Computational Imaging of Dynamic Scenes from Optical Scattering Data

The project is to reconstruct three-dimensional images of dynamic scenes from optical scattering data in near real-time. The image reconstruction and rendering tasks combine Monte Carlo sampling of physics-based models (radiative transport) and machine learning algorithms for the rapid approximation of the solution of the inverse problem. The application in the research proposal offers the possibility of exploring the fundamentals of optical imaging (3D tomography), randomised algebra for solving complex problems in real-time and physics-inspired neural networks for learning a low-dimensional dispersion model "on-the-fly" based on landscape topography data.