This video provides other suggestions gets good at around 1:15 ;)

https://www.youtube.com/watch?v=wVSOHGGgKD4

Is there support for this idea? I've seen it in plenty of novels, but I'd want some stats on how it affects a known, relatively modern, deployed gait recognition system.

You could compare gaits between footage of a crime and footage of you in another public place, probably?

I don't think I've heard of it being used though.

We only have discussions of the Chinese rolling out gait tracking widely. Basically you use existing facial databases to match ids to people in observed areas and capture their gait as they pass observed areas. Then it goes into the database. Using partial matching (non ideal observation of gait or face) allows for greater positive matching in non-ideal circumstances.

DARPA projects from more than a decade ago (VSAM/WAMI for arial platforms like Gorgon Stare) used arial imagery to capture ground shadows for gait tracking purposes.

From chatting with some of the researchers many years ago my understanding is that it usually wasn't accurate enough for unique identification and the gait shadow was dependent on shoe type and clothing, so a persistent gait shadow database wouldn't have been useful. But it could be correlated with ground-based surveillance for identification, for example person A and B were identified on a ground-based security camera entering a building, then gait tracking could be used to monitor where they went after they left the building even if they avoided ground-based security cameras after that point.

Most people carry an accelerometer-equipped smartphone around in their pocket or bag, and there is already precedent for that data being collected and transmitted without consent: https://research.google/blog/android-earthquake-alerts-a-glo...