ipisoft.com
https://forum.ipisoft.com/

5th PS Eye kills calibration
https://forum.ipisoft.com/viewtopic.php?f=6&t=9256		 Page 1 of 2
Author:  mlbspike [ Wed Dec 31, 2014 3:04 pm ]
Post subject:  5th PS Eye kills calibration
Added a USB 3.0 expansion card, to test beyond the 4 camera setup, and all 5 cameras are recognized, and I can capture a calibration video. I have the cameras arranged in counter clockwise order. I can unplug ANY single camera, to get back to 4, and the calibration indicates PERFECT for all 4 cameras. Add the 5th camera back, and it will fail on 1,2 or 3 cameras, and never get above GOOD for the other cameras, which seems ... counterintuitive. Looking at the camera positions, they appear no worse than they are when I calibrate 4 cameras ... any attempt to push them into more accurate positions, and calibrate again only makes things worse. I can drag the other computer down to the capture area, and try the new distributed capture, hoping that it's some kind of bandwidth issue, but I'd guess that to be grasping at straws. Suggestions?
BTW ... recent i5 4590 3.3 Ghz processor, on a z97 MB, with Samsung SSD for capture files.

Mike
Author:  vmaslov [ Fri Jan 02, 2015 4:33 am ]
Post subject:  Re: 5th PS Eye kills calibration
Hello

I do not quite understand: whether you get bad camera positions determined by the calibration, or the positions look good, but the indication of result looks erroneous?

If the former, then it can be explained by the greater number of cameras. More cameras => more values to determine => may lead to greater errors. Also, with more cameras, it takes more practice to keep a light marker visible to all cameras, so there are less "good" frames overall. You can try to choose some relatively short region of frames for calibration, where in most of the frames marker is visible to ALL cameras at once.

Regarding indication of result. There is no straightforward way to determine whether the calibration has actually failed. So we use some heuristics derived from the set of calibration records we have to evaluate the results. It may be the case that these heuristics work poorly in your situation and report failed calibration when it has succeded actually.
Have you tried to track some action with such calibration?

Please describe how the cameras are located when calibration fails. Better by picture.
You can provide us your calibration video for testing. Use file sharing service like Dropbox or Google Drive.
Author:  mlbspike [ Fri Jan 02, 2015 2:14 pm ]
Post subject:  Re: 5th PS Eye kills calibration
The camera positions seem reasonable. Marker in view of all cameras virtually 100% of the time. Somewhat suspicious of the FAILED status, but didn't realize I could try using a calibration file that had failed. Will do further testing, and send files if needed. I did finally get one successful 5 cam calibration, but after testing a larger space, having the same tough time calibrating, in the larger space, and when returned to the original space.

My primary setup has been with 4 cams, stationed on 2 sides of a room in a ... Sort of semi circle, with calibration relatively easy. The motions I capture face into the semi circle. The 5th camera is directly behind the actor. Not a recommended configuration. Does that make the calibration tougher? My calibrations were difficult when I did a box around the actor, and much easier in the semi circle, btw.

Related question. I have a 20x30 area ... Garage with open door. Trying to maximize one side of the capture area ... Trying for 7x16-ish. Seems like a semi circle will work. By compromising the semicircle somewhat, I could permanently mount 3 camera to walls, reducing my setup time. However, if the truer semicircle is easier to calibrate, I'd be better off marking tripod positions, and skipping the permanent mounts. So, my question ... How significant is the shape of the semicircle? Non permanent mounts would also make it easier to add the 6th camera (which I already have), so leaning that way.

As you may guess, I am trying to decide which version to purchase. The motions I need are sports related, primarily running, throwing, diving in a single direction ... Diving being a particular question mark, where I speculate that the extra cameras may be helpful.

Thanks for any comments, or suggestions.
Author:  mlbspike [ Sat Jan 03, 2015 9:24 pm ]
Post subject:  Re: 5th PS Eye kills calibration
As I posted elsewhere, I was able to calibrate 6 cameras successfully several times, and am testing captures. Control of glare, and a brighter bulb for my Maglite seemed to have helped (along with the new distributed processing). Still interested in any suggestions I can get regarding the best setup for my 20 x 30+ space, as regards maximizing the distance of a capture for some running motions.
I'm thinking kind of a:
<-------->
< >
<-------->

configuring 3 cameras in a v on each opposing wall with a 20 ft long capture area (10 ft wide)
Can't seem to figure a way to have a camera on the side without killing the length of the capture volume, or pushing the volume
so far against the opposite wall, as to cause more problems than I help. Since the v on each end will be 18 ft across, vs the
20 foot long capture area, I hope it will be ok.

