Refraction Corrected Pose Estimation for 3D reconstruction of underwater objects

Abstract

Retrieval of 3D geometry from underwater images overthrows single view point assumption and classical structure from motion as the images are distorted due to the refraction of light at the interface between the propagation through different media. Most of the underwater imaging systems suffer from two level refraction due to the propagation of light ray through two media, water and glass. We propose a single layer refraction method where the glass is thin and its effect is neglected. Initial pose is estimated through an adaptive pinhole model and is then refined using refractive fundamental matrix. The ground truth pose estimated with calibration pattern is used to measure the pose error. Our method produced better pose estimate compared to the standard structure from motion.