Hi, Greenlaw! Thaks for the answer! I tried the 4 pseye set up, but still couldn't make it work. I got a better result than before, the cameras were at the correct height (matched the real world location), but the 3D actor wouldn't allign with the actor from the recorded video in all cameras.

I went ahead with the plan B and used the dual kinect. I got nice results, but I feel like coming short in my results... I made a investment with all the cameras to get the best result possible and couldn't use it.

Michael Nikonov, as I stated before, I'm working on my college's studio, so I have to remove the cameras at the end of the day. After I finish the recordings with the kinects, I'll try to work on the ps eye again.

But last time, I assembled the cameras on a hexagonal set up. Camera 1 was facing camera 4, camera 2 facing 5 and 3 facing 6.

All of them are set to the blue dot.

There are a couple of things that may be causing the problem:
*One of the walls doesn't have a clear distinction between wall and floor, since its is meant to be used as a infinite background for photographs (there is a ramp between the floor and wall). I've paced a carpet and some boxes to determine the "ground plane".
*I had to tilt some cameras downward, so that I could keep the floor in frame. Should the cameras be perfectly parallel to the floor?

I managed to get a correct calibration from your video (please see attached file) but this was achieved with some manual tweaks with a modified version of software.

The problem is definitely with camera placement. Almost all cameras are on the same height, and some cameras are placed three in line. This is not recommended - not only because it is difficult to calibrate, but also because such geometry is disadvantageous for tracking.

In the future, you can improve the results by adhering to the following guidelines:

1) Try to place cameras at different heights. For example, place three cameras at 1.6m and three other cameras at 2.5m.
2) When using 6 cameras, try to place them more so that they make up a circle (in top view), not a rectangle.

In the future, we will improve calibration algorithm, but as I already mentioned it will always be preferable to place cameras in a more geometrically advantageous configuration.

As a temporary workaround (until we improve the calibration algorithm), you can send your calibration videos to us and we will help process them.

Thanks, Michael Nikonov!
I'll open the files as soon as I get back at my pc that has the ipi license active.

But I'm having trouble understanding how to set the cameras... should I alternate the heights, like so Cameras 1, 3, 5 at 1,60m and Cameras 2, 4, 6 at 2,05m?

Here is roughly what I did last time (this is what I meant when I said a hexagonal shate. imagine evey vertice being a camera):
http://etc.usf.edu/clipart/43400/43447/6c2_43447_lg.gif

It doesn't have 3 cameras at the same line.

Yes.



Yes, if you place cameras like that it should be fine.