OpenCV CameraParams representation

[AlexeySrus]

Dear developers, thanks for a great toolkit.
I faced the problem of converting matrices from frames metadata to OpenCV CameraParams representation.

I read the matrices with the following code:

def read_matrix(_metadata: dict) -> np.ndarray:
    loaded_mat = np.array(_metadata['cameraPoseARFrame']).reshape(4, 4)
    y_rows = loaded_mat[1, :].copy()
    z_rows = loaded_mat[2, :].copy()

    loaded_mat[1, :] = z_rows * -1.0
    loaded_mat[2, :] = y_rows
    return loaded_mat

v_matrix = read_matrix(frame_metadata)
proj_m = np.array(frame_metadata['projectionMatrix']).reshape((4, 4))
mvp1 = proj_m @ np.linalg.inv(v_matrix)
intrinsics = np.array(frame_metadata['intrinsics']).reshape(3, 3)

Then I convert it to a OpenCV CameraParams class:

cam = cv.detail_CameraParams()
r, _, t, _, _, _, _ = cv.decomposeProjectionMatrix(mvp1[:3])
cam.R = r
cam.t = t
cam.focal = intrinsics[0, 0]
cam.ppx = intrinsics[0, 2]
cam.ppy = intrinsics[1, 2]
cam.aspect = 1

But unfortunately, I didn't get the desired results. Could you help me with camera converting question?

--
Best Regards, Alexey Kovalenko

[ghy0324]

@AlexeySrus Hi! I met the same problem, have you solved it?

[cclaan]

Maybe you could share more code of what you are trying to do with the camera in opencv? I could add an example based on that. It seems like you are exchanging y/z rather than negating them. You might also need to ignore the image orientation exif when using imread ( cv2.IMREAD_IGNORE_ORIENTATION )