Facial recognition works primarily on mapping the distances between key features, kind of like a bitmask. The more features you have the longer the bitmask and the fewer matches.
It's not about being "un-matchable" it's about having so few data points you match with enough people that whoever's deciding what "good enough" is for the application has set the bar above how ever good your match is.
For something like sending advertising the bar is super low, who cares if you send out some extra spam the cost of doing so is low. The advertisers don't care if they can't precisely target the mystery man because gaining or losing one lead isn't important when you have thousands. They'll either just not send anything or send something to every possible match.
For something like insurance and credit the bar is higher because if you get the risk wrong then you lose profit margin or the ability to price competitively. This is the one area where the mystery man can win. The people selling data to the creditors can't just set the "hangs out with poor people" bool to true for each big group of people that is a partial match with a mystery person. That approach doesn't scale.
For something like IDing the guy who got a good look at the guy who placed the suitcase that had the bomb the bar is low because while the cost of interviewing a bunch of people is high the available resources are high.
If you want to stay under the radar of the analytics companies then license plate scanning should worry you a lot more than facial recognition (in the US at least).
Can you elaborate on this?
Not really, it's a classic example (at least, one I've seen several times) to show recognition even when obscured.
That said, I just searched for an example, and found this paper [0] claiming highly effective tortoiseshell glasses to block recognition or deceive to be another. Though it certainly doesn't satisfy your other requirement of:
> not making me look somewhat silly
[0] - https://www.cs.cmu.edu/~sbhagava/papers/face-rec-ccs16.pdf
The real point is that: I believe there are some images in the search space that any recognition algorithm will not see a face in, but which your brain will and vice versa.