Well, let me tell you what ARToolKit does. With the help of a webcam, it can do limited computer vision. The only thing it can see are these special glyphs that are specifically designed to be easy for it to pick out of a scene. However, it can see them fairly reliably, and it can also tell where they are in space and how they’re oriented. You can use this to infer the position of larger objects: let’s say you glue a glyph to, oh, a remote control car. The system can see the glyph, and from that, it can figure out where the car is. Then, you can put the webcam output on the screen, and superimpose 3D graphics. You can even make it look like the 3D objects are riding in the real physical RC car.
If you’ve seen the TV ads for the game “Eye of Judgement,” that’s using augmented reality - it can see the glyphs on the cards, and it can superimpose dragons and things.