Change font size
It is currently Wed Jul 01, 2026 1:50 am


Post a new topicPost a reply Page 1 of 2   [ 11 posts ]
Go to page 1, 2  Next
Author Message
 Post subject: Poor recording results
PostPosted: Tue Jun 11, 2013 10:46 am 

Joined: Tue Jun 11, 2013 10:11 am
Posts: 5
Hello,

I'm using iPi Soft in combination with two Kinects and two Playstation Move Controllers. I had the possibility to use them in a big hall (since there are no obstacles in the recording area I was hoping to get satisfying results). But all recordings are bad, the tracking is lost after a few seconds. Also, the depthinformations from both camera differ from each other (the character 'from depth informations' is much fatter than I am ^^).

Both kinects are positioned in an angle of 180° Degrees to one another.

I guess the problem is, that the kinects cannot recognize the floor etc., since nearly everything in the scene (except of the character) is yellow:

Image


Here are two pictures of the kinects, in the first picture you can see the plate that was used for calibration (by the way, while the plate was laying on the floor, both kinects hat no difficulities to recognize it perfectly):

Image

Image

What was going wrong? Any why was the scene predominantly yellow? It was a normal wooden floor, not glossy or polished.

Or is the position of both kinects the problem? When both kinects were positioned in an angle of 90° Degreed to one another, the result was nearly the same...

I'm appreciative for any help :)


Top
 Profile  
 
PostPosted: Wed Jun 12, 2013 1:01 am 
iPi Soft

Joined: Wed Jan 11, 2012 6:12 am
Posts: 2355
Location: Moscow, Russia
red51 wrote:
But all recordings are bad, the tracking is lost after a few seconds.

It is normal for tracking to lose a single limb, for example, when it's not very distinct or is obscured. But it should usually restore after some manual corrections and Refit Pose.
Can you provide us a couple of your recordings to see the problems more clearly?

red51 wrote:
Also, the depthinformations from both camera differ from each other.

How they differ?

red51 wrote:
I guess the problem is, that the kinects cannot recognize the floor etc., since nearly everything in the scene (except of the character) is yellow:
Here are two pictures of the kinects, in the first picture you can see the plate that was used for calibration (by the way, while the plate was laying on the floor, both kinects hat no difficulities to recognize it perfectly):

Yes, there is definitely something bad for IR with the floor material. From the pictures, it looks like polished with some matte material, or just with shoes :)

red51 wrote:
Or is the position of both kinects the problem? When both kinects were positioned in an angle of 90° Degreed to one another, the result was nearly the same...

In 180 degrees setup there may be the problem with IR interference, when depth picture is filled with bad yellow noise all over. This is usually solved by slightly changing the orientation of one or both Kinects.


Top
 Profile  
 
PostPosted: Wed Jun 12, 2013 9:47 am 

Joined: Tue Jun 11, 2013 10:11 am
Posts: 5
Thanks for your reply! :)

Quote:
It is normal for tracking to lose a single limb, for example, when it's not very distinct or is obscured. But it should usually restore after some manual corrections and Refit Pose.
Can you provide us a couple of your recordings to see the problems more clearly?

I've uploaded two videos of the recordings, see below. Unforunatlely, the .iPiVideo-data is too big.
The first video (captured with my cellphone) might be the result of an insufficient calibration, since all further results were much better than that.
But the second video should be more interesting, this is one of the better results we got. At the beginning, it seemed to be promising, but then (beginning at 0:26) the tracking of the arms get lost.
Is that normal and fixable with manual correction?

Video1 link: http://raiders-reallife.net/ipi/video1.avi
Video2 link: http://raiders-reallife.net/ipi/video2.avi

Quote:
How they differ?

You can see it on this screenshot, the character from depth-informations is much fatter than I am (or look at this giant head or feets) ^^

Image

Quote:
Yes, there is definitely something bad for IR with the floor material. From the pictures, it looks like polished with some matte material, or just with shoes :)

That's it I think ;) So the yellow is indeed not good for recording?

Quote:
In 180 degrees setup there may be the problem with IR interference, when depth picture is filled with bad yellow noise all over. This is usually solved by slightly changing the orientation of one or both Kinects.

We also tried to change the orientation of one kinect a little bit (as you can see in Video 2 I've uploaded), but with no success =/


Top
 Profile  
 
PostPosted: Wed Jun 12, 2013 12:29 pm 
iPi Soft

Joined: Wed Jan 11, 2012 6:12 am
Posts: 2355
Location: Moscow, Russia
red51 wrote:
I've uploaded two videos of the recordings, see below. Unforunatlely, the .iPiVideo-data is too big.

But without source files it is harder to get you exact guidance for overcoming the problems. You can use some file-sharing service to upload the videos, or follow this instruction for direct peer-to-peer sharing using uTorrent.

red51 wrote:
At the beginning, it seemed to be promising, but then (beginning at 0:26) the tracking of the arms get lost.
Is that normal and fixable with manual correction?

