Point Cloud Registration: What It Is and Why It Matters

Point cloud registration is the step that turns many separate laser scans into one point cloud. A terrestrial scanner measures the building from one position at a time, and each station’s points sit in that station’s own coordinate system; registration finds where each station stood relative to the others and brings all of them into one coordinate system, so a wall scanned from three rooms becomes one wall. Everything built on a scan, drawings, a Revit model, a measurement taken months later, depends on how well this step was done. This guide explains what registration is, how it is done, how its quality is reported, and what to check when a point cloud arrives.

Why scans need registering

A scanner on a tripod sees only what is in its line of sight, so a building is scanned from dozens of stations: every room, closet, shaft and stair that has to appear in the deliverable needs a station with a view of it. Each station produces its own cloud, measured from the scanner’s own origin. Until those clouds are aligned, they are a pile of overlapping fragments that cannot be modeled from. The guide to registered vs unregistered point clouds explains what an unregistered delivery looks like and why it is not ready for production.

How registration is done

There are two basic ways to tie stations together.

Targets. Spheres or checkerboards placed in the space are seen from several stations, and their known shapes are matched between scans. Targets give strong ties in spaces with little geometry, at the cost of placing and moving them.

Cloud to cloud. The overlapping geometry itself, matched with algorithms of the iterative closest point (ICP) family that minimize the distance between the two clouds, the walls, floors, columns and ceilings that two stations both see, is matched directly. With enough overlap and enough geometry, no targets are needed. A scanner of the Trimble X7 class registers this way in the field: by its datasheet, each new scan is oriented and aligned automatically with the last or a pre-selected scan, manual alignment is available for difficult cases, and an automatic refinement tightens the whole set at the end. We register without targets on the walls, and check the registration before leaving the site, so a station that does not close, or a corner no station saw, is rescanned the same day.

When a project has survey control, the registered cloud is tied to the surveyor’s control points so it sits in the project’s coordinate system; when it has none, the cloud is delivered in a local coordinate system with the origin and the orientation documented.

How registration quality is reported

Registration quality is not a single number, and it is not the scanner’s accuracy. The scanner’s datasheet accuracy, for a scanner of the Trimble X7 class a 3D point accuracy of 2.4 mm at 10 m (one standard deviation), describes how well one station measures a point. Registration adds the error of placing each station relative to the others. The registration report shows that: by the X7 datasheet, it lists the project and station average error, the overlap between scans and the consistency results. A low average error with good overlap means the stations hold together; a station with high error or thin overlap is where to look first.

The accuracy of the registered cloud, and of whatever is drawn from it, is stated in the U.S. Institute of Building Documentation’s Level of Accuracy terms, as a standard deviation at the 95 percent confidence level. The USIBD guide separates the measured accuracy of the cloud from the represented accuracy of the drawings or the model built on it, and the quote names both.

What to check when a point cloud arrives

• The registration report is included, with the error of each station and the overlap.
• The cloud opens as one dataset in one coordinate system, not as separate stations.
• Walls scanned from both sides line up; floors on either side of a door are at one level.
• The coordinate system is documented: survey control, or a local origin and orientation.
• The formats fit the work: E57, the ASTM exchange format for 3D imaging data, for point cloud software in general, and RCP/RCS for Revit.

If any of these is missing, ask before modeling starts; re-registering later costs more than asking now.

Price and time

A registered point cloud with its registration report starts at about $1,200. With drawings or a model on top, typical ranges from recent projects are $2,000 to $4,500 for a house or an apartment and $3,000 to $9,000 for an office or a commercial floor. A house or a floor is usually scanned in a day, and the registered cloud with its report usually follows within a few days. The 3D laser scanning page describes the scan and its deliverables, and we scan in all 50 states.

Sources

• Trimble X7 3D laser scanning system datasheet: auto-registration, refinement, registration report, 3D point accuracy.
• USIBD Document C120, Level of Accuracy Specification Guide, version 2.0 (2016): measured and represented accuracy, the LOA levels.
• ASTM E2807, Standard Specification for 3D Imaging Data Exchange: the E57 file format.