A piece of glass with tiny little bumps on it can be used to identify objects. The 鈥渟mart glass鈥� could eventually be more compact than using cameras and computers to achieve the same aim.
Machine learning algorithms are becoming good at identifying objects, but using them usually requires a camera and a computer. While this combination fits inside your phone, at Columbia University in New York and his colleagues wanted to make an even smaller object recognition device.
To eliminate the need for a camera, they designed a device to identify objects based on how they scatter near-infrared light.
Advertisement
The team first chose a set of handwritten digits printed on a plastic sheet for the device to recognise. The researchers then created a bespoke piece of glass with tiny bumps. The bumps were designed so that they would manipulate near-infrared light passing through the glass in such a way as to produce a unique pattern. The device was then completed with a detector for spotting the pattern.
Light takes about 0.1 nanoseconds to make it from a digit to being recorded on the detector, and the only parts of the device that need a power supply are the detector and the light source. In tests, the device recognised handwritten digits and whether they were italicised with 90 per cent accuracy.
The researchers also used a computer to simulate a device made of two layers of the glass and found it could recognise 100 illuminated human faces, simulated as greyscale images, with an accuracy of about 80 per cent.
at the Massachusetts Institute of Technology says that the new design could have advantages over more conventional technology because it doesn鈥檛 have to use a camera, which needs time and energy. The device could also plausibly be tweaked to work for visible light bouncing off objects, he says.
However, the new device cannot yet distinguish a set of objects complex enough to make it useful, says at the University of Washington in Seattle.
Yu says the team is working to improve the device鈥檚 object recognition capabilities without making it bulkier in the hope that it could be added to credit cards for facial recognition as a security measure or to small medical devices that identify objects inside of the body.
Reference:
