Real-Time Monocular Visual Odometry for Turbid and Dynamic Underwater Environments

Area of Science:

• Robotics
• Computer Vision
• Oceanography

Background:

• Underwater localization is challenging due to visual degradation from water properties.
• Current methods often rely on expensive sensors or struggle with turbidity and dynamism.
• Pure visual methods show promise but require robustness improvements.

Purpose of the Study:

• To develop a novel visual odometry method for robust underwater localization.
• To address the limitations of existing methods in turbid and dynamic aquatic environments.
• To enhance the capabilities of underwater robots, particularly for archaeological missions.

Main Methods:

• Proposed a new visual odometry algorithm designed for robustness against visual perturbations.
• Assessed the algorithm's performance using both simulated and real underwater datasets.
• Compared the method against state-of-the-art terrestrial visual SLAM techniques.

Main Results:

• The proposed method demonstrates superior performance compared to terrestrial visual SLAM.
• Achieved robust localization even under challenging underwater conditions like turbidity.
• Outperformed existing methods in various simulated and real-world underwater scenarios.

Conclusions:

• The developed visual odometry system offers a robust solution for underwater localization.
• It significantly improves the reliability of visual information in challenging aquatic environments.
• The system is applicable to Remotely Operated Vehicles (ROVs) and other underwater applications.