Autonomous navigation of underwater vehicles

Abstract

The goal of the work is to propose a method to solve with visual means the problem of underwater navigation of an autonomous vehicle moving in deep sea in order to perform inspection of a pipe positioned near the ocean floor. We refer to the ability of an autonomous system to determine its motion with respect to the environment. It is well known that a significant amount of useful information about 3D motion can be obtained from 3D image motion resulting from the projection on the image plane of the camera and 3D objects movements. The method we propose is based on analysis of robust features, extracted from the sequence images, and correspondences between them estimated by applying a method relying only on geometrical invariants. The proposed matching approach consists of an optimization process based on satisfaction of global constraints derived from projective invariance of the cross ratio of any subset of five coplanar points. We match relational graphs characterized by feature-nodes connected by five-order links weighted by cross-ratio values through a nonlinear relaxation labeling process involving high order compatibility measurements.