Using a new algorithm, Stanford researchers reconstructed the movements of individual light particles to see through clouds, fog and other obstacles.
Researchers at Stanford University have developed a kind of X-ray vision, only without X-rays. With hardware similar to that that allows self-driving cars to “see” the world around them, researchers have improved their system with an algorithm. extremely. efficient that can reconstruct hidden three-dimensional scenes, based on the movement of individual light particles or photons.
In tests performed and published in Nature Communications, their system managed to reconstruct the hidden shapes after a 1-inch layer of foam.
“There are a lot of imaging techniques that make images more beautiful, less ‘noisy,’ but this time, what we’ve developed is really a way to make the invisible visible,” said Gordon Wetzstein, assistant professor of engineering. Stanford and the lead author of the article.
“It really pushes the boundaries of what could be possible with any type of detection system. It’s like a superhuman vision. “
This technique complements other vision systems that can see through microscopic barriers – for medical applications – because it focuses more on large-scale situations, such as navigation in cars with automatic fog driving or heavy rains and satellite images of the Earth’s surface. and other planets through cloudy atmospheres.
Scattered light vision supercapacitors
To see through media that diffuses light in all directions, the system combines a laser with an ultra-sensitive photon detector that records every bit of laser light that strikes it. When the laser scans an obstruction like a foam wall, a photon will be able to pass through the foam, hit objects hidden behind it, and climb back through the foam to reach the detector.
The software supported by the algorithms then uses these few photons, along with information about where and when they hit the detector, to reconstruct the hidden objects in 3D.
It is not the first system capable of revealing hidden objects through broadcast media, but it circumvents the limitations associated with other techniques.
Some systems require information about the distance to the object of interest or use only information from ballistic photons, which are photons that move to and from the hidden object through the scattering field, but do not scatter along the way.
“We wanted to be able to create images through diffuse media without these assumptions and collect all the photons that were scattered to reconstruct the image,” said David Lindell, a graduate student in electrical engineering and lead author of the article.
“This makes our system particularly useful for large-scale applications, where there would be very few ballistic photons.”
To make their algorithm compatible with the complexity of the light scattering phenomenon, the researchers had to closely design their hardware and software, even if the hardware components they used were only slightly more advanced than is known. find. currently in cars with automatic driving.
Depending on the brightness of the hidden objects, it took between a minute and an hour to analyze their tests, but the algorithm reconstructed the hidden scene in real time and could even be run on a laptop.
“You couldn’t see through the foam with your own eyes, and even if you look at the photon measurements at the detector, you really can’t see anything,” Lindell said. “But with just a handful of photons, the reconstruction algorithm can expose these objects – and you can see not only what they look like, but also where they are in 3D space.”
In the future, a top-down version of this system could be sent into space to other planets to see through frozen clouds, deeper layers and surfaces. In the shorter term, researchers want to experiment with different delivery environments to simulate other circumstances in which this technology might be useful.