Using Active Illumination for Accurate Variational Space-Time Stereo

This paper addresses the problem of space-time stereo with active illumination and presents a formulation of this problem in the variational framework. Variational problems of this scale are computationally expensive to solve directly. We overcome this challenge by showing that speed-improving techniques, as the full-multi-grid and the multi-leveladaptation techniques, can be applied. We evaluate the performance of our method on 3 ground-truth datasets. The experimental results for synthetic and real datasets show that the combination of active illumination and variational space-time stereo improves the quality of the reconstruction on average by up to 3.1 times compared to a reconstruction from a single passive stereo image pair without active illumination.