I have finally got my 4 cams to calibrate, but the video plane in the studio is laying back to be in line with the camera angle, is this correct? I hope what they say is true in version 3 this will be much easier to calibrate multiple cams. After I take a test record into the studio and try to line the actor with the video footage, following the tutorials by the word, analyze the actor and when I click refit actor, it just goes bonkers and twists into a pretzel figure. It is virtually impossible to get the actor to move to the same location as me in the video and align them. I have my 4 cams set at approx the same height off the floor at *8 ft* which shouldn't be a problem and did finally calibrate after several attempts and using the re calibrate method as described in the tutorial. My workable capture area is 8 ft x 8 ft in a 21 ft x 12 ft total area, cams set in all 4 corners, capture area in all 4 views is centered and looks fine (tape on floor for center spot and perimeter). It says align actors through 1 cam view as close as possible, and I have tried that, as well as aligning in more than 1 view and still pretzel man on refit. When trying to align actor the height shrinks as the actor is moved into position and there is no way to make him larger as the scaling doesn't go that high. Any clues as to what the issue could be? I will send calibration file to anyone that wants to see it and add their own capture video and test it, if that's possible. I will keep attempting it and changing clothing so refit better recognizes and wait for any responses ... Thanks!

Also, the image plane is halfway below the floor grid in the studio even though the calibration for the cams showed mark as floor to be right on, cornfusing to say the least, I am sure once I get it, like any program, it will fall in place, but geezalou, shouldn't be that hard :)

The most often reasons for the tracking to get completely wrong are:

• Bad calibration results. How do you evaluate that your results are good?
• Incorrect background in the video. For example, you have evaluated the background in recorder and then, before recording an action, switched the lighting.
• The colors of actor's clothes are very similar to the background.

Also, the height of 8 feet (looking down) for all cameras is not very good for tracking legs, which are less visible from such POV. In general, we recommend to set most of the cameras at about 4-5 feet looking horizontally (that is near the half-height of an actor with raised hands), and only few of them at extra-high (or low) positions for additional POVs, depending on the kinds of motions.

Your videos are welcome to help you with the process. Use some file sharing service to provide the files.
You may also post some screenshots of Mocap Studio with calibration/action projects to get quick feedback.

Ok, thanks for the reply...

I am trying a way to set up the half circle method, but looks difficult in a long narrow area I have to work in, possibly will be ok.

I set 3 cams at roughly 1.5 m each and 4th left at 8 ft. same location.

Area has consistently lit. no variance there that would affect it.

May have to move outside and make a 16 x 16 ft. PVC pipe frame and cover it with landscape fabric for normalizing the natural light, easily taken down when not needed and stored.

Let me try this cam config. and re calibrate them, if it doesn't work, I will stick to the 2 Kinects until V3 is released and hope the Sony eyes are more easily calibrated.


Thanks.

Ok...

Resetting cam config. worked to get actor to line up and analyze properly, but not the best set up for my particular area, in test for a 360 turn, arms were occluded and lost tracking as well as legs, so probably just going to use the dual Kinects, until I can get an area suitable for 4 Eyes and maybe by then version 3 will be out with better/easier set up for them.

I am going to try one more config. to minimize the occlusions, and thanks for the reply.

Its was much better for me with all cams high and I didn't see a viewing issue with the legs, it was just I couldn't get the actor to line up and refit to the video properly.

Seems if at least 1 video plane isn't pretty vertical, it can't align well, which is what I was realizing after attempting, but didn't hurt to try, maybe 3 high and 1 @ 1.5 m just for aligning actor will work, higher is a good view to limit occlusion, and no issues with stretched out arms at any point in video view for my area.

I do plan on using them more after I see what V3 brings for them.

Yes, calibration was fine, wasn't the problem, I just needed that 1 cam for the floor to appear better. It ran through the steps about the same with this set up.

