54 
large space structure components using self-calibration (Shortis, 1989). In parallel with the film-based photogrammetry, 
research at LaRC has also focussed on utilising CCD based systems to provide access to hostile environments and to 
capture dynamic events. Initial experiments utilised analog video cameras (Burner et al, 1985), but these were quickly 
replaced with solid state cameras to avoid vibration, environmental and interference problems. often encountered with 
applications in wind tunnels. Burner et al (1987) describes the evolution of a CCD camera system currently installed in 
the National Transonic Facility. Childers et al (1994) describes a recent application of CCD cameras to characterise the 
motion of a free-flight model in a large wind tunnel. 
The-requirement for pre- and/or post-calibration of CCD cameras at LaRC is a direct result of applications such as wind 
tunnel testing. The placement of cameras and target arrays is severely limited within the test zones of wind tunnels and 
self-calibration is rarely possible due to the constraints on network geometry. To capture the images in real time during 
model tests there is mandatory requirement for multiple, fixed cameras, so simple resection/intersection techniques are 
in use at LaRC as appropriate to the task at hand. However, there is continuous demand on wind tunnel instrumentation 
to improve the accuracy of measurement in accord with more stringent engineering tolerances. The modelling and 
elimination of systematic errors in the complete image capture and measurement process, including the deterioration 
caused by, for example, recording media (Shortis et al, 1993), is an integral part of all CCD camera processing at LaRC. 
THE CALIBRATION APPROACH 
The use of target fields to calibrate CCD cameras has been successfully implemented by a number of researchers 
(Beyer, 1987; Bósemann et al, 1990; Gustafson, 1988). In order to confidently derive the calibration parameters, the 
geometry of the network should be essentially the same as that required for self-calibration. Multiple, convergent 
photographs of the target field are taken, preferably with a range of camera to object distances and a variety of roll 
angles to reduce correlations between parameters. The calibration parameter set typically comprises primary physical 
parameters to model the imaging system characteristics and high-order additional parameters to model image non- 
linearities and image plane unflatness (Faig and Shih, 1988). The target images are observed manually or measured by 
digital image processing semi-automatically and the network analysis is based on the principle of collinearity and an 
iterative solution by least squares estimation (Shortis, 1989; Shortis et al, 1991). 
However, self-calibration of CCD cameras using the targeted test field technique does have potential problems. The 
small size of the sensor area and the relatively long principal distances give rise to strong correlations between internal 
and external orientation parameters. Typical of these is the projective coupling between the principal point location, 
decentring distortion and (for example) the tip and tilt of the camera, as small changes to any of these parameters 
results in similar changes within the image space. Parameter constraint is the usual remedy, taken to the extreme of 
. parameter suppression in some cases. Because it is rare for two parameters to be 100% correlated there is always 
some unavoidable degradation of the accuracy from the network adjustment, caused by unmodelled systematic errors. 
To reduce the level of correlation internally between the primary physical parameters, and between these parameters 
and external orientation parameters, an initial plumb line calibration can be carried out. As previously mentioned, a 
plumb line calibration obtains the lens distortion parameters effectively independently of all other calibration 
parameters. The results of the plumb line calibration can be used to constrain the subsequent self-calibration. 
Parameter constraints are conveniently incorporated into the least squares estimation process via sequential 
adjustment, also known as a priori weights (Case, 1961). 
There are two restrictions which must be placed on the rigorous implementation of a combined calibration approach. 
First, the plumb line calibration is not totally independent of the results of the self-calibration. The location of the 
principal point, as well as any non-linearities or unflatness of the sensor, must be considered. As the lens distortions 
are computed relative to the principal point, any substantive error in the assumed location will affect the results and 
invalidate the constraints applied to the self-calibration. Non-linearities or unflatness of the sensor will similarly affect 
the plumb line calibration, albeit at a much lower level, and must be accounted for in the analysis. The strategy which 
must be adopted, in the absence of an integrated solution, is an iterative computation. The plumb line data is first 
analysed using an assumed principal point location. The computed lens distortion data is then used as constraints in 
the self-calibration network computation. The principal point coordinates are updated and any significant high order 
calibration parameters are incorporated into a re-computation of the plumb line calibration. The results of this second 
computation of the: plumb line calibration are then used as constraints in a second self-calibration network computation. 
The iteration process, in most cases, will be short-lived. Exceptions will be experienced only when the initial knowledge 
of the camera and sensor is grossly erroneous. 
Second, the plumb line and self-calibration images must be recorded at the same focus setting and within a reasonable 
time delay in order to characterise the lens whilst it is a consistent state. It is also worthy of note that the plumb line 
images would normally have a single plane of focus, whilst self-calibration images of a test range are likely to have a 
significant depth of field due to the three dimensional nature of the test range or the convergent photography, or both. 
This raises the issue of change in lens distortion for test field targets imaged out of the plane of focus. However, due to 
the small size of the CCD sensors and relatively long focal length lenses used in this study, the magnitudes of the lens 
distortions are generally small, which should lead to commensurately small changes in lens distortion out of the plane of 
focus (Fraser and Shortis, 1992). 
CAMERA TYPES 
Fourteen different types of CCD sensor and lens combinations have been calibrated at LaRC using the technique. The 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995 
  
IA