Real-time Monocular Visual Odometry for Turbid and Dynamic Underwater Environments (1806.05842v3)

Abstract: In the context of robotic underwater operations, the visual degradations induced by the medium properties make difficult the exclusive use of cameras for localization purpose. Hence, most localization methods are based on expensive navigational sensors associated with acoustic positioning. On the other hand, visual odometry and visual SLAM have been exhaustively studied for aerial or terrestrial applications, but state-of-the-art algorithms fail underwater. In this paper we tackle the problem of using a simple low-cost camera for underwater localization and propose a new monocular visual odometry method dedicated to the underwater environment. We evaluate different tracking methods and show that optical flow based tracking is more suited to underwater images than classical approaches based on descriptors. We also propose a keyframe-based visual odometry approach highly relying on nonlinear optimization. The proposed algorithm has been assessed on both simulated and real underwater datasets and outperforms state-of-the-art visual SLAM methods under many of the most challenging conditions. The main application of this work is the localization of Remotely Operated Vehicles (ROVs) used for underwater archaeological missions but the developed system can be used in any other applications as long as visual information is available.

• The paper presents the UW-VO algorithm that leverages optical flow and retracking to overcome feature loss in challenging underwater conditions.
• It employs a keyframe-based system and Bundle Adjustment to minimize drift and maintain consistency in the reconstructed trajectory.
• Evaluation against ORB-SLAM and SVO demonstrates UW-VO's superior performance in handling visual degradation and dynamic occlusions.

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

This paper presents an innovative approach to visual odometry (VO) specifically designed for operation in turbid and dynamic underwater environments. It addresses significant challenges in underwater robotic localization, particularly in contexts such as archaeological missions where Remotely Operated Vehicles (ROVs) lack access to traditional GPS and rely heavily on costly navigation systems. Through the development of a monocular VO system named UW-VO, the authors aim to leverage the ubiquitous camera systems on ROVs to provide effective localization without the need for expensive additional sensors.

Challenges and Methodology

The paper begins by elaborating the inherent difficulties of underwater environments, which include strong light absorption, turbidity due to suspended particles, and visual occlusions from marine life attracted to artificial light sources. Traditional VO and VSLAM methods often falter under these conditions, necessitating a system that can handle visual degradation and dynamic occlusions.

To address these challenges, the authors propose a monocular VO method that utilizes optical flow for feature tracking, alongside a retracking mechanism that is robust to temporary feature loss, a frequent occurrence in dynamic underwater contexts. The system is keyframe-based and employs Bundle Adjustment to maintain consistency in the reconstructed trajectory while minimizing drift.

Evaluation and Results

The performance of UW-VO was extensively evaluated against leading terrestrial SLAM algorithms, such as ORB-SLAM and SVO, on both simulated and real underwater datasets. The results indicated that UW-VO consistently outperforms these algorithms in terms of robustness, particularly in scenarios involving high levels of turbidity and visual occlusions. Notably, UW-VO demonstrated superior capabilities in tracking features in visually degraded environments, a testament to its careful adaptation to underwater conditions.

The paper provides several quantitative evaluations, highlighting that the UW-VO achieves lower translation drift compared to ORB-SLAM and SVO, especially as turbidity and dynamism increase. This performance underscores the method's potential to enhance autonomous navigation and localization in deep-sea explorations where cost constraints prohibit the use of high-end navigational sensors.

Implications and Future Directions

The proposed UW-VO algorithm is poised to significantly benefit underwater archaeological missions by providing a reliable visual-only localization method. Beyond archaeology, the system can be adapted to a variety of underwater applications, wherever submeter accuracy is sufficient and operational budgets preclude the use of high-end sensors.

Future advancements may involve integrating additional sensory data such as inertial measurements to address the scale ambiguity inherent in monocular systems. The paper suggests exploring loop closure mechanisms to evolve the VO system into a full SLAM solution capable of reducing drift in large-scale underwater exploration.

This research lays a foundational framework for future AI developments in underwater robotics, with potential applications extending across various domains reliant on underwater navigation and mapping. Through its innovative approach to tackling underwater visual challenges, UW-VO exemplifies the progress and adaptability of visual odometry systems in increasingly demanding environments.