I don't remember all the details for sure sure but I read some research a while back about using some form of radar "thing" to identify people using a details like gait, ear shape and size, neck length and shape, upper back shape and width, arm length and a few other things that I really can't remember.
I think gait was the main factor along with height and then rest were used as filters.
Edit: Forgot to mention this was pretty crude 10 years ago. I'm assuming op is talking about a practical application for something similar.
Yup, exactly. Gait is the main thing I've seen too. But there are a whole grab bag of papers about these applications with different methodologies. Stuff like topological data analysis and machine learning are much more fleshed out these days, so these applications don't need to rely as much on rigid models to be accurate. Could be that they're using subtle, noisy information from breathing patterns, for example.
"Device free", "localization", "identification" and "mm wave antenna arrays" are some useful keywords for checking this stuff out on Google scholar. Some other alternatives for "device free" are "passive" and "adversarial". I think I remember one paper where they were identifying dozens of people at once with like 95%+ accuracy. I could be getting the details wrong on that one but it was along those lines.
For the high level overview though, you can just read the 5G industry whitepapers and it pretty clearly spells out that there are privacy implications and I'm sure that these kinds of applications are exactly why.
Remember, higher frequency = more information density = finer resolution. Which is also why these radio waves can't travel through as much stuff... They simply interact with more; more stuff is opaque to them, and as such they carry information about more interactions in their image.
