Thread: Image Processsing - Road Signs

yeah, just do that...

Originally Posted by konnichiwa

i was considering it for number plates.

Yeah, i'd seen your previous posts regarding that.

This is for something non-privacy destroying

Originally Posted by konnichiwa

that image is WAY too big to do real time image processing on.

You'd need an ARM9 chip hooked up to the Camera or similar and write something efficient in ASM to give you back relevant portions of the picture (eg, the road sign)

then on your PC, have a program apply a neural net to work out exactly what the image is.

Well, what's the likely hood of something 320x240 being able to pick up something like that?

I guess a P-M 2 gig wouldn't cut the mustard?

Couple thoughts,

downsample the capture, find the interesting portion, then process that area in the larger pic. Meaning look for signs first, not numbers. I'll bet it's easier to pick a sign out then text.

Usually they are perpendicular to your view point and squarish, that would probably help.

Usually traffic signs are painted with highly reflective paints. I think they have some sort of metal in the paint. Maybe you can take advantage of this somehow with a lens filter or software filtering.

Got a nitemax? Take some pics of some roadsigns, in the day. See how they show up relative to the rest of the scene.

yea, you could take the picture, downsample it to 2 colours, black and white, and then it should be a cinch to recognize the numbers. All road signs that display a speed limit are usually made up of 2 contrasting colours, so even taking the pic and reducing it to black and white based on hue. I'll bet that even some scanner's text recognition should be able to recognise that

The red sign in that picture would be pretty easy.

Just filter out everything not red:


Then threshhold it and youll find the circle:


YOu can find the rest of the text pretty easily with a few other tricks, the hard part is going to be doing that in a generic case though. Assuming you want to do this with just a camera while your driving down the road, you'll have to get extremely creative to filter out false positives. i.e. the red lights on cars or other things would totally screw with the recognition for the case above. You'll have ot think about every possible thing that could look similar and find a way to filter it out. i.e. restaurant signs, license plates, partially obsured signs, etc.

I think kamakazee has the best solution.

use a B/w camera and then use OCR software, trouble is , it is pretty slow, so you have to have a way of telling it to select an image and then process.