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

Greenlaw · **Posted:** Mon Mar 05, 2012 1:32 am

Hi Michael and Andrew,

I'm not 100% certain about what I'm looking at but I felt these results should be brought to your attention. Below is an image of screen caps from a calibration project created using build 126. In the first three images you can see that there appears to be an offset between the calibration planes from camera #1 and camera #2. I'm not sure if this is a temporal or spacial offset. I believe 125 gave me the same results but I'm uncertain about builds just prior to that one.

In the fourth image, I opened a calibration project that was created with 121 (October 2011) in build 126. When the plane is viewed from a similar angle, the two views appears to be merged better. I'm not completely sure this is the case but it does look that way to me. FWIW, when I recalibrate this scene in 126, I do not get as good a result as I did with 121--meaning, with 126 I get many yellow and red tracker dots and with 121 I got nearly all green tracker dots.

At least, all green is what I remember from using 121. I suppose 126 could be reporting the same results just more accurately. I think need to regress to 121 to confirm. Would it be possible to post the Installers from last October (121 and 122) to help me confirm these results? (I may have these installers archived here but I haven't checked yet.)

Hmm. It just occurred to me that this could be a Recorder issue rather than a Studio issue, but I don't know how to check that.

Anyway, if these results look normal to you, please let me know. I haven't run DMC since last October (when we shot Happy Box) so I'm not completely sure if what I'm looking at is 'normal'.

I'll complete my testing tonight and see if the final results are what they should be. It's possible that what I'm seeing is okay but I'll know for sure soon.

Thanks in advance for your assistance guys!

Greenlaw

Greenlaw · **Posted:** Mon Mar 05, 2012 1:51 am

Just a thought: the main difference between my shoot from last October and tonight (besides build versions,) is the position angle of my Kinects. The Kinects from tonight's shoot are spaced about 10 or more feet apart and set to about 90 degrees. In the shoot from last October the Kinects were placed closer together and the angle was much narrower--possibly 60 degrees.

I'll try a narrower angle and see if I get similar results with build 126 to what I got with build 121 from October.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 1:55 am

Oh, I just remembered: the heights are very different too. The Kinects tonight are positioned at about 5 ft high (just below eye level to me,) and the Kinects last October were only about 3 ft off the floor.

If I don't get expected results, I'll try lowering the devices tonight in a separate test.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 2:25 am

Okay, I think I understand the problem now. I moved the two Kinects closer to where I positioned them last October (though I left them at 5 ft high,) and now I'm now getting calibration results in 126 that may be similar to what I got when I used build 121.

I'm pretty sure the two Kinects were not angled over 90 degrees but it was close to 90. I'm wondering what will happen if I move the devices apart again and switched Studio to calibrate the data in 100 - 180 degrees mode. I don't think I have time to test this tonight but may get to it tomorrow evening.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 2:53 am

Another important piece of the puzzle: I lowered the Kinects to about 4 ft high, shot a new calibration video, and I think the calibration results are even better now--much closer to what I got last October.

So, maybe this is not a problem?

To recap:

If you use the 90 degree configuration like I do, try angling the Kinects to less than 90 degrees and 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.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 3:00 am

BTW, I think there is still a slight offset occurring but I can't tell if it's a temporal or spacial thing. FWIW, I now think this slight offset probably existed in 121 but I just didn't notice it back then. Maybe it doesn't matter? I'll know for sure after I run the tracking tests and see the results.

Looking forward to insight from the iPi team.

G.

Greenlaw · **Posted:** Mon Mar 05, 2012 3:15 am

On closer examination, I think the offset happens more in the blue zone and less in the green zone, so this may be a distance issue. That seems to make sense--I would assume that capture accuracy decreases with distance.

I probably didn't notice this offset when we shot Happy Box because the characters didn't need to walk around in that film.

Okay, it's getting late (2:18 am) so I'm calling it a night. I think feel a little better about build 126--that is, I don't think it's behaving much differently from what I was using last October (build 121,) but I still hope hear some insight from Team iPi Soft about this.

Thanks in advance guys for any helpful info. :)

G.

andrew · **Joined:** Wed Apr 07, 2010 4:33 am **Posts:** 468

Greenlaw wrote:

I would assume that capture accuracy decreases with distance.

You're absolutely correct!

Greenlaw · **Posted:** Mon Mar 05, 2012 3:40 am

Yes, I'm still awake. I just had to try the 100 - 180 degree mode with the full 90-degree video. Well that certainly didn't work--but I guess it should have been obvious why it wouldn't. I think I've been up way too long today. :p

Good night all.

G.

Greenlaw · **Posted:** Tue Mar 06, 2012 1:24 pm

Ran first pass tracking and it seems to be okay--very few errors and the clips are actually quite long. This was done using original capture session with the widely set Kinects (about full 90 degrees) and the apparent volume mismatch reported above.

Maybe this is a non-issue? I'll know better after I run Jitter Removal but so far it looks pretty good.

I'm expecting the later sessions to track more accurately. Should be interesting to compare results.

G.

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

Who is online