aruco: conflicting documentation in estimatePoseSingleMarkers

[crackwitz]

• OpenCV => 4.5.5

Documentation says:

The marker corrdinate [sic] system is centered on the middle of the marker, with the Z axis perpendicular to the marker plane.

This agrees with the implementation.

But it also says:

The coordinates of the four corners of the marker in its own coordinate system are: (0, 0, 0), (markerLength, 0, 0), (markerLength, markerLength, 0), (0, markerLength, 0)

This, to my mind, disagrees with the previous statement. (0,0,0) is the origin, which is the center of the marker, not any of its corners. Perhaps "its own coordinate system" refers to a different coordinate system... which it should not.

I would suggest the following:

The coordinates of the four corners of the marker in its own coordinate system are: (-markerLength/2, -markerLength/2, 0), (markerLength/2, -markerLength/2, 0), (markerLength/2, markerLength/2, 0), (-markerLength/2, markerLength/2, 0)

Or alternatively, to show with a diagram where and in what order the corners are.

I am doubting myself here, because recently someone reported that their code puts an axis triad at the corner of the marker... and I don't see why their code would do that (the python equivalent puts it in the center), but it would agree with the second sentence from the documentation.

[AleksandrPanov]

@crackwitz, to make the pose estimation for the single marker and boards (estimatePoseSingleMarkers(), estimatePoseBoard(), estimatePoseCharucoBoard()) uniform, the coordinate system has been moved to the top left corner.

Thanks for the issue, I've updated the documentation and tutorial. For compatibility with old codes, I can add a mode for the algorithm.

[AleksandrPanov]

To draw the axes in the old format, you can use this sample:

void drawCustomFrameAxes(Mat img, Mat cameraMatrix, InputArray distCoeffs, InputArray estimatedRvec,
                         InputArray estimatedTvec, float axisLength, int thickness = 3,
                         Point3f center = Point3f(0.f, 0.f, 0.f), Point3f axes = Point3f(1.f, -1.f, -1.f)) {
    vector<Point3f> ax = {center, center + Point3f(axes.x, 0.f, 0.f)*axisLength,
                          center + Point3f(0.f, axes.y, 0.f)* axisLength,
                          center + Point3f(0.f, 0.f, axes.z)*axisLength};
    Scalar colors[3] = {Scalar(0, 0, 255), Scalar(0, 255, 0), Scalar(255, 0, 0)};
    std::vector<Point2f> imagePoints;
    projectPoints(ax, estimatedRvec, estimatedTvec, cameraMatrix, distCoeffs, imagePoints);
    for (size_t i = 1ull; i < ax.size(); i++)
        line(img, imagePoints[0], imagePoints[i], colors[i-1], thickness);
}

drawCustomAxes.cpp.txt

Original image for example

[AleksandrPanov]

PR #3256 fixes docs/tutorial and add mode for old coordinate system

[crackwitz]

Does the second picture imply that now Z points into the marker?

Which order of corners is considered counter-clockwise? I'm concerned that the identifiers in #3256 are mixed up. Looking at a marker, it always used to be TL, TR, BR, BL, which I consider clockwise. Now, if X right, Y up, Z near ("old"), that's going negatively around Z. If X right Y down Z far ("new"), that's going positively around Z.

I wouldn't recommend using "clockwise"/"counterclockwise" there at all. The difference isn't the order of corners but the placement (orientation) of the coordinate system.