Re: [Ros-kinect] Interesting systematic noise in calculated

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: [Ros-kinect] Interesting systematic noise in calculated

Michael Krainin
Hi Johan,

My first guess just having looked at the attached image is that this
is related to the depth quantization from the Kinect. I would try
using a slightly larger neighborhood size when computing the normals
and see if that helps.

-Mike

From: Johan Granberg <[hidden email]>
Subject: [Ros-kinect] Interesting systematic noise in calculated
       normals
To: [hidden email]
Message-ID:
       <AANLkTinFEynfyJbV-rRRbwjoQtN==[hidden email]>
Content-Type: text/plain; charset="iso-8859-1"

Hello. I am working on a project in wich i will be using the surface normals
from a pointcloud to do some machine learning. I have used the class
pcl::IntegralImageNormalEstimation to do get the normals, however im getting
some interesting noise in the normals.

Above i have appended an image in which i have put the normals of a point
cloud(normal.normal_x = red channel, normal.normal_y = green channel and
normal.normal_z = blue channel).

The image contains a flat wall and the side of a bookshelf. If you see the
image it is pretty obvious that there is some kind of systematic noise in
the image( wavelike and diagonal). Simmilar noise is present in all the
images of my enviorment that i have taken.

So now to the question: What is the reason for this problem?

My suggestions:
The kinnect has some small systematic error in the depth data?
My kinnect is not calibrated well enough?
I did something stupid?

What do you think?

Cheers, Johan Granberg
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://code.ros.org/mailman/private/ros-kinect/attachments/20110228/c2b34c77/attachment.htm
-------------- next part --------------
A non-text attachment was scrubbed...
Name: no_smooth.jpg
Type: image/jpeg
Size: 31620 bytes
Desc: not available
Url : https://code.ros.org/mailman/private/ros-kinect/attachments/20110228/c2b34c77/attachment.jpg

------------------------------
_______________________________________________
Ros-kinect mailing list
[hidden email]
https://code.ros.org/mailman/listinfo/ros-kinect
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: [Ros-kinect] Interesting systematic noise in calculated

Johan Granberg
Thank you for the response Mike! I have tried sampling with a larger neighbourhood and smoothing on the normal-image(a few different kernels) and it is getting slightly better but its still not very good(waves are still there).

You may very well be correct that quantization is a big part of why i get all of this noise, however depending on what kind of quantization is going on inside the kinect, I suspect that it doesnt fully explains why these waves appear. I will have a look at the original depth image and probably work on that to hopefully get "better" normals.

Cheers, Johan Granberg

2011/2/28 Michael Krainin <[hidden email]>
Hi Johan,

My first guess just having looked at the attached image is that this
is related to the depth quantization from the Kinect. I would try
using a slightly larger neighborhood size when computing the normals
and see if that helps.

-Mike

From: Johan Granberg <[hidden email]>
Subject: [Ros-kinect] Interesting systematic noise in calculated
      normals
To: [hidden email]
Message-ID:
      <AANLkTinFEynfyJbV-rRRbwjoQtN==[hidden email]>
Content-Type: text/plain; charset="iso-8859-1"

Hello. I am working on a project in wich i will be using the surface normals
from a pointcloud to do some machine learning. I have used the class
pcl::IntegralImageNormalEstimation to do get the normals, however im getting
some interesting noise in the normals.

Above i have appended an image in which i have put the normals of a point
cloud(normal.normal_x = red channel, normal.normal_y = green channel and
normal.normal_z = blue channel).

The image contains a flat wall and the side of a bookshelf. If you see the
image it is pretty obvious that there is some kind of systematic noise in
the image( wavelike and diagonal). Simmilar noise is present in all the
images of my enviorment that i have taken.

So now to the question: What is the reason for this problem?

My suggestions:
The kinnect has some small systematic error in the depth data?
My kinnect is not calibrated well enough?
I did something stupid?

What do you think?

