Image hashing for loop closing in underwater visual SLAM

Abstract

[eng] This article presents an experimental assessment of a hash-based loop closure detection methodology specially addressed to Multi-robot underwater visual Simultaneous Localization and Mapping (SLAM). This methodology uses two different top quality image global descriptors, one learned (NetVLAD) and one handcrafted (HALOC). Complete tests were done to compare the performance of both hashing techniques applied in an extensive set of real underwater imagery.