A virtual camera sees around corners
It might seem impossible to take a picture of something behind a wall or around a corner, when the direct line of sight is blocked. But you can actually get information about the object from the light that bounces off the walls. Although walls don’t make very good mirrors, and they scatter incident light all over the place, researchers still have ways to reconstruct the image. But non-line-of-sight (NLOS) imaging has limitations: The simple mathematical models used for complicated scattering only work well for isolated objects without much range of depth.
Now Andreas Velten of the University of Wisconsin–Madison and his colleagues have a new approach for NLOS imaging that circumvents those limitations. Their method uses phasors, or phase vectors—complex numbers that encode the amplitude, frequency, and other properties of a wave. The light moving through the scene to be imaged is reframed as a phasor field undergoing the same diffraction as any physical wave. NLOS imaging thus turns from a complicated scattering problem into a classical optics problem and can borrow from conventional line-of-sight (LOS) imaging models.
Using a known light source, the team measures how the phasor field changes as it moves from the wall to the scene and back again—a step not unlike other NLOS methods. But the rest of the imaging is done computationally using a virtual light source and camera to perform LOS imaging.
Velten and colleagues’ method can mimic any existing LOS imaging technique, and because it does so computationally, it’s not limited by the hardware. On top of that, existing diffraction integral solvers can perform the calculations quickly. The team’s NLOS imaging can handle complex scenes and ambient light (as seen in the figure) and create a three-dimensional image of a room 2 meters deep and filled with objects made of different materials. The future applications are plentiful and include robotics, defense, medical imaging, and autonomous vehicles. The next step, perhaps, is seeing around two corners. (X. Liu et al., Nature, 2019, doi:10.1038/s41586-019-1461-3 .)
