A Czech student from the British University of Surrey Zdenek Kalal developed the Tracking-Learning-Detection (aka Predator) algorithm for tracking objects in a self-learning video stream (recognition accuracy improves with each frame).
The program works fine in a single Intel Core 2 Duo 2.4 GHz stream, 2 GB of RAM.
On the page Zdenek on Youtube you can find other videos.
The author says that he was going to publish the source codes of the project, but when he received hundreds of letters from interested parties asking him to do this, he changed his mind. It turned out that this project is much more interesting than it was originally intended, so there was a chance to earn something.
Such algorithms can be used in tracking systems (for example, for unmanned vehicles), computer and gaming interfaces (virtual mouse), in photo and video cameras, etc.
In fact, this technology is hardly as unique as the author believes. As far as I know, Russian developers from Intel R & D in Nizhny Novgorod are doing something similar (at one of the Intel conferences, they showed a demo of their system).
UPD 04/05/2011 12:23.The article had an effect and Russian developers also began to move.This alogirtm from German Bukharov does not look so impressive, but it can even work on smartphones.