An accurate fit of a semantic model to the actual scene is of crucial
importance for correct reconstruction of the motion in a videophone sc
ene. A new, fast and efficient algorithm for automatic wire-frame fitt
ing of the important facial features is proposed.