Stereo and LiDAR Loosely Coupled SLAM Constrained Ground Detection

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Geospatial Mapping

Background:

• Accurate 3D mapping is essential for robotic navigation and obstacle avoidance.
• Existing projection-based and voxel-based 3D point cloud processing methods have limitations.
• Single-sensor systems often struggle with robust localization and mapping.

Purpose of the Study:

• To propose a hybrid localization and mapping method using stereo vision and LiDAR.
• To develop a pose optimization model by fusing ground information from both sensors.
• To improve the accuracy and robustness of 3D mapping for robotic applications.

Main Methods:

• Utilized stereo vision to extract ground features and LiDAR tensor voting data.
• Fused visual and LiDAR data to establish coplanarity constraints for pose optimization.
• Employed graph-based optimization and local window optimization for pose refinement.

Main Results:

• Demonstrated significant improvements in trajectory accuracy and robustness compared to ORB-SLAM3, F-LOAM, LOAM, and LeGO-LOAM.
• Successfully generated clear, dense 3D point cloud maps.
• Validated performance using the KITTI dataset.

Conclusions:

• The proposed hybrid stereo vision and LiDAR approach offers superior performance for 3D mapping and localization.
• This method overcomes limitations of traditional single-sensor systems.
• Enables the creation of high-definition 3D maps crucial for advanced robotic tasks.