Microsoft's research team in Cambridge has published a paper outlining several ways the speed and accuracy of Kinect can be improved. Lucky Kinect is also made by Microsoft!
The team's research does not build on Kinect's existing technology; rather, it strips it away at the base level and replaces it with a new way of "thinking". Rather than using a recognition system built on tracking limbs, the researchers built one that could identify body parts in an image, pixel-by-pixel.
This would not only remove the need for Kinect to repeatedly calibrate, but would also make its detections speed a lot faster.
That's the good news! The bad news is that this wasn't an official Kinect project, just a research paper, so there's no telling when such findings could be implemented on a consumer level, if they ever can at all.
Kinect's AI breakthrough explained [I Programmer, via GI.biz]