Point Cloud to Mesh: Methods, Cleanup and When to Use a Mesh

Point Cloud to Mesh: Methods, Cleanup and When to Use a Mesh

A laser scanner does not capture surfaces. It captures points: millions of separate measurements, each with a position and often a color and intensity value. Between those points there is nothing. To get a continuous surface that can be 3D printed, rendered, shared as a lightweight file or used as the base for reverse engineering, the point cloud has to be turned into a mesh.

This article explains what a mesh is, how point cloud to mesh conversion works, which steps decide the quality of the result, which file formats are used, and when a mesh is the right deliverable instead of a CAD drawing or a BIM model.

What Is a Mesh?

A mesh is a surface made of small flat faces, almost always triangles. Each triangle connects three vertices, and neighboring triangles share edges. Together they form a continuous skin that follows the shape of the scanned object.

The difference from a point cloud is simple:

  • A point cloud is a set of separate samples. You can see the object in it, but it has no surfaces, so it cannot be printed, measured as an area or volume, or textured as one object.
  • A mesh connects those samples into faces. It has surfaces, normals that show which side is “outside”, and, if it is closed, a volume.

The difference from a CAD or BIM model matters just as much. A mesh describes shape only. It does not know that a group of triangles is a wall, a pipe or a bolt hole, and it has no parameters such as thickness, diameter or material. It is a very detailed digital copy of a surface, not an engineering model.

How Point Cloud to Mesh Conversion Works

Converting a point cloud to a mesh is not a single button. The result depends far more on preparation than on the meshing command itself. A typical workflow has five steps.

Step 1: Register and Check

Most objects need several scan positions. Before meshing, all scans are joined into one point cloud. If two scans are misaligned, the mesh will show it as doubled surfaces, steps or ridges along the joint. Registration is checked on section cuts and overlap areas before any surface is built. We explain this stage in how point cloud registration works.

Step 2: Clean

Every scan contains points that do not belong to the object: people walking by, reflections from glass and polished metal, mixed points along edges, dust, vegetation, tripods and stray points in the air. A meshing algorithm treats every point as part of a surface, so each of these becomes a lump, spike or false wall. Cleaning removes them and, where needed, reduces noise on surfaces. The process is described in more detail in our article on point cloud cleaning.

Step 3: Estimate and Orient Normals

A normal is a direction that shows which way a surface faces. Many meshing methods need a normal for every point to know which side is inside and which is outside. Normals are estimated from neighboring points, and their orientation is checked. If part of the normals point inward, the mesh can come out inside-out or broken in that area. Data from terrestrial scanners has an advantage here: each scan position is known, so normals can be oriented toward the scanner.

Step 4: Build the Surface

This is the meshing itself. The main methods are compared in the table below.

Step 5: Edit, Reduce and Check

The raw mesh is then repaired, simplified to a usable size, filled where required and compared with the original point cloud. These steps are covered in the sections that follow.

Meshing Methods Compared

Different algorithms suit different data. The names vary between programs, but most tools offer some version of these four.

MethodHow it worksBest forWatch out for
Direct triangulationConnects neighboring points into triangles; the surface passes through the measured pointsDense, clean scans of objects where every detail should stayKeeps noise; leaves holes where data is missing
Poisson reconstructionFits a smooth, closed surface to the points using their normalsOrganic and sculpted shapes, heritage objects, closed models for printingCan invent surface across gaps; sharp edges get rounded; needs good normals
Voxel or fusion methodsDivides space into small cubes and builds the surface from themNoisy or uneven data, quick previews, large scenesDetail is limited by voxel size; fine edges soften
Alpha shape and similarConnects points within a set radius; the radius decides how big a gap is closedSimple objects and quick envelopesWrong radius either leaves holes or closes real openings

For engineering work, a common choice is direct triangulation on a well-cleaned cloud, because the surface stays on the measured points. For sculpture, ornament and objects that must be printed, Poisson reconstruction often gives a better looking, closed result, but every area where it filled a gap should be known and checked.

Decimation: Making the Mesh Usable

A mesh built from a dense laser scan can contain tens of millions of triangles. That is often more than a viewer, a game engine, a slicer for 3D printing or a modeling program can handle comfortably.