It did make it easier to align the actor when I faced directly into one of the cams and stood at proper place on floor, trial and error thing.

Thanks

OH YEAH!!! That was the ticket for me!

3 cams at 2.438 m (8 ft.)

1 cam at 1.5 m (51 in.) for fitting the actor to video.

Tracking was very good on first pass even, full 360 degree turns all at 60 fps.

Sweet, might have to use these cam after all, perfect set up for my area this way, gives me a lot more movement area and full hands over head.

Foot and leg tracking was best of all, so that height doesn't impede legs or feet at all in my case.

Whole body tracked great, a few jitter passes and should be awesome... NICE!

Still working on it, but this way my cams can stay up and no constant re calibrating.. I hope.

Video link to subsequent test, refined only with iPi Studios tools. Small jitter from recording software, it's not present in scene on computer.

Custom untextured character imported to iPi for demo.

Running i5 3.70 GHz processor 8 G RAM and ATI Radeon 7700 GPU.

Post processing rendering was very reasonable rate, not complaining.

Once I got a handle on the process (4 cam set up for my area) all fell into place, can't wait for V3 release :)

YouTube link:

http://youtu.be/omVdXDePkE0

I do wish FULL 2 person capturing was available on Basic version, but I will work on work-arounds to possibly make it happen.

Two person setup sounds nice but it does have serious hardware requirements. I understand that sometime down the road for v3, iPi is hoping to support recording with multiple computers so maybe that will help.

In the meantime, I do my multi-character setups in separate passes. I find that if I have an audio track to perform to, especially with dialog or music and lyrics, it's fairly easy to synchronize actors. In the two cg 'Brudders' projects we have posted on Vimeo, that's how we handled our character interactions.

On a live action feature film I worked on last year, I needed to animate two soldiers interacting in the background. Soldier A was climbing out of a landing military helicopter and soldier B was waiting on the ground. Soldier A then opens the door of the helicopter, hops out and walks a few meters to shake the hand of soldier B, who also walks forward to meet Soldier A. In this case, I didn't bother with an audio track, I just counted off the time for when the handshake occurs in my head. Then, in Motion Builder, it was easy to slip the motions a few frames and reposition them to get the handshake in sync. The motion was captured in my living room early in the morning and I had it dropped into an animation scene at work by lunch time.

G.

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

Watch a one minute excerpt on Vimeo now!

Yes, this is what I was thinking might work, let me know if you have attempted it.

If I record the two actors in one recording, then only track video as a single actor (saving each persons tracking info.as its own file), at least the synch would be close and on import to Motion Builder I can move the characters to proper location in scene first if needed, but should appear imported at the files coordinance from recorded scene, not both at 0,0,0?

I do most of my animation for game engines and each file has to be loaded individually anyway, I haven't tested this theory yet, but testing a simple couple action today.
If it works, is an alternative to 2 actor synching actions and will test more complex actions.

Motion Builder can do a lot, but to get 2 separate takes of a couples dance to be synched close enough to be feasible would be difficult.(without more clean up than I wanna look forward to). Basic interactions would be no problem that way.

As of now, with V3 being released soon, I don't know what that will bring, I was really pleased with my very first saveable take to turn out so well once I got cam config. situated, but more testing of extreme positions and actions to find performance limitations, but do like the 60 FPS output.

This way 4 cams can perform close to 6 with very little occlusion in my particlular room area constraints (unless of course a body part is completely hidden from view) and less drag on computer. (My Theory)

I can see where IR of kinect would confuse this theory, but Eye cams dont have that limitation, the program can very clearly differentiate the 2 actors from one another.

I am fairly new to iPi use, but not to motion capture, most other programs of reasonable cost won't handle over 1 camera, so I am now digging into iPi.

I will test my theory and post results (if workable or not), while I wait on V3

I am in expired trial mode, so very limited to what I can try now and 2nd Kinect is a few days out for delivery for testing.

Thanks for the reply.