Skip to content

BlogGuides

What Is a Point Cloud? Data, Capture, Formats and Uses

What Is a Point Cloud? Data, Capture, Formats and Uses

A point cloud is a set of data points in 3D space, each with X, Y and Z coordinates and often color and intensity, produced by laser scanners or photogrammetry to record the surfaces of an object or a building. Together the points form a 3D record of the shape that was scanned. This guide explains what point cloud data contains, how it is captured, what density and accuracy mean, which file formats store it, and how a point cloud becomes a 3D model.

What is a point cloud

A point cloud is raw measurement. Each point is one place where the scanner or the camera found a surface, recorded as a position in three dimensions. A single scan position records millions of points, depending on the scanner and its settings, and a building scanned from many positions adds up to far more. Seen from a distance the points read as walls, floors, pipes and furniture; seen up close they are separate dots with gaps between them.

The points are not connected to each other. A point cloud has no faces, no edges and no objects. It does not know that a group of points is a wall or a duct. That is the difference between a point cloud and a mesh or a BIM model, which are built from it (see the table below).

What is point cloud data

Every point carries a position, and most carry more. The common attributes are:

  • X, Y, Z: the coordinates of the point, in a stated coordinate system and unit.
  • Color (RGB): taken from photographs the scanner or a camera captured, so the cloud looks like the room.
  • Intensity: how strongly the laser pulse returned, which gives a grayscale picture of the surface even when no color was captured.

The ASPRS LAS specification, which defines the format used for airborne and terrestrial lidar, lists intensity as the “integer representation of the pulse return magnitude” and provides RGB color channels for each point. The ASTM E57 format describes the same content for 3D imaging systems: 3D point data, attributes such as color or intensity, and 2D imagery.

How a point cloud is captured

There are three common ways to capture a point cloud of a building. The first two use a laser, the third uses photographs.

Terrestrial laser scanner. A scanner on a tripod measures distances to surfaces in every direction from a fixed position, then moves to the next position. This is the method used for survey-grade documentation of buildings (3D laser scanning), and the one this article assumes unless it says otherwise. A scanner of the Trimble X7 class holds a 3D point accuracy of 2.4 mm at 10 m and 3.5 mm at 20 m (one standard deviation) by its datasheet. That figure is one standard deviation, while a USIBD Level of Accuracy is stated at the 95 percent confidence level, so the two numbers are not compared directly.

Mobile scanning (SLAM). A handheld or backpack scanner is carried through the space while it measures, and software works out its own path from what it sees. It is fast and suits long corridors and large sites, usually with lower accuracy than a tripod scanner.

Photogrammetry. Software matches features across many overlapping photographs and calculates a 3D position for each match. It produces colored clouds from a camera and is common for exteriors, roofs and drone work. In all three cases the output is a point cloud.

Scans taken from different positions each start in their own coordinate system. Registration aligns them into one cloud, and the guide to registered and unregistered point clouds explains what that means for the file you receive.

Point cloud density and accuracy

Two numbers describe the quality of a point cloud, and they are not the same thing.

Density is how closely the points are spaced on a surface, often stated as point spacing, for example one point every 5 mm at a given distance. Higher density shows smaller features, such as a molding profile or a bolt, and produces larger files. Density falls with distance from the scanner, so a wall 5 m away is sampled more finely than a ceiling 20 m away.

Accuracy is how close each measured point is to its true position. A cloud can be dense and still inaccurate, for example when registration drifted between scan positions. In the United States, accuracy requirements are commonly written with the U.S. Institute of Building Documentation (USIBD) Level of Accuracy specification. It defines five levels, LOA10 to LOA50, specified at the 95 percent confidence level. LOA30 is a range of 15 mm to 5 mm, LOA40 a range of 5 mm to 1 mm, and LOA20 a range of 5 cm to 15 mm. The guide separates measured accuracy (the data as captured) from represented accuracy (the model or drawing made from it), and the represented accuracy can be equal to or worse than the measured accuracy, never better.

