cameras allows the storage of a large number of images 
without the need to download them (up to 48 in the Apple 
QuickTake). Although this is a necessary and useful 
function, the compression techniques modify the image. 
The modifications are not noticeable to the human eye, 
but can affect the results from photogrammetric 
measurements. Digital still cameras usually provide the 
option of low resolution images (high compression) and 
high resolution images (low compression). All images 
taken for calibration purposes have been acquired as 
high resolution images. 
4. PROCEDURE 
The calibration procedure developed would best be 
described as a self-calibrating test range calibration, as 
opposed to laboratory or on-the-job calibration (as 
described by Osborn, 1994). A target plate was 
manufactured (see Figure 1), being very similar to target 
plates used recently in other calibration procedures (for 
example: Clarke et al, 1995; van der Vlugt, 1995). 
However, instead of using relatively small targets (2mm 
dia) which would produce images of targets of 5-7 pixels 
diameter, the targets were larger (5mm dia) to produce 
images of up to 30 pixels diameter, depending on the 
object distance used (ranging from 0.8m to 1.5m). This 
approach follows that of Shortis et al., (1995b), as it 
provides more reliable results for the centroiding of the 
targets. 
NI 
E d 
w 
® 
æ 
  
Figure 1. Target Plate. 
The targets were punched from a sheet of retro-reflective 
material following research by Zumbrunn (1995), who 
showed that punching out the material was the preferred 
method of target manufacture. If a target is made by 
masking the retro-reflective material, the image of the 
target would be systematically shifted when viewed at a 
large angle (over 30°), thus making targets produced by 
that method unsuitable for this procedure. 
The target plate was spray painted matt black to provide 
contrast to the retro-reflective targets. The effectiveness 
of this can be seen in Table 2 in Section 5 of this paper, 
which shows the background in the images (average 
minimum grey value) to be very low and the image of the 
targets (average maximum grey value) to be extremely 
high. The grey values (GV) can range from O to 255. 
The target plate was coordinated using 1” theodolites to 
an accuracy determined to be 0.3mm. This was 
140 
confirmed during the bundle adjustment and also by the 
photogrammetric tests of a fellow researcher (Mustaffar, 
1995). 
An extension to this target plate was constructed, to 
provide flexibility to accommodate the different focal 
lengths of the cameras. The original target plate as 
shown in Figure 1, fitted in the centre of the extension 
plate, which measured 700mm by 500mm. The 
calibration of the Philips cameras only required the 
original target plate to be imaged, whereas the Logitech 
Fotoman and the Apple QuickTake required the larger 
target plate to be used, due to their shorter focal lengths. 
Images of the target plate were acquired using each of 
the four cameras to be calibrated. The images were 
acquired from nine different camera stations positioned 
around the target plate at 45? intervals and also from 
vertically above. The approach taken follows the self- 
calibration procedure described by Shortis et al., (1995a). 
For each of the two analog video cameras, the camera 
was set up on a tripod, warmed up for a sufficient period 
of time and then a set of ten images was taken at each of 
the nine camera stations. These images were averaged 
to provide a final set of images to be measured for use in 
the bundle adjustment. As a check, one set of the 
original images of the Right Philips was also measured to 
determine the effect of the averaging and is referred to in 
the results as Single Frame. 
PC-based software programs were written to average the 
images, to make radiometric corrections to the images 
and to measure the image coordinates of the targets. A 
literature survey showed that there is usually no 
radiometric correction done to images after they have 
been acquired for errors such as pixels in the CCD array 
which show black when they should show white or the 
reverse. 
As an adjunct. to the calibration procedure, an 
investigation was carried out to determine whether 
corrections for such ‘bad’ pixels should be incorporated 
into the procedure. Such pixels could include isolated 
ones, blocks of pixels or the array as a whole not 
responding uniformly. This has been referred to as Non- 
Uniform Pixel Response by Beyer (1992). The result of 
the investigation is given in Section 5 of this paper. 
A PC-based computer program (see Fryer, 1994 for a 
description) was used for the measurement of targets. 
The program was menu driven and displayed the digital 
image to be measured. To measure a target, the 
operator defined a rectangular area around the target, 
ensuring no other target encroached on that area. This 
area was processed to find the centroid of the target. 
The centroiding involved accessing each grey value (GV) 
in the given window, finding the average GV and the 
maximum GV and establishing a threshold level. All 
pixels with a GV below the threshold level were ignored. 
The average threshold level for each target is shown for 
each camera in Table 2 in Section 5, along with the 
average number of pixels falling above the threshold level 
for each target. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996 
  
Thre 
usec 
sque 
one 
pixel 
valut 
grey 
valu 
effec 
mett 
each 
resu 
Sect 
grap 
met! 
The 
files 
bein 
usec 
ever 
was 
Prioi 
adju: 
P2 
Thes 
adju: 
the 
men 
The 
tabu 
calib 
the | 
and 
for ‘I 
cam 
The 
Tabl 
imac 
offse 
para 
free( 
it ca 
cam 
RMS 
how: 
the i 
foca 
cam 
relat 
The 
5-1C 
plun 
The 
Sect 
thes 
anal 
sign