Multi-view 3D reconstruction in underwater environments: Evaluation and benchmark
Eduardo Iscar · Katherine A. Skinner · Matthew Johnson-Roberson
TL;DR
A quantitative benchmark of how well in-air 3D reconstruction methods work underwater, including a comparison of low-cost and high-cost cameras.
BRIEF
Underwater mapping supports work from coral reef monitoring to surveying submerged archaeological sites. Many teams now apply 3D reconstruction methods built for use in air, but those methods assume constant brightness and a simple pinhole camera, and neither assumption holds underwater. Few studies measure how much those violations hurt accuracy. This work offers a quantitative evaluation of vision-based 3D reconstruction methods in underwater settings, and it compares low-cost and high-cost camera systems to show what is gained or lost at each price point.
Abstract
Mapping of underwater environments is a critical task for a range of activities from monitoring coral reef habitats to surveying submerged archaeological sites. In recent years, optical reconstruction methods developed for terrestrial (in air) applications have increasingly been applied to the underwater environment by the marine science community. However, assumptions such as the brightness constancy constraint (BCC) and traditional camera pinhole models, which hold in air, do not apply in the underwater environment. There is a lack of literature about how the violation of these assumptions affects the accuracy of the reconstruction. The main contributions of this paper is to develop a quantitative evaluation of computer vision-based 3D reconstruction methods applied to the underwater domain. Comparisons are presented between low- and high-cost camera systems to quantify the tradeoff between equipment cost and performance.