This paper presents a modular localization and mapping system for intervention autonomous underwater vehicles working in semi-structured environments with known landmarks. The system is divided in several modules to make it as generic as possible. Two visual detection algorithms can be used to compute the position of known landmarks by comparing the images taken by the vehicle against an a priori known template. Navigation data, provided by standard navigation sensors, is adapted and merged together with landmark positions by means of an extended Kalman filter. This filter is capable of estimating vehicle position and linear velocity as well as the position of detected landmarks in real-time. Experiments performed with the Girona 500 AUV in a water tank demonstrate the proposed method.
Two-dimensional forward-looking sonars such as Blueview or DIDSON are becoming a standard sensor in both remotely-operated and autonomous underwater vehicles. Registration of imagery obtained from this sensors is of great interest since it constitutes a key step in several applications like the generation of acoustic mosaics or the extraction of vehicle motion estimates from sonar imagery, specially on poor visibility conditions. However, the characteristics of these sonar images, such as low signal-to-noise ratio, low resolution and intensity alterations due to viewpoint changes pose a challenge to the traditional registration techniques applied on optical images. In this paper, the performance of popular registration methods commonly used in photomosaicing are evaluated on real sonar data, including feature-based methods and an areabased approach. Experiments are carried out on different environments, from man-made structured scenarios to more natural and featureless areas, and under challenging conditions such as viewpoint changes and the presence of different sonarspecific artifacts. After assessing the impact of all these factors on the different registration techniques, we show that Fourierbased registration method stands as the more robust option to register acoustic imagery.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.