Camera AI identifies objects at light rate

The team at Princeton University and Washington University developed a compact camera prototype for computer visual technology – a form of AI helping the computer identify objects in photos and videos, Interesting Engineering On 3/3 reported.

The camera prototype shows a new approach to computer vision, based on light instead of electricity, thereby consuming much less energy than traditional computers, while identifying objects at light rate. New research published in magazines Science Advances.

According to Felix Heide, Associate Professor of Computer Science at the University of Princeton, a member of the research group, the idea of ​​the camera formed when he started to learn super surface – artificial material -shaped material with close -up characteristics. With a unique geometric structure, the super surface does not bend the light through the glass or plastic like the traditional lens. Instead, they make light diffraction around the tiny structures, like the way the light spreads through a narrow slot.

Heide’s group cooperates with experts from Nano Washington to design camera and chip manufacturing. Instead of traditional glass or plastic lenses, they use 50 super flat, lightweight lenses, using extremely small nano structures to control light.

These super lenses act as optical nerve network, a AI system inspired by the brain, and bring great benefits. Extremely fast camera, can identify and classify images 200 times faster than traditional artificial neural networks, and also very energy -saving because of light that comes in instead of electricity.

Next, the research team developed a system capable of identifying objects in the image that only used less than 1% of the calculated capacity compared to the old methods. Super surface lens processing up to 99.4% of work volume.

This system introduces a new model, performing hundreds of millions of instantaneous calculations. While traditional artificial nerve network applies mathematical filters to extract data, requires multiple calculations even with a few pixels, a new model to perform complex filtration process when light passes, allowing a few large filters to analyze the entire image at the same time.

“The research really has many widespread applications, from self -driving cars, self -driving trucks, robots to medical and smartphones. Today, every iPhone is equipped with AI or visual technology. This building is still in the early stage, but in the future, all of the above applications can benefit from what we are growing,” Heide said.