If you have the coordinates of a person it should just be a matter of an approx constant z-offset? That is, unless the person is lying down, where you would need to use his view direction. Of course, there are different approaches, you could get coordinates of models with face textures and so on as well.