No disrespect to Kinect, but I have always thought that a system proposed to catch the diversity of human movement in a 3D space would probably need an X, Y, and Z sensor optimised in locations. So therefore, a sensor to both left and right, front and back, and possibly above. That way you would be able to build up a true 3D representation. How can a sensor only located in front of you possibly detect the movement accurately as seen from the side? A lot of extrapolation would be required with situations where a 50/50 assumption would have to be made.

For the sitting dilemma, they would possibly use a primer of the body to say, ignore the lower half of the avatar from a certain point; track only above but then again, for a driving game, the foot needs tracking too.

The other way of doing this is by using a suit on the person that has broadcasting sensors but that defeats the point of not having to attach anything (controller or otherwise) to the subject.

I think a sensor in front of you could realistically see you twisting your hand even if you're pointing it right at the sensor, which is the same as twisting the wrist unless you're some sort of weird contortionist. Whether they have enough resolution / processing power to detect that is another question entirely.

if you look at the tech demo's for kinect like this one it is clear why the game doesn't detect wrist "spin"

http://www.nextgn.com/2010/04/15/this-is-how-project-natal-camera-sees-the-player-shaky-cam-video/

Kinect only "sees" a depth map, a contrast/colour map and from that it uses shape recognition to generate a "skeleton" in 3D space. it can't see things like we do, it only "sees" a 3D stick figure and the hand is just a point off the wrist (and a silhouette of course). so as far as I can tell there is no accurate way for a game to detect the twist of the arm.

hmm very disappointing, people who want to be more accurate in detail will find this annoying.

I still think, that because Kinect can only ever detect the planar front, albeit with depth sensing, it's almost like it can only ever achieve 2.5D, so to speak. It can't possibly know what is happneing behind that planar barrier that is being dictated by the plane (of sight) which is facing the Kinect sensor.

It would make a lot of sense to have 2 additional side sensors. Rear view would not be that important for the majority of things. The problem is, adding sensors adds cost, complexity of setup due to calibration, complexity with just what space is required and so on. The best solution for tracking a subject would be a suit of some kind (ala Lawnmower Man).

Just occasionally it'd be nice to read what Kinect CAN actually do.  The hating on any small detail is getting rather old already.

@slowmo

From my POV - not hating, more like curiosity. I am just logically trying to think it through. I just rewatched the promo released last year by MS, the famous promo where the entire family use Natal. If that's what is expected then cool. But again, I am just throwing things up here because I will admit, I am skeptical, given that only the front plane of the subject is actually being detected.

This problem is prety much obvious if you look on this images from Natal presentation

This image show how Natal see things. It is easy to see major body parts configuration from it, but almost imposible to understand wrist configuration. Kinects need better depth sensevity to do this.

They would be fools to give you that now how else will they be able to flog Kinect   ???

"If you can pick at it post it"

Noone sits down during bowling at least!