This involves background modelling (being able to "ignore" the background), foreground modelling (being able to extract some features of the foreground objects - how they are "human-like" etc.), and tracking (being able to relate one object in one scene to the the same object in another scene in a different position and pose).<p>

See <a href="http://users.monash.edu.au/~hanzi"/>Hanzi's Homepage</a><p>