Trajectory-based visual localization in underwater surveying missions.

Area of Science:

• Robotics
• Computer Vision
• Navigation Systems

Background:

• Traditional Extended Kalman Filter Simultaneous Localization and Mapping (EKF-SLAM) methods are computationally intensive for autonomous underwater vehicles (AUVs).
• AUVs often have limited sensing and processing capabilities, posing challenges for accurate localization.
• Reliable data association and loop closure are critical for robust SLAM performance.

Purpose of the Study:

• To develop a computationally efficient, vision-based localization system for AUVs.
• To improve the accuracy of pose and trajectory estimation in underwater environments.
• To address the limitations of traditional EKF-SLAM in resource-constrained AUVs.

Main Methods:

• Implemented a trajectory-based schema to reduce computational requirements of EKF-SLAM.
• Utilized a visual odometer for motion prediction and Kalman filter for pose estimation.
• Employed image registration for robust data association and loop closure.
• Evaluated both Extended Kalman Filter (EKF) and Iterated Extended Kalman Filter (IEKF) to mitigate linearization errors.

Main Results:

• The trajectory-based approach significantly reduced computational load compared to standard EKF-SLAM.
• The vision-based system demonstrated excellent performance in real-world underwater experiments.
• Achieved very small errors in both vehicle pose estimation and overall trajectory reconstruction.
• IEKF showed potential in further refining pose estimations by addressing linearization errors.

Conclusions:

• The proposed vision-based localization system offers an efficient and accurate solution for AUV navigation.
• The trajectory-based schema effectively enhances EKF-SLAM for AUVs with limited resources.
• The system's reliability in real underwater conditions validates its practical applicability.