Even though the global idea sounds great, according to the sensors that they have I don't think that such precision is possible. I am based only in guesses, but with a normal camera, and CMOS sensor (for depth sense using a infrared projector) that have not been calibrated, the software can be able to determine different shapes (bodies) that are moving (background subtraction to detect the velocity, for example) but is not able to have enough precision to even distinguish the different fingers of a hand.

As I said I haven't played or used Kinetic, nor I know the software that is running, but it seems to me that they are parameterizing estimated movements and mapping them into the space of allowed movements of your "avatar".