Image Compression for Underwater Multi-Robot Loop Closing

Abstract

[eng] This paper presents an experimental assessment of different image compression techniques used to reduce the amount of visual data exchanged between underwater robots, in a centralized multi-robot visual localization and mapping context. Controller agents of centralized coalitions need to receive images from all the other team partners, to detect inter-session loop closings among them that will be used to join and optimize the different individual maps in a single global underwater visual structure. Underwater robots usually communicate each other through omnidirectional very low bandwidth acoustic links, which impose a significant data package size reduction before they are streamed. Accordingly, images of the sea ground usually need to be downsampled and losslessly encoded before they are transmitted. The evaluation of the different compression techniques has been made using a marine image set taken in diverse scenarios of the Balearic Islands, and focused on 2 main issues: compression rate and information loss.