estimateCameraParameters object points of irregular planar shape

I would like to use estimateCameraParameters because I am using image and object point correspondences from a non chessboard pattern. However, my image and object points are still retrieved from a planar surface. The only difference is that the points that lie on the plane may not always be in a rectangular shape. The problem is that in the Coordinate Systems page, it seems as though the object point coordinate system must always start at the top left corner at (0,0). However, this doesn't make sense for a non-rectangular shape. For example, what would (0,0) correspond to on a circle?

I don't see any algorithmic limitations to this since a homography estimation (which I presume is what estimateCameraParameters is doing) only requires at least 4 planar correspondences which don't necessarily have to lie in a rectangle.

Thanks for any clarifications.

[Florian Morsch]

0,0 refers to your starting point. For you example with the circle you could say 0,0 is your center, every other point detected is based on that center coordinate. Same goes for a checkerboard, if you have a simple checkerboard with lets say 4x3 squares (25 mm each), then your first square edge would be 0,0 and the second 25,0 then 50,0 followed by 75,0 and then 0,25, 25,25 and so on. Now if you turn that checkerboard by 45 degree your first point is still 0,0 followed by the next detected point.

So for a selfmade shape your 0,0 coordinate will be the point you detect on the most left/right top/bottom side, depending on your implementation. Every following point coordinates refers to the 0,0 as starting point. This has also nothing to do with the pixel value itself, a point on pixel [435 755] can still be coordinate 0,0 for your calibration pattern