Point cloud file formats

Scanner makers each have their own format, so clouds are usually exchanged in a neutral one. The short list:

Format Extension What it is
E57 .e57 The ASTM E2807 standard for 3D imaging data exchange. Stores points, color, intensity and images. The usual neutral format between scanner software.
LAS / LAZ .las, .laz LAS is the ASPRS binary format for point data, common in survey and GIS. LAZ is its lossless compressed form.
RCP / RCS .rcp, .rcs Autodesk ReCap formats. RCS is a single scan; RCP is a project file that points to the individual scan files. Revit and AutoCAD read these.

E57 is the neutral exchange format to keep and to pass between programs; RCP or RCS is the form Autodesk software works with.

How a point cloud becomes a 3D model

A point cloud is a measurement, and a model is an interpretation of it. There are two common routes.

Mesh. Software connects neighboring points into triangles, which produces a continuous surface. A mesh looks solid and is good for visualization, volumes and fabrication of single objects. It is still one undifferentiated surface: a wall, a pipe and a chair are the same kind of geometry, and none of them can be selected, scheduled or edited as a building element.

BIM. A modeler places real building elements, such as walls, floors, doors, ducts and columns, against the cloud in software like Revit. Each element has a type, a size and parameters, and the model can be quantified, coordinated and edited. This is scan-to-BIM, and it is mostly manual interpretation of the cloud: automatic classification exists, but its output has to be checked element by element. The level of development is set by a stated LOD, such as LOD 300. The scan-to-BIM service covers what is delivered, and the guide to importing a point cloud into Revit shows how the cloud is brought in as a reference.

Point cloud Mesh BIM model
What it is Measured points Points joined into triangles Building elements with data
Made by Scanner or photogrammetry Meshing software from a cloud A modeler working against a cloud
Objects None None Walls, doors, ducts, columns
Editable Cropped and cleaned only Limited Yes, as building elements
Shows Everything the scanner saw Smoothed surfaces What was modeled, to the stated LOD
File size Very large Large Moderate
Typical use Measurement, verification, reference Visualization, volumes, single parts Design, coordination, quantities

What a point cloud can and cannot show

A point cloud records what the scanner or camera had a line of sight to. It shows exposed surfaces: walls, ceilings, floors, pipes in an open ceiling, structure, equipment. It does not show what is behind a surface. Framing inside a wall, a buried pipe, rebar in concrete and the space above a closed ceiling are not in the data unless they were exposed during the scan. The post on whether lidar can see through walls covers this limit.

Two practical consequences follow. Anything hidden in the model was drawn from other documents or from openings, and a careful deliverable marks it that way. And the cloud has shadows: areas behind furniture, above ceiling tiles or under stairs have no points unless the scanner was set up to see them, which is why the scan plan matters as much as the instrument.

A point cloud also does not contain judgments. It will show a wall that leans and a floor that slopes exactly as they are, which is the reason to scan existing buildings, and why drawings traced from it can differ from the original plans.

FAQ

What is a point cloud in simple terms?

A set of millions of measured points in 3D space that together outline a building or an object. Each point has X, Y and Z coordinates and usually a color and an intensity value.

What is point cloud scanning?

It is the capture of a point cloud with a laser scanner, a mobile scanner or by photogrammetry. For buildings it is usually done with a tripod laser scanner placed in several positions, and the scans are then registered into one cloud.

Can you turn a point cloud into a 3D model?

Yes. The cloud can be meshed into a surface, or a modeler can build a BIM model against it in Revit, with walls, floors, doors and ducts as editable elements. The scan-to-BIM workflow is the second route.

What file format is a point cloud?

Common formats are E57 (the ASTM E2807 exchange format), LAS and LAZ (ASPRS), and RCP and RCS (Autodesk ReCap). Scanner makers also use their own native formats.

How accurate is a point cloud?

It depends on the scanner and the registration. A scanner of the Trimble X7 class has a 3D point accuracy of 2.4 mm at 10 m by its datasheet, and a project states its requirement as a USIBD Level of Accuracy such as LOA30 or LOA40.

Can a point cloud show what is inside walls?

No. It records only surfaces the scanner could see. Hidden elements have to come from drawings, openings or other methods.

What is the difference between a point cloud and a 3D model?

A point cloud is unconnected measured points with no objects. A 3D model, whether mesh or BIM, is built from the cloud and adds surfaces or building elements.

Sources

Ready when your site is.