Aerial photogrammetry: how drone photos become survey maps

Aerial photogrammetry is the measurement of the ground from photographs taken from the air. For most of the last century it meant aircraft with large-format survey cameras mapping whole regions. Drones made it cheap enough to map a single construction site every week, and that changed what it is used for. This guide explains how it works, what it produces, how accurate it is, and when a laser scanner is the better tool.

The idea in one paragraph

Look at a point on the ground from two places and its position in three dimensions can be worked out from the difference between the two views, the way two eyes judge depth. Aerial photogrammetry does this for every point on a site, from hundreds or thousands of overlapping photos, and solves where every camera was at the same time. Photography records what the ground looks like. Photogrammetry measures where it is.

From photos to a map, step by step

1. Overlapping photos. The drone flies a grid, usually with 75 to 80% overlap along each line and 65 to 70% between lines, so every point on the ground appears in many photos taken from different positions. (How to fly for it.)
2. Tie points. Software finds distinctive features in each photo (a stone, a tyre track, the corner of a slab) and matches the same feature across photos. Thousands of these matches tie the photos together.
3. Structure from motion. From the tie points alone, the software solves where each photo was taken, which way the camera pointed, and the lens's own distortion. A bundle adjustment refines all of it together until the rays from every camera to every tie point meet as closely as possible.
4. Georeferencing. The solved model is placed on the earth using the GPS position written into each photo and, for survey work, ground control points or RTK or PPK positions, which fix it to within centimetres. (GCPs, RTK and PPK compared.)
5. Dense matching. With the cameras known, the software matches nearly every pixel between neighbouring photos, producing a dense point cloud: millions of measured points, each with a colour.
6. Surface model. The point cloud becomes a digital surface model (DSM), a grid of heights of whatever was seen from above, and often a 3D textured mesh.
7. Orthomosaic. Each photo is projected onto the surface model so that every pixel is seen straight down at true scale, and the photos are blended into one map image.
8. Terrain model. Points on vegetation, vehicles and structures are classified and removed, and the gaps filled, to give a digital terrain model (DTM) of the bare ground. (The three outputs explained.)

How accurate it is

Two separate accuracies matter. The shape of the model (relative accuracy) is excellent even with no positioning at all: distances within a site are usually right to a few centimetres. Where the model sits on the earth (absolute accuracy) depends on positioning: with ground control, RTK or PPK it is typically one to three times the ground sample distance horizontally and two to three times vertically, so a few centimetres; with only the drone's ordinary GPS, a metre or more. Checkpoints, surveyed targets the model is not told about, are how you prove which you have.

Photogrammetry or LiDAR?

For construction sites, quarries and earthworks, which are mostly open ground, photogrammetry is usually all that is needed and gives the photographic map people actually look at. LiDAR earns its cost in forest, dense vegetation, power line corridors and anywhere the ground hides under something.

Where aerial photogrammetry is used

• Construction and earthworks: progress, cut and fill against design, as-built surfaces. (Progress monitoring with drones.)
• Quarries, mines and aggregate yards: stockpile inventory, faces, haul roads. (Stockpile inventory.)
• Land development and topography: existing ground for design, drainage, contours.
• Agriculture, forestry, insurance, heritage: crop health, damage assessment, records of sites and buildings.

Doing it yourself, or having it done

Processing is the computationally heavy step. It can be done with open-source software on a strong computer, with desktop software you buy, or by a hosted service that takes the photos and returns the maps. (What is free and what is not.) Clarity's Enterprise plan processes your drone photos into the orthomosaic, surface and terrain models, point cloud and mesh, with ground control and PPK handled before processing starts; on Studio you bring maps you have already processed and measure them in the browser.