3D scan, Laser survey

Accuracy of a 3D survey: how to set tolerances and monitor quality

‘Millimetre accuracy’. This phrase appears in almost every 3D survey quotation.

Taken on its own, it doesn’t mean much. Does it refer to the accuracy of the device, the point cloud or the delivered plan? Is it measured between two adjacent walls, or relative to national coordinates?

And above all: who has verified it, and how?

For an architect or engineer, these questions have direct implications for the project. Here are some guidelines for setting out a clear requirement, and then checking that it has been met.

What ‘accuracy’ means in a survey

From the instrument to the deliverable

The accuracy specified by a manufacturer relates to a single measurement, under ideal conditions.

In the field, this accuracy deteriorates at every stage: setting up survey stations, cleaning the point cloud, and modelling. A plan or a BIM model remains an interpretation of the point cloud, with its own simplifications.

For the client, therefore, it is the accuracy of the final deliverable that matters.

Relative accuracy and absolute accuracy

Relative accuracy describes the internal consistency of the survey: the distance between two walls, the alignment of a façade, the height of a room. Absolute accuracy describes the building’s position within a coordinate system.

In Switzerland, this system is the MN95 reference frame, which has been official since 2016, using the standard NF02 elevations.

The pitfall is well known. According to swisstopo, the old MN03 frame can deviate from MN95 by up to 1.6 metres. A survey that is highly consistent internally may therefore be out of alignment with the cadastral map, simply because it was referenced to the wrong framework.

Factors influencing accuracy

The quality of a survey depends less on the equipment than on how it is used in the field. The main factors are as follows:

  • the distance and angle between the scanner and the surface, as a wall targeted from a distance or at an angle is measured with less precision
  • the nature of the surfaces; glass, polished metal and very dark materials reflect the laser signal poorly
  • the number and distribution of measurement points, which determine the overlap between scans and thus the robustness of the data set
  • on-site conditions: vibrations from a building in use, temperature fluctuations, and the movement of people or vehicles
  • the geometry of the building itself, with long corridors and stairwells being more prone to measurement errors than compact spaces

Registration: aligning stations without drift

A building comprises tens, or even hundreds, of stations. Registration involves assembling them into a single, coherent point cloud. It is at this stage that small errors can accumulate.

Using targets or point cloud-to-point cloud

Target-based registration relies on spheres or grid markers placed on site, visible from several stations. It is robust and easy to verify. However, it takes time to set up and is not always suitable for fragile surfaces.

Point cloud-to-point cloud registration directly aligns common geometries between neighbouring scans. It is fast and contactless. However, it requires good overlap and may drift in long or detail-poor areas.

In practice, the two methods are often combined.

Linking the survey to known points

For a large building, a site comprising several buildings or a survey carried out over several campaigns, the point cloud is tied to points measured separately, either using a total station or via GNSS. These points establish the absolute position and limit drift over long distances.

The swisstopo guidance on the accuracy of points in the official survey makes a useful distinction here. An anchor point influences the result. A control point, on the other hand, serves only to measure the deviation.

Keeping control points out of the re-alignment process is the basis of a fair verification.

Setting tolerances in the specifications

A tolerance is only meaningful in relation to a specific application. Demanding the same level of precision for a feasibility study as for the installation of prefabricated elements amounts either to paying for unnecessary precision or to lacking the precision that really matters.

Intended use Level of requirement Typical deliverable
Feasibility study, area estimates Moderate; general consistency is sufficient Simplified 2D plans, point cloud
Preliminary design and planning application Good, with reliable dimensions for each component CAD plans, sections and elevations
Construction drawings, MEP coordination Elevation drawings, particularly showing recesses and openings BIM model, detailed drawings
Prefabrication, installation of bespoke elements High-precision, checked point by point Dense point cloud, dimensions checked
Monitoring of deformations over time Very high and repeatable from one survey campaign to the next Point cloud comparisons, deviation reports

The numerical values are determined on a project-by-project basis, in consultation with the contractor, depending on the building and access conditions. For a 3D model, they go hand in hand with the Level of Detail (LOD) selected.

The specifications must include three elements: the expected tolerance, the deliverable to which it applies, and the verification method.

Quality control: how to verify the result

A stated level of accuracy is only valid if it is demonstrated. This can be achieved through several checks:

  • Residual deviations from the adjustment indicate how the stations align with one another. They provide information on internal consistency, but not on absolute position.
  • Independent control points, measured using a total station and excluded from the re-alignment, allow the point cloud to be compared with an external reference.
  • Dimensions measured on site (diagonals of a room, ceiling height) serve as a simple verification method, understandable to all stakeholders.
  • A comparison between the final product and the point cloud reveals what the modelling has simplified.

The Swiss official survey provides a good benchmark for assessing a deviation. The swisstopo guideline cited above requires accuracy to be verified retrospectively, point by point. The tolerance threshold is set at three times the standard deviation: beyond this, the measurement is considered to be subject to error.

These rules apply to the cadastral survey, not to building surveys. Their rationale, however, is readily applicable: a stated level of accuracy must be verified by independent measurements, and points falling outside the tolerance limits must be flagged.

All of this should be included in an accuracy report submitted alongside the final deliverable. It should specify the adjustment method, the coordinate system, the control points used and the deviations observed.

Five questions to ask before commissioning a survey

  1. What will the survey be used for? The intended use determines the tolerance and the deliverable, not the other way round.
  2. What tolerance is required, and to which deliverable does it apply: the point cloud, the plans or the model?
  3. Does the survey need to be tied to the MN95 national coordinate system and NF02 elevations, or is a local system sufficient?
  4. What checks are planned, and how many independent control points?
  5. Will an accuracy report be provided with the deliverable, and what will it contain?

Key points to remember

  • The required accuracy is that of the final deliverable.
  • Internal consistency and positioning within the national coordinate system are two distinct requirements.
  • Re-alignment is the critical stage, and control points excluded from the re-alignment process enable this to be verified.
  • A tolerance is set according to standard practice and is specified in the specifications along with the method for verifying it.
  • A precision report should accompany each deliverable.

A survey delivered with its checks

At AB 3D Scanning, each survey is defined according to its intended use and is then delivered with the checks needed to assess its quality. If you are drawing up a specification or wish to have an existing survey verified, we would be happy to discuss this with you.

Sources