Integration of range images in a multi-view stereo system

Abstract

A novel method for integrating multiple range images in a multi-view stereo imaging system is presented here. Due to self-occlusion an individual range image provides only a partial model of an object surface. Therefore multiple range images from differing viewpoints must be captured and merged to extend the surface area that can be captured. In our approach range images are decomposed into subset patches and then evaluated in a "confidence competition". Redundant patches are removed whilst winning patches are merged to complete a single plausible mesh that represents the acquired object surface.