Anyway, I kind of want to program the vision software myself, to learn from the experience. And I want to aomething a little more advanced than simply blob detection.
Ehhh, thats like saying 'I want to do the really hard stuff first, so that I can learn the basics'

Do blob detection first, and then you can do more advanced stuff like masking, edge detection, face detection, etc. Get the groundwork down first . . . And blob detection is more advanced than you think

What is the speed of the CMUcam3's processor?
google

Oh and matlab has image processing stuff worth looking at.