Visual odometry method for navigation and orientation of an underwater robotic system with a modified image depth estimation algorithm

Abstract

The research focuses on a visual odometry algorithm for the navigation of an underwater robotic system operating in proximity to a planar seabed. The objective of this work is the development and investigation of standard and modified methods for estimating the altitude above the seafloor. To achieve this, the following tasks were completed: the design of a computer vision navigation system utilizing a stereo pair of Basler acA750-30gm cameras; the development of an experimental testbed simulating the motion of an underwater vehicle over a flat seabed; a comparative analysis of standard scene depth estimation algorithms; and the development of modifications to existing methods for altitude determination. The implementation was carried out in C++ utilizing the ROS 2 framework and the OpenCV computer vision library. Original methods for calculating the altitude above the seafloor were developed, leveraging matched feature point pairs from the visual odometry pipeline, which enable enhanced computational accuracy while maintaining system real-time performance.