Cheers, Johan Granberg
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://code.ros.org/mailman/private/ros-kinect/attachments/20110228/c2b34c77/attachment.htm
-------------- next part --------------
A non-text attachment was scrubbed...
Name: no_smooth.jpg
Type: image/jpeg
Size: 31620 bytes
Desc: not available
Url : https://code.ros.org/mailman/private/ros-kinect/attachments/20110228/c2b34c77/attachment.jpg

------------------------------
_______________________________________________
Ros-kinect mailing list
[hidden email]
https://code.ros.org/mailman/listinfo/ros-kinect


_______________________________________________
Ros-kinect mailing list
[hidden email]
https://code.ros.org/mailman/listinfo/ros-kinect
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: [Ros-kinect] Interesting systematic noise in calculated

rusu
Administrator
Johan,

Depending on your application you could try a pcl::MovingLeastSquares surface resampling method first. That should get
rid of the discretization problems.

Cheers,
Radu.
--
http://pointclouds.org

On 03/01/2011 07:08 AM, Johan Granberg wrote:

> Thank you for the response Mike! I have tried sampling with a larger neighbourhood and smoothing on the normal-image(a
> few different kernels) and it is getting slightly better but its still not very good(waves are still there).
>
> You may very well be correct that quantization is a big part of why i get all of this noise, however depending on what
> kind of quantization is going on inside the kinect, I suspect that it doesnt fully explains why these waves appear. I
> will have a look at the original depth image and probably work on that to hopefully get "better" normals.
>
> Cheers, Johan Granberg
>
> 2011/2/28 Michael Krainin <[hidden email] <mailto:[hidden email]>>
>
>     Hi Johan,
>
>     My first guess just having looked at the attached image is that this
>     is related to the depth quantization from the Kinect. I would try
>     using a slightly larger neighborhood size when computing the normals
>     and see if that helps.
>
>     -Mike
>
>     From: Johan Granberg <[hidden email] <mailto:[hidden email]>>
>     Subject: [Ros-kinect] Interesting systematic noise in calculated
>            normals
>     To: [hidden email] <mailto:[hidden email]>
>     Message-ID:
>     <AANLkTinFEynfyJbV-rRRbwjoQtN==[hidden email] <mailto:qwGP_VKGQQGK%[hidden email]>>
>     Content-Type: text/plain; charset="iso-8859-1"
>
>     Hello. I am working on a project in wich i will be using the surface normals
>     from a pointcloud to do some machine learning. I have used the class
>     pcl::IntegralImageNormalEstimation to do get the normals, however im getting
>     some interesting noise in the normals.
>
>     Above i have appended an image in which i have put the normals of a point
>     cloud(normal.normal_x = red channel, normal.normal_y = green channel and
>     normal.normal_z = blue channel).
>
>     The image contains a flat wall and the side of a bookshelf. If you see the
>     image it is pretty obvious that there is some kind of systematic noise in
>     the image( wavelike and diagonal). Simmilar noise is present in all the
>     images of my enviorment that i have taken.
>
>     So now to the question: What is the reason for this problem?
>
>     My suggestions:
>     The kinnect has some small systematic error in the depth data?
>     My kinnect is not calibrated well enough?
>     I did something stupid?
>
>     What do you think?
>
>     Cheers, Johan Granberg
>     -------------- next part --------------
>     An HTML attachment was scrubbed...
>     URL: https://code.ros.org/mailman/private/ros-kinect/attachments/20110228/c2b34c77/attachment.htm
>     -------------- next part --------------
>     A non-text attachment was scrubbed...
>     Name: no_smooth.jpg
>     Type: image/jpeg
>     Size: 31620 bytes
>     Desc: not available
>     Url : https://code.ros.org/mailman/private/ros-kinect/attachments/20110228/c2b34c77/attachment.jpg
>
>     ------------------------------
>     _______________________________________________
>     Ros-kinect mailing list
>     [hidden email] <mailto:[hidden email]>
>     https://code.ros.org/mailman/listinfo/ros-kinect
>
>
>
>
> _______________________________________________
> Ros-kinect mailing list
> [hidden email]
> https://code.ros.org/mailman/listinfo/ros-kinect
_______________________________________________
Ros-kinect mailing list
[hidden email]
https://code.ros.org/mailman/listinfo/ros-kinect
Loading...