Convex Solutions to SfT and NRSfM Under Algebraic Deformation Models.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Geometric Modeling

Background:

• Existing Shape-from-Template (SfT) and Non-Rigid Structure-from-Motion (NRSfM) methods often rely on impractical assumptions like known optic flow or inextensibility.
• These limitations hinder accurate 3D shape recovery from 2D imagery.

Purpose of the Study:

• To develop novel nonlinear formulations for SfT and NRSfM that overcome limitations of prior work.
• To faithfully exploit isometric, conformal, and equiareal deformation models for improved 3D reconstruction.
• To resolve ambiguities using maximal depth and maximal isometry heuristics without requiring optic flow.

Main Methods:

• Proposed nonlinear formulations for SfT and NRSfM utilizing isometric, conformal, and equiareal deformation models.
• Developed solution methods based on Semi-Definite Programming (SDP).
• Introduced an adapted opposite-depth parameterization to address conflicts between SDP and maximal depth heuristics, reducing the relaxation gap.

Main Results:

• The proposed methods require only point correspondences, eliminating the need for optic flow.
• Experimental results on benchmark datasets demonstrate superior accuracy compared to existing SfT and NRSfM techniques.
• The adapted parameterization showed a reduced relaxation gap in SDP models.

Conclusions:

• The novel nonlinear formulations provide a more robust and accurate approach to 3D shape recovery using SfT and NRSfM.
• The reliance on point correspondences and novel heuristics offers practical advantages over previous methods.
• The study advances the state-of-the-art in non-rigid 3D reconstruction.