Large-scale mosaicking with spectral registration based simultaneous localization and mapping (iFMI-SLAM) in the Ligurian Sea

Abstract

Underwater image mosaicking is an important tool for visual surveys, object detection, and as a means to control the underwater robot if done online. In this paper, novel results of an spectral registration method on image data collected by a remotely operated vehicle in the Ligurian Sea are presented. The registration method computes 2D translation, rotation, and scale, along with a covariance matrix for these parameters. An appearance-based place recognition method is used to detect loops. These methods are embedded in a graph-based Simultaneous Localization and Mapping framework to provide a globally consistent and optimized map of the photographed ground.