Re: Barcodes & error detection

In lugnet.robotics, Philippe Hurbain wrote:

> Error correction is good, but no error
> is probably better...

   Absolutely, but in any real world application there is almost always some
error rate.

>> a '1' (black)... read as a '0' (white).
>
> This may be caused by specular reflexion on a
> too shiny black stripe. We had that problem...
> solved by tilting the light sensor...

   John Barnes also suggested that to me after I posted, and I'm embarassed  to
admit that the thought hadn't occured to me. I need to test (again), but there
are at least two reasons why I'm not overly optimistic on this being the
solution. First, there is slight variation in the barcodes themselves that may
negate this (light sensor inclined 2 degrees to the west will still be
perpendicular if a section of the barcode is inclined 2 degrees to the west;
rare, yes, but the error rate is already only 3%). Second and more critically,
to distinguish all three "colors" (white, black, and highly reflective) the
reflective "shiny" tape needs to give as bright as return as possible -
essentially that same shiny, specular reflection that needs to be avoided at
other times.
   Using retroreflective white tape or reflective but slightly diffusive tape
(an aluminum duct tape slightly crinkled and flattened again; I tried it as flat
as possible as well) would seem to work, but actual tests show that it often
can't be distinguished from white.

-- 
Brian Davis