GICP-Based Registration Flow Improvement and Planar Consistency Evaluation for Heterogeneous Multi-LiDAR Systems in...

Area of Science:

• Robotics and Automation
• Computer Vision
• Geospatial Sensing

Background:

• Single LiDAR sensors in grain warehousing face limitations like blind spots and occlusions.
• Heterogeneous LiDAR systems present registration challenges due to differing characteristics, leading to sparse-dense point cloud issues.
• Reliable 3D environment sensing is critical for efficient grain storage, operational safety, and maintaining grain quality.

Purpose of the Study:

• To propose an improved registration workflow for heterogeneous multi-LiDAR systems in intelligent grain warehousing robots.
• To address challenges in point cloud registration arising from differing LiDAR specifications.
• To enhance the accuracy and reliability of 3D environment sensing for autonomous grain warehousing operations.

Main Methods:

• A GICP-based registration flow improvement method is proposed.
• Techniques include overlap-region cropping, voxel downsampling, and a star-topology registration strategy.
• A novel point-to-plane evaluation metric with cross-LiDAR planar consistency verification is introduced.

Main Results:

• The proposed method significantly reduces point-to-plane error in registration tasks (e.g., 0.1487 m for L0-L1, 0.1090 m for L1-L2).
• Performance surpasses existing methods like ICP, point-to-plane ICP, and NDT.
• Acceptable computational efficiency is maintained, providing reliable geometric support for robot perception and navigation.

Conclusions:

• The engineered GICP-based workflow enhances structural alignment quality for heterogeneous multi-LiDAR systems.
• This method provides reliable geometric support for multi-sensor perception, mapping, and autonomous operations in grain warehousing.
• The workflow is optimized for plane-dominated, semi-static environments, validated for static or low-speed multi-LiDAR registration tasks.