How to identify 3D reference points of X (working on own text data)

[SubhashPavan]

I was trying to implement your version1 Opencv implementation and wanted to try out the algorithm with new data.

These are the steps that I have followed, please let me know how do I proceed?

1. I have the Camera Intrinsic Params that I have obtained from Opencv

2. I have taken three images of X posed at different angles from different poses/angles

3. Now my question is how do I populate model.txt file. For instance point (175,0,0) how do I read this? I am assuming that this is in camera coordinates and 175 doesnt indicate any specific UoM

Please help!

[nbhr]

In short, the points in model.txt are identical to the object points in OpenCV camera calibration.

I guess you used a chess pattern for your intrinsic camera calibration by OpenCV with providing the 3D positions of the chess corners as object points and their projections as image points.

Similarly, our calibration algorithm requires the 3D positions of the model points as model.txt and their 2D projections via a mirror in different poses as input?.txt. So if a chess pattern is used, model.txt should be identical to object points in OpenCV. If another object is used, please provide its geometry by yourself.

Hope this helps.

[TNA8]

How to flip the detection result for input?.txt?

[nbhr]

It actually depnds on your implementation. If you are not sure, I would suggest using ChArUco pattern instead of the traditional chesspattern. It does not find the pattern if the image is mirrored (I guess). So you first need to flip the entire image anyway, find the pattern, and flip it back to get the origial corner position.

Anyway the key point is to make the points provided in input{1,2,3}.txt be the projections of the same line in model.txt.

[TNA8]

Thanks for your reply. I have other questions.

• You provided 2 versions of code. Version 1 is not accurate than version 2? I am using version 1 because version 2 has not opencv version. Is it okay to use version 1?
• The calibration matrix(R, T) depends on model.txt. What units did you use in model.txt? I have 3d positions in camera coordinate system in meter and want to convert it in monitor coordinate system in pixel. In this case, model.txt must be in meter or pixel?

Thanks in advance.

[nbhr]

Hi, while we recommend v2 basically, v1 should be ok as well. The unit in model.txt is arbitrary. if you describe the 3D point positions in meters, you get the mirror poses in meters as well. In your case, if you provide the 3D chess corner positions in pixels, your camera / mirror poses will be descibed in pixels.

[TNA8]

Got it. Thanks