This AI camera protects your privacy by recording only specific targets

AI Privacy Camera

Scientists from UCLA are trying to address privacy issues by developing a new artificial intelligence camera that records only specific targets and actively erases everything else.

As digital cameras have become almost ubiquitous, privacy protection issues have somewhat risen. as explained Science BlogSome have chosen to address these concerns with obfuscation or data encryption, but they do not address the problem of displaying data because raw snapshots are still captured before they undergo data processing.

But UCLA professor Aydogan Ozcan and a group of fellow scientists have developed a new camera that bypasses this problem in a smart camera system that records only desired objects or subjects and actively and instantly scans all kinds of other objects at the point of capture without the need for anything. Additional processing afterwards.

in research paperthe scientists describe a camera design that performs class-specific imaging of target objects with instant comprehensive erasure of other classes of objects — in short, a camera that only sees and records what it is told to search.

The new camera design uses what scientists call “reflective computing,” which visualizes a target or class of objects at high resolution while erasing all other objects that don’t match that target.

The camera consists of a series of diffraction layers, each of which is optimized through deep learning. After they are trained, each layer is stitched together in a three-dimensional system that forms what scientists call a “computational illustrator” between the input field and the output plane.

“This camera design does not rely on a standard point spread function, instead, the aggregated 3D diffraction layers collectively act as an optical mode filter that is statistically optimized to pass through key modes of target classes of objects, while filtering through the dispersal of key representative patterns,” the scientists explain. for other object classes (learned through the data-driven training process).”

If any object that does not match the required input is placed in front of the camera, it will be optically erased and reduced to what are described as low-density non-informational patterns – essentially, what photographers would describe as noise.

The camera was tested by presenting it with a set of numbers and asking the camera to find only the second number. As shown below, the final five-layer output captures only the desired number while reducing the other numbers in the frame to stale visual noise.

Scientists say that not only does this make the camera much better at protecting privacy, but it also uses less power than is required with current methods because no post-treatment is necessary.

The camera presented in the research requires that each layer be meticulously trained and then stacked on top of each other, making it impractical for large-scale deployment at present. Scientists seem to understand this, but they say the research they have done could guide the design of camera systems in the future.

“Reflexive camera design teachings can inspire future imaging systems that consume orders of less computing and transmission power as well as store less data, helping to meet our global need for modern, mission-specific, data-efficient and privacy-sensitive imaging systems,” the researchers say.

The hope is that this technology can be used in place of any video that currently requires someone to actively blur areas of the image for privacy by never recording it in the first place.

The full research paper titled To shoot or not: Class-specific refractive cameras with complete optical erasure of unwanted objects It can be read on Springer Open.

Image credits: Bijie Bai, Yi Luo, Tianyi Gan, Jingtian Hu, Yuhang Li, Yifan Zhao, Deniz Mengu, Mona Jarrahi, Aydogan Ozcan

Leave a Comment