Image Processing Reference
with a set of intrinsic parameters (focal length, principal point, skew of axis) and its orientation
is expressed by extrinsic parameters (rotation and translation). Both intrinsic and extrinsic
parameters are estimated by linear or nonlinear methods using known points in the real world
tion pattern with known geometry, usually a flat chessboard.
Many studies have given atention to the camera calibration area, most of them are ded-
icated to the parameters' estimation phase and the refinement location of calibration points
dressed. Tsai [ 7 ] and Zhang [ 8 ] are examples of the most cited papers related to this area. They
propose closed form solutions for the estimation of intrinsic and extrinsic parameters using
about some related works.
Camera calibration is a much discussed topic but the lack of robust algorithms for features
detection diicults the construction of automatic calibration process. Calibration patern re-
cognition is a hard task, where lighting problems and high level of ambiguities are the prin-
cipal challenges. For this reason, applications for camera calibration often require user inter-
vention for a reliable detection of the calibration points. The hand tuning of points is tedious,
a complete identiication of the calibration patern. This is a severe constraint due to illumina-
tion or occlusion problems.
Currently, there is an increasing demand for systems with multiple cameras, such as aug-
mented reality applications and 3D reconstruction, which makes the manual calibration an
impracticable task or time consuming. Some tools for automatic camera calibration are avail-
The application asks the user to define four extreme points that represent the area where an
algorithm searches for the calibration points given the number of rows and columns of the
matic way to detect chessboard paterns in images using the findChessboardCorners() function.
The method performs successive morphological operators until a number of black and white
regions' contours be identified and, subsequently, four corners are extracted of each contours,
comprising the calibration point set. The patern is recognized only if all rectangles are iden-
tiied. In an online system this restriction causes a considerable loss of image frames, since is
not always possible to detect all the chessboard rectangles.
pattern. The described methodology is robust to noise and it is not necessary to identify the
entire calibration patern. In the other hand, the markers are complex and require a special al-
intersections of lines. The methodology uses a combined analysis of two consecutive Hough
transforms to ilter the collinear points inside the patern. The assumption that all points of
interest are collinear makes this algorithm very sensitive to distortions, limiting its use only to
cameras with low radial distortion.
squares within a geometric mesh. Furthermore, the system must to detect three circles to de-
termine orientation of the patern. Harris corner detection is time consuming, sensible to noise,
needs an empirical threshold to select interesting points, and does not produce good results to
that do not belong to the calibration patern, especially for images with complex backgrounds.