3-D shape reconstruction from perspective images and weak-perspective camera by error weighting

摘要

One of the most popular methods of reconstructing 3-dimensional (3-D) shape from multiple 2-D images employs weak perspective (i.e. scaled orthographic) camera models and Moore-Penrose (MP) pseudo-inverse matrices. Thus, the method cannot avoid reconstruction error due to perspective camera images. This report proposes a more precise reconstruction method which recursively weights reconstruction error with discrepancies between perspective and orthographic projections. For each aspect of an observed object, the discrepancy has its particular property. The discrepancies are evaluated by a kind of orthogonal projection matrices and represented by covariance matrices. Many numerical simulation results show that our method is effective for wide ranges of image noise level and distance between camera and object.