Template-Based Monocular 3D Shape Recovery Using Laplacian Meshes.

Area of Science:

• Computer Vision
• Computer Graphics
• Geometric Modeling

Background:

• Monocular 3D shape reconstruction of deformable surfaces is an ill-posed problem.
• Existing methods struggle with outlier elimination and real-time performance.
• The Laplacian formalism is used in computer graphics for mesh regularization.

Purpose of the Study:

• To transform monocular 3D shape reconstruction into a well-posed problem.
• To develop a robust and efficient method for deformable surface reconstruction.
• To enable real-time 3D surface reconstruction from monocular images.

Main Methods:

• Extending the Laplacian formalism for deformable surface reconstruction.
• Utilizing correspondences with a reference image.
• Solving a linear least squares problem for initial shape estimation.
• Refining the initial estimate via constrained optimization.

Main Results:

• The extended Laplacian formalism significantly improves the well-posedness of the reconstruction problem.
• Outliers are reliably and quickly eliminated through a linear least squares solution.
• An accurate initial 3D shape estimate is obtained, with accurate 2D projections.
• The final surface reconstruction is achieved through constrained optimization.

Conclusions:

• The proposed method provides a robust and accurate solution for monocular 3D shape reconstruction of deformable surfaces.
• The approach effectively reduces the dimensionality of the problem without compromising accuracy.
• The method facilitates real-time implementation, opening possibilities for interactive applications.