CAR-LOAM: color-assisted robust LiDAR odometry and mapping for solid-state LiDARs

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• LiDAR odometry and mapping (LOAM) is crucial for robot perception, enabling simultaneous pose estimation and high-precision environmental mapping.
• LOAM accuracy is compromised by point correspondence outliers, a significant issue for solid-state LiDARs with irregular sampling.

Purpose of the Study:

• To develop a robust visual-assisted LOAM framework, CAR-LOAM, specifically designed for solid-state LiDARs.
• To address and mitigate the impact of outliers on LOAM accuracy.

Main Methods:

• Incorporation of a perceptually uniform color difference weighting strategy to filter color correspondence outliers.
• Utilization of a robust error metric based on Welsch's function to suppress positional correspondence outliers.
• Integration of visual data to assist LiDAR-based odometry and mapping.

Main Results:

• CAR-LOAM achieves accurate robot localization.
• The system reconstructs dense, precise, colored, and three-dimensional (3D) point cloud maps.
• Experimental results demonstrate superior accuracy compared to existing LOAM methods in challenging scenarios.

Conclusions:

• CAR-LOAM effectively enhances the accuracy and robustness of LiDAR odometry and mapping for solid-state LiDARs.
• The proposed outlier exclusion strategies significantly improve localization and map quality.
• This framework offers a promising solution for advanced robot perception systems.