calibrate cameras from any position

[cpatel245]

Hi @DarienBrito

Do you already have the solution for calibrating cameras and merging the point clouds when kinects are placed more than 90 degrees? if yes, could you please elaborate on the solution or maybe include it here in the repo?

Thank you in advance.

[DarienBrito]

Hi there!

I have a solution indeed but I have not published it yet. The solution implies creating intermediary matrices in between the transformations so that when the angles are larger they can be computed disregard of the position. The mechanic for this is rather simple. Take snapshots of a given point cloud state and move the cameras progressively towards the desired final position.

I have not had the time to put this in shareable form, but since I have an upcoming project involving many Kinects I will probably be doing this anyways, in which case I will upload a new version to account for these cases.

[cpatel245]

Hello @DarienBrito

Were you able to calibrate the cameras placing them at/more than 90 degrees to each other (e.g., placed at the corners of the square/rectangle)? I was able to do that, but edges of the objects in the scene seems to degrade the (visual) quality of merged point clouds probably due to color/depth shifts. #1058 #1588

If you have tried the calibration for mentioned set-up, could you please tell me if you have faced the same issue?
And if you would be able to share some results here, that would be also great !!!
Thanks in Advance!

[DarienBrito]

Hello @crypto-g245

Sorry for late response. Have been tremendously busy. Yes, I was able to do so and I recall having some degradation indeed. In my case some of it was due to the amount of machines involved (I was using a single machine for 4 devices), once I moved to a split of 2 machines with 2 devices each, the glitches dissaperad. The angular degradation as far as I can remember is probably due to distortion of the space, not much that can be done on that regard... unfortunately I do not have a setup at this moment where to test and did not keep snapshots of it... maybe this coming year I'll rig them up again.

Curious to know your method too. Was it intermediary matrices as I described above?

[cpatel245]

Hello @DarienBrito ,

No worries.
Yes, the method is quite similar to what you mentioned.

I think the distortion was also due to interpolated depths in colour camera's view. I was transforming depths to colour camera's coordinate system to get higher resolution (more no of points), but in depth cam's view, quality is much better. But the edges at the stitching lines are still there and I think that's because of the depth-colour shift as I mentioned in the previous comment. because these points on the edges often have the colours of the background surfaces.

And as you said, there is not much that I can be done to resolve this issue completely as of now. because issues related this particular problem (colour-depth shift) are still open and there is no satisfactory solution as of now.

[DarienBrito]

Hi @crypto-g245,

Yeah, I guess such distortions will be addressed at some point and hopefully resolved as far as possible. I'm scheduling some time in February to tweak with the Kinects again for a project. I'll probably update the library a bit more and perhaps add the intermediary matrices technique to that update :)