Decimation reduces the number of triangles while keeping the shape. Good decimation is adaptive: it removes many triangles on flat areas, where they add nothing, and keeps them on edges, curves and fine detail. The key setting is the allowed deviation from the original surface, not a percentage. A mesh reduced to a small fraction of its triangles can still follow the scan closely if the simplification was controlled.

Two practical rules:

  • Keep the full-resolution mesh as the master and create lighter versions for each use: one for web viewing, one for rendering, one for printing.
  • Agree on the target file size or triangle count with whoever will use the mesh before delivery.

Holes, Watertight Meshes and Repair

A scanner records only what it can see. Undersides, deep recesses, areas behind other objects and very dark or shiny surfaces often leave gaps. A mesh built from such data has holes.

A watertight mesh is one with no holes: a fully closed surface that separates inside from outside. It is also manifold, meaning every edge is shared by exactly two triangles, with no stray faces, self-intersections or flipped normals. Watertight meshes are needed for:

  • 3D printing, because the slicer has to know what is solid;
  • volume calculations;
  • some simulation and analysis tasks.

For visualization and many measurement tasks, a mesh does not have to be watertight.

Hole filling is useful, but it is an assumption, not a measurement. A filled patch shows a plausible surface where no data exists. On a heritage object or a part for reverse engineering, it is good practice to mark filled areas, or deliver them as a separate layer, so nobody later measures a surface that was never scanned.

Typical repair tasks also include removing small floating pieces, fixing flipped normals, smoothing noise without losing edges and splitting the mesh into logical parts.

Color and Texture

Laser scans often carry color from the scanner’s camera. On a mesh, that color can be stored in two ways:

  • Vertex color: each vertex has a color. Simple, but the color detail drops when the mesh is decimated.
  • Texture: an image is wrapped onto the mesh through a UV map. The geometry can be light while the surface still looks detailed. This is the usual choice for visualization and heritage presentation.

Mesh File Formats

The format should match where the mesh is going. These are the most common ones.

FormatStoresTypical use
OBJGeometry, UV coordinates; textures through a separate material fileVisualization, exchange between 3D programs, textured heritage models
STLTriangles only, no color or units3D printing, many CAD and manufacturing tools
PLYGeometry with per-vertex color and other attributesScan data exchange, archiving colored meshes
FBXGeometry, materials, textures, scene dataAnimation, game engines, visualization
glTF / GLBGeometry and textures in a compact, web-friendly formOnline viewers, AR and web presentation

Two things to check with any format: the units (STL in particular does not store them, so a model in meters can arrive a thousand times too small or too large) and the coordinate system, if the mesh has to sit in the same position as other project data.

How to Check a Mesh Against the Scan

A mesh always differs slightly from the point cloud. Triangles are flat, decimation removes detail, smoothing moves surfaces, and hole filling adds surface where there were no points. The question is whether these differences stay within what the project needs.

The usual check is a deviation analysis: the mesh is compared with the original point cloud, and the distance between them is shown as a color map. Areas where the mesh departs from the scan stand out immediately and can be reviewed. The scan itself sets the limit: our FARO laser scans have a typical accuracy of ±2–3 mm, and a mesh cannot be more accurate than the data it was built from.

Mesh vs CAD vs BIM: Which Deliverable Do You Need?

A mesh, a CAD model and a BIM model can all be built from the same scan. They answer different questions.

MeshCAD modelBIM model
What it isTriangulated surface that copies the scanned shapeClean geometry: lines, solids, NURBS surfaces with exact dimensionsBuilding elements with geometry and information
Follows real imperfectionsYes, including wear, damage and deformationUsually simplified to design intentUsually simplified to agreed modeling rules
Editable as designHard: changing a hole size means editing many trianglesYes, by dimensions and featuresYes, by element type and parameters
Carries informationShape and color onlyDimensions, sometimes featuresElement types, materials, parameters, .ifc export
Best forVisualization, heritage records, 3D printing, deviation checks, base for reverse engineeringManufacturing, drawings, fabrication of replacement partsDesign, renovation, coordination and facility management of buildings

In many projects the answer is not “one or the other”. A mesh is often an intermediate step: a reverse engineering specialist uses it as the reference for building a parametric CAD model, or a heritage project keeps a detailed mesh of ornament while the building itself is modeled in Revit. If you are deciding between drawings and models, our article on point cloud to CAD explains the CAD route.

