Hi guys!

I started recording and tracking of some mocap data with 4 PS eye cameras. now i ran into a calibration issue with the auto-calibration, but i was able to fix 3 of the 4 cameras pretty well with manual adjustments.

when i started the tracking process, i realised that the software tried to match the bad-calibrated camera (which was camera 2) more than the well-calibrated cameras (camera 3 and 4). i know good calibration is important for every camera, but in this particular case i just couldnt get camera 2 right. my impression was, that the software first looks at cam1, then cam2, then (maybe?) cam3, and if the body part (e.g. left hand) is present in those cameras, the other cameras are ignored and the software matches the average hand-position to fit those cameras.

not really an issue to fix, but i'm just trying to understand a little bit better whats going on under the hood and why i get the tracking results that i am getting ;)

cheers and thanks!

Hi

If you mean that you have fixed the camera positions after the calibration is completed, than you should not do that. You always need to get good computed positions.
First, you must ensure that marker misdetection value for each of the cameras is not higher than 20%. You can see this value in the "Scene" tab after the calibration is complete. If not, you should record more quality calibration video, in which the marker will be less obscured or brighter, depending on the cause of misdetection.
Second, if "Auto detect initial camera positions" option does not produce good results, you should disable it. Then make manual adjustments to make camera positions approximately correspond to the reality (camera order matters here). And after that run the calibration again.

Tracking algorithms gives no priority to any of the cameras. It uses all the information available to get the best average fit of the model to the actor's image from all the cameras.

thanks for the quick reply.actually yes, i manually moved the cams after the computed positions, but i'll give it another shot today with a 6-camera setup and a brighter maglight. the initial position that the software computed was quite ok, but always some cm off. (sometimes around 7cm, which then again is quite much)

actually by now i had always only around 1% (max) of red points on each camera, as well as 5 ground points. i'll try to calibrate today once again with the new stuff, and post my results back here. in case the calibration wont work well, i'd be glad if someone more experienced could have a look at the video and maybe tell me what i'm doing wrong.

cheers and thanks!

hi again!

aaalright, so this time i used 6 cameras in a half-circle setup and did 5 ground-points, plus, the maglite is always visible to all the cameras (0.057% misdetections. i used the old maglite since the new one turned out to be way too bright for calculation...the "dot" was almost as big as a light-bulp, and sometimes bigger than the tracking-square ;) ..just as info ).

i tried to go for the "cage" method (as seen in a tutorial by manymotion/greenlaw - thanks by the way - which is apparently also suggested by most users which are aktive here in the forum). i started the video with the plain background but set my ROI for the calculation from frame 424 to 3908 (since there, my marker is visible to all cameras).

my results after placing the groundpoints:

Distance cam01 to cam02: measured: 4m, computed: 3.96m (4cm off)

cam01: measured: 1.57m, computed: 1.57m (set the scene scale to match it)
cam02: measured: 2.41m, computed: 2.40m (1cm off)
cam03: measured: 1.74m, computed: 1.74m (baam!)
cam04: measured: 1.61m, computed: 1.64m (3cm off)
cam05: measured: 3.01m, computed: 3.04m (3cm off)
cam06: measured: 1.56m, computed: 1.61m (5cm off)

all in all i'm quite happy with the results (although 5cm off is quite a bit...), but i'm curious why it turned out to be quite accurate this time, since i actually didnt make anything different than the last times, but using more cameras.... any ideas?

thanks guys!

PS: heres an image of the result:

Nice work--it looks perfect!

I wish I had that much space for the stage in our garage. :)

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

thanks! yeah indeed, i'm quite happy with the amount of space we're able to use over here :). thanks again for the videos you put up online and the "documentations" you're wirting here in the forum. they're a big help and really useful!

perhaps you have any more hints on calibration, to get rid of those last 3-5cm which are off?

once, i managed to get a calibration which was a maximum of 2cm off for every camera, and i had the feeling that the tracking result was MUCH better. the 3d character overlapped our actor almost perfectly in every view. unfortunatly, for some reason i havent been able to reproduce this successful calibration once again, and i have absolutly no clue how i could improve the calibration results even further.

with this calibration (3-5cm off) a view cameras seem to be quite off, and the 3d character cant really match the actor like he should.

we're thinking about using ipidmc in production, so i'd guess, an almost perfect calibration is a MUST. i studied all the wikis and forum entries, yet, i'd be really thankful for any further hints of what i could do better to improve the calibration results.

I saw that image and thought, that looks TASTY!
Great job man!

Oli

thanks, oli! also for the effort in putting up some tutorials! highly appreciated! :)

unfortunatly, for me the automatic calibration (even after manual adjustments, turning off the "initial positions" and hitting the "calibrate" again) is sometimes for a few cams around 5-7cm off, and i'm sure that this reduces the tracking accuracy a lot. yet, i'm always getting only green dots in my calib videos.

so maybe someone has any hints on fine-tuning the cam positions? i'm kind of stuck here on this one.

thanks mates!

If you're using six cameras, I wouldn't worry too much about that--the data is going to get averaged out in the end anyway, especially after you apply Jitter Removal.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!