The words on the rest of this site
Point cloud vocabulary, as a practitioner writes it.
Point cloud
The set of measured 3D points a laser scanner or a photogrammetric process returns, each one carrying a position and usually further values. It is the raw material every other term on this page describes.
ALS
Laser scanning flown from an aircraft or a drone, looking down over a wide area. It is what a national mapping programme or a power line corridor survey is usually flown with.
MMS
A laser scanner mounted on a vehicle, capturing the street or the track from the level it is driven along. It produces far higher point densities than an airborne flight, over a much narrower strip.
TLS
A laser scanner set up on a tripod, scanning from a fixed position. Several positions are registered together to cover a site.
Point density
How many measured points fall on a square metre of surface, written pts/m² or ppm². It is the first thing a classifier asks about a file, because a model trained at one density does weaker work at another.
Intensity
The strength of the laser return a point came back with, recorded alongside its position. Surfaces reflect differently, so intensity separates things geometry alone cannot, and every standard Pointly classifier needs it in the file.
Return
One laser pulse can come back more than once, from a leaf, then a branch, then the ground beneath. Each of those is a return, and the return number is what lets a classifier see the ground under a canopy.
LAS
The open binary format point clouds are exchanged in, holding the points, their attributes and a class per point. Pointly reads LAS 1.2 and 1.4, and 1.4 is the one to send.
LAZ
LAS, compressed without loss. Every value survives and the file moves at a fraction of the size, which is why it is the format to hand over for a large survey.
CRS
The system a file states its coordinates in, so that a number means a place on the earth. A file that carries one gets converted to metres automatically; one without is read as though it were already in metres.
EPSG code
A number that names one coordinate reference system, for instance in the metadata of a LAS file. It is the short way to say which CRS a dataset is in.
Tile
A rectangular piece of a survey, cut so that a large capture can be processed and handed over in parts. An even tiling is what makes processing scale, and one tile is enough to see how a classifier lands on your data.
Classification
Giving every point in a cloud a class, ground, vegetation, building, conductor and so on, so the cloud can be read by object rather than by point. The original file comes back untouched with the classification added as a dimension.
Vectorization
Turning classified points into geometry: a line for a conductor, a point for a pole, a polygon for a building footprint. It is what makes a survey joinable to a CAD drawing or an asset register.
Segmentation
Cutting a cloud into coherent pieces without saying what each piece is, where classification says what it is. In practice the two run together: the segments are what a class gets assigned to.
Class catalog
The list of classes a project works to, with an id and a name for each, agreed before anything runs. Ours are published on the workload pages so you can hold them against your own specification rather than discover them in a quote.
ASPRS classes
The standard class numbering the LAS specification defines, where 2 is ground and 5 is high vegetation. Every catalog on this site starts from it and says where it extends beyond it.
DTM, DSM and CHM
Three surfaces derived from one classified cloud: the terrain under everything, the surface over everything, and the canopy height as the difference between them. Which one a project needs is usually the first question after the classes.
LOD2
A building model with roof shapes but without facade detail, which is the level most city models are built at. It is derived from a classified cloud rather than measured, so what it is worth depends on the classification under it.
Strip adjustment
Fitting the overlapping flight lines of an airborne survey to each other so the same roof does not sit at two heights. It happens before classification, and a cloud that has not had it will classify worse for reasons that are not the model.
Ground truth
A piece of data classified by hand and held back, so a model can be measured against something it has never seen. Accuracy figures without one are opinions, which is why ours name the tile they were measured on.
MPoints
The unit this site prices in: one MPoint is one million points, and a BPoint is a thousand of those. It is the honest unit for classification because the work scales with points rather than with area or with file count.
Air gap
A machine with no route to the internet at all, not a firewall rule but an absent connection. It is the condition some surveys are captured under, and the reason the classification stack is available as an appliance you run yourself.
Digital twin
A model of real infrastructure kept current enough to be worked from rather than looked at. A classified and vectorised point cloud is the layer most digital twins of built infrastructure are measured from.
See it on your own data
One tile is enough to show how the classes land on your data.