Depth Coordinates Mismatch

Typically: "How do I... ", "How can I... " questions
Post Reply
deepgrasp
Posts: 5
Joined: 27 Jun 2017, 13:29

Depth Coordinates Mismatch

Post by deepgrasp »

Hello,

I am trying to combine 2D and 3D depth information from a Vision Sensor but I noticed some strange mismatch between the collected coordinates. I give you an example of the data I collected and combined (the first number is just a line index, the 3 numbers after that are the x,y,z depth coordinates and the last one is the handle which I got from the 2D image):

9. -674.66896772385,-699.45961236954,-1499.9959468842,10
10. -434.63253974915,-452.66792178154,-970.74943780899,10
11. -432.64237046242,-452.66631245613,-970.74604034424,24

As you can see, these lines don't match because the depth information notices a "jump" from line 9 to 10 but the handle information gets the "jump" from line 10 to 11. I compared this behaviour with the 2D depth image of the Vision Sensor where the "jump" is also one line later than the depth coordinates suggest. This mismatch even propagates such that in the later lines there is a mismatch of 4 or 5 lines.
I'm collecting the depth coordinates by using the "Extract coordinates from work image"-Filter (after putting the original depth image into the work image). Is there maybe a way to look into this filter to understand how it is working and if this mismatch problem comes from the computation of the depth coordinates? Or is this some kind of bug?
I would really appreciate it if someone could have a look at this and give me a short reply on how to solve this problem.

Thanks a lot!

Best,
deepgrasp

coppelia
Site Admin
Posts: 10336
Joined: 14 Dec 2012, 00:25

Re: Depth Coordinates Mismatch

Post by coppelia »

Hello,

yes, the reason is because the coordinate extraction algorithm will evenly subdivide the angular space of a vision sensor that works in perspective projection mode, i.e. the extracted points will be evenly spaced around the vision sensor view arc.
On the other hand, the rgb image from the vision sensor has the pixels evenly spaced in front of the vision sensor.
We have modified the way the vision sensor coordinate extraction filter works, and now you can specify how your want the extracted points to be spaces (evenly in angular space, or evenly in euclidean space). This will be available in next release (i.e. V3.4.1). But you may contact us for a beta version.

Cheers

deepgrasp
Posts: 5
Joined: 27 Jun 2017, 13:29

Re: Depth Coordinates Mismatch

Post by deepgrasp »

Hello,

thanks for your quick reply. Is there any possibility for me to work around such that I still can use the current version? So e.g. by getting the depth image via the coordinate extraction algorithm and then finding the corresponding pixels in the rgb image? Or maybe somehow transform the angular space to euclidean space or vice versa?
If not, when will the next release approximately be published? If there is no other possibility, I will probably contact you for a beta version. Thanks for giving me this opportunity!

Best,
deepgrasp

coppelia
Site Admin
Posts: 10336
Joined: 14 Dec 2012, 00:25

Re: Depth Coordinates Mismatch

Post by coppelia »

I think the easiest would be to try the beta version. The next release will probably come out in 1-2 months of time or so.

Cheers

Post Reply