Drone photogrammetry turns overlapping aerial photographs into 3D models, orthomosaics and elevation maps, and it has transformed surveying, construction and archaeology. The principle is simple, the same point photographed from many angles can be located in 3D, but the practice needs careful flight planning and good ground control. This guide covers the basics, from planning to processing.

How Photogrammetry Works

Photogrammetry builds 3D models by finding the same features in many overlapping photographs and triangulating their positions. The drone flies a grid pattern, photographing the ground with 70-80% overlap between images, so every point appears in several photos. Software matches the features, calculates the camera positions, and builds a dense point cloud, then a mesh, then an orthomosaic and a digital elevation model. The more overlap and the better the photos, the more accurate the model.

Planning the Flight

Plan the flight before you take off: set the altitude for the resolution you need, lower for more detail, higher for more coverage, and set the overlap to at least 75% forward and 65% side. Fly in consistent light, ideally overcast, because changing shadows confuse the matching software. Use a flight planning app to automate the grid, and check the battery: a large site may need several flights. A well-planned flight saves hours of processing time.

Ground Control Points: The Accuracy Key

Ground control points, markers on the ground with precisely measured coordinates, are what turn a relative model into an accurate survey. Without them, the model's absolute accuracy is limited to the drone's GPS, which is metres, not centimetres. With them, accuracy improves to a few centimetres, which is what surveyors and engineers need. Place five to ten GCPs spread across the site, measure them with an RTK GPS or a total station, and mark them clearly so they are visible in the photos.

Processing: Software Choices

The main processing options are Pix4D, Agisoft Metashape, and the open-source OpenDroneMap, which is free and capable. Processing takes hours for a large site, and it needs a powerful computer with plenty of RAM and storage. The output is an orthomosaic, a distortion-free aerial image, a digital surface model, and a 3D mesh, which you can import into QGIS or CAD software. Start with OpenDroneMap to learn the workflow, then consider paid software when you need support and speed.

Accuracy and Limitations

Photogrammetry accuracy depends on the flight, the ground control, and the terrain: flat, featureless areas like water and fresh snow produce poor models, because the software cannot find matching features. Vegetation is a problem too, because the model shows the top of the canopy, not the ground beneath it. LiDAR is better for terrain under trees, while photogrammetry is better for texture and colour. Know the limits, and the tool will serve you well.

Photogrammetry and the Real World

Photogrammetry is used far beyond drones: archaeologists reconstruct ancient sites, engineers inspect bridges, and filmmakers create digital sets from real locations. The principles are the same, and the skills transfer: a drone pilot who understands photogrammetry can work in any of these fields. The technology is becoming cheaper and more accessible, and the applications are growing. The 3D model is the output, but the understanding of how it was made is the skill, and the skill is what employers pay for.

Key Takeaways

- Photogrammetry builds 3D models from overlapping photos with 70-80% overlap - Ground control points improve accuracy from metres to centimetres - OpenDroneMap is the free processing option; Pix4D and Metashape are paid - Photogrammetry struggles with water, snow and dense vegetation

Frequently Asked Questions

How accurate is drone photogrammetry? With good ground control points, accuracy reaches a few centimetres. Without them, accuracy is limited to the drone's GPS, which is metres.

Photogrammetry turns photos into maps. Plan the flight, place ground control, and let the software build the model.