What Is Photogrammetry? How We Turn Drone Footage Into Accurate 3D Maps

Photogrammetry is one of those words that sounds like it was designed to make people stop asking questions. It's long. It has 'gram' in the middle. It sounds technical and inaccessible.

It's actually a beautifully simple idea.

Photogrammetry is the science of extracting measurements from photographs. Specifically: if you take many overlapping photographs of the same object or surface from different positions, a computer can calculate the 3D geometry of that surface with extraordinary accuracy — because the differences between the images tell you exactly how far away each point is, from multiple angles, simultaneously.

That's it. That's photogrammetry.

What makes it powerful in the drone context is that a UAV can take thousands of these overlapping images of a large area very quickly, very precisely, and from positions that would be impractical or impossible for a camera on a tripod or a person on foot.

How It Actually Works: Step by Step

Step 1 — The Drone Captures Overlapping Images

GarudX's survey drones fly a pre-planned grid pattern over the area being mapped. Every image overlaps with the adjacent images by at least 70–80% — this redundancy is not waste, it's the mechanism that makes the whole process work.

Why so much overlap? Because photogrammetry software needs to identify the same physical feature in multiple images taken from different positions. The more images that capture the same point, the more accurately the software can triangulate its 3D position.

For a flat agricultural area, the drone might fly 50–80 metres high, covering 5 hectares in a 20-minute flight with 400 images. For a steep river valley or building face, we fly lower and slower — more images, more detail, more processing time.

Step 2 — The Software Finds Matching Features

After the flight, the images go into photogrammetry software (we use Agisoft Metashape and DJI Terra depending on the project). The software identifies thousands of distinctive features — a rock texture, a painted road marking, a corner of a building — that appear in multiple overlapping images.

From the differences in where each feature appears in each image (called disparity), the software mathematically calculates the 3D position of every feature in space. This step is called Structure from Motion (SfM), and it produces a sparse point cloud — a constellation of 3D points representing the identified features.

Step 3 — Dense Matching Creates the Full Point Cloud

From the sparse feature points, the software generates a dense point cloud — running the same matching algorithm at full image resolution across every overlapping pair. A 400-image survey of 5 hectares might produce 50–100 million 3D points. Each point has an X, Y, Z coordinate and an RGB colour value from the original image.

Step 4 — Ground Control Points (GCPs) Anchor It to Real-World Coordinates

A point cloud without reference to real-world coordinates is floating in space. To anchor it accurately, we place Ground Control Points — physical markers (usually painted crosses or checkerboard targets) at known GPS coordinates measured with a DGPS or RTK GPS receiver.

The software fits the point cloud to the GCPs, correcting any accumulated drift from the drone's onboard GPS. With 5–8 well-placed GCPs, we achieve 3–5cm absolute accuracy. With RTK-corrected drone positioning, we can sometimes eliminate the need for physical GCPs while maintaining similar accuracy.

Step 5 — Deliverables Are Generated

From the point cloud, we generate whichever deliverables the project requires:

Orthomosaic: A flat, georeferenced aerial photograph stitched from hundreds of drone images and corrected for perspective distortion. Every point in the image has a real-world coordinate. This is what you use like a map — you can measure distances, areas, and draw boundaries.

Digital Surface Model (DSM): The elevation of everything the drone saw — buildings, trees, vehicles, and the ground surface.

Digital Terrain Model (DTM): The elevation of the bare ground only, with buildings and vegetation mathematically removed. Critical for flood modelling, infrastructure design, and topographic analysis.

3D Mesh / Point Cloud: The full 3D model. Can be imported into GIS software, AutoCAD, Revit, or a web viewer. Used for volumetric calculations, structural inspection, heritage documentation.

What Accuracy Can You Expect?

This is the question engineers ask most often, and the answer has two parts: relative accuracy (how consistent the model is with itself) and absolute accuracy (how close the model's coordinates are to real-world GPS coordinates).

Standard photogrammetry with GCPs: 3–5cm absolute horizontal, 5–8cm vertical. Sufficient for land surveys, construction monitoring, terrain modelling.

RTK/PPK corrected photogrammetry: 1.5–3cm horizontal, 3–5cm vertical. Suitable for engineering design and formal submissions.

Combined UAV + DGPS: <2cm absolute, independently verified at checkpoints. Required for legal cadastral surveys and regulatory submissions.

What Photogrammetry Cannot Do

Photogrammetry sees what the camera sees. Under dense forest canopy, the camera sees only the top of the canopy — the ground below is invisible. This is where LiDAR becomes essential: laser pulses can penetrate gaps in vegetation to measure the ground surface beneath.

For heavily forested terrain, powerline corridors under tree cover, or any application where bare-earth elevation under vegetation is needed, LiDAR is the right tool. We offer both.

Why This Matters for You

If you've ever needed to know: how big is this piece of land, how much soil has been moved, where exactly does this property boundary sit, how has this construction site changed since last month, or what does the ground look like in a place too steep or remote to walk — photogrammetry is how you answer that question, accurately, repeatably, and far faster than any ground-based survey method.

That's the practical value of a word that sounds complicated but describes something elegantly simple: measure the world by looking at it from above, from many angles, with a lot of overlap.

If you have a survey project in Nepal, talk to us. We'll tell you whether photogrammetry is the right tool, how accurate it will be, and what the deliverables will look like — before you commit to anything.