Database for licenseplate with Haartraining?
Anyone know of a database for an implementation of Haartraining on license plates? It would need around 5000 positive images (with license plates on them) and 2500 negative pictures (without license plates). The license plate region isn't important.
"The license plate region isn't important." -- hmm, you might be wrong about that.
Why reinvent the wheel? There are quite decent open source license plate detectors out there trained by using the OpenCV framework.
@berak - what I meant was that the training images could be from anywhere, not necessarily one region @StevenPuttemans - mostly for educational purposes, but I didn't know of the existence of openaplr, I'll check em out for some pointers
atill, stick with plates from one region. mixing chinese, arabian and whatnot plates for training will only result in noise.
I agree with @berak, there is a reason why ALPR systems have models for specific regions. There is to much difference, hence noise, between regions, leading to a model that is not descriptive enough. It will lead to a model that detects each rectangular object with decent ratio dimensions as a license plate...
you might be out of luck even if you restrict it to your country . in general, - the more diversity there is, the worse the results will be.