High-Precision Registration of Point Clouds Based on Sphere Feature Constraints.

Area of Science:

• Computer Vision
• Geometric Computing
• 3D Metrology

Background:

• Point cloud registration is crucial for multi-view 3D measurements, but precision is challenged by non-overlapping areas and curvature-invariant surfaces.
• Existing methods struggle to achieve high accuracy in these difficult scenarios, impacting overall measurement reliability.

Purpose of the Study:

• To introduce a high-precision point cloud registration method utilizing sphere feature constraints.
• To address the limitations of current registration techniques in scenarios with limited overlap or featureless surfaces.

Main Methods:

• A novel approach employing sphere feature constraints to create virtual overlapping areas.
• Utilizing these virtual areas to generate more accurate corresponding point pairs and mitigate noise.
• Employing an optimization method with a weight function to solve for transformation parameters.

Main Results:

• The proposed method effectively generates accurate corresponding point pairs even with non-overlapping data.
• Significant reduction in the influence of noise on registration accuracy was observed.
• High-precision registration was achieved, validated through simulations and experimental tests.

Conclusions:

• The sphere feature constraint method offers a robust solution for challenging point cloud registration scenarios.
• This technique enhances the precision of 3D measurements by improving registration accuracy.
• The method demonstrates practical applicability and effectiveness in real-world applications.