Mesh Generation

The Marching Cubes algorithm is the definitive method for extracting a polygonal mesh from a 3D scalar field, such as a Signed Distance Function (SDF) or a Truncated Signed Distance Function (TSDF). It works by 'marching' a cube through the volumetric grid and, at each cell, consulting a pre-computed lookup table to generate triangles based on the field values at the cube's eight corners.

• Core Function: Converts implicit surfaces (defined by an isovalue) into explicit triangle meshes.
• Primary Use: Essential for surface reconstruction from volumetric representations common in RGB-D SLAM and neural implicit representations.
• Limitation: Can produce ambiguous topology and often requires post-processing like mesh smoothing to reduce stair-step artifacts.

Poisson Surface Reconstruction is an algorithm that creates a smooth, watertight mesh from oriented point clouds (points with surface normals). It formulates surface reconstruction as solving a Poisson equation, interpreting the oriented points as samples of an indicator function whose gradient is a vector field.

• Mathematical Basis: Solves for the implicit surface whose gradient best matches the vector field defined by the oriented normals.
• Key Strength: Excellent at creating closed, manifold surfaces from noisy, incomplete data, making it robust for photogrammetry and LiDAR scans.
• Output: Produces a smooth, implicit function that is then meshed using an algorithm like Marching Cubes.

The Ball-Pivoting Algorithm (BPA) is a surface reconstruction method that operates directly on a point cloud. Conceptually, a ball of a user-defined radius 'rolls' over the point set, connecting three points to form a triangle whenever the ball can touch them without containing any other point.

• Mechanism: A geometric, region-growing approach that builds triangles based on local point density.
• Ideal For: Dense, uniformly sampled point clouds, such as those from structured light scanning or high-quality Multi-View Stereo (MVS).
• Consideration: The chosen ball radius is critical; it must be larger than the largest hole to be bridged but smaller than the spacing of any surface feature that should remain a hole.

This approach uses computational geometry to create meshes from unorganized points. A Delaunay Triangulation in 3D creates a tetrahedralization of the point cloud that maximizes the minimum angle of all tetrahedra. Alpha Shapes then define the surface by filtering these tetrahedra based on a scale parameter (alpha).

• Process: First, compute the 3D Delaunay triangulation. Then, remove tetrahedra whose circumsphere radius is greater than the alpha value. The remaining triangles form the surface mesh.
• Flexibility: The alpha parameter controls the level of detail—small alpha captures fine details, large alpha creates a convex hull.
• Application: Useful for reconstructing surfaces with sharp features and complex boundaries from point clouds.

Screened Poisson Reconstruction is an advanced variant of Poisson reconstruction that incorporates point constraints into the linear system. This allows the solver to better preserve the positions of the original input points, leading to more accurate geometric fidelity compared to the standard method.

• Enhancement: Adds a screening term that penalizes deviation of the reconstructed surface from the input sample points.
• Benefit: Reduces the over-smoothing tendency of classic Poisson reconstruction, better preserving sharp edges and fine details from Structure from Motion (SfM) outputs.
• Typical Workflow: The method takes oriented points, often with confidence weights, and outputs a higher-fidelity implicit function for meshing.

Raw reconstructed meshes are often excessively dense. Mesh simplification (or decimation) reduces the polygon count while preserving the overall shape, boundaries, and error tolerance. This is critical for real-time applications in AR/VR and spatial computing.

• Quadric Error Metrics (QEM): The most common algorithm. It iteratively collapses edges, choosing the contraction that minimizes the geometric error, defined by a quadric matrix for each vertex.
• Goal: Produce Level-of-Detail (LOD) meshes for efficient rendering and transmission.
• Post-Processing Step: Often paired with mesh smoothing and re-meshing to ensure clean topology and even triangle distribution after simplification.