No matter how many twists and turns they encounter, a team of University of Wisconsin-Madison engineers can see around blind corners—all the while standing in the same spot.
Pairing cutting-edge sensing technology with powerful new algorithms, the team—led by Talha Sultan (PhD ’25) and Andreas Velten, an associate professor of electrical and computer engineering and biostatistics and medical informatics—developed a method of “seeing” around multiple corners to image hidden scenes. In principle, the method could allow researchers to look around five, six, even 10 corners, as long as there is enough light bouncing between them.
The technique, which collects information from light-reflecting surfaces that act as “virtual cameras,” could eventually image geologic features like caves, reveal inaccessible archaeological sites, or aid search and rescue teams and military personnel. Insights from the research also suggest new techniques for imaging through clouds, smoke, water and other light-scattering phenomena.
“Seeing a hidden object is only part of the story,” says Sultan. “We also want to understand how light travels between hidden surfaces. From measurements taken at a visible wall, we can computationally focus illumination on a chosen hidden spot and reconstruct how light spreads through the scene over time.”
Sultan, Velten and their collaborators described their method in a paper published by the journal Nature Communications on July 22, 2026 and more information is available via GitHub.
Velten has researched non-line-of-sight imaging for most of his career, exploiting the fact that photons, or discrete particles of light, carry much more information than conventional imaging systems can pick up. Traditional digital cameras, for instance, average the brightness and color of multiple photons, representing them as a single pixel. Combining all of those pixels produces an image, but contains little other data.
For their imaging, Sultan and Velten use more advanced sensors called single-photon avalanche diode arrays (SPADs). These sensors capture an immense amount of data, including the flight time and intensity of a photon signal.
To see around corners, the team first flashes laser pulses onto a reflective surface, which can include a typical wall or other solid material. The light bounces and illuminates any hidden objects, with some of that light reflecting back. The team focuses their photon detector on the reflective surface, which serves as a “virtual camera,” providing the time-of-flight and angle of individual photons. Using specialized algorithms that essentially work backward, the team reconstructs the hidden scene using the data provided by the photons.
In previous work, Velten and his students used a single SPAD sensor that slowly scanned a reflective surface, collecting information on light particles one at a time, a process that took hours. In this new work, the team used improved algorithms and a 16×16 array of photon detectors fabricated by colleagues at Politecnico de Milano, in Milan, Italy, giving them 256 sensors (plus or minus a few that burned out) to work with. Not only did this array improve the speed and resolution of the hidden images, it also gave the researchers the ability to “focus” their virtual camera, spotlighting areas in the hidden scene.
The upgraded system let them test the hypothesis that, with enough light and resolution, they could process data from multiple “virtual cameras” looking around more than one corner. “The question was, if we can place a virtual camera on one surface, could we focus our laser light and detector on a spot in the hidden scene and place another virtual camera to look around another corner?” says Velten.
The answer turned out to be yes, though the second image suffered from resolution and noise issues. Still, it was a proof of concept. With algorithmic improvements to streamline data processing and with a larger SPAD array, the team believes the technique could create higher resolution images around multiple corners. With enough resolution and light, the team believes the system could image more complex scenarios, like a building corridor or cave system.
It may take a while before that happens. While the development of single-photon detectors is maturing quickly, creating arrays of the detectors with high spatial and time resolution is still technically challenging. “We only had a 16×16 SPAD array, which was pretty low resolution,” says Velten. “But we can see from this research what we could do with a high-resolving detector. It’s just a matter of waiting for the technology to catch up.”
In the meantime, Velten is pursuing other facets of this research. The virtual cameras in this research, he explains, are basically regions of scattered light, and the processing is reconstructing information and images from this light. Velten says it may be possible to apply similar techniques to imaging systems to peer through fog or clouds, see underwater and even look into human tissue.
“Our work is basically just understanding this process of light and imaging through scattering. We’ve shown that we can image through any number of scattered layers, like reflections on walls,” he says. “Understanding how information gets lost during scattering and how you might try to recover it is a very interesting problem. But fairly manageable if we can understand it.”
Other UW-Madison authors include Eric Brandt and Khadijeh Masumnia-Bisheh.
Other authors include Simone Riccardo and Alberto Tosi of Politecnico di Milano in Milan, Italy, and Pavel Polynkin of the University of Arizona.
The UW-Madison authors acknowledge support by the Air Force Office for Scientific Research (FA9550-21-1-0341, FA9550-26-1-B169) and the Defense Advanced Research Projects Agency (DARPA) through the DARPA REVEAL Project HR0011-16-C-0025.