Sort of. I deal with realtime facial motion capturing. The first step of my work is easy: finding a region of interest, which is actually in my case the face, obviously.

I'm currently testing RGB, HSV and LAB color spaces about which is suited best to assign each pixel a probability if it is a skin pixel (research showed HSV is the best). Then I group 'em and together with a motion and a geometry cue I hopefully get the face bounding box.