It may help though that like LA Noire, Quantic likely will limit themselves to actual faces of their actors, which may help as then perhaps more can be automatically detected/mapped.
Yes, that can be done, just as in Heavy Rain. Which also raises the question why they wouldn't go with team Bondi's tech... the explanation might be that they want to capture body and face at the same time.
Obviously, Cage also didn't say they were at Avatar quality now, but thought they could be in a few years, and I don't think that is totally unrealistic, if, as you indicate, you make a bit of a rough comparison overlooking some of the 'details'.
It
is a completely unrealistic claim. Facial capture is a very complex issue with multiple aspects and it can't be compared to traditional motion capture.
With body movement, you read the pose of the limbs, and in some rare cases the muscle or tendon tensions add an extra layer to that. But the primary motion you want to capture is what's under the skin, the rotations of the joints. It's still not totally straightforward, because CG characters rarely share the proportions and build of the performer and a lot of remapping and editing is involved. But it's very hard to notice small problems - if the general dynamics of the movement are right and there aren't any obvious problems like intersections or feet sliding on the ground then we accept it as realistic human movement.
The problem with faces is that the underlying motion of the jaw joint is just a part of the entire result. We care a lot more about the deformation of the surface, the folds, wrinkles, the shape of the mouth, whether we see teeth or the white of the eyes, and so on. This is the
primary component of the facial performance, and even if there's a 100% match between the actor and the CG character, it is still not enough to place markers on the surface and capture what happens.
So either you capture the entire surface, including things like the eyes with the bulge of the cornea, and the inner mouth and tongue and everything - or you will have to reproduce every square millimeter's behavior in the computer somehow. LA Noire choses to capture everything, at the cost of losing sync with the body capture, and losing the ability to edit the facial performance in any reasonable way. Others try to somehow capture and translate the deformations from the actor's face to a CG character's facial rig.
This is the part where it gets complicated because you can't see the inner mouth, and you can't track every square millimeter unless you're using dozens of highres cameras, so you need an animator to create the missing movement and finetune the rest. And you need a rigger to reproduce the surface deformations, the tiny ticks and the large skin movements and folds and so on. And this takes a LOT of time, research, money and so on.
I've just attended some talks on CG characters in movies and thus I can give a few exact figures.
Dobby and Kreacher in HP7 took 60 people and 8 months for that few minutes; of that, more than 6 man-months were spent just on creating Dobby's facial model and deformation rig. Granted, there was no capture involved for the animation and that's what took most of the team to create.
Clu in Tron Legacy took a similar effort, two people spent more than 6 months just on building the head and preparing it for animation. Capture using 4 cameras took them 50-80% of the way and the rest was keyframe animation. They also had a lot of trouble to compensate for the differences in the body performance of John Reardon and the facial performance of Jeff Bridges.
In both cases very large teams of animators were involved with creating the final performances, just as with Avatar.
I believe this should already should prove both of my points - that games can't compete with the scale of the effort put behind these realistic characters, and that performance capture on its own won't replace animators and thus it might not be worth the effort. For example, is Crysis 2's facial animation any better than UC2 or 3? Because they are using facial capture, and yet I personally prefer UC's hand animated stuff.