Panoramic annular SLAM with loop closure and global optimization.

Area of Science:

• Robotics
• Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Visual SLAM systems are crucial for robot navigation and mapping.
• Panoramic annular lenses offer a wide field of view, beneficial for SLAM.
• Existing SLAM methods face challenges in loop closure detection and accumulated error.

Purpose of the Study:

• To propose a novel visual SLAM system, PA-SLAM, utilizing a panoramic annular lens.
• To improve loop closure detection accuracy and robustness.
• To reduce accumulated error and scale drift in SLAM.

Main Methods:

• Developed a PA-SLAM system employing a panoramic annular lens.
• Implemented a hybrid point selection strategy for key point repeatability and bag-of-words loop closure.
• Utilized geometric verification and Sim(3) relative pose estimation for pose graph optimization and global bundle adjustment.

Main Results:

• The hybrid point selection strategy ensures reliable loop closure detection.
• Global optimization significantly reduces accumulated error and scale drift.
• PA-SLAM achieves state-of-the-art accuracy with high robustness and efficiency.

Conclusions:

• PA-SLAM offers a robust and efficient solution for visual SLAM.
• The proposed methods enhance loop closure and minimize localization errors.
• PA-SLAM demonstrates superior performance on real-world datasets.