DPRG List  

[DPRG] RE: How much do our choices effect our results?

Subject: [DPRG] RE: How much do our choices effect our results?
From: Sluggy! sluggy9912 at swbell.net
Date: Wed Aug 29 17:15:14 CDT 2007

On Wed, 2007-08-29 at 16:38 -0500, John Abshier wrote:

>  Humans usually do not use odometry to navigate.

Della Bea Robinson: Let me call you a cab. 
Ray Charles: I got it. Three blocks up, left for two, right for one,
fifteen giant steps and I'm at the Crystal White Hotel. Hello!

Of course, Ray Charles was an unusual case in many ways.

> I would put the emphasis on vision.

Vision has several advantages in that, as a single sensor technology, it
is VERY adaptable to specific tasks. See a line, detect a line. See a
wall, detect a wall. See a can, detect a can. I saw a web page once in
which a hobbyist was digitally playing 45 RPM records by analyzing a
high resolution scan of the record.

Most vision projects these days work off the analysis of an entire
captured frame. This is both the power and the limitation of the
technique. You can analyze a video frame 900 ways. On the other hand,
you *have* to analyze the frame in some way to get any useful
information out of it, then you have to deal with the flood of
information as up to (or in excess of) 30 of these frames come at you
each second. Even if you drop and ignore many frames, much of the data
is only good of you watch trends, so you have to not only see, but store
a buffer of analyzed frames. Even if you analyze it in hardware, taking
the load off your CPU, you still have to *deal* with the analysis in
close-to-real time. It doesn't take very long for the sensor tail to be
wagging the robo-dog. :)


More information about the DPRG mailing list