LIO-SAM++: A Lidar-Inertial Semantic SLAM with Association Optimization and Keyframe Selection.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Current lidar-inertial SLAM (Simultaneous Localization and Mapping) algorithms primarily use geometric features for point cloud alignment.
• This approach is vulnerable to dynamic objects, occlusion, and environmental changes, leading to incorrect feature associations and pose estimation errors.

Purpose of the Study:

• To develop a robust lidar-inertial SLAM system that enhances pose estimation accuracy by integrating semantic and geometric constraints.
• To improve keyframe selection and association optimization within the LIO-SAM framework.

Main Methods:

• The proposed system enhances the LIO-SAM framework by incorporating semantic information alongside geometric constraints.
• It mitigates erroneous matching points by comparing the consistency of normal vectors in local regions.
• An adaptive keyframe selection strategy based on semantic differences between frames is introduced.

Main Results:

• The integrated semantic and geometric constraints significantly improve matching accuracy.
• The adaptive keyframe selection enhances the reliability of generated keyframes.
• Experimental results on the KITTI dataset demonstrate a substantial improvement in pose estimation accuracy compared to existing systems.

Conclusions:

• The novel lidar-inertial SLAM system effectively addresses the limitations of purely geometric approaches.
• Combining semantic and geometric information leads to more accurate and reliable robot localization.
• The proposed method offers a significant advancement for real-world SLAM applications.