Game Development Reference
In-Depth Information
application to application. All existing camera calibration approaches can be
categorized from several different points of view.
Given the underlying assumptions about the parameters to be estimated, the
following three general types of calibration exist:
Passive (Fixed) Calibration: All camera parameters are assumed fixed and,
therefore, the calibration is performed only once. This approach often uses
images of objects for which the accurate geometric and photometric
properties can be devised, and reconstructs the relationship between such
properties and the recorded images. From these relations all camera
parameters can be estimated quantitatively through some linear or nonlin-
ear optimization process (Tsai, 1987; Triggs, McLauchlan, Hartley &
Fitzgibbon, 1999).
Active Calibration: For active vision purposes, some of the intrinsic camera
parameters (typically focus and zoom) are assumed to vary actively, while
the extrinsic parameters are either changed on purpose in a controlled
fashion (e.g., pure rotation) or not at all (Willson, 1994). The investigation
of the relationship between zooming (and/or focus) and other intrinsic
parameters is at the center of this approach.
Self-Calibration: Depending on the application requirements, all or some of the
camera parameters are assumed to vary independently, while the remaining
ones are unknown constants or parameters that have already been recov-
ered via a pre-processing step (e.g., through a passive calibration tech-
nique). The purpose of self-calibration is to be able to recover the camera
parameters at different settings of the optical and geometrical configura-
tions (Pollefeys et al., 1999).
By considering the appearance of the camera parameters, we can identify:
Explicit Calibration: This represents the traditional approach (Slama, 1980),
according to which the values of all individual camera parameters are
calculated explicitly.
Implicit Calibration: In this case, only certain relations of the camera
parameters are recovered, for example, the projection matrix in equation 4.
These relations must contain enough information to support subsequent
calculations. The exact value of each individual parameter is not made
explicit (Wei & Ma, 1994). However, all individual camera parameters may
be calculated from the recovered relations.
Search WWH ::




Custom Search