Dual Kinect Calibration Problems

mitchfx · **Joined:** Fri Jan 20, 2012 8:23 am **Posts:** 23

Hello everyone,

I'm using a dual Kinect setup with iPi Basic and running into some calibration problems. My capture space is approx 18x18' with one camera about 8ft off ground (tilted down) and the other about 4ft high (slight tilt up) in a 180 degree config. The calibration card is a 20x30" foam board, and I've followed all the advice here, on the wiki and by others such as Joe from TrueBones. However, my overlap between the two point clouds doesn't hold up very well as the actor moves across the capture space. After running the calibration on the card video, i usually end up with an avg angle error of approx 4 degrees and a position error of about 0.025. This is with around 500 good frames found.

A couple of interesting thing I've noticed might or might not be a clue as to what's wrong. First of all, I find that if I try to manually tweak the camera 1 a bit, it "pops" or snaps rotation when the initial manipulation starts. It's as if the roll or gimbal locks. This also happens if I try to navigate the Camera 1 viewport (without moving the actual camera), but ONLY the first time It's touched after calibration.

The other odd thing is that my cloud data from each camera seems to be off by a small temporal amount (time shift). Each 3d point cloud appears to be from very slightly different points in time (like 1/2 frame) which is most noticeable when the actor is moving fast. I know my Kinects were purchased about a year apart, so I'm wondering if any changes in manufacturing might be causing some kind of discrepancy.

Thanks,
Mitch

kahalany · **Joined:** Thu Mar 01, 2012 4:38 am **Posts:** 1

Hi, I got the exact same problem here. It seems there are both a temporal lag and a spatial error. No matter the setup I choose (30-90 or 180) or drivers (Primesense / MS), and even though I have enough capture space and a big cardboard I always get a 3-4 angle error and a 0.025 - 0.030 position error. The consolidated footage never match up to a satisfactory level. What happens is that both kinects seem to work AGAINST each other instead of complementing each other during the track, and so I end up with an unstable, inaccurate, jittery result.

Ironically, I find the single Kinect setup much more accurate and predictable. Although I have to manually compensate for occlusion, which is a real pain.

Help would be greatly appreciated!

All the best,

Royie

Greenlaw · **Posted:** Thu Mar 01, 2012 7:55 pm

I wonder if this is an issue with the latest update. The previous version worked well for me but I haven't tried this recent version yet. I have to record some motions tonight so I will let you know how it goes.

G.

GR1F1ThH · **Joined:** Fri Apr 17, 2009 2:45 pm **Posts:** 44

I've gotten the same degree of error in a 90degree setup. The extra limit in space and decrease in accuracy is not worth the extra effort of calibrating the dual Kinect system at the moment.

Greenlaw · **Posted:** Sun Mar 04, 2012 7:55 pm

I finally did a test with the latest update and I too am experiencing a lag with one camera. There are no dropped frames but the data from one Kinect is definitely a few frames out of sync. I can calibrate a scene successfully because I tend to 'hold' the calibration plane for a few beats when rotating it but this is obviously a showstopper when it comes to tracking.

This is probably a new bug because I don't think I ran into this issue back when we did 'Happy Box' last fall. I'll look at the problem more closely tonight and post more info about this problem in the Issues section. There's one possibility I need to check to be sure the problem isn't on my end but if that's not the cause, then this is probably a new bug.

Ugh. The timing for this problem is bad--we have to produce a test for a potential freelance project and this is going to hold us up. I may need to regress to older build; I hope I still have the installer from last September-October here.

Greenlaw

Greenlaw · **Posted:** Sun Mar 04, 2012 11:57 pm

I was putting together screen caps of the issue when I noticed that Studio just got another update (126). Did you guys fix this already? If so, thanks!

I haven't checked this closely yet but at a glance the results with this build seem to be better. Will post again soon after I've worked with it a little more and know for sure.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 12:07 am

No, I guess not. 126 was actually released a couple of weeks ago--I must have been running 125 or earlier.

In any case, the temporal offset is still there. I'll record another round of tests tonight to confirm if this is repeatable. If the error shows up again I'll post screen caps in the Issues section later tonight.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 1:36 am

Hi,

Could you guys take a look at my screen caps here?:

Build 126 (and earlier) time/space offset with dual Kinect?

Is this the offset problem you're seeing? I just want to be sure we're talking about the same thing.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 2:56 am

I'm cross-posting from the 'Issues' thread because I think this info will be useful to readers of this thread:

If you use the 90 degree configuration like I do, try angling the Kinects to less than 90 degrees and position them about 4 ft off the ground. Position the calibration plane to the middle of the 'green zone'. If you do this, the volume generated from the two Kinects should merge pretty closely.

Hope this helps other users. I posted more info over in the Issues thread linked above.

G.

Dual Kinect Calibration Problems

Who is online