The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1060 
one needs to check whether the estimated internal 
characteristics of these cameras from a calibration session 
remain stable over time. The question that arises from such a 
need is what are the tools and standards that can be used to 
evaluate the stability of a given camera? 
This paper will thus focus on these issues in relation to amateur 
digital cameras. First, an automated method for an indoor 
camera calibration procedure is introduced. The main objective 
of such a procedure is to provide mapping companies using 
these cameras with a simple calibration procedure that requires 
an easy-to-establish test field. More specifically, a test field that 
is comprised from a set of linear features and few point targets 
will be used to carry out the camera calibration procedure. The 
determination of the interior orientation parameters will be 
based on the observed deviations from straightness in the image 
space linear features as well as the measured distances between 
the point targets. In addition to having a simplified calibration 
test field, we will outline an automated procedure for the 
extraction of the linear features and point targets from the 
captured imagery. The simplified test field and the automated 
extraction of the linear features and point targets are the key 
factors for enabling the data providers to effectively carry out 
the calibration procedure, which is general enough to handle 
any amateur digital camera. The paper will then discuss the 
concept of how to evaluate camera stability, which will be 
followed by the introduction of a set of tools for its evaluation. 
The stability analysis tools will be based on quantitative 
evaluation of the degree of similarity between the reconstructed 
bundles from temporal calibration sessions. The bundle 
similarity is used since the camera calibration procedure aims at 
generating a bundle that is as similar as possible to the incident 
bundle onto the camera at the moment of exposure. 
Following the calibration and stability analysis discussions, the 
paper will deal with several related questions as follows: 1) 
How to develop meaningful standards for evaluating the 
outcome from the calibration procedure, 2) How to develop 
meaningful standards for evaluating the stability of the involved 
camera, 3) Is there a flexibility in choosing the stability analysis 
tool, which is commensurate with the geo-referencing 
procedure to be implemented for this camera, and 4) Can the 
stability analysis be used for evaluating the equivalency of 
different distortion models. These questions will be discussed in 
turn, and some experiment results from datasets captured by 
two amateur small format digital cameras are presented. 
2. CAMERA CALIBRATION 
Deriving accurate 3D measurements from imagery is contingent 
on precise knowledge of the internal camera characteristics. 
These characteristics, which are usually known as the interior 
orientation parameters (IOP), are derived through the process of 
camera calibration, in which the coordinates of the principal 
point, camera constant and distortion parameters are determined. 
The calibration process is well defined for traditional analogue 
cameras, but the case of digital cameras is much more complex 
due to the wide spectrum of designs for digital cameras. It has 
thus become more practical for camera manufacturers and/or 
users to perform their own calibrations when dealing with 
digital cameras. In essence, the burden of the camera calibration 
has been shifted into the hands of the data providers. There has 
thus become an obvious need for the development of standards 
and procedures for simple and effective digital camera 
calibration. 
Control information is required such that the IOP may be 
estimated through a bundle adjustment procedure. This control 
information is often in the form of specifically marked ground 
targets, whose positions have been precisely determined 
through surveying techniques. Establishing and maintaining this 
form of test field can be quite costly, which might limit the 
potential users of these cameras. The need for more low cost 
and efficient calibration techniques was addressed by Habib and 
Morgan (2003), where the use of linear features in camera 
calibration was proposed as a promising alternative. Their 
approach incorporated the knowledge that in the absence of 
distortion, object space lines are imaged as straight lines in the 
image space. Since then, other studies have been done by the 
Digital Photogrammetry Research Group (DPRG) at the 
University of Calgary, in collaboration with the British 
Columbia Base Mapping and Geomatic Services (BMGS), to 
confirm that the use and inclusion of line features in calibration 
can yield comparable results to the traditional point features. In 
order to include straight lines in the bundle adjustment 
procedure, two main issues must be addressed. The first is to 
determine the most convenient model for representing straight 
lines in the object and image space, and secondly, to determine 
how the perspective relationship between corresponding image 
and object space lines is to be established. In this research, two 
points were used to represent the object space straight-line. 
These end points are measured in one or two images in which 
the line appears, and the relationship between theses points and 
the corresponding object space points is modelled by the 
collinearity equations. In addition to the use of the line 
endpoints, intermediate points are measured along the image 
lines, which enable continuous modelling of distortion along the 
linear feature. The incorporation of the intermediate points into 
the adjustment procedure is done via a mathematical constraint 
(Habib, 2006a). It should be noted, however, that in order to 
determine the principal distance and the perspective centre 
coordinates of the utilized camera, distances between some 
point targets must be measured and used as additional 
constraints in the bundle adjustment procedure. 
To simplify the often lengthy procedure of manual image 
coordinate measurement, an automated procedure is introduced 
for the extraction of point targets and line features. The steps 
involved in the procedure are described in detail in Habib 
(2006a) and are briefly outlined in the following sub-section. 
2.1 Automated Extraction of Point and Line Features 
The acquired colour imagery is reduced to intensity images, and 
these intensity images are then binarized. A template of the 
target is constructed, and the defined template is used to 
compute a correlation image to indicate the most probable 
locations of the targets. The correlation image maps the 
correlation values (0 to ±1) to gray values (0 to 255). Peaks in 
the correlation image are automatically identified and are 
interpreted to be the locations of signalized targets. Once the 
automated extraction of point features is completed, the focus is 
shifted to the extraction of linear features. The acquired 
imagery is resampled to reduce its size, and then an edge 
detection operator is applied. Straight lines are identified using 
the Hough transform (Hough, 1962), and the line end points are 
extracted. These endpoints are then used to define a search 
space for the intermediate points along the lines. Once the point 
and linear features have been extracted through this automated 
procedure, they are incorporated into the bundle adjustment, 
according to the method outlined in Section 2, to determine the 
camera IOP.