Visual SLAM for underwater vehicles: A survey

Abstract

Underwater scene is highly unstructured, full of various noise interferences. Moreover, GPS information is not available in the underwater environment, which thus brings huge challenges to the navigation of autonomous underwater vehicle. As an autonomous navigation technology, Simultaneous Localization and Mapping (SLAM) can deliver reliable localization to vehicles in unknown environment and generate models about their surrounding environment. With the development and utilization of marine and other underwater resources, underwater SLAM has become a hot research topic. By focusing on underwater visual SLAM, this paper reviews the basic theories and research progress regarding underwater visual SLAM modules, such as sensors, visual odometry, state optimization and loop closure detection, discusses the challenges faced by underwater visual SLAM, and shares the prospects of underwater visual SLAM. It is found that the traditional underwater visual SLAM based on filtering methods is gradually developing towards optimization-based methods. Underwater visual SLAM presents a diversified trend, and various new methods have emerged. This paper aims to provide researchers and practitioners with a better understanding of the current status and development trend of underwater visual SLAM, while offering help for collecting underwater vehicles intelligence.