Turning rough surfaces into non-line-of-sight cameras

Overcoming the spatial mixing of reflection and scattering from rough surfaces to see a hidden scene is a difficult challenge with a wide range of applications. Existing methods counter the mixing effect using active time-of-flight or coherence-based measurements but at the cost of expensive and dedicated optical systems. Here, we derive a model to accurately describe the microscale scattering properties of a rough surface and show it is well-conditioned for inversion. This enables computational passive non-line-of-sight (NLOS) imaging with unprecedented capabilities. With an ordinary monochrome camera, we demonstrate real-time and high-resolution passive NLOS imaging with spatial resolution of sub-mm, temporal resolution of 25 fps, field of view of near 90°, and full-color recovery. Furthermore, our approach permits high-quality keyhole NLOS imaging through a mm-scale keyhole and, notably, allows for non-invasive calibration and imaging without any manipulation of the hidden scene. These results open a new avenue for the development of NLOS imaging, scattering imaging, and relevant applications to real-world conditions.