Surpassed the usual si: created a “smart” camera of a new type of size less than 1 mm (video)

< IMG SRC = "/Uploads/Blogs/64/88/IB-FQB9B0QP_9316240d.jpg" Alt = "Outrded the usual shi: created & quot; quot; smart & quot; quot; quot; quot; quot; quot; quot; quot; quot; quot; quot; quot; quot; quot; < P > According to scientists, a new camera with machine vision recognizes objects at a speed of light and also uses light instead of electricity.

< p > Researchers at the University of Princeton and Washington University made a breakthrough, developing a new type of camera with computer vision & mdash; type of artificial intelligence (AI) that helps computers recognize objects in photos and videos. About this & nbsp; writes & nbsp; Interesting Engineering portal.

< p >< iframe width = "560" Height = "314" SRC = "//www.youtube.com/embed/lfgzagpwhv4" Alowfullscreen = "ALLOWFULLSCREN" >0 ~/IFrame >0 ~/P > < p >

< P > It is reported that the chamber of the size of a grain of salt (less than half a millimeter wide) can make clear, full -color images with unprecedented detail, as well as & nbsp; recognizes objects at speed of light. Thus the device uses light instead of electricity.

< P > “This is a completely new way of thinking about optics, which is very different from traditional optics. This is a through -projection, where optics are operating in conjunction with the computing block”, & mdash; The Doctor of Philosophy, Professor of Electrical Engineering, Computer Engineering and Physics at Washington University and one of the authors of the study of Arch Madzhumdar.

< p >In his turn, researcher Felix Heide, noted that the idea was formed when he began to explore the meta -terminal & mdash; artificial leaf -shaped materials with subvivale characteristics.

< p > Metapover does not refract light through glass or plastic like traditional lenses. Instead, they refract the light around tiny structures, just as light spreads when passing through a narrow gap.

< p > to design the camera and make Chip Hide and his students combined efforts with the experts of the Washington Nanotechnology Laboratory. Instead of using a traditional glass or plastic lens, the team integrated 50 folded flat and light metal metals that use microscopic nanostructures for light control. These metalins duplicate the optical neural network, the AI ​​system inspired by the brain.

< p >Exploring this technology, scientists have found that the light coming out of the pillars arrays did not have to resemble the original image. Instead, the columns acted as specialized filters, sorting optical data by categories, such as edges, light and dark areas, or even features that go beyond human perception, thereby giving computers pre-processed, structured information.

~ ~ < p > “We have realized that we do not need to record the perfect image. We can only record certain characteristics that can then be combined to accomplish tasks such as classification”, & mdash; noted Felix Gide.

< P > then the command developed a system capable of identifying objects on images using less than 1% of the computing power required for traditional methods, with a meta -top lens made with 99.4% of work load.

~ ~ < P > If traditional neural networks use mathematical filters (kernels) to extract data, requiring numerous calculations even for several pixels, then the new system naturally conducts complex filtration during light, allowing several large filters to analyze all the images immediately. According to scientists, this technology can make a revolution in several technologies.

< p > “This, of course, still has to see, but here we have shown the first step. And this is a big step forward compared to all other available optical realities of neural networks”, & mdash; summed up the Madzhumdar Arch.