All:
I basically agree with Osman's comments below:
With regards to high performance OpenCV, I think we need to focus
on platforms that already run it well rather than hoping that we
can "buy first" and "run later". I think that OpenCV is heading
in the general direction of OpenCL/CUDA, but I do where they are
at the moment. If anybody has additional knowledge to suggest what
a reasonable high performance OpenCV platform is, I'm all ears.
With regards to the RasPi, I view it as a Wi-Fi/Ethernet connection
(via a Wi-Fi dongle), USB host, and some GPIO. It can run Python
and C/C++ and ROS. I have no expectation that the RasPi graphics
processor will provide any additional kick to OpenCV. I'm pretty
confident that the ROS core will run on it with acceptable
performance. As long as ROS can be compiled for it, I'm happy.
Regards,
-Wayne
10/11/2012 08:24 AM, Osman Eralp wrote:
> It is true that the wifi stopped working on my PandaBoard. One day it
> was working, and the next day it wasn't. I didn't change anything. I put
> in an old SD card with a known good image and found the wifi didn't work
> with the that image too, so it looks like a hardware problem. I searched
> with Google to see if this is a common problem. It is not a common
> problem. Maybe Wayne and I have just been unlucky.
>
> My requirement for a mid-range processor board is that it must run
> OpenCV reasonably well. For me, "reasonably well" means tracking a
> colored object at 10 fps. If the PI can do that, I'm all in.
>
> I see my robot on-board processor options aligning like this:
>
> * For simple tasks, such as reading sensors, use a ARM Cortex Mx.
> * For behavior and simple computer vision tasks, use a Raspberry Pi.
> * For complex computer vision tasks, use an Atom or Core processor.