Fast Visual Odometry for a Low-Cost Underwater Embedded Stereo System ^†

Area of Science:

• Robotics
• Computer Vision
• Marine Technology

Background:

• Underwater surveys require robust navigation and site mapping systems.
• Existing systems often lack real-time feedback and efficient processing for low-resource environments.
• Accurate visual odometry is crucial for autonomous underwater vehicles (AUVs).

Purpose of the Study:

• To present the hardware and software design of an embedded stereo system for underwater imaging.
• To introduce a novel, lightweight visual odometry method tailored for the underwater environment.
• To enable real-time navigation, site coverage mapping, and quality feedback during underwater surveys.

Main Methods:

• Development of a stereo embedded system with real-time image acquisition and processing capabilities.
• Implementation of a lightweight visual odometry algorithm using stochastic pose representation and semi-global optimization.
• Integration of a novel stereo matching approach optimized for underwater conditions and system-attached lighting.

Main Results:

• The system provides real-time visual feedback on image quality, aiding survey operations.
• The visual odometry method successfully generates navigation data and site coverage maps.
• The system demonstrates robustness and suitability for low-computational resource platforms in real-world underwater tests.

Conclusions:

• The developed embedded stereo system effectively supports underwater surveys.
• The novel visual odometry method offers a promising solution for real-time navigation in underwater environments.
• The system's efficiency and robustness indicate significant potential for future autonomous underwater exploration.