All articles · 6 min read

What is a point cloud — and what can you use it for?

A point cloud is the raw measurement points from a scan: millions of individual measurements with x, y, z and a colour, with no interpretation of what is a wall and what is a sofa.

Point cloud, floor plan and 3D model are three different things

When a home is scanned with LiDAR, three layers of reality appear, and it is worth keeping them apart:

• The point cloud is the measurements. Raw, unedited, with gaps wherever you didn't point the phone, and with noise wherever the surface was shiny or dark.
• The floor plan is the interpretation in 2D: here someone — software or a person — has decided that these particular points make up a wall, and drawn a straight line through them.
• The 3D model is the interpretation in 3D: walls with thickness, doors with frames, floors and ceilings, built from the floor plan.

That means the floor plan is always tidier than the point cloud, and the point cloud is always more honest. A crooked wall stands crooked in the point cloud and straight in the floor plan. If you need to document how the home actually looked on the day, the point cloud is your evidence.

How a point comes to be

The LiDAR sensor in an iPhone or iPad measures distance by sending out pulses of light and timing how long they take to come back. The result is a depth image: a grid of distances, one figure per pixel.

To become a point cloud, each of those distances has to be converted into a position in space. That requires the phone to know where it itself was standing, and which way it was facing, at the moment the measurement was taken. It tracks that with the camera and the motion sensors. The distance plus the phone's position and orientation gives one point in the world — and the colour is picked up from the camera image at the same spot.

Do that for every depth image as you walk around the home, and you end up with hundreds of thousands or millions of points in one shared coordinate system. That is the point cloud.

And that is why the unforgiving rule holds: anywhere you didn't point the phone has no points. Behind the sofa, on top of the cupboard, inside the cupboard — there is a gap. The point cloud cannot guess; that is its entire virtue.

Thinning, range, and why points get discarded

A raw point cloud from a walkthrough of a home is usually far too dense: many millions of points, most of them sitting on top of each other because you filmed the same wall from three angles. So it is almost always thinned before being written to a file.

The usual method is a grid: the space is divided into small cubes, and one point is kept per cube. A 2 cm grid gives a cloud where no two points sit closer than 2 cm apart — plenty to see the shape of the room and to measure by, and a file that fits in an email instead of taking up several gigabytes.

Points are also cut by range. Measurements closer than roughly a quarter of a metre are typically your own hand, or the door frame you walked past. Measurements further away than roughly five metres are noisy enough on a phone's LiDAR that they do more harm than good. HouseSense therefore keeps to that range when the point cloud is written out — better to have fewer points you can trust than many that lie.

Worked example: A room measuring 4.12 × 3.05 m with a 2.50 m ceiling has roughly 62 m² of surface (floor, ceiling and four walls). At one point per 2 × 2 cm, that gives 62 ÷ 0.0004 = roughly 155,000 points — if every bit were hit. In reality it comes out lower, because furniture casts shadows and you never quite cover every square centimetre.

The file formats: PLY, XYZ, PTS and LAS

Point clouds are exchanged in four formats you will very likely run into:

• PLY (Stanford PLY) — the most common for point clouds from scans. Can be saved in binary, so the file takes up less space, and carries a colour per point. Opened by CloudCompare and MeshLab.
• XYZ — plain text, one line per point: x, y, z and optionally r, g, b. The simplest format there is, readable by anything, but the files get large because numbers written as text take up room.
• PTS — the surveyor's text format. Similar to XYZ, but has the point count on the first line and typically expects Z to point upward, the way surveying tools reckon.
• LAS — the binary format from surveying and airborne LiDAR. Compact, well-defined, and what a surveyor or an engineer will usually ask for. Here too, Z points upward.

Note the difference in orientation — it is the most common mistake when a point cloud gets passed on. Graphics software normally assumes Y points upward, while the surveying world assumes Z points upward. If the house arrives lying on its side at the recipient's end, that is almost always what happened, and it is fixed with a single 90-degree rotation.

HouseSense writes the point cloud in all four: PLY and XYZ are oriented the same way as the 3D model, while PTS and LAS are written with Z pointing upward, as surveying tools expect.

So what do you actually use a point cloud for?

Four things, in falling order of how often they come up in an ordinary home:

1. Documentation. This is what the home looked like when I took it over. That is worth something at a handover, in a damp-damage case, in an insurance claim, or before a renovation where something gets torn down.
2. Checking a measurement after the fact. Forgot to measure the alcove? You can measure it in the point cloud afterwards — even six months later, once you no longer have access to the home.
3. A basis for a consultant. An architect or engineer can lay their own drawing over the point cloud and see whether it matches reality, instead of having to trust your drawing.
4. Crooked and curved things. Curved walls, vaults, old timber-frame houses and sloped ceilings, which a floor plan necessarily simplifies. The point cloud does not simplify.

On the other hand, you should not send a point cloud to a painter or a flooring fitter. They need an area and a floor plan in PDF, not a file with hundreds of thousands of points they have no program to open.

A point cloud cannot tell you what things ARE

The most important thing to understand about a point cloud is that it knows nothing. The points on the wall aren't labelled "wall". There is no distinction between the curtain and the window behind it, between the floor and the rug on it, between the load-bearing wall and the bookcase standing against it.

Anything that looks like knowledge — room names, areas, wall thicknesses, doors — is something a program or a person has laid on top of the points afterwards. And the question of whether a wall is load-bearing cannot be answered by the point cloud, the floor plan or the 3D model. That needs a building surveyor or a structural engineer.

But precisely because the point cloud does not interpret, it is the one file in the stack where you can go back and see what was actually measured. That is why it is worth keeping.