Yes, this should be easily fixed. When you first notice such error - pause the tracking, manually correct the pose with Rotate and IK tools, hit Refit Pose, track backwards to correct missed frames, and continue tracking forward from the last good frame.

red51 wrote:
You can see it on this screenshot, the character from depth-informations is much fatter than I am (or look at this giant head or feets)

This is the consequence of bad calibration: the distance between two Kinects was not determined correctly. This may be caused by not sufficient of calibration board or the yellow floor. It would be good to have your calibration video as well.

red51 wrote:
That's it I think ;) So the yellow is indeed not good for recording?

Yellow is bad on the actor and on the floor. Other areas don't matter much.


Top
 Profile  
 
PostPosted: Wed Jun 12, 2013 3:00 pm 

Joined: Sat Nov 12, 2011 12:40 am
Posts: 16
hi red51,
1. move your "Kinect" to the left (from photo)/area that's a bit dark.
2. position your body against the wall / somewhat closer to the wall.
3. position your "Kinect", one in front of you, and the other slightly behind you (avoid interference).
4. try to covering the floor / area you are working on with something.
:)


Top
 Profile  
 
PostPosted: Thu Jun 13, 2013 3:59 am 

Joined: Tue Jun 11, 2013 10:11 am
Posts: 5
dyhvn2 wrote:
hi red51,
1. move your "Kinect" to the left (from photo)/area that's a bit dark.
2. position your body against the wall / somewhat closer to the wall.
3. position your "Kinect", one in front of you, and the other slightly behind you (avoid interference).
4. try to covering the floor / area you are working on with something.
:)

Hi dyhvn2,

thanks for the hints! I'll give it a try when I have the next time the opportunity to use the hall.

vmaslov wrote:
But without source files it is harder to get you exact guidance for overcoming the problems[...]

Alright, I've uploaded the records that were shown in the second video I posted before, furthermore I've uploaded the calibration video. You find the files here: http://raiders-reallife.net/ipirecord/
Please tell me if something is missing or not working.

vmaslov wrote:
Yes, this should be easily fixed. When you first notice such error - pause the tracking, manually correct the pose with Rotate and IK tools, hit Refit Pose, track backwards to correct missed frames, and continue tracking forward from the last good frame.

I've never done it on that way before, thank you, I'll try it out! :)

vmaslov wrote:
This is the consequence of bad calibration: the distance between two Kinects was not determined correctly. This may be caused by not sufficient of calibration board or the yellow floor. It would be good to have your calibration video as well.

Perhaps you can tell me - after checking the uploadad records - if it might be caused by insufficient calibration?


Top
 Profile  
 
PostPosted: Thu Jun 13, 2013 5:20 am 
iPi Soft

Joined: Wed Jan 11, 2012 6:12 am
Posts: 2355
Location: Moscow, Russia
Alright, I've downloaded the files - 2 videos, 1 project and 1 scene. I'll check them when be back in office in a few days.


Top
 Profile  
 
PostPosted: Mon Jun 17, 2013 2:58 pm 
iPi Soft

Joined: Wed Jan 11, 2012 6:12 am
Posts: 2355
Location: Moscow, Russia
Finally got to your files.

Well, I must say the movements of the calibration board in the video are totally wrong. You give too much rotation instead of translation, and this makes the rectangle shape completely unrecognizeable. Please see a sample motion pattern in this video:
http://vimeo.com/bassill/sample-motion-pattern-for-calibration-board

I've also worked with your action video to check whether it can be cleaned up. And, in spite of bad calibration, I was able to get some "satisfactory" results. I touched only arms, because the errors with them were pretty obvious. Mostly I used tracking backward. Several times it was required to edit the arm pose with IK, and even with Rotate tools (then Refit Pose, and track forward or backward). The project file is attached.

I recommend you first to get familiar with calibration and basic tracking and cleanup techniques. And then add more complexity with PS Moves.


Attachments:
File comment: Cleaned up action
2013-05-28_20-47-00-cleaned-up.zip [786.49 KiB]
Downloaded 639 times
Top
 Profile  
 
PostPosted: Tue Jun 18, 2013 7:58 am 

Joined: Tue Jun 11, 2013 10:11 am
Posts: 5
Thank you for having a look at the files! :)

I assumed that the calibration was wrong. The motion pattern you've posted is a great help, indeed the calibration procedure in the video is very different from the way I did it.

I'm also suprised that you was able to get a satisfying result from the action video, I must admit that I was not very familiar with cleanup techniques.

Anyway, many thanks for your help and your advise, I'm now very confident to get much better results the next time! :)


Top
 Profile  
 
PostPosted: Thu Sep 05, 2013 11:31 pm 
iPi Soft

Joined: Thu Apr 09, 2009 6:44 am
Posts: 203
The Walls are beyond the kinects range..
Thats why the color is solid yellow.
Can I suggest one kinect lower (front)
The other higher for the back

_________________
iPi Soft


Top
 Profile  
 
Display posts from previous:  Sort by  
Post a new topicPost a reply Page 1 of 2   [ 11 posts ]
Go to page 1, 2  Next


Who is online

Users browsing this forum: No registered users and 49 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