ICP registration with DCA descriptor for 3D point clouds.

Area of Science:

• Computer Vision and Geometry
• 3D Data Processing

Background:

• Point cloud registration is crucial for 3D modeling and reverse engineering, aligning point clouds from different viewpoints.
• The standard Iterative Closest Point (ICP) algorithm often fails due to local minima, requiring a good initial guess and minimal occlusion.

Purpose of the Study:

• To develop a more robust and efficient point cloud registration method.
• To overcome the limitations of the traditional ICP algorithm regarding initial values and data quality.

Main Methods:

• A novel 3D local feature descriptor, Density, Curvature, and Normal Angle (DCA), was designed.
• The DCA descriptor combines density, curvature, and normal information for robust feature description.
• The algorithm establishes correspondences and an initial registration using DCA features, then refines with ICP.

Main Results:

• The proposed ICP algorithm demonstrates improved registration accuracy and robustness on public datasets.
• The method achieves a fast running speed compared to traditional approaches.
• The DCA-based initial estimation effectively prevents ICP from converging to local minima.

Conclusions:

• The DCA feature descriptor significantly enhances the performance of ICP for point cloud registration.
• This improved ICP algorithm offers a reliable solution for 3D modeling and reverse engineering applications.
• The method provides accurate, robust, and efficient point cloud registration, even with challenging datasets.