Thread Subject: OCR on Tank Cars - Any Experience?

My company is considering the development of an image
processing systems for reading numbers on tank cars. A
number consists of eight digits. The images will be taken
outside any buildings, so the environment will vary through
the day and the year. Also, the distances of the digits
vary and the background is not homogeneous. The numbers can
be light or dark. There can be snow, dirt, rust and de-
coloration on the tank. Plants with snow on them can be
present in an image, too.

Has anyone worked on such a problem so far? If so, how long
did it take to develop such an image processing system? How
long would it take if one single person must do all the
experiments and programming?

I am afraid to answer your question. We expect answers from experts not questions! According to me, I would like to put like this as a response to your answer that, to develope a recognition system of your reuirement may require not more than infinite years.

"OkinawaDolphin " <OkinawaDolphin@Hotmail.com> wrote in
message <ftkqb3$ka$1@fred.mathworks.com>...
> My company is considering the development of an image
> processing systems for reading numbers on tank cars. A
> number consists of eight digits. The images will be
taken
> outside any buildings, so the environment will vary
through
> the day and the year. Also, the distances of the digits
> vary and the background is not homogeneous. The numbers
can
> be light or dark. There can be snow, dirt, rust and de-
> coloration on the tank. Plants with snow on them can be
> present in an image, too.
>
> Has anyone worked on such a problem so far? If so, how
long
> did it take to develop such an image processing system?
How
> long would it take if one single person must do all the
> experiments and programming?

it would be cheaper and more reliable to stick on an rfid
tag and buy an off the shelf reader. next best would be
to paint or stick on a barcode, though then you are back
into some of the image processing problems, though the
recognition part is easier and off the shelf.

unless of course you are looking for a long term research
and design project.

On Apr 10, 12:18=A0pm, "David " <d...@bigcompany.com> wrote:
> "OkinawaDolphin " <OkinawaDolp...@Hotmail.com> wrote in
> message <ftkqb3$k...@fred.mathworks.com>...
>
>
>
>
>
> > My company is considering the development of an image
> > processing systems for reading numbers on tank cars. A
> > number consists of eight digits. The images will be
> taken
> > outside any buildings, so the environment will vary
> through
> > the day and the year. Also, the distances of the digits
> > vary and the background is not homogeneous. The numbers
> can
> > be light or dark. There can be snow, dirt, rust and de-
> > coloration on the tank. Plants with snow on them can be
> > present in an image, too.
>
> > Has anyone worked on such a problem so far? If so, how
> long
> > did it take to develop such an image processing system?
> How
> > long would it take if one single person must do all the
> > experiments and programming?
>
> it would be cheaper and more reliable to stick on an rfid
> tag and buy an off the shelf reader. =A0next best would be
> to paint or stick on a barcode, though then you are back
> into some of the image processing problems, though the
> recognition part is easier and off the shelf.
>
> unless of course you are looking for a long term research
> and design project.- Hide quoted text -
>
> - Show quoted text -

You should look at some character recognition methods that are used in
number-plate recognition. A famous system in London records
registration plates of vehicles entering the toll zone incredibly
accurately.
I would recomend using 2D cross-correlation to find your numbers if
they are of a standard font. Use the 'regionprops' function to pick
out the character objects in the image and you can use im rotate and
imresize to make the characters aligned with your template to match.

Adam

OkinawaDolphin wrote:
> My company is considering the development of an image
> processing systems for reading numbers on tank cars. ...

Is this for in house use or as a commercial product?

Besides the problems you and others have briefly itemized on the
difficulty, unless you're limiting the system to a single RR and/or
style of car, there's even more variation than those mentioned in the
paint schemes.

Altho I don't know for sure, I would think there would have been quite a
lot of work on RR car ID in general for the switching/tracking problem
of the railroads so first thing I would do would be to search in that
area for what one might find already.

Then I'd reconsider the idea in light of that knowledge gained.

Others have made some other reasonable comments as well...


--

The OCR on tank cars will be applied for a commercial
product. I have only few images of tank cars and the print
quality strongly varies. The camera position varies also. I
have the impression that only a small percentage of tank
car numbers can be recognized correctly.

In article <ftl7d9$94r$1@fred.mathworks.com>,
OkinawaDolphin <OkinawaDolphin@Hotmail.com> wrote:
>The OCR on tank cars will be applied for a commercial
>product. I have only few images of tank cars and the print
>quality strongly varies. The camera position varies also. I
>have the impression that only a small percentage of tank
>car numbers can be recognized correctly.

If the numbers are on tank cars rather than on box cars, then will
there not also be the additional complication of curvature of
the printing?

--
  "I buy more from my grocer than he buys from me, and I bet it's
  the same with you and your grocer. That means we have a trade
  deficit with our grocers. Does our perpetual grocer trade deficit
  portend doom?" -- Walter Williams

Walter Roberson wrote:
> In article <ftl7d9$94r$1@fred.mathworks.com>,
> OkinawaDolphin <OkinawaDolphin@Hotmail.com> wrote:
>> The OCR on tank cars will be applied for a commercial
>> product. I have only few images of tank cars and the print
>> quality strongly varies. The camera position varies also. I
>> have the impression that only a small percentage of tank
>> car numbers can be recognized correctly.
>
> If the numbers are on tank cars rather than on box cars, then will
> there not also be the additional complication of curvature of
> the printing?

The size of a RR car is such that will be, imo, an effect far less than
any of the others already mentioned... :)

I just looked as a I came by the packing plant on the way home from town
at the three or for sitting on the siding--three cars; three _totally_
different lettering styles, location, and size plus one was almost
obliterated w/ grime from an obvious overfill scenario sometime. That
doesn't even consider the grafitti sprayed all over! :)

This is a toughie at best. Good idea, would undoubtedly have some
application if it were done, but the developer better have some deep
pockets methinks.

--

OkinawaDolphin wrote:
> The OCR on tank cars will be applied for a commercial
> product. I have only few images of tank cars and the print
> quality strongly varies. The camera position varies also. I
> have the impression that only a small percentage of tank
> car numbers can be recognized correctly.

GENERAL ELECTRIC COMPANY [US/US]; One River Road, Schnectady, NY 12301 ...
Title: RAIL CAR TRACKING SYSTEM
Abstract:
A method and system for tracking a rail car (22) having an on-board
communication system (12) including a location determining system (24,
33) and a transceiver (26) for receiving and transmitting rail car data.
The communication system including a processor (28) responsive to an
executable program for enabling operation of the transceiver (26) to
transmit rail car data during a reporting event and having a memory for
storing data and instructions. ...The rail car (22) includes an AEI
reader (33) attached to the rail car for reading AEI tags (33) on other
rail vehicles passing by the reader, the reader being in communication
with the rail car message system (12) for transmitting data indicating
at least a location of the other rail vehicles.


Would seem there would be some competition to beat... :)

Of course, an optical recognition system could still be useful and
perhaps cheaper if somewhat more limited. I think the above patent
preceded the widespread availability of (relatively) cheap GPS units
such as what UPS, etc., are using.

I didn't find much just on OCR recognition, quite likely owing to the
already perceived difficulties and the assured (albeit maybe more
expensive per car) type of solution outlined above.

There are several operations in business I found using similar systems,
this patent abstract was the most succinct (altho it is OT, I apologize
for that it did seem kinda' an interesting question).

--