Efficient Registration for Human Surfaces via Isometric Regularization on Embedded Deformation.

Area of Science:

• Computer Vision
• Geometric Modeling
• Computational Geometry

Background:

• 3D registration is crucial for surface correspondence, often using Iterative Closest Point (ICP) based non-rigid deformation.
• Embedded deformation methods accelerate real-time applications but struggle with intricate surfaces due to simplified structural regularization.
• Existing methods often assume uniform surface rigidity, leading to suboptimal performance and slow convergence without extensive parameter tuning.

Purpose of the Study:

• To develop a novel mesh alignment solution that decouples regularization from the deformation model.
• To explicitly manage the rigidity of vertex clusters for improved accuracy in 3D registration.
• To enhance efficiency and performance in non-rigid surface alignment tasks.

Main Methods:

• Proposed a novel solution decoupling regularization from the underlying deformation model by managing vertex cluster rigidity.
• Designed an efficient two-step solution alternating between isometric and embedded deformation with cluster-based regularization.
• Enabled region-adaptive regularization through cluster refinement for efficient execution.

Main Results:

• Demonstrated effectiveness in mesh alignment tasks, even with large-scale deformation and imperfect data.
• Achieved superior numerical and visual performance compared to state-of-the-art methods.
• Showcased efficient execution and support for region-adaptive regularization.

Conclusions:

• The proposed method effectively addresses limitations of traditional and embedded deformation techniques in 3D registration.
• Explicitly managing vertex cluster rigidity leads to more accurate and robust mesh alignment.
• The efficient two-step approach offers a significant advancement for real-time and complex surface registration applications.