I did a 6 camera PS3 calibration, and I notice that from

Camera 1 - the two cameras it can see (4 & 5) are shifted about 3 feet to the right and maybe 6-8 inches up.

Camera 2 - both cameras it sees (4 & 5) are perfect.

Camera 3 - one camera is perfect (#6) , one (#5) is maybe 6 inches to the right.

camera 4 - one (#2) is perfect, one (#1) is 2 feet to the left,

camera 5 - two (#2, #3) are perfect, one (#1) is 2 feet to the left.

Camera 6 - the one camera (#3) it sees is perfect.

When I use the calibration data to track, it does seem to mostly track ok, but I had to input the actor's height at 2.48 (the max).

I guess I'm not sure exactly what to expect as there is some variance between the actor and the skeleton.

I would guess it would track better if all the camera's positions were correct?

BTW - would be nice if we could easily upload photos on this board? I have to use something like tinyPic - and I will only use that site with a box outside of my network as I don't like turning on javascript for a site like that - so it's a real pain to upload images.

Just curious but have you tried setting up an arrangement that's less than a circle? Like 3/4 or half-circle even?

I still haven't gotten around to using six cameras but when I do, the two new PS3 cameras will have to fit within my current 'slightly wider than semi-circle' setup (because of limited space.) I'm already getting fairly good tracking results even with fairly complicated motions like walking and jogging in a complete circle, but I think the extra cameras will help reduce the excessive jiggles I'm seeing in the body. Might be a couple of weeks before I can try this. (Spending most of my time building environments and props now.)

G.

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

Watch a one minute excerpt on Vimeo now!

No, I haven't tried a 3/4 or half-circle. I'm trying to capture dance moves which pretty much require a full view.

When you have finished your calibration, are all the cameras lining up exactly?

I'll have to review my previous ones, as I wasn't looking to see whether the cameras lined up before, but I think there may be some variability in calibrations for some reason, as I know I had to do one calibration over again previously. Maybe I need to do much wider loops with the flashlight or something?

BTW - what is the large yellow ball for that iPi seems to generate? Does it think that is a light source?

I can understand why you wish to keep the full circle. That said, I'm getting pretty good results with a slightly wider than 1/2; basically, the cameras are seeing enough around me to figure out what's going on. Also, I'm generally performing to the 'front', so I try to keep important motions visible to the 'audience.'

I am curious, however, to learn what the actual trade off is between full and semi-circle. With a full circle, you get full coverage of course. But with a semi-circle you may get more precision since the cameras are closer together. Will have to test that one of these days.


I'm not exactly sure what you mean by 'lining up'. If you mean, do my virtual cameras correspond to the positions and angles of the real camera, then yes. Not all are exactly to the cm, but they're pretty close. (Remember I'm still using four cameras though, and haven't tested six yet.)


I should post one of my calibration videos as I've been consistently getting excellent iPi DMC calibration ratings everytime. My pattern is pretty simple: I draw a box starting with the farthest corner (from the 'front',) touching the ground for about a second for each bottom corner, and then I draw a spiral from the top of the 'box' down to a point at the bottom. This gives me five ground points. When I draw the box, I make very certain that the entire box is visible to all the cameras all the time. This is another advantage of doing a smaller the full circle setup--it means you don't have to do any crazy contortions to keep from occluding the maglite.

But as I've said, I'm still very curious to learn if a full circle will give me better results.

Yes, that's the program's light source. You should set this up under the Scene tab. I don't think it affects calibration, but if you set it up to match the lighting in your scene, it may help the tracker better understand what it's looking at. In my case, I'm lighting the performance almost equally from both sides and front so I position the light up and to the front.

Hope this helps.

G.

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

Watch a one minute excerpt on Vimeo now!

Yes, I noticed that I can change the selected camera, and see these 'virtual' cameras mostly in the correct positions.

I just had a tracking session lose an arm at a place where nothing was really changing, not even in a difficult position, so I bet it would be best to have them all line up. I could not fix this problem, no matter what I did with IK, backwards tracking, etc. Frustrating as I've seen iPi deal with arms behind the back, crossed arms, all sorts of stuff - strange to lose an arm that seems less difficult to track.

