Advancing Global Pose Refinement: A Linear, Parameter-Free Model for Closed Circuits via Quaternion Interpolation

Area of Science:

• Robotics
• Computer Vision
• Geomatics

Background:

• Global pose refinement is crucial for Simultaneous Localization and Mapping (SLAM) systems, especially those using Light Detection and Ranging (LIDAR).
• Drift in pose estimation, caused by successive 3D point cloud registration, leads to inaccuracies in platform path calculation.
• Existing methods often struggle with significant pose differences and computational efficiency in large-scale environments.

Purpose of the Study:

• To propose a novel, linear, and parameter-free model for global pose refinement in LIDAR-based SLAM.
• To address the challenge of trajectory drift correction and improve the accuracy of 3D pose estimation.
• To introduce an efficient coarse-to-fine pairwise registration technique.

Main Methods:

• A linear, parameter-free model utilizing a closed circuit for global trajectory corrections.
• Mapping rotations to quaternions and employing Spherical Linear Interpolation (SLERP) for rotational transitions.
• Least Squares (LS) method for rotation closure constraints and a distinct linear phase for translation adjustment.
• A coarse-to-fine pairwise registration integrating Fast Global Registration and Generalized ICP with multiscale sampling and filtering.

Main Results:

• The proposed model demonstrates effectiveness in 3D pose estimation across diverse point cloud datasets (Mobile and Terrestrial Laser Scanners).
• Significant pose differences were successfully managed, indicating robust performance.
• Efficient pose optimization was observed, particularly in larger trajectory circuits.

Conclusions:

• The developed linear model offers a robust and efficient solution for global pose refinement in LIDAR-SLAM.
• The integration of SLERP and LS-based constraints provides accurate trajectory corrections.
• The coarse-to-fine registration method enhances the overall performance and applicability of the SLAM system.