"In tests involving 197 participants, the researchers said the system identified individuals with nearly 100% accuracy. The recognition remained effective regardless of viewing angle or how the participants walked."
These things always have carefully controlled setups. If you read the article, the participants had to wear non-baggy clothes and walk past the routers in a prescribed way (although they did at least have them do a couple of different walk styles, e.g. carrying some water).
Perhaps more importantly, they train the model on known people doing the same normal walk past the routers 16 times.
It's definitely impressively accurate but still quite far from the real world (and the authors acknowledge that).
Perhaps more importantly, they train the model on known people doing the same normal walk past the routers 16 times.
It's definitely impressively accurate but still quite far from the real world (and the authors acknowledge that).