Bib Detection

I’m trying to write some code to detect the numbers on Bibs for races.
I’ve been working with MBS’ OCR plugin but it’s not going as well as I need.
I think the problem is that they are very complicated scenes and not just paper through a scanner.
When complete I’m going to want to automate it so I’ll have to add a pattern recognition, eg discard “Chicago”, “Exit”, but keep “26016”. The discard/keep is easy enough, but maybe gives you a sense of complexity of the images.

Has anyone else tried something like this?
Any tips or hints would be greatly appreciated.

may be you should try with opencv ?

Did you try to set the white list of characters to recognize?

@Christian Schmitz , I did.

Oh, if you contact Jim Meyer he did a wonderful presentation at XDC this year about using Machine Learning from AWS and Google to detect things on an image. I would think this would be a wonderful use of such a technology.

@Christian Schmitz I’m reading online that this engine’s whitelist isn’t working correctly with some versions. I don’t know which your plugin uses.

@Bob Keeney, thanks for the suggestion. I don’t know Jim. Is he in this forum?

He’s a lurker and I don’t think he posts a lot.

OK, complete failure with MBS OCR (could be my fault), and I can’t get the OpenCV example to compile.
Other suggestions?

There are commercial products but they require uploading everything through a web server, and with the quantity and size of the images I’ll be dealing with, it’s just not an option.

I’m considering hiring a squad of high school kids to simply type them in!

That’s the cheaper option, no doubt.