exclusively sensor fulfilling exposures and a mean value for 
target size of less than 6 pixel. The reference points and scale 
points (Hubbs-targets) have a size of 6.35mm. The RMS values 
for the reference points are much better as they are in the centre 
of the system. A problem constitutes the adjustment of the 
reference points in two horizontal levels. Hence, this yields to 
almost sloped and small ellipses. However, there was no better 
alternative in view of getting the reference coordinates 
measured with an independent measuring system with an 
appropriate accuracy. 
-29—