3D Facial Animation

Transcribed from the printed issue; not yet reviewed. Read it in the PDF of the issue, page 60

I really can’t do justice to this one; you just have to see the videos to get a proper grasp of what Keith was driving at. The particular problem he was attacking was the difficulty of achieving realistic ‘lip-synch’ in animations such as the ITV series ‘Spitting Image’.

Some of the important characteristics of the face relevant to animation are:

• 14 bones and 100 muscles.
• the positioning and spacing of 20 key elements.
• ‘micro expressions’ lasting 1/40th of a second or less, which are missed by 24 frames per sec animation.
• the more realistic the face, the more critical your viewers!

The classic approach to the problem is to take 3D photographs of a face (on which a polygonal pattern has been drawn), and digitize these with a wire-frame diagram. The problem with this is the enormous amount of data needed for the ‘in betweening’ between snapshots; the computer run times for even very brief animations were quite massive.

Keith had started down this track, but was now exploring a quite different approach: using a dynamic ‘muscle model’ to build expressions analytically rather than from a vast database of stored positions. The animations done with this were quite fascinating; by tensing just two muscles in the cheeks (and computing the effects on the affected facial surface) he generated a remarkably convincing smile. Because no interpolation was needed (the step size could be as fine as he liked) the smile genuinely ‘flickered’ in just the way a real smile can.

In short I really think Keith is on to something. Whether it is relevant to APL systems this side of the millenium is another question; I suspect it is, in the kind of ‘right brain’ way that Phil Smith was showing us at APL-86. Thank God the equipment worked for this one!!