Engineering news
Previous systems developed at Stanford University in California have captured objects, textures and even movement around corners by bouncing lasers off walls and capturing the reflections from scenes, but they have had low resolution and impractical scan times.
Now a team led by Christopher Metzler, also of Stanford University and Rice University in Texas, appear to have overcome those issues by harnessing the power of deep learning – with some caveats.
“Compared to other approaches, our non-line-of-sight imaging system provides uniquely high resolutions and imaging speeds,” said Metzler. “These attributes enable applications that wouldn't otherwise be possible, such as reading the licence plate of a hidden car as it is driving or reading a badge worn by someone walking on the other side of a corner.”
The new system can reportedly distinguish sub-millimetre details of a hidden object from 1m away – very close for something that could be used on a car. It is nonetheless designed to combine with other imaging systems that produce low-resolution room-sized reconstructions, such as one developed by electrical engineer Gordon Wetzstein and colleagues at Stanford.
With further development, the new system could let self-driving cars look around parked cars or busy junctions to see hazards or pedestrians. Other applications suggested by the researchers included installing it on spacecraft to capture images around corners or inside caves on asteroids. There could be other uses in medical imaging, robotics and defence.
The system uses a commercially-available camera sensor and a “powerful but otherwise standard” laser source. The laser beam bounces off a visible wall onto the hidden object and then back onto the wall, creating an interference pattern known as a ‘speckle pattern’ that encodes the shape of the hidden object.
Reconstructing hidden objects from speckle patterns requires solving a challenging computational problem. Short exposure times are necessary for real-time imaging but produce too much noise for existing algorithms to work. To solve this problem, the researchers turned to deep learning.
“Compared to other approaches for non-line-of-sight imaging, our deep learning algorithm is far more robust to noise and thus can operate with much shorter exposure times,” said paper co-author Prasanna Rangarajan from Southern Methodist University. “By accurately characterising the noise, we were able to synthesize data to train the algorithm to solve the reconstruction problem using deep learning without having to capture costly experimental training data.”
The researchers tested the technique by reconstructing images of 1cm letters and numbers hidden behind a corner, using an imaging set-up about 1m from the wall. Using an exposure length 0.25s, the approach produced reconstructions with a resolution of 300 micrometre, or millionths of a metre
The work is part of Darpa's Revolutionary Enhancement of Visibility by Exploiting Active Light-fields (Reveal) programme, which is developing a variety of techniques to image hidden objects around corners. The researchers, some of whom also work at Princeton University, are now working to make the system practical for more applications by extending the field of view so that it can reconstruct larger objects.
The research was published in Optica, the Optical Society’s journal.
Want the best engineering stories delivered straight to your inbox? The Professional Engineering newsletter gives you vital updates on the most cutting-edge engineering and exciting new job opportunities. To sign up, click here.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.