Evaluating Structure-from-Motion on shiny and non-textured surfaces in borescope videos

Abstract

To aid in damage assessment, creating 3D reconstruction from borescope videos of jet engines could be very beneficial. However, jet engines often have shiny and non-textured surfaces, and the performance of 3D reconstruction methods is unknown in this case. This paper aims to qualitatively and quantitatively evaluate Structure from Motion (SfM) on these borescope videos. SfM is a technique for 3D reconstruction that uses collections of images to create 3D models. An evaluation was done on borescope videos with differing characteristics using SIFT, SuperGlue, and ground truth for feature detection. Even though small experiments with the global SfM approach produced insufficient results, more extensive experiments using incremental SfM show promising performance on borescope videos and potential for accurate damage assessment, especially when combined with multi-view stereo.