Microsoft Research is currently having a Techfest at Redmond where it is showing off a lot of new work. The latest work on the Kinect uses the same sort of machine-learning approach to distinguish between an open hand and a clenched fist. Although there are no details, its general method was to use a large number of images of people's hands and supervised training to distinguish between open and closed hands. The learning algorithm is based on a forest of decision trees, which is the same general method used to implement the skeleton tracking. Being able to detect an open or closed hand might not seem to be much of an advance, and certainly not as good as a multi-gesture touch screen interface, but it is enough to allow the user interface to distinguish a "pick up" or "grip" gesture. So you can move the hands within an image, close both hands to grip the image points and move apart to zoom. You can't get the software at the moment, but it has been promised for the next version of the Kinect SDK for Windows along with the long awaited 3D scanner Kinect Fusion.