jit.openni: Remove background wall from the usermap generator shape

Oct 22, 2013 at 12:34pm

jit.openni: Remove background wall from the usermap generator shape

Hi folks and hi Dale :)

I am stuck with another problem with jit.openni. Whenever I get too close to a background wall, my usermap shape starts getting pixels from the wall behind me (see attached pic)

Are there any possibilities I can remove that background wall? Is there a function I should know about so that I may do this?

Thank you for looking into this!!
ygreq

Attachments:
  1. Screen-Shot-2013-10-22-at-10.17.13-PM
  2. Screen-Shot-2013-10-22-at-10.16.43-PM
#268782
Oct 22, 2013 at 2:26pm

The only possible solution that comes to mind is trying to mask the usermap using the depthmap, cutting at the wall’s distance.
this should work:
<code>

– Pasted Max Patch, click to expand. –

</code>

#268796
Oct 24, 2013 at 4:04am

Wow! It works great.
Thank you so much for your help. Your patch does wonders!

The only problem that remains is the fact that I wanted to put the kinect a bit higher than the position where it would create a perfect perpendicular in relation to the wall.

I mean if I put the kinect straight in the middle of the background wall (in relation to up and down, left and right), I can cut it out of the mask easily with your patch as it is perpendicular to the wall. But if I move it a bit higher and the perspective changes, I think I need to calculate something else and I don’t know exactly what. Or is it even possible? Any pointers?

Once again, thank you so much for your help!
ygreq

#268983
Oct 24, 2013 at 4:46pm

OpenNI user recognition…well…its not as good as the Kinect SDK. If you are on windows, I recommend you switch to dp.kinect. The recognition is better *and* you can turn on automatic realtime reorientation of all the data based on gravity; its a cool feature I added several months back.

If stuck on Mac, the usermap is poor. Now much can be done about that. The depthmap you could also rotate pixel by pixel using rotation matricies, trig formulas, etc. See wikipedia for all the fun.

#269055
Oct 31, 2013 at 10:23am

“rotation matricies, trig formulas, etc.” That’s what I figured. ;))

Yeap, I know you made a better object for kinect, but, unfortunately, I am stuck on Mac. Still waiting for a OSX version, though.:D

I already created another post trying to figure out another thing related to data coming out of jit.openni (http://cycling74.com/forums/topic/filter-repetition-of-numbers-from-a-matrix/). But I am asking here as well just in case.

Any idea how I can get a list of all the users present? I don’t want to use the data coming from [skeletons, floor info output], but from the [usermap generator] output. I know I get a matrix with different numbers relating to the users being detected. What I wanted was a simple list, not the matrix. Any ideas how I can get that out of the object or how may I process the matrix? (I have a few ideas but I pretty much stuck – please check the link above).

Thank you all! And a big thank you to LSKA and Dale for the much needed info!
ygreq

#269767
Nov 1, 2013 at 11:11am

Using the usermap matrix to get users is costly. Scanning through 307200 integers is highly discouraged. Instead, use the data coming from the /user or /skel messages. The data they present will match the data in the usermap. Check this thread for suggestions http://cycling74.com/forums/topic/jit-openni-user-id-problems/

I doubt you’ll see Microsoft supporting Kinect on Osx anytime soon. They’re focused on Windows and Xbox. However, if they do, then I can port dp.kinect like I did with jit.openni.

#269857

You must be logged in to reply to this topic.