Pairwise Registration Algorithm for Large-Scale Planar Point Cloud Used in Flatness Measurement.

Area of Science:

• Computer Vision
• Metrology
• Geospatial Data Analysis

Background:

• Accurate flatness measurement is critical in various industrial applications.
• Existing 3D point cloud registration methods can struggle with large-scale planar datasets, impacting measurement precision.
• Laser profilometry generates dense point clouds requiring robust processing for metrology.

Purpose of the Study:

• To develop and validate an optimized 3D pairwise point cloud registration algorithm for precise flatness measurement.
• To achieve fast and accurate six-degrees-of-freedom (6-DoF) pose estimation of large-scale planar point clouds.
• To improve the performance of flatness measurements obtained via laser profilometry.

Main Methods:

• Proposed an algorithm that extracts point cloud boundaries for enhanced feature descriptors.
• Implemented neighborhood evaluation to eliminate invalid keypoints and establish initial matching pairs.
• Utilized clustering with geometric consistency constraints for coarse registration, followed by Iterative Closest Point (ICP) for fine registration.

Main Results:

• The proposed algorithm demonstrated superior boundary extraction compared to existing methods.
• Experimental results confirmed enhanced registration performance, leading to more precise flatness measurements.
• The algorithm achieved fast and accurate 6-DoF pose estimation for large-scale planar point clouds.

Conclusions:

• The optimized 3D point cloud registration algorithm effectively improves flatness measurement accuracy and efficiency.
• Boundary extraction and geometric consistency constraints are key to robust registration of planar point clouds.
• The developed method offers a significant advancement for metrology applications relying on laser profilometry data.