Variational Level Set Evolution for Non-Rigid 3D Reconstruction From a Single Depth Camera.

Area of Science:

• Computer Vision
• Geometric Modeling
• 3D Reconstruction

Background:

• Real-time 3D reconstruction of dynamic, non-rigid surfaces is challenging.
• Existing methods struggle with topological changes and rapid motions.
• Accurate surface reconstruction requires robust handling of data association and geometric consistency.

Purpose of the Study:

• To develop a novel framework for real-time 3D reconstruction of non-rigidly moving surfaces.
• To enable accurate reconstruction despite topological changes and fast motions.
• To improve geometric consistency and data association in 3D surface reconstruction.

Main Methods:

• Utilizes a variational level set method to warp truncated signed distance fields (TSDFs).
• Employs gradient flow optimization with a data term for voxel-wise alignment.
• Incorporates regularizers like Killing vector fields and Sobolev space gradient flow for geometric consistency.
• Leverages Laplacian eigenfunctions for implicit and explicit voxel correspondence estimation.

Main Results:

• The framework achieves real-time 3D reconstruction of non-rigid surfaces.
• It successfully captures rapid motions and topological changes, outperforming related techniques.
• Demonstrates high geometric reconstruction fidelity and voxel correspondence accuracy.
• Offers strategies for both moderate and large motion scenarios.

Conclusions:

• The proposed framework offers a robust solution for real-time 3D surface reconstruction.
• It effectively handles complex scenarios involving fast motions and topological changes.
• The use of Laplacian eigenfunctions provides a powerful mechanism for data association in dynamic scenes.