Real-time positioning and tracking for vision-based unmanned underwater vehicles

Abstract

Unmanned underwater vehicle (UUV) is a key technology for marine resource exploration and ecological monitoring. How to use vision-based active positioning and three-dimensional perception to realize UUV underwater autonomous navigation and positioning is the basis for UUV's underwater operations. The complexity and unstructured characteristics of seawater bring new challenges to vision-based underwater high-precision positioning. Traditional visual localization algorithms mainly include geometric-based visual localization algorithms (such as ORB-SLAM2) and deep learning-based visual localization algorithms (such as DXSLAM). In this paper, based on the typical marine environment (low brightness, dynamic fish interference, underwater light spot, high turbidity), the experimental analysis and comparison of different visual positioning methods of UUV is carried out, which provides a reference for realizing the real-time localization of UUV, and further provides a better solution for UUV underwater measurement and monitoring operations.