> Meaning, to deploy this tech, you need access to the space first to train this model.
Government secret services have gone and re-built the houses of high-profile raid targets before. And it's only a matter of time until simulation technology becomes powerful enough to do a "good enough" digital recreation from blueprints.
> multiple people moving was not evaluated by this tech/probably harder to achieve
At least one scenario in the article mentions "a room with several people". People moving around is just a question of speeding up data collection and analysis.
> With that in mind, I think privacy-invasion capability of this technology is exaggerated by some comments here.
The key thing is, what is public research grade now, is likely already a developed asset in government toolkits. And now that the general public has access to such technology, it will - more likely than not, given it can be done on 50$ low-cost devices - be commoditized, particularly where there is a financial interest in tracking people. I think a good candidate will be supermarkets and similar stores - at least in Europe, stuff like running analytics on surveillance cameras is pretty frowned upon under GDPR, but something like wifi-based tracking should be relatively unproblematic.