Compute normals

A normal is the direction a surface faces at a point. LiDAR measures positions, not surfaces, so normals have to be estimated: Phytograph fits a small plane through each point's nearest neighbors and takes the direction perpendicular to it.

Computing them once and storing them on the cloud means you can color by them, export them alongside the points, and inspect surface shape without recomputing anything.

Run it

1. Open Compute Normals (tool column, or Tools → Pre-processing → Compute Normals).
2. Under Clouds, check the clouds to compute normals on. The list opens with whatever is selected in the Scans pane already checked; the box beside the Clouds label checks or clears them all. A cloud that was not imported from a file is listed but can't be checked, because normals are stored with the imported cloud.
3. Set the options below, then click Compute Normals.

The checked clouds are processed one after another, all with the same options, and the panel names the cloud it is on. Cancel stops the run; clouds that already finished keep their normals. Checking a hidden cloud shows it while it stays checked, and hides it again when you uncheck it or the panel closes.

Each cloud is recolored by curvature when the run finishes, which is usually the most informative first look: smooth bark and ground go dark, foliage and edges light up.

Options

Neighbors — how many nearby points are fitted to estimate each normal. Larger values give a smoother, noise-tolerant result but blur fine detail like twigs and leaf edges; smaller values follow detail but are noisier. The default of 30 suits most scans.

Limit search radius — off by default. A plain neighbor count adapts on its own to a terrestrial scan, where return density falls with distance from the scanner: a fixed radius that is right at 5 m finds almost nothing at 40 m. Turn it on, and set a radius in meters, when you need to stop the search bridging across a gap — between two leaves that are close in space but not part of the same surface, for instance.

Orientation — a plane fit gives a direction but not which way along it the surface faces, so the result has to be oriented:

• Toward sensor (default) flips each normal back along the beam that measured it. This is the physically correct answer for a scan, and Phytograph uses the best sensor information the cloud carries — the per-pulse beam origins if they were imported, otherwise the scan origin. If nothing records where the scanner was, the cloud's own center is used as the viewpoint instead, so normals face inward toward it — the same convention as a real sensor, which sees a surface from one side.
• Up (+Z) points every normal to the upper hemisphere. Use it for ground, terrain, or a DEM-like surface.
• Leave unoriented keeps the raw plane fit. Fine for curvature and verticality, which ignore the sign, but not for meshing.

What it writes

Five per-point columns, all available in the Display → Color by picker and in the export column picker:

Column	Meaning
nx, ny, nz	The normal's three components.
curvature	Surface variation — near 0 on a flat patch, rising at edges and in foliage. Cannot exceed 1/3.
verticality	Angle of the normal from vertical, in degrees, folded to 0–90. 0 is flat ground, 90 a vertical trunk or wall.

nx/ny/nz are the standard PLY names for normals, so a cloud exported as PLY carries its normals into CloudCompare, MeshLab or anything else that reads them — and a PLY with normals imported back into Phytograph is recognized as carrying normals rather than three unrelated scalar columns.

Large clouds

The compute splits the cloud into overlapping tiles and runs them across every available core, so it scales to hundreds of millions of points without needing the whole cloud in memory at once. Each tile overlaps its neighbors by enough that a point near a tile edge still sees its full neighborhood — the result is identical to processing the cloud whole.

Expect roughly 40 seconds for 10 million points on a modern laptop. Above 5 million points the recoloring is handed to a background job so the columns are usable immediately, and you can keep working while the viewer catches up. Cancel stops a run at any point and leaves the cloud untouched.

Normals and later edits

Normals describe a point's neighborhood, so cropping, erasing or filtering the cloud afterwards changes what the right answer would be for every surviving point near the cut.

Phytograph does not throw the columns away when this happens — away from the cut they are still a good approximation, and recomputing a large cloud is not free. Instead, reopening Compute Normals on an edited cloud shows a note that the normals may be out of date, and the button offers Recompute Normals. Recompute when you are about to rely on them; ignore it when you are not.

Related

• Clean a point cloud — crop, filter and resample before computing normals, not after.
• Triangulate a mesh — meshing estimates its own normals internally, with the orientation each method needs.
• File formats — which formats carry normals out of Phytograph.