This is an implementation of the face tracking algorithm used in OpenCV, though it's from scratch rather than a port. The base algorithm is described by Viola and Jones. The tracking algorithm is my own idea. Basically it's the Viola/Jones algorithm restricted to a window in searchspace near the previous found window. At the moment, tracking only works in single face mode.
My nature of the algorithm, this is not restricted to face finding. The actual smarts of how to find things are encoded in the haar classifier cascades, loaded from xml files. The same haar classifier xml files used in OpenCV. At the moment, tilted features are not supported, and neither are trees. Tree support is coming soon.
These xml files are very large (the one used in this demo is 923k, be patient). However, they compress very well, to < 100k. I'll implement a FAR loader soon.