Self-Correction of 3D Reconstruction from Multi-View Stereo Images

Abstract

We present a self-correction approach to improving the 3D reconstruction of a multi-view 3D photogrammetry system. The self-correction approach has been able to repair the reconstructed 3D surface damaged by depth discontinuities. Due to self-occlusion, multi-view range images have to be acquired and integrated into a watertight non-redundant mesh model in order to cover the extended surface of an imaged object. The integrated surface often suffers from "dent" artifacts produced by depth discontinuities in the multi-view range images. In this paper we propose a novel approach to correcting the 3D integrated surface such that the dent artifacts can be repaired automatically. We show examples of 3D reconstruction to demonstrate the improvement that can be achieved by the self-correction approach. This self-correction approach can be extended to integrate range images obtained from alternative range capture devices. ©2009 IEEE.