DOE-LVI: Tightly Coupled LiDAR-Visual-Inertial SLAM System with Dynamic Object Elimination.

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Simultaneous Localization and Mapping (SLAM) systems face significant challenges in dynamic environments due to moving objects.
• Existing SLAM methods often exhibit reduced accuracy and robustness when encountering dynamic elements.

Purpose of the Study:

• To develop an advanced, tightly coupled LiDAR-Visual-Inertial SLAM system capable of effectively handling dynamic objects.
• To enhance the robustness and accuracy of SLAM in complex, real-world dynamic environments.

Main Methods:

• Proposed Dynamic-Object-Elimination LiDAR-Visual-Inertial SLAM (DOE-LVI) system integrating Visual-Inertial System (VIS) and LiDAR-Inertial System (LIS).
• VIS extracts depth from LiDAR, correlates with visual features, and minimizes visual/IMU residuals for pose estimation.
• LIS refines dynamic point removal using range images, ground fitting, and submap scan matching; employs global LiDAR descriptors for loop closure.

Main Results:

• DOE-LVI demonstrates robust localization and mapping performance in dynamic environments.
• The system effectively mitigates the impact of moving objects on SLAM accuracy.
• Evaluations on KITTI and UrbanNav datasets confirm superior performance compared to existing methods.

Conclusions:

• DOE-LVI offers a significant advancement in SLAM technology for dynamic scenarios.
• The proposed method provides reliable and accurate localization and mapping in challenging real-world conditions.
• DOE-LVI enhances the applicability of SLAM in robotics and autonomous systems operating in cluttered and moving environments.