Out-of-Core Efficient Blending for Underwater Georeferenced Textured 3D Maps
M. Johnson-Roberson · M. Bryson · B. Douillard · O. Pizarro · Stefan B. Williams
TL;DR
An out-of-core texture blending method turns underwater stereo images into high-resolution, georeferenced 3D seafloor maps without exceeding memory limits.
BRIEF
Underwater robots collect thousands of overlapping stereo images of the seafloor, which can be combined into 3D maps. Painting those images onto the 3D surface at full resolution is difficult because the data is too large to hold in computer memory at once. This work describes a system that builds georeferenced 3D maps from an underwater vehicle's stereo cameras and blends the textures out-of-core, processing the data in pieces. Careful model parameterization keeps distortion low while preserving the resolution of the source images. Tests used synthetic data and real maps from the Sirius AUV.
Abstract
This paper presents a system for the creation of georeferenced 3D maps projectively textured with visual data gathered with an underwater robot. Using optical stereo cameras, maps are reconstructed and textured. Within the paper we propose a novel out-of-core 2D texture blending process that allows for high resolution texturing of complex 3D structure. Through the use of state-of-the-art model parameterization and texture at lasing the distortion of the final result can be minimized while the resolution of the original source imagery is maintained. We demonstrate both synthetic and real texturing results on 3D maps gathered with the Sirius Autonomous underwater vehicle (AUV). We discuss the implications for insufficient resolution when observing benthic features. Finally we conclude and discuss future directions for underwater 3D blending.