Article

How accurate is your point cloud, really?

Engineer in training (CPI), geomatics

Share

You need a point cloud, or you’re about to order one. You may have been told it has “2 mm accuracy.” But what exactly does that number mean—and does it really match what you’ll receive? This article explains, in plain language, what “accuracy” means for a point cloud, why a visually appealing point cloud isn’t necessarily a reliable one, and what questions to ask your supplier to get data you can truly rely on to make decisions.

This visual comparison contrasts the theoretical and ideal accuracy claimed by a manufacturer (on the left, narrow and simple) with the actual uncertainty of the georeferenced output (on the right, wider and more complex). The latter results from the accumulation of external factors such as the method, georeferencing, and processing.

“2-mm accuracy”: What the number doesn’t tell you

The “2 mm” listed in a technical specification describes the instrument under perfect, almost laboratory-like conditions: a single measurement, an ideal target, at a short distance. Your site, however, is nothing like a laboratory. Between that catalog figure and the actual precision of your final deliverable, there’s a whole process involved—and many factors that can affect the result. Confusing the two leads to unrealistic expectations—and sometimes to disputes.

Precision, accuracy, resolution: three different things

People often confuse words that don’t mean the same thing:

  • Accuracy (repeatability): If you measure the same point twice, do you get the same value?
  • Accuracy: Does the measurement land in the right place in the real world?
  • Resolution (density): the number of dots per unit area—not to be confused with accuracy.

A cloud can look very dense and “beautiful” on screen… even while being misaligned or noisy. So the real question isn’t “How accurate is the scanner?”, but “What is the uncertainty of MY output, given WHAT I want to do with it?”

Where Does Uncertainty Come From? The Concrete Factors

There are several factors at play, and most of them have nothing to do with the technical specifications:

  • Distance and angle of sight. The farther the scanner is from a surface, or the more it is aimed at it “at an angle,” the more noise there is in the measurement. And when the same surface is scanned from multiple positions, the data points overlap: their small discrepancies accumulate and “thicken” the point cloud at that location.
  • Temperature and the environment. Moving from a heated indoor environment to a cold outdoor one (or vice versa), direct sunlight, vibrations, dust, or condensation—all of these factors affect the measurement. A reading that alternates between indoors and outdoors is not taken under the same conditions from start to finish.
  • The instrument. Its calibration and the width of its laser beam.
  • The Method. The number and arrangement of positions, overlap, and the drift of a mobile system (SLAM) over a long distance.
  • The georeferencing. The quality of the control network on which everything is based.
  • Treatment. Alignment (matching the scans together) and noise filtering.
  • Outliers. Reflections, moving objects, dust: false spots that need to be removed.
This infographic visually breaks down how different sources of error add up: instrument imperfections (angle of incidence), the capture method (positions and drift), georeferencing (control network), and outliers (reflections) all contribute to widening the final uncertainty envelope.

If managed properly, these factors can be controlled. If ignored, they will silently undermine your deliverable.

How a Reliable Supplier Gets It Right—and Proves It to You

Rigor is essential before, during, and after the survey: calibration, a redundant control network, closure loops, and noise filtering according to clear rules. But above all, a good provider doesn’t ask you to take their word for it—they document everything. Specifically, this is done in two steps:

  1. A report on checkpoints. First, we establish the reliability and accuracy of the reference network itself—the foundation on which everything else rests.
  2. A report on assembly and realignment. We then demonstrate how the various scans were stitched together, and how the cloud was registered to these control points.
This image shows how quality is validated by comparing the raw point cloud (on the left) to certified external reference points. The process generates a visual report showing the deviations (in the form of three-dimensional error vectors) and visually confirms whether the deliverable meets the tolerance defined for the project.

These reports turn “trust me” into “here’s the proof.” You’ll then know, backed by numbers, whether your cloud meets the tolerance requirements you need.

A real example. On a recent survey of airport infrastructure, 14 targets were used to register the point cloud and 34 others were kept independent, for checking only. The first group gave a 3D root mean square error of 15 mm; the second, 40 mm. The honest figure is the second one. A flattering figure can also hide a registration error: our sample quality control report shows a link that displayed 1 mm yet only closed to 15.6 mm. Monitoring calls for the same rigour: at Montréal’s Gesù Church, about 40 measurement sessions tracked the building’s movements during the underpinning of its foundations.

Why It Protects You

Documented uncertainty provides assurance. It tells you whether the deliverable is precise enough for your purposes (design, manufacturing, inspection, asset management); it prevents costly decisions based on inaccurate data; and it resolves disagreements should they arise. Conversely, a cloud delivered without any indication of uncertainty is a black box: impressive to look at, impossible to defend.

This analytical visualization illustrates how scientific research transforms raw scanner data (beam and angular grid) into mathematically rigorous uncertainty models. It shows the transition from a raw data point to a complex three-dimensional error volume, establishing reliability measurement as a professional standard.

Even the research community is getting involved

A sign that this is a serious issue: much of the research for 2026 (ISPRS, Toronto Congress) focuses specifically on how to measure and standardize this uncertainty. What might have been considered a minor detail is becoming a true professional standard. A vendor who can explain this to you in simple terms is a vendor who takes your data seriously.

Conclusion

The accuracy of a point cloud isn’t listed on a spec sheet—it’s determined by the methodology and substantiated by documentation. Before ordering a survey, ask a simple question: “Can you provide documentation of the actual uncertainty of my deliverable, for my specific use?” A provider who answers “yes”—with supporting reports—offers you much more than just a scan: it’s data you can rely on to make decisions with complete confidence.

FAQ

“2 mm accuracy” on the technical data sheet—is that my accuracy?

No: This figure describes the instrument under ideal conditions. Your actual accuracy also depends on distance, angle of sight, temperature, method, georeferencing, and processing.

Why are some areas of my cloud “noisier” than others?

This is often because they were observed from a distance, at an angle, or from multiple positions where the observation points overlap. Distance and the angle of incidence increase the noise.

How can I tell if a cloud falls within my tolerance?

Through inspection reports: first, a report on the inspection points, then a report on the assembly and realignment based on those points.

What is an outlier?

A false point (reflection, moving object, dust) that, if not filtered out, degrades the geometric quality of the cloud.

Why require documented uncertainty?

To make decisions based on reliable information, meet your contractual tolerances, and protect yourself in the event of a dispute.

Do you need to order a point cloud and want to be sure you’ll get the right level of accuracy for your project? Topo3D documents the uncertainty of its deliverables based on your tolerance, with supporting reports. Let’s discuss your requirements.

About the author

Engineer in training (CPI), geomatics

Justin is an engineer in training (CPI) with a degree in geomatics engineering from Université Laval. Before joining Topo3D, he spent four years at Québec’s Ministry of Transportation, where he planned, operated and processed mobile LiDAR surveys, automated processing workflows (FME, ArcGIS) and implemented ISO-compliant quality procedures. Since 2024, he has acquired and processed LiDAR data for Topo3D’s construction and infrastructure projects and automated their processing. He takes part in the annual mobile LiDAR monitoring of a road in the Northwest Territories.