Cubature Split Covariance Intersection Filter-based Point Set Registration.

Area of Science:

• Computer Vision
• Robotics
• Geometric Computing

Background:

• Point set registration is crucial for many computer vision tasks but remains challenging due to simultaneous error sources like noise and outliers.
• Existing methods often struggle with combined error types or extreme conditions, limiting their real-world applicability.
• Robustness and precision are key requirements for reliable point set registration algorithms.

Purpose of the Study:

• To develop a robust point set registration algorithm capable of handling multiple simultaneous error sources.
• To improve the precision and reliability of point set registration in the presence of noise and outliers.
• To introduce a novel filtering framework that addresses the limitations of previous registration methods.

Main Methods:

• The point set registration problem is modeled as a non-linear state space model.
• A split covariance intersection filter (SCIF) is employed to capture correlations between state transitions and observations.
• A recursive cubature split covariance intersection filter is derived, approximating non-linearity with a third-order term.

Main Results:

• The proposed algorithm demonstrates enhanced robustness and precision compared to existing filtering-based methods.
• Experimental validation on public datasets confirms the algorithm's effectiveness against noise and outliers.
• The algorithm outperforms state-of-the-art methods in specific aspects of point set registration.

Conclusions:

• The developed recursive cubature split covariance intersection filter provides a robust solution for point set registration.
• This approach effectively handles simultaneous dependent and independent error sources, improving registration accuracy.
• The algorithm offers a significant advancement for computer vision applications requiring precise point set alignment.