A point cloud is what a 3D scan of the world looks like before it’s a model. Every point holds a 3D position, a color, and often classification metadata. Photogrammetry gives you them for free. LiDAR gives you the version that can see through trees.
§ 01Try it
A live, low-density point cloud preview. Toggle between elevation, RGB, and intensity coloring; drag the density slider. Real surveys produce 50–500 million points; we’re showing 2,400 here so the rotation stays smooth.
Color by
Real surveys produce 50M–500M points. Each point holds (x, y, z) plus attributes — RGB, intensity, classification.
FIG.05Sparse point cloud preview. Try different colorings — same geometry tells different stories about elevation, surface, or material.
POINT CLOUD
A 3D set of points, each with at least an (x, y, z) position and typically additional attributes — RGB color, return number, intensity, classification. The most fundamental 3D data product in mapping.
§ 02Two ways to make one
There are two production paths to a usable point cloud, and they’re good at very different things:
Photogrammetry
Triangulate from many overlapping photographs. Free if you already own a camera drone. Sees only what the camera sees — treetops, not understory.
LiDAR
Pulse a laser many times per second and time the returns. Penetrates vegetation. ~$30k payload + ~$10k processing tools.
§ 03LiDAR’s advantage — multi-return
Each LiDAR pulse can produce multiple returns as it pierces a tree canopy: leaves bounce some of the light back, branches more, and the ground (if visible) returns the last echo. Photogrammetry only ever sees the first return — the treetops.
This is why LiDAR is the only honest choice for forestry work, terrain under vegetation, or anywhere a true bare-earth DTM is required.
§ 04The LAS file format
Both LiDAR and photogrammetry tools settle on the same exchange format — LAS (or its compressed LAZ variant). The structure is simple:
Public Header
Version, point format, scale, offset, bounding box, point count.
VLRs
Variable Length Records — most importantly, the CRS / EPSG code.
Point Records
The actual data. ~34 bytes per point. 100M points ≈ 3.2 GB.
EVLRs
Extended VLRs at file end — full waveform data, optional.
§ 05Classification
Each point in a LAS file has a 1-byte classification code (the ASPRS standard). These are how a DTM gets generated: filter to ground points only, ignore the rest:
1 · Unclassified
The default — points the classifier hasn't labeled yet.
2 · Ground
Bare earth. Used to generate DTMs.
3 / 4 / 5 · Vegetation
Low / medium / high vegetation tiers.
6 · Building
Roof and façade returns.
7 · Noise
Outliers — birds, dust, sensor errors.
9 · Water
Water surface returns.
§ 06When to use which
Construction earthworks
Photogrammetry. Open ground, lots of texture, photogrammetric is enough.
Roof inspection
Photogrammetry. RGB is the actual deliverable; LiDAR adds little.
Forestry / canopy
LiDAR. Photogrammetry literally cannot see the ground under leaves.
Power line clearance
LiDAR. Vegetation encroachment requires bare-earth DTM under the lines.
3D modelling for marketing
Photogrammetry. Color photo texture is what makes the model look real.