Iterative Differential Entropy Minimization (IDEM) Method for Fine Rigid Pairwise 3D Point Cloud Registration: A...

Area of Science:

• Computer Vision
• 3D Data Processing

Background:

• Point cloud registration aligns 3D data but traditional methods like RMSE and ICP struggle with noisy, low-overlap, or unevenly dense point clouds.
• Existing methods often require a fixed reference point cloud and extensive preprocessing, limiting their real-world applicability.

Purpose of the Study:

• To introduce a novel objective function for fine rigid pairwise 3D point cloud registration that is robust to common data imperfections.
• To develop a registration method that does not rely on a fixed point cloud and demonstrates superior performance in challenging scenarios.

Main Methods:

• A new differential entropy-based metric was developed as the objective function for an optimization framework.
• The proposed method, Iterative Differential Entropy Minimization (IDEM), was implemented for fine rigid pairwise 3D point cloud registration.
• Performance was evaluated through case studies comparing IDEM against RMSE, Chamfer distance, and Hausdorff distance.

Main Results:

• IDEM demonstrated effective alignment even with significant differences in point cloud density, presence of noise, holes, and limited overlap.
• The proposed metric consistently identified the optimal alignment, unlike RMSE which sometimes failed in adverse conditions.
• IDEM showed robustness and outperformed traditional metrics in challenging registration scenarios.

Conclusions:

• Differential entropy provides a robust metric for 3D point cloud registration, overcoming limitations of Euclidean distance-based methods.
• IDEM offers a more reliable and less preprocessing-intensive solution for aligning imperfect 3D point clouds in computer vision applications.