I want to draw 3D Positions in a webcam Image using OpenCV's projectPoints function. During testing I noticed, that I always have a certain offset from the real object. This is most obvious when trying to project the origin (0,0,0) to the image center. The image shape is (2988, 5312, 3), the big red dot is the image center at (1494, 2656) and the small red dot with the lines attached is the projection of the origin (0,0,0) with NO translation and NO rotation leading to (1476, 2732).

The main question: Why is the projected point not in the middle of the image?

I determined the projected point like so:

origin_2d, jacobian = cv2.projectPoints(
    np.array([(0.0, 0.0, 0.0)]),
    np.array([(0.0, 0.0, 0.0)]),
    np.array([(0.0, 0.0, 0.0)]),
    mtx,
    dist,
)

So the origin (0,0,0) should be projected to 2D with no translation and no rotation and thus appear in the image center, right?

I obtained the camera mtx and dist by executing a camera calibration routine as described here OpenCV docs:Camera Calibration using the chessboard you see in the image. Is there something wrong with my calibration? Or am I misunderstanding something in the projection process?

Thanks and best regards, Felix

I want to draw 3D Positions in a webcam Image using OpenCV's projectPoints function. During testing I noticed, that I always have a certain offset from the real object. This is most obvious when trying to project the origin (0,0,0) to the image center. The image shape is (2988, 5312, 3), the big red dot is the image center at (1494, 2656) and the small red dot with the lines attached is the projection of the origin (0,0,0) with NO translation and NO rotation leading to (1476, 2732).

The main question: Why is the projected point not in the middle of the image?

I determined the projected point like so:

origin_2d, jacobian = cv2.projectPoints(
    np.array([(0.0, 0.0, 0.0)]),
    np.array([(0.0, 0.0, 0.0)]),
    np.array([(0.0, 0.0, 0.0)]),
    mtx,
    dist,
)

So the origin (0,0,0) should be projected to 2D with no translation and no rotation and thus appear in the image center, right?

I obtained the camera mtx and dist by executing a camera calibration routine as described here OpenCV docs:Camera Calibration using the chessboard you see in the image. Is there something wrong with my calibration? Or am I misunderstanding something in the projection process?

Thanks and best regards, Felix

• python
• opencv
• computer-vision
• camera-calibration
• coordinate-transformation

• crosspost: forum.opencv./t/projectpoints-origin-projection-off-center/… – Christoph Rackwitz Commented Nov 20, 2024 at 12:47

1 Answer 1

The image shape is (2988, 5312, 3), the big red dot is the image center at (1494, 2656)

You calibrated that camera, right? Then that's not the optical center. The optical center is in the camera matrix, along with the focal length.

(1476, 2732)

That looks more like it, although I'd have expected these numbers to also have digits after the decimal point.