Non-line-of-sight imaging (NLoS), which is the task of recovering image data from objects hidden from direct observation or hidden by corners, can be approached in several ways, frequently requiring techniques to recover data from highly scattered or diffuse optical signals.
In recent years, methods for reconstructing images using the intrinsic spatial correlations encoded in scattered light have joined techniques that use time-of-flight measurements to track a signal’s path. However, these methods are limited in photon-starved conditions in part because of the inadequately understood noise properties of that correlation.
A project has made substantial progress by improving the noise model for recovering obscured objects from indirect reflections and training a deep neural network to crunch the numbers.
This non-line-of-sight imaging device offers remarkably high resolutions and imaging speeds compared to other methods. These characteristics allow for applications that would not otherwise be possible, such as reading a badge on a person walking around a corner or reading the license plate of a concealed vehicle while it is moving.
According to the project team, the imaging system bounces a continuous-wave laser at an angle off a visible wall to illuminate an item hidden behind a corner and out of direct view. The scattering pattern that results when the reflected light strikes a different area of the same wall is then noted.
Related Content: Terahertz Laser – Imaging, Security, Communication