MIT Researchers Use Radio Waves to Peer Through Walls, Track Gestures
Mapping radio-based detection to a virtual skeleton lets researchers track people behind walls and in complete darkness.
Researchers at the Massachusetts Institute of Technology (MIT) have developed a means of monitoring not only the position of humans stood behind walls but also tracking their gestures, from waving to answering a phone call, using a combination of radio- and vision-based tracking algorithms.
The idea that people can be tracked behind closed doors by monitoring how their bodies affect radio signals isn't new, but the system developed and demonstrated by the MIT team goes a stage further: Data from the radio signals is mapped onto a simple skeleton model, which can then make use of vision-based data sets to allow a machine learning algorithm to track gestures based on the skeleton's movement.
"We introduce a neural network model that can detect human actions through walls and occlusions, and in poor lighting conditions," researcher Tianhong Li explains in a post-publication interview with MIT Technology Review on the project. "By translating the input to an intermediate skeleton-based representation, our model can learn from both vision-based and radio frequency-based datasets, and allow the two tasks to help each other."
The resulting system is capable of recognising a range of gestures which would previously have taken vision-based data to detect, working in complete darkness or even through walls. The result is claimed to be equivalent to a vision-based system in accuracy, detecting everything from playing a game on a smartphone to attacking another occupant.
Where the system falls down compared to a vision-based equivalent, of course, is that there's no real way to identify an individual from their radio-generated skeleton data — unlike a vision-based system, which could use facial recognition. Li, however, positions this as a potential positive, pointing out that the system could be used to track an occupant while assuring their privacy in scenarios such as monitoring an individual at risk of falls in their home.
A more detailed write-up is available on MIT Technology Review, while the paper itself can be found on arXiv.org.