Source: The Atlantic
City dwellers spend nearly every moment of every day awash in Wi-Fi signals. Homes, streets, businesses, and office buildings are constantly blasting wireless signals every which way for the benefit of nearby phones, tablets, laptops, wearables, and other connected paraphernalia.
When those devices connect to a router, they send requests for information—a weather forecast, the latest sports scores, a news article—and, in turn, receive that data, all over the air. As it communicates with the devices, the router is also gathering information about how its signals are traveling through the air, and whether they’re being disrupted by obstacles or interference. With that data, the router can make small adjustments to communicate more reliably with the devices it’s connected to.
But it can also be used to monitor humans—and in surprisingly detailed ways.
As people move through a space with a Wi-Fi signal, their bodies affect it, absorbing some waves and reflecting others in various directions. By analyzing the exact ways that a Wi-Fi signal is altered when a human moves through it, researchers can “see” what someone writes with their finger in the air, identify a particular person by the way that they walk, and even read a person’s lips with startling accuracy—in some cases even if a router isn’t in the same room as the person performing the actions.
Several recent experiments have focused on using Wi-Fi signals to identify people, either based on their body shape or the specific way they tend to move. Earlier this month, a group of computer-science researchers at Northwestern Polytechnical University in China posted a paper to an online archive of scientific research, detailing a system that can accurately identify humans as they walk through a door nine times out of ten.
The system must first be trained: It has to learn individuals’ body shapes so that it can identify them later. After memorizing body shapes, the system, which the researchers named FreeSense, watches for people walking across its line of sight. If it’s told that the next passerby will be one of two people, the system can correctly identify which it is 95 percent of the time. If it’s choosing between six people, it identifies the right one 89 percent of the time.