Hi all,

Finally I've been able to test this software !
For the moment I have only 2 cams tough, so I didn't expect too much.

And so... I'm stuck at calibrating stage.

First, the setup. In my garage, one cam is on front and the other on the side, not at a 45° angle but more like 35-40 (space restrictions). Don't know if its a good setup.

For the moment I use an old HDV cam and my beloved HMC 150. I can synchronyse the timecodes, so no problem. And I can shoot at 50 fps.

Once I shooted, I downconverted the videos and "stitched" them.

When I load it into ipi Desktop, first time I calibrated it some tracks were red. As I've read in the forum, there's a random seed applied to the tracking algorithm, I think... ? So I closed and tried again.

Yay, finally a clean solve. So I save the scene, ready to do some mocap... But there I see that the calibration doesn't really match. I can't align the actor and the 3d model. I tried to adjust it manually, but after that the auto-analyzing to fit the actor acted weird, and I had the legs floating in front of the actor. No good...

So my question is ; should I just wait for my 4 PS Eye to arrive, or can I still test something with only 2 cams ? Does the software need a 3rd cam to really estimate correctly the 3d space ? I don't expect too much, but if I could make it work with 2 cams, I think it would save me some time later.

Maybe there's a better placement for them ? Could lens distortion be an issue ?

Hi,

I'm not familiar with these cameras, but since they are different it may be difficult for iPi DMC to match them.
iPi Soft recommends that all the cameras be the same models. Also, you need to be sure you have the correct FOV data entered for the cameras.

I don't have any experience with two camera setups, just three and four, but my guess is that they need to be at least 45 degrees apart, not less. 75 - 90 degrees is probably closer to ideal for capturing depth.

Hope this helps

Greenlaw

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

Watch a one minute excerpt on Vimeo now!

Earlack, can you upload or send some screenshots to illustrate your question?

I used the calibration tools to find the FOV, seems correct on that point. I use those cams just for testing purposes, in a few days I'll switch to 4 PS eye.



I think that may be the main problem. The first camera seems correctly estimated, the second is close but not exactly correct. I guess if it were at 90° it would be easier.




I just spotted those red dots floating everywhere... At one moment during the calibration stage, my marker turns off, thats why. Maybe it's the problem ?

I know the HMC150 has quite a bit of distortion, should I undistort the footage first ? And I didn't White balanced the cams, their WB are completely different, is that a problem too ?

Such amount of red dots is acceptable. I most cases this is due to marker incorrect detection (some other point on image is detected as a marker). Thus this is not problem.



It depends on the value of distortion. The results of calibration looks ok. Likely this is not problem too.



Definitely not.



Likely calibration is ok.



Can you provide some screenshots?
Or (much better) send to us your video and you projects (nondisclosure guaranteed).

Thanks for your help.

The videos are freakin' huge, cause it's almost uncompressed. I'm re-encoding them and I'll upload them when it's done.

Here are some wmv's of the videos.

First, only a T-pose to roughly align the cams :

http://www.megaupload.com/?d=USZNA6JJ

Then the calibration process and the mocap itself (I know I do lots of weird stuff)

http://www.megaupload.com/?d=Q7NEVY9U

It's a rough test ; I plan to use a cleaner background with more diffuse lighting later.

Earlack, thank you for uploading, but can you convert it to the format, that can be open in iPiStudio?

The following comments are based on viewing the video, but not on its processing:

1) You move marker too fast. Try to move it much slower!

2) You didn't mark ground, did you? If you want to track feet, you should put a marker on the floor at several points, the most distant from each other. At least three round points should be (not in one line!). Then use these points to mark ground in iPiStudio after calibration.

Your screenshot from previous post shows that you have marked some of the points as the ground, but the video does not show when you put the marker to the floor.

3) In part of frames marker is not visible in one of the cameras. Likely, the red dots are just these frames.

4) Note that there is no need to turn off light during calibration video recorder. Marker detector in iPi Studio is rather clever to distinguish marker from background and actor.

5) As to performance video, it is very difficult lighting conditions: too dark, many shadows, many differences in brightness. Dark shadows on the floor... :( Is it possible to provide more scattered light?

Ah, I converted to wmv which is my less favorite format, but didn't checked if it worked. Would you recommand one ? Cause uncompressed Avi is fine to work for me, but uploading 15 GB might be long :)

Thanks for the input.



Oops, wasn't sure about that. I'll go slowly next time. BTW, are there some "moves" or figures recommended ?



Yeah, I've read it only AFTER shooting the video... I tried to mark anyway, hoping I could play with the floor height but of course, it doesn't work. I'll reshoot a proper one.



Yeah, at one moment my marker switches off because it's only 2 batteries taped on a light bulb... Couldn't seem to find a sorta "bulb-candle" here.



I know there's no need for that. My first try was on the opposite side of the garage, and some little spots of light coming through the doors made the calibration go crazy. So I shot on the other side, but in the dark to be on the safe side. I'll retry with lights on.
I guess if I had a more flatly-lit background, it would help too...



I thought it would be the real problem. Well that was a test anyway. In this environment, it will be hard, the ceiling is really low, I can bring some spots to give more light and lessen the shadows, but the main garage light is like right above my head so it will be probably hard... Might have to test somewhere else.

Outdoor on a cloudy day should give me a good, diffused light, but what about calibration ? Can Ipi Desktop pick up the marker even if it isn't super-bright ?
For the moment the weather is really changing fast here... morning it's cloudy and it rains every 30 minutes or so, and then when it stops its a superbright sunshine making hard shadows everywhere...

I'll give it a shot tomorrow.

When my PS Eyes will arrive, I'm counting on the wider FOV to be able to shoot indoor, in another room.

I see. We'll try to convert ourself. But it can take some additional time.




This is the root problem!




You could obtain acceptable results this environment if you'll:

• pay attention that first two seconds of performance video don't contain actor (even part of actor) - this is important for correct background calculation
• add light
• use black T-shirt and red sleeves
• play simpler movements or use 4 cameras

Yes. It is enough that marker is distinguishable from background.




You can repeat shot in the previous environment but pay attention to ground marking, lighting, absence of shadows on actor body (black T-shirt should help!), abs. free background during first several seconds of performance video.

We hope that you make friends with this technology:)