what is new camera matrix in cvundistortpoints.?

Question

I am facing a problem in understanding what the new camera matrix in cvUndistortpoints. We get a camera matrix of camera in stereocalibration itself and why we have to pass a new cameramatrix in cvstereorectify which we obtain from cvundistortpoints. How exactly new camera matrix different from camera matrix after the output of cvUndistortPoints?

void cvUndistortPoints(const CvMat* src, CvMat* dst, const CvMat* cameraMatrix, 
         const CvMat* distCoeffs, const CvMat* R=NULL, const CvMat* P=NULL)

The reference link is here: https://docs.opencv.org/3.4/da/d54/group__imgproc__transform.html#ga55c716492470bfe86b0ee9bf3a1f0f7e

Heddie

Actually, P, the new camera matrix, is an -input-. cvUndistortPoints does not create it. – Crankshaft

Answer 1

The answer is one page up in the documentation you provided. P is given to you by StereoRectify.

Here is what it says about it under undistort in opencvrefman.pdf, which, my friend, you should be reading too:

newCameraMatrix – Camera matrix of the distorted image. By default, it is the same as cameraMatrix but you may additionally scale and shift the result by using a different matrix.

Now, I am pretty new to this, and have not had occasion to use the StereoRectify functions yet, but I am guessing that you might want to undistort an image to look like what a different camera would see.

Crankshaft

I would be very happy if you could clarify me, why we transfer the new camera matrix into cvUndistortpoints. What is the specific use in passing this new matrix.. – Heddie

The confusing part in the documentation is the phrase "camera matrix of the distorted image". It's NOT the matrix of the distorted image, but of the hypothetical "undistorted" image - i.e. the output image. A camera matrix has 4 important fields 2 focal lengths (for x and y), and 2 shifts - the coordinates of the "principal point". First two allow for scaling, the latter can shift the viewpoint. I also tried to rectify the image in one go, by multiplying the inNewCameraMatrix by inverse of rotation matrix from the extrinsic parameters of camera calibration (to look exactly along the z-axis). – Hardily

Answer 2

In the case of monocular camera newCameraMatrix is usually equal to cameraMatrix , or it can be computed by GetOptimalNewCameraMatrix for a better control over scaling.

Check out http://opencv.willowgarage.com/documentation/cpp/camera_calibration_and_3d_reconstruction.html#cv-getoptimalnewcameramatrix

Edifice

Any chance you can post the relevant code? The link is dead. – Valval