DPRG List  

DPRG: RoboVision

Subject: DPRG: RoboVision
From: R. Steven Rainwater srainwater at ncc.com
Date: Fri Jun 20 15:29:47 CDT 1997

On Fri, 20 Jun 1997 15:21:45 -0700 you wrote:

>I used to work in the Advanced Algorithms Section of the Image
>Processing Lab at TI. I used to tie up a VAX 785 all night doing image
>processing on a few frames.  I do not think there is much you can do
>with a 8051 even if you could get the image in its 64K of RAM space.  In
>real (fielded) systems we had to resort to custom hardware for the front
>end processing. Then high speed signal processors for the back end
>processing.  I doubt you could be very successful with a 33 MHz 386. A
>486 is starting to get in the useful processing range, and a pentium is
>better. For image processing the faster processor and more memory the

Yep, I'd agree with most of what you've said there.  I've done some
image processing at work that needed to run in "real-time" (30fps).
We use a TI C44 DSP to do the work.  It does very simplistic types of
processing on each frame and we ended up using only about 2 meg of
static RAM on each system.  Even so, the C44 is just barely adequate
at 275 MOPS to get the each frame done before the next one shows up.

What I'm looking to do for my robot in terms of vision, though, is
nothing as complex as traditional image processing algorithms.  I'd
be happy if I could get some basic motion information out of the 
system.  I've seen other systems that did this using relatively 
simple neural networks.  I'm trying to dream up some way of using
really cheap processor to get some basic information out of the
system like motion detection of up-down-left-right-in-out...  

- -Steve

| email: srainwater at ncc.com             | "As you struggle to save humanity  |
| Web:   http://www.ncc.com/            | be sure to avoid electrodes in     |
| Tel: 972-650-2002  Fax: 972-650-1929  | your path"          -Robotron 2084 |


More information about the DPRG mailing list