Structure from plenoptic imaging

Plenoptic cameras allow to recover the structure of a scene from a single image. Recent works on scene structure estimation recover the disparity associated with each pixel but do not consider metric depth. In this work, we propose a methodology for scene metric reconstruction that estimates initial disparity values by analyzing the epipolar plane images. A dense estimate of disparities is obtained by regularization. These disparity estimates are then converted to 3D metric information using the intrinsic parameters of the plenoptic camera. Experiments have been conducted both with simulated and real cameras yielding promising results.