Change font size
It is currently Tue Jul 28, 2026 1:13 am


Post a new topicPost a reply Page 2 of 4   [ 33 posts ]
Go to page Previous  1, 2, 3, 4  Next
Author Message
PostPosted: Sun Mar 06, 2011 11:48 am 

Joined: Sat Feb 12, 2011 12:38 pm
Posts: 85
Your video looks really good. I'm not sure quite how to handle a much larger capture area (more rectangles? more spirals?). I think it might be an issue with cameras in the round instead of just being on one side of the actor.

I did take my own advice and created a new video with a thresholded value so that the background was completely black, and the only thing showing was the LED. The data was exceptionally clean, and it still came out with a really crummy calibration. One of the cameras was 75m off the ground, and one was below it. At this point the only thing I can think of is that the pattern of moving the light I used isn't good.

I'll try to do as simpler pattern in my living room, but the issue with DMS not working well with camera roll might be an issue there too.

Back to the drawing board,

Michael


Top
 Profile  
 
PostPosted: Sun Mar 06, 2011 5:23 pm 

Joined: Mon Aug 03, 2009 1:34 pm
Posts: 2423
Location: Los Angeles
nofarb17 wrote:
I'm not sure quite how to handle a much larger capture area (more rectangles? more spirals?). I think it might be an issue with cameras in the round instead of just being on one side of the actor.

I would just make a bigger table shape, maybe with more legs, and a broader spiral or some random pattern inside. Keep is simple but cover a large enough area to make up for the greater distance between camera. Making a complex pattern won't neccessarily improve accuracy; more likely it will just cause DMC to take a lot more time to solve the calibration.

As for camera arrangement, if you only have four cameras, I recommend a semi-circle like the one I use. The relative closeness of the cameras helps DMC understand the relationship between the images. With six cameras you can widen the circle for greater coverage or keep them within a semi-circle for greater precision within that semi-circle. The advantage of a full circle is that there is less chance of self-occlusion and the advantage of a semi-circle is that you get data that is more focused and accurate when performing 'to the camera'. The formation you choose should depend on the type of action you want to capture. The size of your capture space may also help you decide; I chose a semi-circle formation because of my relatively small shoot space (a two-car garage.) With a semi-circle, I can get about 6 to 7 feet of useable performance area, which is good for many actions including short walks, jumps, martial arts and tumbles. I might even be able to get a small run cycle if I plan on looping the action in Animeeple or Motion Builder (not ideal but doable.) But if I used a full circle in my current shoot space, the performance area would be much, much smaller--too small to bother with in my opinion. If I had to shoot in a very small space (like a living room or single car garage,) I would seriously look to the Kinect Sensor for capture.

As a matter of fact, we ordered a Kinect this weekend specifically to use at home. We figure this would give us a quicker way to capture simpler actions without leaving the house. (We will still to transport our 'mocap computer' to our off-site stage for more complicated motions.)

Quote:
...but the issue with DMS not working well with camera roll might be an issue there too.

Roll really should not be an issue because the PS3 Eye cameras themselves do not physically roll, and I believe the software expects the cameras to be set up in a fairly straightforward manner (height and tilt only.) My suggestion is to keep your camera setup simple; just mount the cameras on vertical stands (tripods or c-stands) for height and use the camera's base for tilt to look up or down; do not 'dutch' the cameras. If you're positioning the cameras in unusual angles this may just confuse DMC because it may not be what's expected by the software or what the developers intended.

BTW, I modified the base of my PS3 Eye cameras using a Helicoil kit so that they can be mounted like a regular camera. I'll post a video for how this is done; the mod is easy to do and I highly recommend it for improved stability. Alternatively, in the past I've used plastic clamps and velcro to 'lock down' the cameras; not as stable as screw mounts but okay for short term setups.

Hope this info helps. Remember, keep things simple; don't make DMC work harder than it needs to and you'll get the superior results.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)
Image
Watch a one minute excerpt on Vimeo now!


Top
 Profile  
 
PostPosted: Mon Mar 07, 2011 3:46 pm 
iPi Soft

Joined: Tue Mar 31, 2009 9:13 pm
Posts: 805
It would help if you could send us the problematic video or at least screenshots. (We promise we will use it only for debugging).

Screenshots should be the easiest way for you. You can send screenshots to me by email.

The recommended way for sending the video: encode the video into an AVI file compressed with Xvid codec. This should result in a reasonably small video file size. You can use VirtualDub ( http://www.virtualdub.org ) to do the encoding. Xvid can be downloaded from http://www.xvid.org/Downloads.15.0.html . Both Xvid and VirtualDub are free. Other formats/codecs are less preferable. After encoding the video, please upload it to some file sharing server like http://www.filefactory.com and send me a link.

By the way, increasing the contrast of calibration video should not make any difference. Our calibration algorithm does not use any hard-coded threasholds. Instead, we calculate a correlation with 2D gaussian pattern in every frame. So if your maker is visible on video and is moving, it will be detected. I think the problem in your calibration video is in something else. My guess it is caused by incorrect initial camera position. Our algorithm for automatically calculating initial camera position is not very reliable in current version. So you may need to set initial camera positions manually, and this can be a bit tricky - it is easy to set wrong camera order (like moving the left camera to the right etc.) Another possible problem is not moving marker through sufficiently large portion of capture volume - this makes it difficult for software to analyze the scene.


