Actually, here I just rephrase what's writtein in the OpenCV reference manual. object_points is array of the feature points of the calibration patern in the pattern coordinate space. The array is duplicated as many times, as the number of the pattern views captured. For example, if the pattern is planar, such as the checkerboard, X and Y axes are usually set parallel to the board sides with the coordinate origin at one of the board corners, and Z axis is orthogonal to the board plane. In this case object_points is{(X_k,Y_k)=(square_size*(k mod ), square_size*(k / )}. image_points is the projection of the object_points to the image view. It should be retrieved using a specialized algorithm, such as cvFindChessboardCorners. There is see opencv/tests/cv/src/acameracalibration.cpp, opencv/filters/calibfilter/calibfilter.cpp and opencv/cvaux/src/cvcalibfilter.cpp for examples how to use cvCalibrateCamera2. There will be also a simple example added to opencv/samples/c/ that wil demonstrate the function use.
