Tracking moving targets with wide depth of field behind a scattering medium using deep learning

Abstract

When light propagates through a scattering medium, imaging of an object hidden behind the scattering medium is difficult due to wavefront distortion. Scattering imaging is a technique for reconstructing images by solving the problem of complex reconstruction from speckle images. Tracking moving targets behind a scattering medium is a challenge. Scattering imaging using deep learning is a robust technique that learns a huge number of pairs of ground-truth images and speckle images. Here, we demonstrate tracking of moving targets with an extended-depth of field (DOF) behind a scattering medium based on deep learning of speckle images acquired at different depths. We found that it was possible to track moving targets over a wide axial direction by increasing the number of trained positions.