Sure, there's some correlation, but a correlation can just mean that if one's getting better, the other probably is too. Just knowing that a correlation exists doesn't help us much.

The reason why self-driving cars could be almost perfect even if face recognition still had problems is that self-driving cars don't need to have great detection rates on people - it's enough to know where the road is and where other stuff is, the nature of that other stuff is only of secondary concern. To find out where stuff is, self-driving cars don't have to use still images of the environment - they can use things like multiple cameras, fancy laser range-finders, and motion parallax.