In computer graphics, we pride ourselves on performant and accurate forward simulation of light (rendering). Here, we use graphics as methodology to solve problems beyond the generation of pretty pictures. By marrying extremely efficient forward renderers to numerical optimization methods, we develop novel analysis-by-synthesis approaches that push the boundaries on challenging problems like 3D surface reconstruction, real-time tracking of occluded objects from indirect diffuse light reflections, and light field imaging through uncalibrated free-form optics.
Julian Iseringhausen, Bastian Goldlücke, Nina Pesheva, Stanimir Iliev, Alexander Wender, Martin Fuchs, Matthias B. Hullin
ACM Transactions on Graphics 36(4) (Proc. SIGGRAPH), 35:1--35:11, 2017.
Raindrops on a window heavily distort the view of the scene. We show that a fully calibrated 4D light field can be recovered from a single photograph taken under such adverse conditions.
Piergiorgio Caramazza, Alessandro Boccolini, Gabriella Musarra, Matthias Hullin, Roderick Murray-Smith, Daniele Faccio
Computational Optical Sensing and Imaging, 2017.
We demonstrate the use of machine learning to classify temporal histograms of the light-echoes backscattered from bodies hidden from view around a corner, captured by a SPAD camera.
Jonathan Klein, Christoph Peters, Jaime Martín, Martin Laurenzis, Matthias B. Hullin
Scientific Reports (Nature Publishing Group), 6, 32491; doi: 10.1038/srep32491, 2016.
We demonstrate the tracking of objects outside the line of sight from 3rd-order indirect diffuse reflections, captured using a regular laser pointer and a 2D camera.