Innovate
USF tech can see around corners, through obstacles
Typical X-ray machines, including those found in hospitals and airports, can scan objects from about three feet away. Researchers at the University of South Florida have achieved 97% accuracy at a mile.
Concepts like X-ray vision from extraordinary distances, and the ability to see around corners, were once relegated to the realm of science fiction. Local research could make the theoretical a reality.
The Defense Advanced Research Projects Agency (DARPA) has tasked John Murray-Bruce, an associate professor at the University of South Florida, with developing an X-ray imaging system that can discern obstructed objects from miles away. The National Science Foundation (NSF) also recently selected the researcher to create a mathematical algorithm that detects threats hidden around corners.
Murray-Bruce is pushing the limits of what machines can see and interpret from his Information Science and Computational Imaging Lab at the USF Bellini College of Artificial Intelligence, Cybersecurity and Computing. He calls it “extreme imaging.”
“You don’t think you can form images in these types of scenarios,” Murray-Bruce said. “But it turns out, if you think carefully about the mathematics, you really can.”
A system that interprets incomplete data and provides a picture of what is around the corner is already “pretty successful.” Murray-Bruce said the technology can be “tremendously useful” for motorists.
The system, in a controlled environment, can detect if multiple people are in a room. Users can also tell if they are moving, and potentially map their trajectories.
Urban warfare is exceedingly dangerous as soldiers must breach buildings and clear rooms without knowing what or who is on the other side. Murray-Bruce said the technology, in ideal conditions with little external light, could potentially identify a weapon or anticipate danger.
“Have we deployed it into practice? No,” he added. “That’s where my job stops. I’m a researcher.”
Doctoral student Robinson Czajkowksi creates a hidden scene. The research team will use shadows and algorithms to see around the wall.
Murray-Bruce could not share sensitive information related to the DARPA project. He said the X-ray imaging system is still in its infancy, and the federal agency known for creating the internet wants to “explore what’s possible.”
Typical X-ray machines, including those found in hospitals and airports, can scan objects from about three feet away. The computational imaging researcher has achieved 97% accuracy at a mile and 75% at nearly three miles.
“You get these incredibly noisy measurements – they look like they contain no information,” he explained. “They just look like specs of white dots. We’re developing approaches that can interpret what’s beneath those white dots. There is information.”
Quickly verifying incoming cargo at sea could prevent a small nuclear device from reaching a port. Murray-Bruce noted that X-ray photons can penetrate nearly anything, outside of a kilometer-thick lead block.
“In certain scenarios, we’re doing well,” Murray-Bruce said. “In other scenarios, not so well.”
He and his students typically combine physics with machine learning in their research. Murray-Bruce said their approach has led to some surprising results and more accurate images.
The team at USF is also “up against” extremely well-funded defense contractors. Murray-Bruce called it a “real honor” that DARPA has entrusted them to work on the X-ray imaging project.
“It does speak very well about the quality of work that is going on at USF and within my lab,” he said.
Murray-Bruce believes his team brings a unique perspective to a table that is typically not reserved for academics. Their latest research stems from reconstructing 3D images of obstructed scenes from ordinary photographs.
Murray-Bruce said DARPA projects are notoriously challenging. “You don’t solve the problem directly, but a wealth of other kinds of ideas and innovations come about.”
Phase 2 of the DARPA program begins in a couple of months. Murray-Bruce said it will further push technological limits by introducing scenarios at greater distances and significant motion blur.
He is cautiously optimistic that the agency will select his team to continue their research. The five-year NSF project began July 1.
Murray-Bruce said he had to rethink the “mathematical pipeline that leads to these algorithms that extract information” from hard-to-read or incomplete data. The NSF study will provide a “more complete theory for how to form images with computation.
“Perhaps we can get the same quality images, but you don’t have to sit in an MRI scanner for 20 minutes,” Murray-Bruce explained. “You can sit for two seconds. We think there are new potential modalities of imaging systems that we could actually discover using our approach.”
