With the Internet of Things slated to have tens of billions of connected devices by 2020, one of the most crucial design considerations for internet-connected products is figuring out how to seamlessly integrate these devices into everyday life. In this respect, teaching machines how to identify the individuals they are interacting with is paramount—it will allow for the total personalization of everything that is promised by the IoT. Rather than just having internet-connected light bulbs and refrigerators that are sitting around waiting to get hacked, these devices will be able to recognize you and interface with you according to your preferences (something that devices like the Xbox One are already doing via facial recognition).
So far there have been a number of proposed methods for integrating human identification into smart objects, ranging from the creepy and invasive (think RFID chip implants or facial recognition) to the limited and cumbersome (like fingerprint scanners). In the quest for a non-invasive yet ubiquitous mode of human identification, a team of researchers from Northwestern Polytechnic University figured out a way to use WiFi signals to ID individuals moving around in a room—with an ID accuracy upwards of 90 percent.
As the team detailed in a paper posted to arXiv earlier this month, their novel approach to human identification—which they’re calling FreeSense—uses interruptions in WiFi waves to identify individuals based on body shape and motion patterns. This is accomplished by monitoring changes in the WiFi’s channel state information (CSI), which is a fancy way of saying the fine-grained data about how a WiFi wave is propagating in a given space.
“Due to the difference of body shapes and motion patterns, each person can have specific influence patterns on surrounding WIFI signals while she moves indoors, generating a unique pattern on the CSI time series of the WIFI device,” the team writes in its report. “FreeSense…is nonintrusive and privacy-preserving compared with existing methods [of human identification].”
WiFi channel state information has already been successfully deployed as a motion sensor in other contexts, such as detecting when someone in a room has fallen or hearing what they are saying when they speak. Prior to this new research, none of these applications have been able to leverage WiFi CSI to identify individuals, however.
To test their new methods, the team members used a normal laptop and off-the-shelf WiFi router in a 30 square meter “smart home environment,” complete with typical home furnishings. The enlisted nine volunteers to function as a representative family that might be operating in this smart home, with the goal of using WiFi CSI to identify these individuals as they navigated the space.
To begin with, the researchers trained their system to classify individuals based on 20 samples of them walking across the space in a straight line. Once this baseline was established, the individuals then navigated the space an additional 20 times to test the system.
When all nine individuals were testing the system (one person in the room at a time), it was able to achieve about a 75 percent accuracy in identifying them; when it was just two individuals, FreeSense was able to identify them nearly 95 percent of the time. The reason for the difference is simple: the more individuals with similar body types or motion patterns you have in the system, the trickier it is to identify them. Still, the team found that the ideal number of people that can be identified by their system is somewhere between 2-6, which would capture the range of most nuclear families—with 6 people moving in the system, they were still able to achieve an 88 percent ID accuracy.
Now that the FreeSense proof-of-principle has been successfully demonstrated, the researchers hope to improve upon the design by testing it when multiple people are in the room at once and seeing how increasing the distance between the WiFi receiver and transmitter affects recognition accuracy.
