Ground plane triangle is not square.

[aweinstein]

Hi:

During the calibration process, after doing the "Set ground plane", the captured triangle is not square.

See this to see what I mean:

http://www.ocam.cl/fotos/optitrack/ground_plane.jpg

I think that by definition the calibration triangle should be captured as square triangle.

Any comments?

[VincentG]

The triangle should be a 90 degree angle, at the base.

Could you send my an image of the calibration square itself, if the reflectors in place.

[aweinstein]

Here is an image of the triangle:

http://www.ocam.cl/fotos/optitrack/triangle.jpg

It is a commercial carpenter square, so I'm pretty sure it is square. (Similar to this one http://tinyurl.com/2q732d).

It must be noted that sometimes, with some camera configuration, I do get the right triangle .

[leith]

off hand, it looks like your cameras are too close together. The more angular disparity there is amongst rays reconstructing a point, the more accurate it tends to be. If your cameras made a 90 degree arc around the square, resulting in two cameras rotated perpendicular to one another, you would likely get a better result.

Both your calibration and the resulting reconstruction will likely suffer from the cameras being too close.

A number of academic computer vision examples of stereoscopic reconstruction use cameras that are relatively close together. however, those cameras are usually very high resolution and have undergone bench calibration. The math is similar but the scenario is not the same.

If you need to get an accurate calibration in this configuration, your best bet is to make sure that when you wand, you not only move the wand across all the camera views, but do it at different distances and depths from the cameras. Its the depth that will keep that kind of setup in check and calibrating correctly.

[Jim]

Also, the surface looks pretty reflective to me, a matte surface, or a black one will help with reflections off the metal.