Acceleration of Non-Rigid Point Set Registration With Downsampling and Gaussian Process Regression

Area of Science:

• Computer Vision
• Computational Geometry
• Medical Imaging

Background:

• Non-rigid point set registration aligns shapes by deforming one point set to match another.
• Accurate registration is crucial for applications in medical imaging, computer graphics, and robotics.
• Existing methods can be computationally intensive, especially for large-scale datasets.

Purpose of the Study:

• To develop an efficient acceleration method for non-rigid point set registration.
• To enable registration of point sets with millions of points.
• To outperform current state-of-the-art acceleration techniques.

Main Methods:

• The proposed method divides registration into downsampling, registration of downsampled sets, and interpolation of deformation vectors.
• A registration algorithm utilizing a motion coherence prior is employed for downsampled point sets.
• Gaussian process regression is used for interpolating shape deformation.

Main Results:

• The algorithm successfully registers point sets containing over ten million points.
• Demonstrated significant reduction in computing time compared to existing acceleration methods.
• The motion coherence prior and Gaussian process regression effectively handle complex deformations.

Conclusions:

• The proposed three-step acceleration method offers a radical improvement in computational efficiency for non-rigid point set registration.
• This technique is scalable to very large point sets, opening new possibilities in various scientific fields.
• The approach provides a robust and faster alternative to current registration acceleration algorithms.