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 4:37 pm 

Joined: Sat Feb 12, 2011 12:38 pm
Posts: 85
Michael Nikonov wrote:
It would help if you could send us the problematic video...



Hi Michael:

I'm using virtualdub to "fix" the camera order. I managed to make the video match the initial configuration in the scene tab fairly well (except I don't have cameras looking straight at the left and right arms in the T-pose). I'll try to see if I can compress the calibration video and post it for you to look at.

In addition, I did a calibration at an earlier time in the shoot, and I've been "fixing" the camera order for that clip today. I'll try to calibrate with that video too and post the result here.

I've got another shoot Wednesday, so I'll be trying somethings like varying the LED movement pattern and turning off one bank of lights (to see if that helps with reflections). Like I posted before, when I created a video with a thresholded LED, there seemed to be much fewer false positives.

Ironically, I did manage to get a good(ish?) calibration in my living room using Greenlaws LED pattern, so I've confirmed that I can see to get the software to work on my system, I'm just not sure what went wrong in the larger space - hopefully, you can see something in the calibration video.

Thanks, Michael 2


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 5:46 pm 

Joined: Sat Feb 12, 2011 12:38 pm
Posts: 85
Michael Nikonov wrote:
...The recommended way for sending the video: encode the video into an AVI file compressed with Xvid codec. This should result in a reasonably small video file size.


How much compression should I use with Xvid? The original file is 25GB and that compresses to 25MB with maximum Xvid compression. However, the video is really bad and shows lots of banding.

Michael 2


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 6:13 pm 

Joined: Sat Feb 12, 2011 12:38 pm
Posts: 85
Hi Michael:

OK, I'm uploading the calibration footage. It's 285MB, and it shows the LED well. Since I used a blue LED, the blue channel is the brightest. My new shoot will use a white LED.

I'm not sure how to tell you how to find the file on filefactory.com, but when it's done uploading, I'll post details about where to access it.

More soon,

Michael 2


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 6:34 pm 

Joined: Mon Aug 03, 2009 1:34 pm
Posts: 2423
Location: Los Angeles
Just curious but how large was your 'off site' shoot space?

The largest space I've tried to shoot in was at my work, which at the time had a fairly spacious 'cafe' area. FYI, I had a lot of problems tracking that shoot but the issues were not related to the size of the space, rather more to do with the polished floor, glass windows and shiny kitchen appliances, and other 'confusing' elements.

One of these days I'm going to ask permission to shoot at my workplace again. The new building containing our screening room has a large open area that's about the size of a warehouse and it might be perfect for running motions if I can light it adequately. Some day I might also try an outdoor shoot in my backyard.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)
Image
Watch a one minute excerpt on Vimeo now!


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 7:58 pm 

Joined: Sat Feb 12, 2011 12:38 pm
Posts: 85
Greenlaw wrote:
Just curious but how large was your 'off site' shoot space?

G.


The room is 21.5 x 24.5 feet with 10 foot ceilings.

Michael


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 8:29 pm 

Joined: Mon Aug 03, 2009 1:34 pm
Posts: 2423
Location: Los Angeles
nofarb17 wrote:
The room is 21.5 x 24.5 feet with 10 foot ceilings.
I think that's about the size of our shoot space, which is basically a two-car garage. We have our performance area in one corner with a 10 x 10 feet foam matte on the floor, although the useable performance space for mocap is a bit smaller; maybe about 7 x 7 feet at most. I have a diagram with measurements somewhere; I will post it when I find it.

FWIW, I don't think I could do a full circle camera setup in this space without cramping the performance area. The semi-circle works very well though.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)
Image
Watch a one minute excerpt on Vimeo now!


Top
 Profile  
 
PostPosted: Tue Mar 08, 2011 8:45 pm 

Joined: Mon Aug 03, 2009 1:34 pm
Posts: 2423
Location: Los Angeles
Found it:

PooperMocapTechNotes

This document was created as part of our script notes for a four camera shoot last November. It doesn't have all the measurements of the space but it does explain quite a bit about our setup.

Hope it helps

G.

Edit: I should have mentioned that I'm not able to fully maximize the distance of the cameras in this space because there is about 3 to 4 feet lost behind both 'walls' of the stage for shelf and a narrow walk space. If it was possible to push the stage back to the actual wall, I'm sure I could get the full 10 x 10 area for performance, maybe slightly larger even.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)
Image
Watch a one minute excerpt on Vimeo now!


Top
 Profile  
 
Display posts from previous:  Sort by  
Post a new topicPost a reply Page 2 of 4   [ 33 posts ]
Go to page Previous  1, 2, 3, 4  Next


Who is online

Users browsing this forum: No registered users and 26 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  


Powered by phpBB® Forum Software © phpBB Group
610nm Style by Daniel St. Jules of Gamexe.net