Instead of just trying to detect the fingertips, this approach uses a 3D model of the hand and then tries to orient and bend it to fit with the data from the Kinect. Because it's a complete model of the hand you not only find the positions for the fingertips, but also all other joints. This makes my work more than obsolete ;-)
Reference:I. Oikonomidis, N. Kyriazis and A.A. Argyros, "Efficient model-based 3D tracking of hand articulations using Kinect", to appear in Proceedings of the 22nd British Machine Vision Conference, BMVC 2011, University of Dundee, UK, Aug. 29-Sep. 1, 2011.
Here's the link to the project site (more videos)