Game Development Reference
In-Depth Information
as important as the flexibility of freely changing the camera configurations. In
such cases, self-calibration is needed. A brief survey is, therefore, devoted to
self-calibration techniques in the fifth section for completeness. To show how
calibration results can be used in specific applications, two such practical and
representative applications have also been presented in the sixth section.
As shown in the fourth section, passive camera calibration has been studied
extensively during the past 40 years. Recently, due to the interest in image-based
modeling (IBM) and IBR, research on self-calibration has intensified. Generally
speaking, passive calibration and self-calibration were developed for different
goals and circumstances. When the influence of the non-linear distortion
component of the camera cannot be neglected or when highly accurate measure-
ments are to be made based on the recovered camera geometry, passive
calibration with deliberate modeling of the distortion is necessary. With a fixed
camera set-up, as all parameters are recovered beforehand by passive calibra-
tion, a real-time vision system can be built (Lei & Hendriks, 2002). On the other
hand, if the camera configuration is not fixed and the change is unpredictable,
self-calibration is needed to get the values of all parameters whenever needed.
However, due to the difficulty of robust feature extraction and correspondence
estimation, self-calibration is carried out off-line and, thus, real-time processing
cannot be guaranteed.
Within the passive camera calibration approach, different techniques can be
applied for different applications. It may or may not be necessary to model
distortion. If the accuracy offered by the linear model is acceptable to the
problem at hand, distortion does not need to be considered for higher efficiency;
otherwise, it has to be estimated and corrected. The complexity of modeling
distortion also differs from application to application. This mainly depends on the
required accuracy. In some cases, distortion can be estimated in advance of the
calibration of other camera parameters, but, in others, it is necessary to estimate
all camera parameters and distortion coefficients simultaneously to get highly
accurate results. The former is more efficient and versatile while less accurate
than the latter. The values of all camera parameters could be revealed explicitly
or only certain intermediate expressions need be calculated. It depends on what
the subsequent processing requires. From the above discussion, it is easy to see
that which calibration technique is adopted for a specific application completely
depends on the requirements of accuracy, the vision set-up, and the computation
resources. Therefore, of course, compromises can be made between available
calibration techniques and the application requirements. Two example vision
applications requiring different calibration forms were already introduced in
section 6. Because a high level of accuracy is required in both cases, distortion
is modeled in both of them. However, because we want to later on utilize dynamic
stereo set-up in the face model reconstruction application, which would be
calibrated by self-calibration, we estimate the distortion in advance of the
Search WWH ::
Custom Search