The Sapienza computer scientists say Wi-Fi signals offer superior surveillance potential compared to cameras because they're not affected by light conditions, can penetrate walls and other obstacles, and they're more privacy-preserving than visual images.

[…] The Rome-based researchers who proposed WhoFi claim their technique makes accurate matches on the public NTU-Fi dataset up to 95.5 percent of the time when the deep neural network uses the transformer encoding architecture.

you are viewing a single comment's thread
view the rest of the comments
[–] 42 points 1 year ago (12 children)

they're more privacy-preserving than visual images.

hhhhwat. How can they identify you and also be privacy preserving? 🤔

  • source
  • hideshow 12 child comments
  • [–] 2 points 1 year ago (2 children)

    They know you are a person and they can call your a certain UUID, but there will be a hard time matching you to your name etc.

    Camera's can do face recognition (if your face is even in the database) to know who you are.

    This only works until the point where they have your form in a database which they can check...

  • source
  • parent
  • hideshow 2 child comments
  • [–] 5 points 1 year ago (1 child)

    We have heard this non-sense before, only to find it's trivially easy to connect to your PID.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 1 point 1 year ago

    Never said that it wasn’t easy, it’s just harder than with facial recognition. In theory you could do it correctly in a way that it isn’t indentifiable.

    Also this works in places where faces are protected

  • source
  • parent
  • [–] 1 point 1 year ago (4 children)

    They can see you're a person but not exactly who you are.

  • source
  • parent
  • hideshow 4 child comments
  • [–] -1 points 1 year ago (3 children)
  • [–] 2 points 1 year ago* (2 children)

    Well they can identify you are the same person but not your identity.. So it's like a disenbodied fingerprint.

    I suppose they could potentially make some database and train an AI on it someday to match to actual identities, but usefulness would be pretty limited at only 95% accuracy. That's a false reading 1/20 times, so I suspect it would fail bigly to accurately recognize people from large data sets.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 1 point 1 year ago

    I'd imagine it's like online advertisers: they convert your fingerprint to a token to try to sell you shit, but they allegedly don't know who exactly you are or where you go. So visiting animatedllamaporn.com is still your little secret...

  • source
  • parent