Towards Three-Dimensional Heterogeneous Imaging Sensor Correspondence and Registration for Visualization
M. Johnson-Roberson · O. Pizarro · S. Williams
TL;DR
A pipeline using spin image feature matching registers range data from two AUV views into visually consistent 3D seafloor meshes.
BRIEF
Mapping the seafloor in 3D from an autonomous underwater vehicle means stitching together range data collected from different viewpoints. This work looks at the practical issues in building large 3D reconstructions of the benthos from two distinct views. It matches 3D features, using spin images, to estimate the Euclidean transformation between views, then refines that estimate. The pipeline processes range points robustly and outputs registered meshes that look consistent to a human viewer.
Abstract
The registration and 3D mosaicing of range images presents an interesting challenge for the underwater domain. This work attempts to lay out some of the issues involved in generating large 3D reconstructions of the benthos from two distinct views gathered from an AUV. Specifically, we address using 3D feature correspondence to estimate a Euclidean transformation between views and then refine that estimate to generate visually consistent meshes for human viewing. To achieve this goal we implement spin image feature matching in a pipeline that robustly processes range points to output registered meshes.