Hi folks:

I've recently done my first shoot with 8 cameras, and the calibration program fails miserably. The Marker misdetection statistics are 50-80% in one run I did. The video looks good based on my experience with a 6 camera setup. I did find calibration easier and more reliable in version 1 than in version 2. The 6 camera shoot I did in version 2 took much more work to calibrate including a manual edit of the xml scene file to get good initial camera positions. I have also had to edit the xml scene file this time to get good initial positions (the mouse based move controls in DMC for the camera did not allow proper camera placement).

Can you please help me solve the problem with the current calibration? I did this calibration shoot as part of a 1-day workshop and the video cannot be reshot. Workshop participants are depending on me for the results.

Thanks,

Michael Miles

I will upload the mjpeg video files to a share so ipisoft can look at them if they wish, Michael

Most likely, calibration problems are caused by incorrect auto-detection of initial camera positions (Auto-detect initial camera position checkbox selected).
Better results you achieved with v1 on 6 cameras are explained not by difference between v1 and v2, but by difference in cameras number.
Fails of initial camera positions autodetection are more frequent for 8 cameras.
We'll improve this in the next version (v3) which hopefully will come out around September.

Before then, the only way to solve this problem is to turn off "Auto detect initial camera positions" checkbox and manually set initial camera positions. You do not need to do this very precisely, the main requirement that you have correct mutual configuration of cameras, all cameras looking to the center of capture area.

If this does not help, please share your calibration video, we'll look at it.

_________________
iPi Soft

Hi pas:

I tried several passes without the auto-detect set, and they all go crazy and don't work. I will email the file download link to whomever is relevant - I sent the files to Andrew last time - is he still the one who will handle file download?

Version 2 is more problematic than version 1 using since I used exactly the same number of cameras (6) in both versions.

Thanks for helping with this,

Michael

Please send your project to our tech support e-mail ipi@ipisoft.com

_________________
iPi Soft

Hi folks:

OK, I was in contact with ipisoft tech support (great support there guys!), and they told me what was wrong with the calibration video. Unfortunately, I did not have another calibration video from the 7/16 shoot, so I had to try to figure out a way to salvage what I had. They told me that the calibration light had to be visible in ALL 8 cameras for at least 100 frames. So, when I shot on a new day, I figured that I had to follow these rules:

1) Of course, the maglight is mounted on a stick so I don't occlude it with my body.
2) I check the floor circle size visible in all 8 cameras, this is actually smaller than the capture area (about 1/2 the diameter in my case).
3) Whenever I pass in front of an individual camera, I lift the light up above my head a few inches so all the cameras see it at all times. I approach the left side of the camera, lift the light up, walk to the right side of the camera and then lower the light down to the floor.
4) I use the "table top" calibration pattern to move the light as on the ipisoft wiki.


The problem with my 7/16 calibration video is that I didn't lift the light up above my head, and my body kept occluding the marker (camera angles didn't help also). The solution I came up with was to export the calibration video, convert the exported individual camera videos into a recorder version 1 video with only frames where the light marker was visible in all cameras (this video had 1216 frames). I then used the recorder version 2 to convert the version 1 video into a version 2 calibration video, and then successfully calibrated off of that. I was worried that the markers had to be contiguous, but DMC was able to calibrate with jumps in the marker position.

I will be going into more detail about how I fixed the calibration on the yahoo motion capture group, and may post the Matlab code there (under GPL) if people are interested in it:

https://groups.yahoo.com/neo/groups/motioncapture/info

Regards and thanks to the folks at ipisoft for their technical help, Michael