Geodesic Mapping for Dynamic Surface Alignment.

Area of Science:

• Computer Vision
• Computational Geometry
• 3D Surface Analysis

Background:

• Dynamic surface alignment is crucial for analyzing non-rigid objects in motion.
• Existing methods often rely on appearance or parameterization, limiting their applicability.
• Multiview stereo reconstruction provides 3D manifold meshes of deforming objects.

Purpose of the Study:

• To develop a novel dynamic surface alignment method using geodesic mapping.
• To achieve accurate point-to-point correspondence and smooth mapping based solely on intrinsic geometry.
• To enable applications like texture transfer and motion flow estimation for 3D dynamic surfaces.

Main Methods:

• Geodesic mapping (geodesic diffeomorphism) between surfaces with global geodesic distance.
• Geodesic-based coordinate system for defining correspondence points via a coarse-to-fine strategy.
• Probabilistic framework optimizing a piecewise objective function for complete surface alignment.

Main Results:

• Reliable point-to-point surface alignment with smooth mapping achieved.
• Method demonstrated robustness without requiring surface appearance, shape, or parameterization.
• Compelling results shown on challenging real-world datasets.

Conclusions:

• The proposed geodesic mapping approach offers a powerful, intrinsic geometry-based solution for dynamic surface alignment.
• The technique is versatile, applicable to various computer vision tasks involving non-rigid 3D objects.
• Future work can explore extensions for more complex deformations and object types.