This article originally appeared in TidBITS on 2013-06-06 at 10:45 a.m.
The permanent URL for this article is:
Include images: Off

Researchers Demo Ubiquitous Gestural Control via Wi-Fi Sensing

by Adam C. Engst

We’ve become accustomed to tapping and swiping our smartphones, but what if you could control devices in your home or office with gestures made from wherever you’re standing? WiSee [1], a technology developed by University of Washington computer scientists, can recognize gestures by analyzing how specific body motions disrupt Wi-Fi signals — no special sensors or cameras are necessary. The overall effect is a lot like the Xbox Kinect [2], which relies on cameras and thus works only in a single room.

“This is repurposing wireless signals that already exist in new ways,” said lead researcher Shyam Gollakota, a UW assistant professor of computer science and engineering. “You can actually use wireless for gesture recognition without needing to deploy more sensors.”

The UW researchers created a special receiver that listens to the Wi-Fi signals from all the devices in the home, and looks at the minute Doppler shifts and multi-path distortions that result from human movement within the environment. Using the MIMO (multiple-input and multiple-output) technology inherent to 802.11n, WiSee can differentiate among up to five people within the space. Since you wouldn’t want random movements to be interpreted as gestures, WiSee requires a trigger gesture, after which it watches only that person, presumably until a timer elapses or someone else takes control. The trigger gesture could also serve as a password of sorts. At the moment, only a single device can be controlled at a time, though the researchers are looking at how to control multiple devices simultaneously.

In testing in a two-bedroom apartment and an office environment, WiSee was able to identify nine whole-body gestures with an average accuracy rate of 94 percent (the accuracy rate for random guesses is only 11 percent). False positives — unintentional movements that were interpreted as gestures — happened roughly 2.5 times per hour when two gestural repetitions were required; increasing the necessary repetitions to four essentially eliminated false positives (0.07 events per hour). See “Whole-Home Gesture Recognition Using Wireless Signals [3]” (PDF) for the full paper.

Although WiSee’s proof-of-concept requires a special receiver, the researchers say that the necessary technology could be embedded in consumer-level wireless access points. Also, the paper and demo video [4] say nothing about how WiSee controls things like music volume, television station, and the like — although such hand-waving (hah!) is understandable in a proof-of-concept, a real-world system would have to come up with a coherent method of controlling a potentially wide variety of devices and applications. That’s not impossible, but it is a non-trivial problem.