3D shape reconstruction method by the MMSE linear camera model

Abstract

This report proposes a new iterative method for reconstructing three dimensional shape of an object from many two dimensional images by using generalized inverse matrices and linear camera models. The camera model can best approximate actual perspective camera image in the sense of the minimum mean squared error when a three dimensional shape and an image are given. At the first iteration, the method employs Moore-Penrose's G-inverse matrix. In the following steps, series of G-inverse matrices and MMSE camera models are constructed and employed in order to compensate for reconstruction error due to mainly non-linear distortion of perspective camera images.