Metric three-dimensional reconstruction model from a light field and its calibration

Abstract. The traditional methods of depth reconstruction from light fields have excellent performance in depth estimation; however, the depth information has nonlinear scale ambiguity. We divide the three-dimensional metric information into two easy subproblems instead of a whole question. A two-step method is proposed to obtain the metric depth and the metric lateral information, respectively. The depth model is proposed using the refocus property of the light field and the Gaussian lens formula. The model of the lateral coordinate is reasoned from the relationship between the light field in the scene and the recorded light field. Then, three calibration methods are proposed to determine these five parameters (the center position of the microlens for recovering the four-dimensional light field, the camera principal point, and the initial image distance). The experiments are conducted in our developed light field camera. The precision performance of our method is also confirmed in measuring the calibration board. When the proposed method is adopted to reconstruct the natural scene, the preferable results are obtained.