I'll try your excellent box suggestion for the calibrations.

BTW - looks like they are doing a 'gold' release tonight of LW, so hopefully all the FBX issue have been resolved?

I hope they keep up the furious bug fixes after the 'gold' release. That's what they should have been doing all along.

I tried drawing a giant box, in a very long take, but the calibration was totally messed up.

I tried to calibrate again, this time turning off my soft boxes after I assumed a t-pose to initially align the camera (as I noticed that tracking sometimes got lost due to the light intersecting with a softbox).

I used large arcs this time, and every single virtual camera in the scene lined up perfectly with the real cameras.

However, the ground (and the virtual character), is tilted quite a bit in a few cameras.

I'll try to record tomorrow, and see how the calibration works.

Maybe I need to expand the area (and use a large box) for the ground plane instead of dragging the lite around in a circle with a few intersecting lines in the middle.

I don't remember that clicking on ground points changed the angle of the ground plane in the calibration view, at any rate.

Also, my character is a bit small in the final calibration window, and to make him near the actor's size, I'd have to set the height to the max. I'm unsure if it's a good idea to change the actor's size during calibration - or if that has any effect at all?

I don't actually draw a 'giant' box. It's usually about five by five feet or smaller. This is not so big that I risk going off camera, and it's big enough for me to get a good calibration.

I'm surprised to hear it's not working well for you; for me calibration has become reliable enough that I don't really think much about it anymore. It sounds like something else is really throwing off your calibration. Can you post one of your calibration videos?

G.

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

Watch a one minute excerpt on Vimeo now!

Greenlaw - my calibration.avi is 3.17 GB. I use the highest quality settings as my RAID is fast, so ... probably too large to post.

I'll try a smaller box.

It could be that I went outside the visible area, or it could be a few reflections in one window that I hadn't blocked off.

At any rate, the large arcs seemed to work and get the cameras in the correct position. I just have to worry about the ground planes now.

How do your ground planes look? Are they all at the proper angle?

Maybe iPi can post another calibration tutorial at some point, as there are many things left out of the original tutorials - and go into detail about how to determine when you have a proper calibration - and what might caused issues (like the ground planes at the wrong angle, or the cameras not matching the virtual cameras).

I hope I can get Andrew interested in my Reindeer lighted nose idea, as head tracking would be nice.

I spoke with a guy today who owns one of the very expensive 8 camera motion capture systems (with all the bells and whistles), and he was actually interested in looking for another solution as it took so long to set it up and capture.

When he heard from a friend that I was experimenting with motion capture, he called to see if I could maybe off-load some of this for him, but I just pointed him towards the ipi site.

For pre-vis and other uses, iPi could be sold to the folks who have these high-end systems, as it's so easy to use.

But, that head tracking would really make a lot of difference

My ground plane is good. When laying down your ground points, it's very important not to have them in a line, as this makes it difficult to calculate a plane. You really only need to define three points in a triangle to get a good ground plane. More points may help but it really depends on where they are placed. The five ground points in my usual tracking pattern seems to work well for me. I'll try to post a video to illustrate when I have time.


Yes. I demoed iPi Studio for my boss and coworkers last summer. I did a quick proof-of-concept for a previs we were working on. We wound up going in a different direction so the test was never actually used, but as a proof, the was successful and stirred some serious interest in the system, especially since it was so quick and easy for me to knock it out. Normally, one of us would just block out a 'quick and dirty' keyframe animation for previs, but this looked much better than the usual blocking; for a client presentation, previs done this way could make a very positive impression.

As for the head tracking, yeah. Previously, I wasn't terribly concerned by this because the head is fairly easy to animate. But now that I'm working on a musical piece I'm find that I'm having to keyframe quite a lot of head bobbing which the iPi system failed to capture, so I changed my mind...now I really want to see head tracking! :)

G.

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

Watch a one minute excerpt on Vimeo now!

I suppose you could draw a big circle on the ground to get a plane. I don't know if this is better or worse than what I'm doing (very few points as opposed to many,) but it will be a lot more data so it may be more accurate. Just thinking.

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

Watch a one minute excerpt on Vimeo now!