GazeTags Give Smart Wearables a New Way to See

Smart glasses and other wearables are getting better at understanding the world around us, but there's a catch. Most rely on cameras and computer vision to recognize objects, which can quickly drain their batteries and also raises privacy concerns. Researchers at Carnegie Mellon University have come up with an alternative that lets these devices recognize and interact with everyday objects without ever needing to take a picture.

The technology is called GazeTags, and it works by attaching small electronic tags to objects around the home or office. Each tag costs just over $2 to make and can run for more than a year on a single coin cell battery. When paired with compatible smart glasses or earbuds, these tags can tell an AI assistant what objects a person is facing and provide useful information about them.

The trick is to use infrared light instead of cameras. A wearable device periodically sends out a narrow beam of infrared light in the direction the user is facing. Any GazeTags caught in that beam wake up and respond with their identity, along with any additional information they have been programmed to provide. In this way, the system can figure out what someone is paying attention to without the processing power normally needed for computer vision.

To demonstrate the idea, the researchers built prototype smart glasses and earbuds using an ESP8285 SoC, an infrared receiver, and a 940 nm infrared LED. The beam covers an angle of approximately 40 degrees, which is narrow enough to provide a useful indication of where someone is looking.

Each GazeTag measures 30 mm across and 18 mm thick, with an ATtiny85 microcontroller, infrared transmitter and receiver, and CR2477 coin cell battery packed inside. To conserve energy, the tags spend most of their time sleeping, waking briefly to listen for incoming signals. This gives them an estimated battery life of about 1.4 years.

So what can you actually do with them? The researchers showed off several possibilities. A tag attached to a potted plant can report its soil moisture level, allowing an AI assistant to remind you when it's time for watering. Another attached to a kettle can tell you how hot the water is, while one on a fire extinguisher can provide its last service date. You could even look at a lamp and say "turn off" to switch it off.

The team put GazeTags through a series of tests to see how well they work. With five tags grouped closely together, the system successfully identified all of them in 99.75% of trials at distances of four and six meters. Performance dropped at longer distances, especially when more tags were added.

GazeTags aren't a perfect replacement for computer vision. They can only detect the direction someone's head is facing, not exactly where their eyes are looking, and objects need to be equipped with tags beforehand. Still, for just a couple of dollars per object, this could be a practical way to make AI wearables more useful without constantly running cameras or worrying about what they're recording.