Originally Posted by Scouse Monkey
I've worked on a few vision algoritms for autonomous vehicle control and computer vision is one of the most difficult problems out there. Just because its easy for humans to do, doesn't imply its easy for computers. Finding a yellow or white box from a good picture head on is not that difficult assuming your background is not noisy and is distinctive, but to do it in a real world enviornment is a whole other problem. You have to take into account that the license plate can be any distance and angle from you and probably out of focus. You might be able to constrain that a bit to only be able to recognize plates of cars directly in front of you, but your biggest problem is going to be to deal with very poor image quality and more importantly very inconsistent images you'll get from a cheap web cam. You'll get over saturation when its too bright out and very dark images when its dark out in addition to the normal image noise you get just because of the poor image sensor quality. When you have poor image capture, your going to get a lot of crappy images and you'll have a ton of things that look very similar to license plates, i.e. headlights, street signs, wheels, street lines, partially obscured license plates, depending on your algorithm you use and the conditions the image sensor of the camera was exposed to, if your just looking for a box, all these things can look similar, particularly when you add in a typical amount of noise and over saturation that is to be expected. After you do find the license plate, you'll still have a ton of problems getting the text on the plate to a point where you can use OCR on it. OCR recognition goes down exponentially when you use skewed or noisy images.Originally Posted by myozone
In any case, most programs used specifically for image processing can't do anything useful out of the box. Most require a significant amount of tayloring to the specific problem your trying to solve and will require an in depth understanding of the image processing techniques they use. To get good recongition, you will have to account for everything from the mounting position of the camera to the image sensor quality and speed you'll be driving at.
If your really serious about doing this, there are a number of image processing code libraries out there made specifically for this, alot free, just google for image processing, but its not going to be an easy thing to do.
Originally Posted by Scouse Monkey
No I'am not PI ;-) Just intersted how it could be done....
you are crazy mon ami, tu es peut etre un policier qui veut prendre les numero de plaque des voyoux!!!
I was thinking about this a few months ago. Primarily to catalog ****bag drivers, and as vision technology gets better you could detect previously known bad drivers from a reasonable distance.
People who I can see benefitting from such technology:
* parents with kids in the car, wanting to avoid known dangerous drivers
* people wanting to be alerted when near friends, a crush, etc
* PI's, investigators, cops, big brother, repomen
* fun and experimental programs/scripts to find "cool" name based license plates (sort of like DXing in the Ham Radio field)
Audi A4 Carputer, 80% setup.
i was wondering, do red-light cameras only work on instate cars?
Lilliput; XPC/FLEX mobo; 1.7 ghz P4 Mobile;512 DDR; 160 gb HDD; opus 150; slot usb dvd-rw
My work log
I worked for ACS (they are/run EZ-Pass for the government). Red light Photo enforcement systems were a big thing we developed the past few years. The camera records the date, time of day, and time elapsed since the beginning of the red signal and the speed of the vehicle. Electronic flash produces clear images of the vehicle license plates under all light and weather conditions.Originally Posted by Peoples
So they take the picture, and humans actually decipher the plates (like my intern friend).
carputer: in progress
LOL. That's awesome OCR software you're using. What I had suggested previously is what is done by the Red Light guys. A picture is taken of the intersection with a high resolution camera and some guy crops and zooms in on the license plate.Originally Posted by bd3521
"It does not understand, if it does not ride."
M10000, 512 Mb, 2,5" 40G, Slim DVD, Opus 90W, Xenarc 7", Leadtek 9532 GPS,
RoadRunner, MapMonkey, Pioneer HU, Pioneer 4 ch 400W, Kappa 6.5, Basslink X 200W
Originally Posted by kevlar
THIS TECHNOLOGY ALREADY EXISTS!! It's called a cell phone.