Depth Estimation and Validation of Plenoptic Light Field Camera

A light field (LF) camera can record both the spatial and directional information of light rays in a three-dimensional (3D) scene. This is different from traditional photography, which sums up all the light rays to record their intensity and discard the directional information. Accurate depth estimation is crucial for 3D sense reconstruction which can be achieved by the directional information provided by the LF camera. In this study, a refocus model has been examined to accurately estimate the metric depth of a 3D scene. A range of camera calibrations has been conducted to determine the refocus model parameters and thus to achieve accurate depth estimation. Ground truth data from a stereo camera is utilized to validate the calibrations. The results indicate that the root-mean-square error is approximately 5cm when the depth range is 23.5cm to 65.5cm. Therefore, it is demonstrated that the proposed refocused model is feasible to estimate the depth information of the 3D scene accurately.