Variational scene flow and occlusion detection from a light field sequence

In this work we propose a variational model for 3D motion estimation from a sequence of light field images captured with a plenoptic camera. Plenoptic cameras provide information about the intensity and direction of light. We formulate an energy functional with a data fidelity term that takes into account the geometric consistency of the images and the occlusions between objects in the scene. The smoothness term is based on the vectorial total variation. The geometric consistency is imposed using the sub-aperture image representation and computing disparity maps between the views by means of a structure tensor method. Finally, we also present experiments over light field image sequences.