Post date: Mar 07, 2012 7:31:0 AM
There seem to be quite a number of different ways to do object recognition. This is probably due to the fact that it is very very hard and people keep trying to come up with better ways to do it.
One of the methods that seems popular and that I'm going to go with is called Haar Cascades.
http://en.wikipedia.org/wiki/Haar-like_features
I will need to create a library of images. Typically with Haar training you need a library of positives, a library of negatives and some background images. The background images are to train in some ability to not misinterpret noise in the background.
Positive and negative images are self explainatory.
When training you are supposed to need a very large dataset in order to produce accurate results. I'm thinking I can get away with a smaller dataset but I should still have a good sized one. For background I think I can pretty much skip since I will be dealing only with a black and white profile and no background