So I already have a followup to my own question.  I have a detector node
that does the face detection and publishes the results using the rein
DetectionArray message type which includes an ROI for each detected
face.  I have a separate display node that displays the raw camera image
as well as any markers that might be generated by the detector (or other
nodes) such as bounding boxes around faces.

My question is this: is it better ROS practice to have the display node
subscribe to the DetectionArray topics from each of the detector node(s)
and then draw bounding boxes (or other markers) from the those messages,
or, should I create services in the display node that draw markers on
the image upon requests from the detector nodes?  I rather like the feel
of the services approach, but since frame rates will be about 20-30 fps,
is this considered on the high side for service-based transactions?

Thanks!
patrick

On 6/9/2011 11:47 AM, Patrick Goebel wrote:
> Hello ROS users,
>
> I was reading up on the ROS REIN framework
> (http://www.ros.org/wiki/rein) which looks really nice and I want to do
> something similar using Python.  I have a good start and will use my
> pi_face_tracker application as a test case.  But before I get too far
> along, has anyone already done this?  In other words, has anyone already
> put together a Python-based image processing pipeline for ROS/OpenCV and
> optionally, point clouds as well?  (Though I realize that PCL does not
> yet have Python bindings.)
>
> Thanks!
> patrick
>
> http://www.pirobot.org
>
>
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>