Global Localization for Autonomous Underwater Vehicles Using Visual Odometry
Loading...
Files
Date
2021-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Path planning and navigation are two of the main challenges for achieving autonomous driving. At the same time, localization has been considered as the footstone for path planning and navigation because the odometry information of a vehicle is required for the high-level planning algorithms. For autonomous underwater vehicles (AUV), expansive navigation sensors, such as deepsea sonar, have usually been utilized to get the location of the robot underwater. This project, however, is intended to introduce the potential application of real-time underwater visual odometry (VO) algorithms based on data retrieved from a monocular camera on both simulated environments and real underwater datasets. Due to the outbreak of the pandemic, this project is primarily developed and tested based on the platform of an underwater simulator named UWSim. Since the working environment for underwater robots is not always perfect, different feature detection algorithms were performed on real open-source underwater footage to determine the most suitable image processing algorithm for this project. The proposed visual odometry algorithm is based on state-of-the-art feature detection algorithms, feature tracking techniques based on optical flow, and projected geometry theories. Comparisons of the ground truth odometry of a simulated underwater robot and the odometry calculated from the visual odometry algorithm will be presented.