Thanks to research from Stanford University, a new type of camera system using artificial intelligence (AI) offers faster image classification and energy efficiency.
Autonomous cars and systems that use image recognition technology use AI-enhanced computers that self-learn object recognition. At the moment, these computers are too slow and large for future applications.
“That autonomous car you just passed has a relatively huge, relatively slow, energy intensive computer in its trunk,” says Gordon Wetzstein, Stanford assistant professor of electrical engineering, and leader of the research.
According to Wetzstein, next-gen tech requires a fast and small system to process image data. The Stanford team recently addressed this problem by mixing two types of computers into one concise design. The result was a hybrid optical-electrical computer primed as an image analysis expert.
The prototype camera’s first layer is an optical computer that doesn’t need to suffer from the power drain of digital computing’s mathematics calculations. The computer physically preprocesses the image data, and filters it naturally as light passes through the optics. Layer one has zero input power, which saves time, energy, and calculations.
“We’ve outsourced some of the math of artificial intelligence into the optics,” says Julie Chang, Standford graduate student.
A digital electronics computer sits at the second layer. Since preprocessing was already handled, it has fewer steps to complete.
“Millions of calculations are circumvented and it all happens at the speed of light,” says Wetzstein.
Even though the current prototype is large, the researchers believe the tech can one day fit inside an aerial drone or handheld device. Experiments have shown accurate and fast idenficiation of objects in natural environments, such as airplanes, animals, and vehicles.
“Some future version of our system would be especially useful in rapid decision-making applications, like autonomous vehicles,” says Wetzstein.
To learn more, read the article, “A convex 3D deconvolution algorithm for low photon count fluorescence imaging,” published in Scientific Reports.
