One important aspect of an interface is its ease of use. A vision-based interface allows the user to interact with virtual environments and information without the need for encumbering sensors, goggles and the associated umbilicus of wires. Christopher R. Wren, Ali J. Azarbayejani, Trevor J. Darrell, Alex P. Pentland (and others) have built a system called pfinder that implements such an interface.
Pfinder embodies the fundamental vision technology behind the Smartroom project, a test-bed for exploring such
unencumbered interfaces. It runs in real-time on a
standard SGI Indy, and has performed reliably on literally thousands of
people in many different physical locations.
The ability to track people in real time is also important for a wide range of applications beyond Human-Computer Interface: very-low bandwidth image coding (for telepresence), automated security systems, and visually guided devices are just a few example applications.
The first step in tracking the user is determining what part of
the input image is the user. We accomplish this using a Likelihood Ratio Test framework combined with
robust estimation techniques..
Once pfinder finds the user, there are several
ways to model them.
For a more formal treatment of this topic, download the publication that talks about the concepts, and math behind pfinder. Or look at an html version of the same report.
Several people have worked on pfinder over the years. A list of contributors can be found in the SigGraph contributors list.