This somewhat depends on if you want to track the head as a static or deforming object. Static would mean you could matchmove markers on non-deforming parts (actor should not be grimacing) from just one sequence as usual but choose move scene instead of move camera on export. This way you would get the basic orientation of the skull as your solve.
If you need to conform to a deforming face with facial expressions changing, at least two witness cams additional to the main camera should be used. MM will do the markers on the face as a mocap track, getting depth information from the difference in perspective of the cams in use.
I take it you also read this thread on a similar subject.
I meant to post a link to examples on HCW in my post there, but the link seems to be missing.
I did manage to do this in Max, but it did not work as it should, I had to use a workaround.