Re: [Ros-kinect] Is it possible to change the resolution of depth image?

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

Re: [Ros-kinect] Is it possible to change the resolution of depth image?

Radu B. Rusu
Administrator
Ivan,

On 12/08/2010 08:53 PM, Ivan Dryanovski wrote:

> On Fri, Nov 26, 2010 at 3:10 PM, Radu Bogdan Rusu<[hidden email]>  wrote:
>> Why would you like to change the resolution from the driver? I suppose we can implement a rough downsampling method in
>> the driver, but it is way easier for you to do it on the client side. Besides, for depth, you don't really want a method
>> that just takes every 2nd measurement on the width x height grid, but rather a true downsampling method using voxel
>> grids. pcl::VoxelGrid<T>  does this efficiently - and with the use of nodelets you don't need to write any code!
>>
>> Cheers,
>> Radu.
>
> Radu,
>
> I experimented a little with the voxel grid filter, and it _appears_
> that by using it, you lose the structure of the cloud data, ie, you
> cannot treat the points as neighbours in a 640x480 image anymore. If
> that is indeed the case, then adding a sparsify option in the Kinect
> driver is a good idea. Or perhaps I'm not using the filter correctly,
> or interpreting the output correctly.
>
> I'm not sure how things will develop now with the new PrimeSense
> driver, but I guess it's something to keep in mind.
>
> Ivan

You're right - I've been toying with the idea of a PixelFilter (we still want the VoxelFilter to be general enough so it
can be used with any 3D data, even if it's a collection of registered 3D point clouds). I have a preliminary
implementation that I could commit to PCL within the next few days. That should keep the structure intact.


Cheers,
Radu.
--
http://pointclouds.org
_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: [Ros-kinect] Is it possible to change the resolution of depth image?

Patrick Mihelich
The OpenNI interface released by PrimeSense has APIs that allow setting the resolution of the depth sensor to QVGA (320x240). Whether the Kinect actually supports this in hardware I don't know yet - we can try it tomorrow - but this would be an ideal option if it works.

Cheers,
Patrick

On Thu, Dec 9, 2010 at 1:24 AM, Radu Bogdan Rusu <[hidden email]> wrote:
Ivan,

On 12/08/2010 08:53 PM, Ivan Dryanovski wrote:
> On Fri, Nov 26, 2010 at 3:10 PM, Radu Bogdan Rusu<[hidden email]>  wrote:
>> Why would you like to change the resolution from the driver? I suppose we can implement a rough downsampling method in
>> the driver, but it is way easier for you to do it on the client side. Besides, for depth, you don't really want a method
>> that just takes every 2nd measurement on the width x height grid, but rather a true downsampling method using voxel
>> grids. pcl::VoxelGrid<T>  does this efficiently - and with the use of nodelets you don't need to write any code!
>>
>> Cheers,
>> Radu.
>
> Radu,
>
> I experimented a little with the voxel grid filter, and it _appears_
> that by using it, you lose the structure of the cloud data, ie, you
> cannot treat the points as neighbours in a 640x480 image anymore. If
> that is indeed the case, then adding a sparsify option in the Kinect
> driver is a good idea. Or perhaps I'm not using the filter correctly,
> or interpreting the output correctly.
>
> I'm not sure how things will develop now with the new PrimeSense
> driver, but I guess it's something to keep in mind.
>
> Ivan

You're right - I've been toying with the idea of a PixelFilter (we still want the VoxelFilter to be general enough so it
can be used with any 3D data, even if it's a collection of registered 3D point clouds). I have a preliminary
implementation that I could commit to PCL within the next few days. That should keep the structure intact.


Cheers,
Radu.
--
http://pointclouds.org
_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users


_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: [Ros-kinect] Is it possible to change the resolution of depth image?

Radu B. Rusu
Administrator
In any case, I still want that PixelFilter developed (better said - a more efficient version of VoxelFilter - if that's
too inefficient for "lesser" CPUs). There's a lot of other systems, including our stereo cameras, where such a
downsampling method would be welcome.

It'll probably take like 1-2 hours to implement and test tomorrow, so I'll give it a shot.

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

On 12/09/2010 01:33 AM, Patrick Mihelich wrote:

> The OpenNI interface released by PrimeSense has APIs that allow setting the resolution of the depth sensor to QVGA
> (320x240). Whether the Kinect actually supports this in hardware I don't know yet - we can try it tomorrow - but this
> would be an ideal option if it works.
>
> Cheers,
> Patrick
>
> On Thu, Dec 9, 2010 at 1:24 AM, Radu Bogdan Rusu <[hidden email] <mailto:[hidden email]>> wrote:
>
>     Ivan,
>
>     On 12/08/2010 08:53 PM, Ivan Dryanovski wrote:
>      > On Fri, Nov 26, 2010 at 3:10 PM, Radu Bogdan Rusu<[hidden email] <mailto:[hidden email]>>  wrote:
>      >> Why would you like to change the resolution from the driver? I suppose we can implement a rough downsampling
>     method in
>      >> the driver, but it is way easier for you to do it on the client side. Besides, for depth, you don't really want
>     a method
>      >> that just takes every 2nd measurement on the width x height grid, but rather a true downsampling method using voxel
>      >> grids. pcl::VoxelGrid<T>  does this efficiently - and with the use of nodelets you don't need to write any code!
>      >>
>      >> Cheers,
>      >> Radu.
>      >
>      > Radu,
>      >
>      > I experimented a little with the voxel grid filter, and it _appears_
>      > that by using it, you lose the structure of the cloud data, ie, you
>      > cannot treat the points as neighbours in a 640x480 image anymore. If
>      > that is indeed the case, then adding a sparsify option in the Kinect
>      > driver is a good idea. Or perhaps I'm not using the filter correctly,
>      > or interpreting the output correctly.
>      >
>      > I'm not sure how things will develop now with the new PrimeSense
>      > driver, but I guess it's something to keep in mind.
>      >
>      > Ivan
>
>     You're right - I've been toying with the idea of a PixelFilter (we still want the VoxelFilter to be general enough so it
>     can be used with any 3D data, even if it's a collection of registered 3D point clouds). I have a preliminary
>     implementation that I could commit to PCL within the next few days. That should keep the structure intact.
>
>
>     Cheers,
>     Radu.
>     --
>     http://pointclouds.org
>     _______________________________________________
>     [hidden email] <mailto:[hidden email]> / http://pointclouds.org
>     https://code.ros.org/mailman/listinfo/pcl-users
>
>
_______________________________________________
[hidden email] / http://pointclouds.org
https://code.ros.org/mailman/listinfo/pcl-users