快猫短视频

Prototype smart glasses can track people鈥檚 eye movements using a technique similar to sonar, an approach that uses 95 per cent less power than other methods.

Virtual reality headsets like Apple Vision Pro听 and Meta Quest Pro use cameras pointing towards the wearer鈥檚 face to track their eye movement. This offers high accuracy, typically estimating the angle of their gaze to within 1 degree, but also consumes a lot of power.

听

at Cornell University in New York and his colleagues created a system called GazeTrak听that tracks eye movements with sound, doing away with the need for cameras entirely.

Tiny speakers, one for each eye, broadcast sound at a frequency above 18 kilohertz 鈥� outside the range of most humans鈥� hearing 鈥� towards the face, which is reflected and picked up by four microphones on each side of the glasses.

Free newsletter

Sign up to The Daily

Sign up to newsletter

An AI model can then interpret these sounds to unpick which direction the wearer鈥檚 eyes are pointing, because of inconsistencies in the shape of their eyeballs and eyelids.

鈥淲hen you move your eyes, your reflected signal will be different,鈥� says Li. 鈥淪o by analysing the differences between the signals, we can track your gaze movement. It鈥檚 like in sonar systems.鈥�

In tests with 20 participants, the glasses managed to track people鈥檚 gaze with an error rate of just 3.6 degrees. This is less accurate than current commercial devices, but accurate enough for most applications in virtual reality, gaming or user interfaces, say the researchers.

鈥淭here鈥檚 still a gap. But I think we also have some room to improve the accuracy,鈥� says Li. 鈥淲e鈥檒l certainly do some follow-up work to improve the performance.鈥�

People who are blind can navigate indoors with a phone in their pocket

Two wayfinding apps use motion sensors and AI to help people who are blind navigate a building, without needing to hold their phone out in front of them and risk theft

Apart from using less power, the technique offers more privacy, as some users may not feel comfortable having a camera continually filming them just to track eye movement, the researchers say. They also expect it to cost less and add less weight to devices.

Because each person鈥檚 eyeballs vary in shape, the AI model used by GazeTrak currently needs to be trained for each individual user. But the researchers hope that if they collect enough data on different users, they can create a universal model that would work for anyone 鈥� which would be key for a commercial product.