Scientists from the University of Zaragoza develop a method to “see through corners”

• The new method, which has just been published in Nature, transforms walls into virtual lenses
• Diego Gutiérrez, Adrián Jarabo and Ibón Guillén, researchers at the Graphics and Imaging Lab of the I3A, have developed this novel computational imaging technique in collaboration with the University of Wisconsin – Madison
• The work allows for the first time to reconstruct complex and large non-visible scenes and opens the door to the use of this technology in real-world applications in security, rescue missions or autonomous vehicles
• This novel technology is now being tested even in NASA's distant survey of craters on the moon.

A new computational imaging technique, which combines ultra-fast imaging hardware at a trillion frames per second with novel computational techniques, allows for the first time to see around corners in real-world scenes not controlled in a laboratory, including geometry and complex materials.

This technique has been published in the prestigious journal Nature, considered the most important in scientific research, and has been developed thanks to the collaboration of scientists Diego Gutiérrez, Adrián Jarabo and Ibón Guillén from the Graphics and Imaging Lab group of the Engineering Research Institute of Aragón (I3A) from the University of Zaragoza, with researchers from the University of Wisconsin - Madison, led by Andreas Velten.

Since Gutiérrez and his team, in collaboration with MIT (USA), developed a camera capable of capturing light at one billion frames per second in 2013, the idea of ​​using this technology to see through corners has been explored by several groups. globally, but until now it was limited to simple objects in ideal laboratory conditions: "Essentially, at a trillion frames per second we can approximate which path each photon has taken as it travels through the scene, and then mathematically reverse the path of millions of photons and reconstruct a scene," says Adrián Jarabo. "However, reversing this path is not very robust, making this technique impractical except in simple situations under very controlled conditions," he adds. These limitations meant that this promising technology could not be applied under realistic conditions.

As Diego Gutiérrez explains, to try to overcome these limitations, it was necessary to adopt a radically new approach: "We realized that traditional cameras do not require complex mathematical models to capture an image. In fact, their basic operation has been known for a few years. 200 years. So we rethought the problem from scratch, and developed a computational model that allows us to transform a wall into a virtual camera that directly captures the hidden scene."

These virtual cameras have the same capabilities as existing optical systems (also including microscopes or telescopes), and are very robust to signals degraded by noise, range, ambient lighting, and even very short exposures, necessary to bring this technology to the real world. : "Right now we capture non-visible scenes in half a minute, but by the middle of next year we hope to be able to capture them in one or two seconds, or even less," says Ibón Guillén.

This new model not only allows us to overcome the limitations of previous models, but also "describes a new class of algorithms that in turn open the door to a multitude of applications in fields such as security, robotics, autonomous vehicles, or rescue situations in disaster situations, where members of rescue teams could assess risks when entering dangerous areas without having to expose themselves," says Gutiérrez.

Proof of the future impact of this technology is the interest on the part of the US Department of Defense, which has co-financed this research (along with, among others, a Leonardo Scholarship from the BBVA Foundation to Adrián Jarabo) within the REVEAL program. In the medium term, NASA is in the process of testing to incorporate this technology for remote exploration of lunar caves.