[ros-users] Depth image REP

[Hordur Johannsson]

+1 

I agree with the usefulness of having a depth image instead
of a point cloud to pass that information around. But here
are some issues that would be good to see addressed in the 
proposal.

It would be useful to include the baseline and precision of
the underlying cameras/projectors if the depth is generated by 
triangulation. In the case of stereo camera the camera info
documentation says that the displacement between the left
and right camera is coded in P. This is not the case when
you look at the camera info of the depth image from the OpenNI
node. In addition the depth image might come from a time of flight
sensor that would have different error characteristics. 

The CameraInfo message has a field distortion_model string
and an associated vector for the model parameters. The error
characterization of different depth sensor could be fit into 
that model. But maybe that is to much abuse of the message
and could cause some confusion. It would be better to have 
something like sensor_model or noise_model.

A possible solution would be to introduce a image_geometry::DepthCameraModel.
Then define the CamerInfo::D as being general model parameters
that would then be correctly interpreted by the DepthCameraModel class.

-- 

Not as part of this particular REP, but at some point I would 
propose having the stereo_image_proc support a depth image.

Thanks,
Hordur

On Tue, 2011-12-06 at 14:51 -0800, Patrick Mihelich wrote:
> Hi all,
> 
> Attached is a draft REP defining a representation for depth images in
> ROS. This standardizes the format used by the ROS OpenNI driver and
> associated processing nodelets, and should be useful to other
> producers/consumers of depth data.
> 
> You can also view it in HTML at
> http://people.willowgarage.com/mihelich/rep-depth-image.html.
> 
> If this interests you, please state your vote (see REP 10 - Voting
> Guidelines) as well as any comments and concerns.
> 
> We seem to be having archival issues, so (light) early feedback for
> the ros-sig-perpip list is reproduced below.
> 
> Cheers,
> Patrick
> 
> Forwarded conversation
> Subject: Depth image REP
> ------------------------
> 
> From: Ethan Rublee <erublee at willowgarage.com>
> Date: Fri, Dec 2, 2011 at 3:56 PM
> To: Patrick Mihelich <mihelich at willowgarage.com>
> Cc: ros-sig-perpip at code.ros.org
> 
> 
> +1
> 
> ----------
> From: Vincent Rabaud <vrabaud at willowgarage.com>
> Date: Fri, Dec 2, 2011 at 4:13 PM
> To: Ethan Rublee <erublee at willowgarage.com>
> Cc: Patrick Mihelich <mihelich at willowgarage.com>,
> ros-sig-perpip at code.ros.org
> 
> 
> +0, it looks good but I don't know much about the previous messages.
> The deprecation of the DisparityImage will happen through the tic-toc
> I guess.
> 
> 
> ----------
> From: Brian Gerkey <gerkey at willowgarage.com>
> Date: Fri, Dec 2, 2011 at 5:47 PM
> To: Patrick Mihelich <mihelich at willowgarage.com>
> Cc: ros-sig-perpip at code.ros.org
> 
> 
> +0.
> 
> I'm not a user of this part of the system, but find the REP to be
> well-reasoned and convincing.
> 
>        brian.
> 
> 
> 
> 
> _______________________________________________
> ros-users mailing list
> ros-users at code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users