How dense does drone LiDAR need to be for automated linework?

Search "lidar point density" and you get specs written for terrain mapping. USGS base specifications, ASPRS density guidance, a 2014 blog post on point spacing. All of it answers one question: how many points do I need to build a good ground model?

That is the wrong question for most drone survey work today. Nobody flies a corridor or a site just for a DTM anymore. They fly it to get linework: edge of pavement, curb, striping, crown, poles, signs, building footprints. And the density that makes a fine DTM is nowhere near the density that lets software draft a curb line for you.

This post is the guidance we give customers when they ask "what should I fly at so your automation actually works?" One framing note first. Across all the data we process, three things decide how much automation can pull from a dataset: calibrated imagery aligned to the cloud, coverage and overlap, and density. For drone LiDAR, density is usually the deciding one, and it is the one flight planning controls most directly.

For automated linework on Mach9, our engineering team's guidance for drone and aerial LiDAR is a minimum of 100 points per square meter, with 400 or more recommended. Results degrade as density drops toward that minimum. Below 100, the platform still processes and classifies the data, but results fall off quickly, so plan on drafting more of the linework by hand.

Here is what the automation drafts from drone LiDAR today:

• Linework from the point cloud. Edge of pavement, painted lines, sidewalk edges, vegetation edges and overhead wires (vegetation edges and wires are in beta). On dense data, roughly 650 points per square meter and up in the drone projects we have run, flown in a cross-grid, separate curb lines, crosshatch and guardrail come through too.
• Building outlines. Drafted from an orthophoto uploaded with the point cloud (beta).
• Ground classification. On every upload, at any density.

What still needs an operator: point features such as signs, poles and manholes, which are extracted from mobile mapping data but not from drone data today, plus QA and the judgment calls. Drone linework can also come out rougher at the edges than mobile, so budget time for review.

Two caveats. Anything with a vertical face (sign faces, pole-mounted equipment) comes through at lower density from the air than from the ground no matter how you fly, so plan on oblique passes or ground capture if the deliverable depends on them. And the aerial models are still expanding, so a feature that automates on mobile data today may not be on for aerial yet.

The pattern is simple. Terrain is forgiving because it is big. Linework is not. A typical curb is a 6-inch break in a surface, and the software needs several points across that break to find it. At 100 points per square meter, points land about 10 cm (4 inches) apart, so a curb gets one or two. At 400, they land about 5 cm (2 inches) apart, and it gets about three.

These two get mixed up constantly, so it is worth separating them.

Accuracy is how close each point is to where it really is. It comes from your GNSS and IMU, your control, your PPK or RTK workflow, and your calibration. A well-run drone LiDAR flight can be accurate to a few centimeters at 50 points per square meter.

Density is how many of those accurate points land on each square meter of ground. It comes from your altitude, your speed, your sensor's pulse rate, and your overlap.

An accurate but sparse cloud is great for volumes and terrain. It is still a bad input for automated linework, because the algorithm cannot draft an edge it cannot see.

Density scales with time over target. Fly lower, fly slower, overlap more. Every sensor has its own numbers, but the direction is always the same.

One customer example. Welch Comer, a civil engineering and surveying firm in Idaho, flies a GeoCue TrueView 540 at 150 to 200 feet above ground and about 5 meters per second on the projects they bring into Mach9. Their survey lead flies as low and as slow as he safely can to build up density, and by GeoCue's published figures those settings work out to roughly 900 to 1,200 points per square meter in a single pass, well above the 400 we recommend. In our joint webinar, he said the extracted linework comes out denser than what they would collect in the field.

A few practical notes from their workflow and others like it:

• Plan for the smallest feature you need. If the deliverable includes curb and gutter, fly for the curb. Terrain will take care of itself.
• Cross-flight the corridor and keep overlap. Edges of the swath are the sparsest part of any pass, and a single flight direction biases both the cloud and the ortho. Perpendicular strips fix both. For aerial LiDAR, 30 to 50 percent side overlap is a reasonable baseline.
• Add an oblique pass if you need vertical features. Pure nadir flying puts very few points on sign faces, pole hardware and building walls. A separate oblique mission is the cheapest way to get them for your operators to draft, since point features are not automated on drone data today.
• Slow down at intersections. Turn arrows, crosswalks and curb ramps are the densest concentration of small features on a roadway job.
• Don't fly high to save battery on a linework job. You will spend the saved flight time, several times over, drawing lines in the office.

Density gets you geometry. Imagery helps the software tell what it is looking at.

For drone work, that means an orthophoto. Upload a georeferenced GeoTIFF in the same coordinate system as the LAS or LAZ tiles, and the platform uses it alongside the points for paint and pavement boundaries that geometry alone cannot separate from the ground, and to draft building outlines.

Point features are the exception. Signs, poles and other objects are recognized from mobile mapping data, where calibrated panoramic imagery (E57 is the cleanest way to deliver it) gives the software a close look at each one. Drone data does not get point features today, so if the deliverable needs an asset inventory, plan a mobile or ground pass.

If you have to choose between flying denser and flying with a good camera, the answer depends on the deliverable. Edges and linework: density. Paint and building outlines: the orthophoto helps most.

You do not always control the flight. Sometimes the data comes from a subcontractor, or was flown for a different deliverable, or the terrain forced you high.

Low-density drone data still processes. The cloud gets classified on upload, you can build a TIN and contours from whatever you draft, and building edges come through if you brought an orthophoto. What you lose is most of the automated linework. Plan the office time accordingly, and use the semi-automated tools (the linear extraction tool along a guideline, drape to snap vertices to the cloud, pick surface to land a click on the averaged surface instead of a stray point) rather than drawing every vertex from scratch.

And before you re-fly, ask what the deliverable actually needs. A site plan with terrain and footprints may be fine as flown. A roadway plan with curbs is not.

Drone LiDAR sensors got good faster than the office side did. A crew can fly a corridor in an afternoon that used to take a week on the ground. Then the data goes back to the office and someone spends three days tracing curbs, because the software they have cannot do it.

The tool is the bottleneck, not the team. Flying for density is the cheapest fix there is, because it turns the office work from drafting into checking.

What point density do I need for drone LiDAR feature extraction? Our engineering team's guidance is a minimum of 100 points per square meter, with 400 or more recommended for automated linework such as edge of pavement and striping. If you need curbs, plan on well above that, flown in a cross-grid. Terrain-only deliverables work at much lower densities.

How accurate is drone LiDAR? Survey-grade drone LiDAR with proper control and a PPK or RTK workflow is typically accurate to a few centimeters. Accuracy depends on your positioning workflow, not your point density.

Does flying lower always give better results? Lower and slower gives more points per square meter, which helps automation find small features. The tradeoff is flight time and battery. Fly for the smallest feature in the deliverable.

Can I extract linework from sparse drone data? Yes, manually or with semi-automated tools that snap to the point cloud. Automated linework needs enough points across each edge for the software to detect it.

Upload a drone dataset and see what the automation drafts before you plan the next flight. Book a demo at mach9.ai/get-started.