As a former employee of Reactrix Systems (may it rest in peace), I know something of this space. We had this working a few years ago (ok, we were on Linux and using the Tyzx 3D camera.)
I'm happy to see that MS bought 3DV. 3DV has been promising that their camera will be available at scale for years. I talked to them at GDC a year and a half ago. "It'll be out this summer." "Who's your manufacturer?" "Oh, we don't have one yet. We'll just get someone in Taiwan or China to do it. Manufacturers are a dime a dozen."
So are companies with unbuildable products. I had to resist the urge to laugh in his face -- they didn't have their sh*t together in February and they'd have product on the shelves in the summer? I pretty much gave up hope on them then.
Which is too bad. They have a nice product -- not as good as Tyzx or PrimeSense, but theoretically it could be far less expensive.
Don't think of the EyeToy. That's a simple 2D differencing system, and not a very good one. I actually think it's a good product overall -- not enough so that I'd bother to buy one, but you could have fun with the right content. The vision system could be done better even with the harsh constraints (working in just about anyone's living room is surprisingly difficult.) But 2D images are fundamentally limited.
A depth-sensing camera like the Z-Cam, Tyzx, or PrimeSense can pick out gestures in front of your body, it can measure direction and acceleration in 3D, and it can pick players out from the background far better.
At the same time, our intuition was (and our experience found) that trying to convert camera input into controller input is just a bad idea. You don't want this for playing Halo 3. You need games built for camera input. It's fun for more than games, too -- we prototyped some nifty photo, video, and 3D model interfaces.
One of our most eyecatching prototypes was a light saber game where you could slice through oncoming hordes of robots. The video is dark and not very well done, sorry.