Slide 001: Mesh Reconstruction from Point Clouds¶
Slide Visual¶
Slide Overview¶
This slide covers the transition from raw point cloud data to triangulated mesh surfaces. Students learn the major surface reconstruction algorithms, their requirements and trade-offs, and how to evaluate the quality of the resulting mesh. This is the essential first step in the post-scan processing pipeline.
Instruction Notes¶
Why Mesh Reconstruction?¶
Point clouds are sets of unconnected 3D points. While useful for visualization and measurement, most downstream applications require continuous surfaces: 3D printing needs a watertight shell to define solid volume; CAD software requires surfaces to fit features against; rendering engines need faces to calculate lighting and shading. Mesh reconstruction bridges this gap by connecting points into a triangulated surface.
Reconstruction Algorithms¶
Poisson Surface Reconstruction The most widely used algorithm for scan data. Poisson reconstruction formulates surface fitting as solving Poisson's equation — it finds a smooth, continuous surface that best fits the point positions and normal vectors. Key characteristics: - Requires oriented normal vectors at every point - Produces watertight (closed) surfaces automatically — but may create geometry in areas with no data - Controlled by an "octree depth" parameter: higher values = more detail but more triangles - Excellent for organic shapes, sculptures, and objects without sharp mechanical edges - Can over-smooth sharp edges and small features
Ball Pivoting Algorithm (BPA) A simpler approach that "rolls" a virtual ball across the point cloud. When the ball contacts three points, it creates a triangle. The ball then pivots to the next point and creates the adjacent triangle. - Does not require normal vectors (though they improve results) - Does not fill gaps — leaves holes where point density is insufficient - Ball radius must be tuned to match point spacing (typically 1.5-2x average point spacing) - Better for preserving sharp edges than Poisson - Produces non-watertight meshes that need hole filling for 3D printing
Screened Poisson Reconstruction An improved version of Poisson that adds a data fidelity term, keeping the surface closer to the actual point positions. This reduces the tendency of standard Poisson to create "phantom" geometry in data-sparse regions. Available in MeshLab and CloudCompare.
Choosing an Algorithm¶
| Factor | Poisson | BPA |
|---|---|---|
| Watertight output | Yes (automatic) | No (holes remain) |
| Requires normals | Yes (critical) | No (helpful) |
| Sharp edge preservation | Poor | Good |
| Gap handling | Fills with estimated surface | Leaves holes |
| Typical use case | Organic shapes, full coverage | Mechanical parts, partial scans |
Practical Workflow in MeshLab¶
- Import point cloud (PLY, PTS, XYZ)
- Compute normals (Filters > Normals > Compute Normals for Point Sets)
- Orient normals consistently (all pointing outward)
- Run reconstruction (Filters > Remeshing > Screened Poisson or Ball Pivoting)
- Clean up: remove disconnected components, trim excess geometry from Poisson "skirt"
- Verify: check manifold status, visual inspection, cross-sections
Key Talking Points¶
- Mesh reconstruction is the critical bridge between scanning and all downstream applications
- Normal vector quality directly determines mesh reconstruction quality
- Poisson produces watertight meshes but may invent geometry; BPA preserves edges but leaves holes
- Algorithm selection depends on the object type and intended application
- Reconstruction is not the final step — cleanup (decimation, smoothing, repair) follows
Learning Objectives (Concept Check)¶
- Explain why mesh reconstruction is necessary for most applications of scan data
- Compare Poisson and Ball Pivoting reconstruction algorithms
- Describe the role of surface normal vectors in mesh reconstruction
- Execute a basic reconstruction workflow in MeshLab or CloudCompare
Last Updated: 2026-03-19