When a Mesh Is the Right Choice

Reverse Engineering

For a part with no drawings, a mesh is the first digital copy of its real shape. It is used to measure the part, check wear and deformation, and serve as the reference surface for a CAD model that can be manufactured. Simple prismatic parts are often modeled directly from the cloud; complex freeform parts usually go through a mesh first. See our reverse engineering services for how scanning and modeling work together.

Heritage and Historic Objects

Carved stone, sculpture, moldings, ornament and damaged surfaces are poorly described by straight lines and parametric families. A textured mesh records their real condition: cracks, losses and weathering. It is used for conservation records, condition monitoring, restoration planning and making replicas. Learn more about our heritage scanning services.

Visualization and Presentation

A decimated, textured mesh opens in web viewers, presentation tools and game engines, where a raw point cloud is too heavy or not supported. Meshes from scans are also used as context in renders, for example an existing building next to a proposed design. For finished images and animations, see our CAD rendering services.

3D Printing and Replicas

Printing needs a watertight mesh in a format such as STL. Scale models of buildings, replicas of museum pieces and replacement decorative elements all start this way.

Deviation and Condition Checks

A mesh of the current state can be compared with a design model or with an earlier scan to show deformation, wear or differences between design and reality as a color map.

When a Mesh Is Not the Best Choice

If the goal is to design a renovation, coordinate MEP systems, produce floor plans or manage a building, a mesh is the wrong tool. Designers need walls, doors, floors and pipes as real elements, which means a CAD drawing set or a BIM model. In that case the scan goes to scan to BIM or CAD modeling, and a mesh is used only where unusual shapes need it.

What to Specify When Ordering a Mesh

  • Purpose: printing, visualization, reverse engineering, heritage record or comparison.
  • Format and units: OBJ, STL, PLY, FBX or glTF; meters, feet or inches.
  • Watertight or open: and whether filled areas should be marked.
  • Size limit: a target triangle count or file size for each version.
  • Color: none, vertex color or texture.
  • Coordinate system: local, or matching other project data.
  • Which areas matter most, so detail is kept where it is needed.

FAQ

What does point cloud to mesh mean?
It is the process of connecting the separate points of a 3D scan into a continuous surface made of triangles. The result, a mesh, can be printed, rendered, textured and measured as a surface, which a raw point cloud cannot.
Which meshing method is best for laser scans?
It depends on the data and the purpose. Direct triangulation keeps the surface on the measured points and suits clean, dense scans for engineering work. Poisson reconstruction gives smooth, closed surfaces for sculpted objects and printing. Voxel or fusion methods handle noisy data and quick previews.
What is a watertight mesh?
A watertight mesh has no holes. It is a fully closed surface with consistent normals, so software can tell inside from outside. It is required for 3D printing and volume calculations, but not for visualization.
What is mesh decimation?
Decimation reduces the number of triangles while keeping the shape within an allowed deviation. Flat areas lose most of their triangles, while edges and detail keep them. It makes heavy scan meshes usable in viewers, engines and printing software.
Which file format should I use for a mesh?
STL for 3D printing, OBJ for exchange and textured models, PLY for colored scan meshes, FBX for engines and animation, and glTF or GLB for web viewers. Always confirm the units, especially with STL.
Can a mesh be used instead of a CAD or BIM model?
For visualization, heritage records, printing and deviation checks, yes. For design, drawings, manufacturing and building coordination, usually no: a mesh has no editable dimensions or element information. In those cases it serves as a reference for building a CAD or BIM model.
How accurate is a mesh made from a laser scan?
A mesh cannot be more accurate than the scan. Our FARO laser scans have a typical accuracy of ±2–3 mm. Decimation, smoothing and hole filling add further differences, so the finished mesh is compared with the point cloud using a color deviation map.
Can you make a mesh from my existing point cloud?
Yes. We first check the cloud’s registration, coverage and density, then agree on the purpose, format and level of detail before meshing.

From Points to a Usable Surface

A good mesh starts long before the meshing command: with a well-registered, clean point cloud and a clear purpose. The method, decimation, hole filling and format all follow from what the mesh is for, whether that is a replica, a heritage record, a presentation or the base for a CAD model. ScanM2 travels to sites in any US state for FARO laser scanning and can also work from your existing point cloud, delivering meshes, CAD models and Revit models from the same data. Learn more about our 3D modeling services, or send us your project details to discuss which deliverable fits your task.

We work in cities all over the world