DPRG
DPRG List  



DPRG: linear features

Subject: DPRG: linear features
From: Kipton Moravec kmoravec at airmail.net
Date: Sat Feb 20 10:27:14 CST 1999

David Philip Anderson wrote:
> 
> Howdy
> 
> Ralph T. writes:
> 
> > Due to extremely limited time, I am unwilling to work on a line-follower
> > until I can characterize the sensors with some semblance of what is
> > expected. Just call me a stickinthemud.
> 
> Dear Stickinthemud:
> 
>         Did you happen to see the "Robots Alive" special from a year ago
> or so, hosted by Alan Alda (of "MASH" fame)?  They featured an autonomous
> car that worked essentially by recognizing "linear features" rather than
> any kind of lines per se.  So it could track on stripes on the hiway,
> curbs, cracks in the asphalt, edges of the road, anything.  Pretty slick,
> all running from a little TV camera mounted on the dash and a standard
> PC sitting in the trunk.
> 
> I bet these are out-of-work defense contractors looking for a new market
> for their wares.  Still, the fact that they can do it with the same kind
> of hardware we have is encouraging.  I've said before, and I truly believe,
> these guys aren't any smarter than us!  (Just better funded.)
> 

Yup.  That is one point I tried to make when I gave my talk 
on image processing at a meeting awhile back.  We were doing 
image processing on a 8 MHz system back in 1983 with some custom
hardware for the preprocessing.  Still we could only process 5 
frames per second but that was enough to hit the target with a 
fast falling bomb. The robots now have faster, more efficient 
processors, and they move slower. 

Added to the fact that there are inexpensive FPGA's that you can
learn quickly to program, it does not take much in terms of hardware to 
get you up and running.

Choosing the right algorithms to impliment are another thing. 

> Most of our line followers are actually edge detectors.  How does one
> truly define a line?  What is required to robustly identify one,
> independent of it's width, color, contrast, etc?  Evidently somebody
> has already solved this problem, with a PC no less.  Can we do the same?
>
Edge detection is the easiest form of image processing.  It can be 
accomplished in hardware with a subtraction and a 2 pixel delay 
(i.e. two 8 bit flip-flops) This is a piece of cake for a pentium class 
machine. Even a micro controller can do it, but the data rate of about
12 
MHz tends to make people want to do it in hardware as it is captured.

------------------------------

More information about the DPRG mailing list