Local Deformable 3D Reconstruction with Cartan's Connections.

Area of Science:

• Computer Vision
• Geometric Modeling
• 3D Reconstruction

Background:

• Deformable object 3D reconstruction from monocular images is challenging.
• Existing methods often rely on simplified deformation models like isometry.
• Current approaches may use discrete points or non-parametric representations with limited constraints.

Purpose of the Study:

• To develop a unified differential framework for Shape-from-Template (SfT) and Non-Rigid Structure-from-Motion (NRSfM).
• To extend 3D reconstruction to a broader class of deformation models beyond isometry.
• To provide a generic SfT solution independent of specific deformation models.

Main Methods:

• Utilizing Cartan's theory of connections and moving frames for a differential approach.
• Applying infinitesimal-level assumptions on surface geometry and mappings.
• Developing a novel skewless deformation model for SfT and NRSfM.

Main Results:

• The proposed framework unifies existing SfT and NRSfM solutions, simplifying their derivation.
• It enables 3D reconstruction for a comprehensive range of algebraic deformation models, including isometry.
• A new skewless deformation model was introduced and solved.
• A generic SfT solution was achieved without requiring explicit deformation modeling.

Conclusions:

• The differential framework offers a complete and unified solution for deformable 3D reconstruction.
• The method demonstrates superior accuracy and computational efficiency compared to state-of-the-art techniques.
• This approach advances the field by handling complex deformations and simplifying existing methods.