Monocular Video Frame Optimization Through Feature-Based Parallax Analysis for 3D Pipe Reconstruction

Abstract

Structure-from-motion (SfM) techniques have been widely used for three-dimensional (3D) scene reconstruction from sequential video frames. However, for reconstructing narrow and confined spaces such as the interior of drainage pipes, selecting geometrically optimal frames is a major challenge, not only to reduce the number of needed frames but also to yield better geometry. This paper introduces a coarse-to-fine method to optimize the selection of monocular video frames based on a geometric criterion called feature-based parallax analysis for 3D pipe reconstruction. The proposed method was applied in two experiments with a monocular camera fixed on a customized robot. Experimental results show that our approach only requires respectively 9.66% and 3.15% of the number of frames. The spatial distribution of the retrieved frames was uniform and reasonable, enabling the successful SfM process to achieve a complete reconstruction of the pipe geometry.