Mike
Author:  mlbspike [ Sun Jan 04, 2015 4:00 am ]
Post subject:  Re: 5th PS Eye kills calibration
You'll have to use your imagination to compensate for my poor attempt at using character graphics, which lost all it's spaces, in the previous post.

BTW, I assume the advertised 20x20 capture space is actually more of a circle ... Somewhere between a 20 ft diameter circle inside that space, and the 28 ft ft diameter circle containing a 20x20 space, and that as you approach this limit, good lighting becomes even more important for the actor at his max distance from a camera ... Though being at his min distance from other cameras will help balance things out.

Mike
Author:  vmaslov [ Sun Jan 04, 2015 9:49 am ]
Post subject:  Re: 5th PS Eye kills calibration
The setup you suggest looks reasonable. You can also experiment with heights depending on specific motions you want to capture. For example, for diving onto the floor, the final stage of a motion is better captured from above, thus placing the 2 cameras at the corresponding end to the ceiling or at increased height should help.

The exact shape of camera setup does not matter much. What matters is how many cameras with considerably different POVs see an actor at any given time. Circular setup with all cams looking to the center is generally an ideal situation for mocap. But of course many other configurations will work depending on the available room shape/size and particular motion to capture. What should generally be avoided is:
  • Placing 2 cameras close to each other and looking in the same direction.
  • Placing 2 cameras so that their fields of view have small intersection.
Author:  mlbspike [ Mon Jan 05, 2015 5:12 pm ]
Post subject:  Re: 5th PS Eye kills calibration
Well ... moved one camera to a better position, and ... dead in the water again.

Here's the video file (6 cameras). I obviously clipped some off the beginning, and end.

https://www.dropbox.com/s/zr9jy4tjyfcja ... Video?dl=0

Very similar to a perfect calibration from the previous day (slightly different position for one camera), but instead of .31 for misdetects, it's 40,50,60% ... pretty much crap.

I've now spent more time trying to calibrate than the software will cost me (at even modest wage rates).

Would appreciate any information anyone can offer. Going to take the risk, and go the for the most expensive subscription, on the assumption that my calibration can be resolved. Saw a couple posts on other sites giving up on ipisoft because the calibration was impossible, but I think it would be more than a couple random posts, if it was really a nightmare. Though ... nightmare pretty much describes it, for me so far. Picked up some painters paper, to cover some bright wall areas, and some canvas drop cloths, and will try again, while I wait for someone to hopefully offer a solution.

Mike
Author:  vmaslov [ Tue Jan 06, 2015 1:39 pm ]
Post subject:  Re: 5th PS Eye kills calibration
Hi

First, let me make some clarification. Our calibration algorithm is stochastic. That is, it uses some random selections from the source data for computations. Therefore, the results may (and usually do) differ between several runs on the same input.
When the data are good enough, the results are very close. When the data are worse, the calibration may show poor results if much of that random selection falls onto the bad data (frames where marker is not detected). In that case, if you see that overall quality of calibration video seems OK, you can just perform another run or two of calibration to check whether you can get substantially better results.

I've checked your video. For me, it calibrates nicely - of 3 runs I've got 2 PERFECT and 1 GOOD results. In a GOOD result, only single camera had questionable value of misdetects above 30%. And in all 3 runs the determined positions for cameras look very close.
Author:  mlbspike [ Tue Jan 06, 2015 2:51 pm ]
Post subject:  Re: 5th PS Eye kills calibration
That's encouraging ... and puzzling. I did get that ... and I've tried multiple runs on the same data, with no success (though I may have adjusted the scale, and turned the auto position toggle on and off before re-running. I have made about 25 failed attempts, since my two successful calibrations. I've run a number of them multiple times (autodetect position set). I will try the example I sent again, to try to eliminate as many variables as possible. Doesn't seem like I could be using different settings (there aren't a lot of settings). AutoDetect on, is about it. Puzzled.

Mike
Author:  vmaslov [ Tue Jan 06, 2015 3:12 pm ]
Post subject:  Re: 5th PS Eye kills calibration
A guess: what if there are glitches in GPU computations in Mocap Studio on your machine? We had several reports of the weird behavior of the program which were solved by the update of video drivers.
What is your video card and do you have the latest drivers?

FYI, I did calibrate with the default options: 'Auto detect positions' on, others off. I'm attaching a project file with the results of the last run.

Attachments:
File comment: Calibration project
2015-01-04_22-01-36.zip [252.97 KiB]
Downloaded 878 times
Page 1 of 2 All times are UTC - 8 hours [ DST ]
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/