Photogrammetry is the science of extracting three-dimensional measurements from two-dimensional photographs. By capturing overlapping images of an object or landscape from different angles, software can calculate the shape, size, and position of everything in the scene. The technique works at every scale, from mapping entire cities with aerial cameras to digitizing a coin on a tabletop.
How Photographs Become 3D Models
The core principle behind photogrammetry is triangulation. When you photograph the same point from two different positions, each image records that point at a slightly different location on the sensor. That difference, called parallax, is the same phenomenon your brain uses when your two eyes judge depth. Software measures parallax across thousands of matching points between overlapping photos, then traces imaginary lines (called image rays) from each camera position through the image and out to the real-world point. Where two or more of those lines intersect, the software pins down the point’s exact position in three-dimensional space.
Do this for millions of points and you get a dense “point cloud,” a constellation of tiny dots that together describe every visible surface. From there, the software connects neighboring points into a mesh of tiny triangles, drapes color information from the original photos on top, and produces a textured 3D model you can rotate, measure, and analyze on screen. The math relies on knowing, or solving for, two things: the internal geometry of the camera (sensor size, focal length, lens distortion) and the camera’s position and angle for each shot. Modern software figures out both automatically from the images themselves.
Aerial vs. Close-Range Photogrammetry
Photogrammetry splits into two broad categories based on where the camera sits relative to the subject.
Aerial photogrammetry uses cameras mounted on aircraft or drones to photograph large areas from above. It’s the go-to method for topographic mapping, urban planning, agriculture monitoring, and archaeological surveys where you need to see a big chunk of terrain at once. Because the camera is far from the ground, each image covers a wide area, but fine surface detail can be limited unless the drone flies low or uses a high-resolution sensor.
Close-range (or terrestrial) photogrammetry places the camera on the ground, on a tripod, or in someone’s hands, typically within a few meters of the subject. Engineers use it to inspect bridges and pipelines, geologists track glacier movement with it, and heritage professionals digitize statues and building facades. The tradeoff is straightforward: you get finer detail but cover less area per session.
The Photogrammetry Workflow
Every photogrammetry project follows roughly four stages, whether you’re scanning a building or surveying a coastline.
Planning. Before taking a single photo, you decide on overlap strategy, equipment, and any reference points (called ground control) you’ll place in the scene. Control points are markers with known real-world coordinates that let you tie the final model to a geographic coordinate system or enforce accurate scale.
Image capture. You photograph the subject from many positions, ensuring each part of the surface appears in at least two (ideally three or more) overlapping images. For aerial work, drones fly predetermined grid patterns. For close-range work, you walk around the object in loops, keeping roughly 60 to 80 percent overlap between consecutive shots. One critical rule: never crop or resize images afterward, because the software needs the original pixel dimensions to calculate camera geometry correctly.
Processing and triangulation. Software detects matching features across the image set and runs a computation called a bundle block adjustment, which simultaneously solves for every camera position and every 3D point in one large optimization. This step catches errors, removes outliers, and produces the initial sparse point cloud. The software then densifies that cloud into millions of points.
Model generation. The dense point cloud is converted into a triangulated mesh, textured with color from the photos, and exported as a 3D model, orthophoto, elevation map, or whatever the project requires.
What You Need to Get Started
At its simplest, photogrammetry requires a camera and software. A modern smartphone can produce usable models of small objects. For professional results, a camera with a larger sensor and a fixed focal length lens gives the software more consistent data to work with. Research comparing drone surveys with different lenses found that focal lengths of 35 mm or shorter produced significantly better 3D reconstructions than longer lenses, because wider lenses capture more overlap and geometric diversity per image. Lenses longer than 50 mm showed noticeable quality losses.
On the software side, several mature tools dominate the market. RealityCapture, now owned by Epic Games, is popular in visual effects and game development for its speed and integration with real-time engines. Agisoft Metashape is widely used in research and surveying for its cross-platform support and one-time purchase model. Artec Studio recently introduced AI-driven reconstruction that can build detailed models from as few as 80 to 100 smartphone photos. Free and open-source options like Meshroom and COLMAP also exist for hobbyists and researchers.
Photogrammetry vs. LiDAR
LiDAR, which measures distance by bouncing laser pulses off surfaces, is photogrammetry’s closest rival for 3D capture. The two technologies overlap in many applications, but their strengths differ in important ways.
Photogrammetry is more cost-effective and accessible. It uses standard photography equipment, while LiDAR requires specialized sensors that can cost tens of thousands of dollars. Photogrammetry also produces full-color textured models by default, since the source data is photographs.
LiDAR generally offers greater precision in complex environments, particularly where dense vegetation or other obstacles block lines of sight. Laser pulses can penetrate tree canopy gaps to map the ground beneath, something photographs alone cannot do. LiDAR also works in low-light conditions, while photogrammetry depends on adequate, consistent lighting.
Many professional workflows now combine both: LiDAR for the geometry and photogrammetry for the color and texture. Some software packages can ingest both data types in a single project.
Where Photogrammetry Is Used
The technique spans a surprising range of fields. In construction and engineering, photogrammetry tracks earthwork volumes, monitors structural deformation, and generates site models that feed directly into building information models. For historical buildings with complex geometry, researchers have shown that photogrammetry can produce structural analysis models comparable to those built by hand, at lower cost and in less time.
Archaeology and cultural heritage rely heavily on photogrammetry to digitize excavation sites, fragile artifacts, and endangered monuments. A drone survey of a dig site can preserve spatial relationships between finds that are lost the moment objects are removed.
Forensic investigators use photogrammetry to create 3D records of crime scenes, autopsies, and accident sites. The model preserves spatial evidence that flat photographs cannot, allowing investigators to revisit and measure a scene long after it has been cleared.
In film and video games, photogrammetry captures real-world textures and environments that artists would otherwise model by hand. Entire game worlds have been built from scanned rocks, trees, and buildings. Surveyors, miners, farmers, and environmental scientists all use variations of the same underlying technique.
Accuracy and Limitations
Professional photogrammetric data in the United States is evaluated against standards published by the American Society for Photogrammetry and Remote Sensing. The latest edition defines accuracy classes for both horizontal and vertical measurements, with the tightest class (Quality Level 0) requiring vertical errors no greater than 5 centimeters. Typical drone survey projects achieve vertical accuracy in the range of 5 to 10 centimeters, depending on flight altitude, camera quality, and the number of ground control points used.
The technique does have blind spots. Transparent, reflective, and textureless surfaces cause problems because the software cannot find distinguishing features to match between images. Glass, still water, and uniform white walls are classic failure cases. Lighting matters too: harsh shadows or changing light between shots can confuse feature matching.
A newer technology called neural radiance fields (NeRF) is starting to address some of these gaps. NeRF uses machine learning to synthesize 3D scenes from photos and has shown better results than traditional photogrammetry on reflective and transparent surfaces. In forensic testing, NeRF models produced more lifelike and detailed visualizations of difficult surfaces. The tradeoff is high computational demand and limited compatibility with standard 3D file formats, so it currently serves more as a complement to photogrammetry than a replacement.