Point Cloud Registration

Point cloud registration is the step that aligns several laser scans of one building, each recorded from its own station, into a single point cloud with one shared coordinate system. A single scanner position sees only part of a building, so a survey needs many positions. Each scan starts with its own position and orientation, and registration calculates the movements that bring all of them into agreement.

The result is only as good as the links between scans. Software finds common ground through the overlap of neighbouring scans, through physical targets placed in the room, or through a combination of both. Once aligned, the scans form one dataset that a designer or modeller can read as a whole. In Slovak practice the process is called registrácia mrakov bodov.

What is point cloud registration?

It is the mathematical alignment of separate scans into one consistent model. Each scan link has a rotation and a translation, and registration solves for them so that the overlapping surfaces of two scans coincide. The aligned data then describes the whole building in a single frame, which is what a designer needs when measuring openings, levels and room shapes.

Registration does not tell you whether the building is correctly placed on the earth. That is a separate step called georeferencing, covered below. Registration only makes the scans agree with each other.

How do target-based and cloud-to-cloud registration differ?

There are two main approaches. Target-based registration places reference objects, usually spheres or flat targets, in the scene and uses their positions to link the scans. Cloud-to-cloud registration ignores targets and matches the surfaces of the scans themselves, typically by finding shared planes, edges and corners. Either method can be followed by a global adjustment that spreads the remaining error across all scan links instead of leaving it at one point.

Aspect	Target-based	Cloud-to-cloud
How scans are linked	Shared physical targets seen from two or more stations	Matching surface geometry in overlapping areas
Preparation on site	Targets must be placed and kept still during the survey	Needs enough overlap and stable geometry such as walls and floors
Independent control	Target centres can be measured separately, for example with a total station	Depends on the surveyor adding independent control points
Typical weak point	Targets can be missed, moved or poorly placed	Repetitive or featureless spaces can match in the wrong place

Neither method is automatically more accurate. Comparative studies of terrestrial scanning have found that the difference between methods can be hard to separate from the error of the registration itself, so the result has to be checked whichever method is chosen. The choice is usually made by the surveyor based on the rooms, the access and the deliverable.

Why do poorly registered scans show double or bent walls?

Every scan link carries a small error in rotation and position. When two scans disagree, the same wall is recorded twice at slightly different positions, so it appears as a doubled surface or a thick band. This is most visible on thin elements such as door reveals, skirting and plaster edges, where even a small shift stands out clearly in the cloud.

Bending is a different failure. When a building is scanned as a long chain of stations, each link inherits the error of the link before it. The error builds up gradually along the chain, so a straight wall scanned from one end to the other curves, and a floor drifts out of level across the rooms. A cloud can look sharp in each room while being distorted across the whole building. This is why registration must be checked across the full project, not only between neighbouring scans. A designer who takes dimensions from a bent cloud will take wrong dimensions for openings, and those errors then flow into drawings.

Symptom in the cloud	Likely registration cause	What to check
Doubled wall surface	Two scans linked with a shift between them	Residuals of the links listed in the report
Bent or curved wall	Error accumulated along a chain of stations	Positions of independent points at both ends of the building
Floor not level between rooms	Height error drifting through the chain	Height control points in several rooms
Model offset from the site	Missing or wrong georeferencing	The coordinate system and the control point list

How is a registered cloud georeferenced to S-JTSK?

Registration gives a consistent model, but that model still has its own local origin. Georeferencing ties it to a known national system. In Slovakia that system is S-JTSK, which is based on the Krovak projection and is commonly referenced by the EPSG code 5514 in software. Georeferencing needs control points: positions measured in S-JTSK by a surveyor and identified in the cloud. The transformation from local to national coordinates is then calculated from those points.

The quality of georeferencing depends on where the control points lie. Points spread across the site, including near the corners of the building and the boundary, constrain the transformation far better than points clustered in one room. Control measured independently, with a total station or with satellite positioning, is the reference against which the cloud is judged. Geodetic work of this kind is covered in the geodetic survey article.

What should a client ask a surveyor for in a registration report?

Ask for a written registration report with the delivery, not only the cloud file. A usable report should state, in plain terms:

• The method or methods used, whether targets, cloud-to-cloud or both, and the software and version.
• The number of stations and how they link, ideally shown in a diagram.
• The residual error of each link and the error of the overall adjustment, stated as figures with their units.
• Which control points were used for georeferencing and which were held back as independent checks, with the difference between their known and computed positions.
• The coordinate system, the height reference and the datum, in writing.
• Any areas not covered and the reason, such as glass, dark surfaces or inaccessible rooms.

Also ask whether the surveyor will check the cloud against an independent measurement before handover. If the cloud will feed a scan-to-BIM model, the report should accompany the model so that the design team can see which parts are reliable.

What are common misconceptions about point cloud registration?

• A registered cloud is automatically accurate. Registration makes the scans consistent with each other, and accuracy depends on control.
• Cloud-to-cloud registration needs no preparation. It needs overlap and stable geometry, and it can drift in repetitive spaces.
• Smooth-looking walls prove good registration. A local match can hide an error that builds up over the whole building.
• Registration and georeferencing are the same step. Registration aligns scans with each other, while georeferencing places the model in a national system.