Point clouds explained: data, registration, software and deliverables

What is a point cloud?

A point cloud is a collection of points in 3D space that sample the surfaces of a scene. Each point has X, Y and Z coordinates, and usually extra attributes such as return intensity, RGB colour, time stamp or classification. Viewed together, millions or billions of points form a detailed 3D picture that can be measured directly.

Point clouds are the raw output of laser scanners, lidar sensors and photogrammetry. A lidar point cloud comes from distance measurements by laser, while a photogrammetric cloud is calculated from overlapping images. Both describe geometry, but they differ in noise, density and how they handle vegetation and dark surfaces.

On their own, point clouds are not drawings or models. They are the measured evidence from which surveyors and designers produce deliverables, which is why point cloud processing is such an important part of any scanning project.

What is in point cloud data?

Point cloud data consists of coordinates plus per-point attributes, stored in a file format suited to the source and use. Knowing the formats helps avoid data loss between tools.

Structured formats such as E57 keep each scan’s position and grid organisation, which helps with registration and normal estimation. Unstructured formats store points as a flat list. Always check that the coordinate system and units are documented with the files.

How point cloud scanning captures data

The capture method determines density, accuracy and coverage. The main point cloud scanners and sources are:

• Terrestrial laser scanners: tripod scanners producing dense, low-noise clouds of buildings and structures. Our article on terrestrial laser scanning explains the method.
• Handheld and mobile SLAM scanners: fast capture while walking or driving, with typically centimetre-level consistency.
• Scanning total stations: survey-grade control and scanning combined in one georeferenced instrument.
• Airborne and drone lidar: large-area terrain and corridor data with multiple returns through vegetation.
• Photogrammetry: point clouds derived from overlapping photos, rich in colour but dependent on texture and light.

A 3D point cloud survey often blends sources, for example drone data for terrain and terrestrial scans for building facades, joined in one coordinate system.

Point cloud registration and georeferencing

Registration aligns many individual scans into one consistent point cloud. Georeferencing then transforms that cloud into a project or national coordinate system.

Registration methods include target-based alignment using spheres or checkerboards, cloud-to-cloud matching of overlapping surfaces (often refined by an iterative closest point algorithm), and survey-style registration where the scanner is set over known points. SLAM data is registered continuously during capture and then optimised in post-processing.

Georeferencing uses control points measured with a total station or GNSS receiver. A good registration report lists residuals for each target or overlap and the overall error. Keep some control points out of the adjustment as independent checks, because residuals alone can hide systematic errors. Every downstream model inherits the registration error, so this step deserves the most care.

Point cloud processing workflow

Most point cloud processing software follows the same core sequence, whatever the source of the data:

1. Import and organise: load raw scans, check units, coordinate system and completeness.
2. Register and georeference: align scans and fit them to control, then review residuals.
3. Clean: remove noise, moving objects such as people and vehicles, reflections and mixed points at edges.
4. Colourise: map camera imagery onto points if colour is required.
5. Subsample: reduce density to a uniform spacing suited to the deliverable, keeping full resolution where detail matters.
6. Classify: separate ground, vegetation, buildings, utilities and other classes.
7. Quality check: compare against check points and inspect sections for doubled surfaces.
8. Export and deliver: produce LAS, E57 or derived products with metadata.

Point clouds are large, so a workstation with plenty of RAM, a capable graphics card and fast SSD storage makes a big difference to processing time.

Point cloud annotation and classification

Classification assigns each point to a category; annotation adds labels to objects or regions. Both turn raw geometry into information.

In surveying, ground classification is the most common task, because a bare-earth surface is needed for terrain models and contours. Other classes include low, medium and high vegetation, buildings, wires, poles and water. The LAS format reserves standard class codes for many of these, which helps exchange between tools.

3D point cloud annotation goes further, labelling individual objects such as signs, pipes, valves or vehicles. Annotated datasets are used for asset inventories and to train machine-learning models that classify future scans automatically. Automated classification speeds up large jobs, but manual review is still needed where errors would affect the deliverable.

How to convert a point cloud to a 3D model

Converting a point cloud to a 3D model means fitting surfaces or objects to the points so the result is lighter, editable and meaningful. The right method depends on the deliverable.

• Meshing: triangulating points into a surface, suited to terrain, heritage objects and visualisation.
• CAD drafting: tracing plans, sections and elevations from slices through the cloud.
• Parametric modelling: fitting walls, slabs, pipes and steel sections, used in scan-to-BIM and as-built workflows.
• Surface models: building a DTM from ground points for levels, volumes and contours.

Before modelling starts, agree the level of detail and the modelling tolerance. Point cloud modelling to a few millimetres takes far longer than modelling to several centimetres, and many projects only need the tighter tolerance in critical areas.

Point cloud services or in-house capture?

Firms with occasional scanning needs often buy point cloud services, while those with regular projects usually benefit from owning a scanner and processing in-house.

Mars Geosystems supplies the capture side. The terrestrial phase-shift scanner records up to 2 Mpts/s with HDR imagery for colourised clouds. The handheld SLAM laser scanner weighs 925 g and achieves point-cloud thickness of 1 cm or less after post-processing, suited to quick building and site capture. The scanning total station adds control and scanning from one setup. All scanners export to open formats for your chosen processing workflow.

For help matching a scanner to your deliverables, send your project details through the quote request page.

Frequently asked questions