23 July, 2018
Using WiFi, a team at Computer Science and Artificial Intelligence Laboratory (CSAIL) is now able to “see” a person on the other side of a wall and precisely track their movements, even if it’s something as subtle as giving a high five, according to new research to be presented at the Conference on Computer Vision and Pattern Recognition next week.
“People have been trying to detect people through walls since the 70s,” said Dina Katabi, a professor at CSAIL and the lead researcher on the project. “Around 2013, we showed that we can track people accurately. What’s new here is that, for the first time, we can create a dynamic skeleton of the person, their posture, and how they’re moving.”
A data set to train the AI didn’t already exist, so the team had to make one by manually creating stick figures of human movements based on images they captured through both the wireless device and a camera. By showing the figures to the neural network along with the WiFi signals, the network was able to learn which radio signals were made by each movement.
“Let’s say the police want to use such a device to see behind a wall,” Katabi explained. “It’s very important to know if somebody is standing in a position that indicates they are holding a weapon, for example. All of that you can’t do with just a blob.”
“Particularly in the current climate, this is an important question,” Katabi said. “We have developed mechanisms to block the use of the technology, and it anonymizes and encrypts the data. And then there is a role of policy that protects the user and doesn’t stifle innovation.”
Privacy is always going to be a concern with new technology, especially something that can see through walls, but it also comes with many possible benefits.