iris was fixed and the automatic exposure function set as iris 
priority. 
Figure 2. The zoom CCD camera 
The frame grabber used for this project was the MRT PCMCIA I 
imaging card. This card can be used directly with a notebook 
computer or with a desktop computer via an adaptor. It digitised 
both fields of the composite video input and generated a digital 
image of 920(H)* 574(V) pixels. Only the luminance signal, 
which was quantised into 8 bits, was used for processing. 
It is intended that the focus and the iris are fixed for both 
calibration and application in order to reduce the possible 
combinations to a manageable level. The focus distance is then 
needed to be determined before calibration. This is obviously 
related to the depth of field, the distance range the system is to be 
used for and the object resolution to be achieved. For this 
calibration the object resolution was specified as that 1 pixel was 
equivalent to 3mm in object. The working distance range was 
from 2.5 m to 30 m, which means the zoom was able to be 
adjusted to achieve a 3 mm object resolution within this distance 
range. To ensure sufficient depth of field throughout for a fixed 
iris of f/22 and 0.2 pixel circle of confusion, the focus distance 
was calculated to be 25 m. The camera was then fixed at this 
focus distance and the f-stop fixed at 22 throughout the course of 
calibration. The detail of this derivation is beyond the scope of 
this paper. This scenario was designed only for the convenience 
of fitting the experiments in the laboratory space. More relevant 
scenarios to city survey will be used in future where longer 
distances will be assumed. 
The calibration was required to determine the interior orientation 
parameters, the lens distortion parameters and the exterior 
orientation parameters to the telescope system at various focal 
settings within the zoom range. It was also required to assess the 
accuracy of calibration and the repeatability of the calibrated 
values against zoom action. 
3 CALIBRATION METHOD - 
THE CAMERA-ON-THEODOLITE METHOD 
The camera-on-theodolite calibration method was used for this 
calibration. Two targets were set 16 m and 27m away from the 
camera theodolite station respectively and almost aligned with 
the station, as shown in Figure 3. 
Figure 3. The calibration set-up 
The telescope of the theodolite was rotated to 25 directions so 
that each target was imaged at 25 different positions evenly 
spread in the image frame. For each image/telescope position, an 
image was captured and the horizontal and vertical angle readings 
were noted. For each of both targets, the 25 target images were 
located by image processing. The corresponding 25 sets of three 
dimensional co-ordinates of the targets with respect to the 
rotating telescope system were calculated using the following 
equations: 
- cosv u sin(/i - h u ) N 
Y 
= D 
sin v sin v u + cos v cos v„ cos{h - h a ) 
cos v sin v o - sin v cos v H cos(/j - h n ) y 
In the equations, D is the distance between the target and the 
theodolite; h Q , v Q are the horizontal and vertical readings while, 
initially, the theodolite is sighting at the target; h and v are these 
readings an image is captured. With these corresponding 2D and 
3D co-ordinates for 50 points (for both targets), the analytical 
space resection with lens distortion models was performed to 
determine the interior and exterior parameters. The exterior 
parameters here were the camera to theodolite parameters. For the 
detail of the camera-on-theodolite calibration method, see the 
author's previous papers (Huang and Harley, 1989) 
The feature of this calibration method of needing no control array 
greatly reduce the cost of the calibration. It does not suffer from 
the problem of having too few targets at the zoom in setting as 
the test field methods do (Wiley and Wong, 1995). The camera to 
theodolite parameters are determined as the natural and direct 
outcome of the method. If the control field method had been used 
instead, those parameters would have had to be determined by 
sighting the theodolite at the targets in the control field. In short, 
this method is most suitable for this particular calibration though 
it is a general camera calibration method. 
4 TARGET IMAGE LOCATION AND ITS ACCURACY 
A ring shaped target is used as shown in Figure 4. The outer and 
inner circles of the ring were used to locate the target centre 
independently then the mean was taken. For each circle, the edge 
is detected first using automatic thresholding, refined to subpixel 
by interpolation secondly and fitted with an ellipse finally 
(Huang & Harley, 1990). The discrepancies between the outer 
and inner circle determinations are systematic and thought to be 
attributed largely to the biased lighting condition. They were 
hoped to cancel out each other to a great extent in their mean 
value, which was always used as the final target image centre. 
Figure 4. The original target image (a) and its edge image (b) 
The accuracy of the mean was assessed using a more objective 
method - the subpixel shift method. A series of images were 
captured. Between two captures, the target was shifted by an 
amount equivalent to 0.1 pixel at the image scale. The detected 
image positions and the positions-should-be were subjected to a 
regression analysis. The standard deviation estimated forms the 
accuracy measure for the image co-ordinates. This accuracy 
measure reflects the pixel phase effect.