3D Scene Modeling from Dense Video Light Fields

Abstract

Light field imaging offers unprecedented opportunities for advanced scene analysis and modelling, with potential applications in various domains such as augmented reality, 3D robotics, and microscopy. This paper illustrates the potential of dense video light fields for 3D scene modeling. We first recall the principles of plenoptic cameras and present a downloadable test dataset captured with a Raytrix 2.0 plenoptic camera. Then we describe a scene depth estimation algorithm from a sparse set of light field views with occlusion handling via a low rank matrix completion method. Finally, the obtained disparity maps are used to construct 3D point clouds.