Fusion: build a smooth approximate mesh from a point cloud

Build a smooth, continuous mesh from one or several noisy point clouds

3 min read
In the app
  • PointsFusion

Fusion builds a smooth mesh from a point cloud. It does not connect the points: it builds a volume around them on a grid of small cubes (voxels) and extracts a new surface from it. The noise of the points is averaged out, so Fusion is a good choice for noisy or uneven scans. Because the surface is rebuilt, it can deviate slightly from the original points, and details smaller than the voxel size are lost. To keep every point as a vertex, use Points to Mesh instead; to compare all methods, see How to Turn Point Clouds into Mesh.

The ribbon tooltip mentions a point cloud "with normals", but you do not need to compute them first: if the cloud has no normals, Fusion computes them itself. If the cloud already has normals, Fusion uses them as they are, so make sure they are oriented consistently (Orient Point Normals).

Fusion: Where to Find

Build a mesh with Fusion

  1. Select one or more point clouds in the Scene tree.
  2. Click Points › Fusion.
  3. Check the Voxel Size and, if needed, the settings under Advanced.
  4. Click Fuse, or press Enter.
Setting What it does Default
Voxel Size The size of the cubic cells (voxels) of the volume. Smaller voxels capture finer details but make the computation slower and the mesh heavier. Proposed from the size of the selected cloud, for example 0.754 mm for a bust about 0.5 m tall
Avg Influence Neighbours (under Advanced) Sets the distance of influence so that on average this many points lie within it. Higher values give a smoother surface but take longer. 20 (from 6 to 999)
Min Influence Neighbours (under Advanced) The minimum total influence of nearby points for a triangle to appear: there must be at least about this many points close by. Higher values leave holes where points are sparse. 5
Fuse All in One (under Advanced) Shown only when several point clouds are selected. On: all clouds are fused into one mesh. Off: each cloud becomes a separate mesh. On
Preserve Point Clouds (under Advanced) Keeps the original point clouds in the scene in addition to the mesh. Off

On systems that support CUDA, Advanced also shows Use NVIDIA GPU: with a CUDA-capable NVIDIA graphics card, Fusion can use it for faster processing.

Fusion: Panel

Result

  • One cloud: a new mesh named after the cloud with fused added, for example Nefertiti Points fused. Several clouds with Fuse All in One: one mesh named Fusion result.
  • If the points have colors, the mesh vertices get colors from them.
  • Without Preserve Point Clouds, the clouds are removed from the scene. With it, they stay in the scene and are only unselected, so hide them to see the mesh.
  • To undo, press Ctrl+Z (Cmd+Z on macOS).

Fusion: Result

Fusion takes longer than Points to Mesh: on the scan in the example below, about five times longer. The time grows quickly as the Voxel Size gets smaller. If it takes too long, increase the Voxel Size, or reduce the cloud with Sample Points first.

Example and tips

On a clean scan of the bust of Nefertiti (1 009 118 points, no normals), Fusion with the default settings made a smooth mesh with 1 799 535 triangles. It also left 193 small holes and 5 components (the yellow spots on the cheek and neck). On such a clean, dense scan, Points to Mesh gives a better result; Fusion pays off on noisy data.

The mesh Nefertiti Points fused made by Fusion: 1 799 535 triangles, 193 holes, 5 components; the holes are visible as yellow spots on the cheek and neck.

  • Holes in the result? Lower Min Influence Neighbours or raise Avg Influence Neighbours, or close the holes afterwards with Fill Holes.
  • Small loose pieces? Remove them with Keep Largest Component (see Separate Components).
  • Mesh too heavy? Use a larger Voxel Size, or reduce the mesh with Simplify.

Customize Consent Preferences

We use cookies and similar technologies, such as local storage. You will find detailed information about all of them under each consent category below.

To learn more about how Google processes personal information, visit: Google privacy policy

Always Active