I try to rectify an image and some points, which are on this image. Rectifying the image works very well (this part of the code is not from me):
(mapx, mapy) = cv2.initUndistortRectifyMap(camera_matrix,dist_coefs,np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]]),newCameraMatrix,(int(resolution*w), int(resolution*h)), cv2.CV_32FC1)
RectImg = cv2.remap(img, mapx,mapy,cv2.INTER_LINEAR )
But when I rectify the points with undistortPoints and the same parameters, I get (for my opinion) wrong coordinates.
point_file = np.loadtxt(fn)
point_matrix = np.zeros(shape=(nr_points,1,2))
# fill Pointfile in n x 1 x 2-Matrix
for each in range(0, nr_points):
point_matrix[each][0][0] = point_file[each][0]
point_matrix[each][0][1] = point_file[each][1]
point_matrix_new = cv2.undistortPoints(point_matrix, newCameraMatrix, dist_coefs)
These are the parameters I'm using (newCameraMatrix & dist_coefs):
[[ 4.93906295e+02 0.00000000e+00 1.24539714e+03]
[ 0.00000000e+00 4.92616567e+02 1.03814593e+03]
[ 0.00000000e+00 0.00000000e+00 1.00000000e+00]]
[ 5.93179211e-01 3.59577119e-02 -3.34062329e-05 -8.92301489e-05
-5.27895858e-04 9.28762999e-01 1.46636733e-01 0.00000000e+00]
I'm also not sure, what the output actually is. Are the resulting values are image-coordinates or sensor-coordinates and in which unit? I could only find a documentation of undistortPoints for cv (http://docs.opencv.org/modules/imgproc/doc/geometric_transformations.html) but not for cv2.
example (image size = 2448 x 2048 pixel / sensor size = 8.6 x 6.6 mm
Input (point_matrix):
[[[ 0. 0.]]
[[ 2448. 0.]]
[[ 0. 2048.]]
[[ 2448. 2048.]]
[[ 1224. 1024.]]]
Output (point_matrix_new):
[[[-6.49236118 -5.42726306]]
[[ 6.39989444 -5.5358653 ]]
[[-6.50237385 5.28935346]]
[[ 6.58981334 5.54673511]]
[[-0.04336093 -0.02874159]]]
Thank you for your help!