- How: software finds shared features across photos and triangulates 3D positions.
- Wins for: objects, drone surveys, AEC, cultural heritage, game-asset capture.
- Loses to splatting for: photoreal walkable interiors — Gaussian splats train faster and render better.
Table of contents
The plain-English definition
Photogrammetry is the science of reconstructing 3D shape from overlapping 2D photographs by triangulating the same features across many images.
The core idea is the same one your eyes use to perceive depth. Two eyes see slightly different views of a scene. Your brain triangulates the offset and computes distance. Photogrammetry generalizes this to dozens or hundreds of camera positions, then solves a large geometry problem to recover where every surface in the scene sits in 3D space.
The output is usually a textured polygon mesh — a model with triangles forming the surfaces and photo textures wrapped across them.
From a pile of photos to a 3D mesh
1. Feature detection
The software scans every photo for distinctive points — corners, textures, edges — using algorithms like SIFT, ORB, or modern learned detectors like ALIKED.
2. Feature matching across photos
It finds which features in photo A also appear in photos B, C, D. This shared-feature graph is what makes triangulation possible.
3. Structure from Motion (SfM)
From the matches, the pipeline solves for where each camera was and where each matched point sits in 3D. COLMAP is the open-source standard here.
4. Dense reconstruction and meshing
The sparse 3D points are densified, then meshed into triangles, and finally textured using the original photos.
Where photogrammetry actually wins
Objects. Sculptures, products, archaeological artifacts. A single object photographed from 60 angles produces an accurate, reusable 3D model.
Drone surveys. Construction progress, agricultural mapping, terrain models. Drones are photogrammetry's killer app.
Cultural heritage. Buildings, statues, ruins. Museums and galleries have used photogrammetry for two decades to digitize collections.
AEC and surveying. Site documentation, as-built modeling, BIM workflows. Where measurement accuracy matters more than render speed.
Game and film assets. Capturing real props, faces, and environments for use in Unreal, Unity, or VFX pipelines.
"Polygon meshes are still the right tool when you need to measure, edit, or print. They are the wrong tool when you need to walk through a room online."
Photogrammetry vs the closest neighbors
vs Gaussian splatting
For photoreal walkable interior tours, Gaussian splatting now produces better-looking, faster-rendering results from the same input photos. The visual ceiling is higher and the training is cheaper. Photogrammetry is still preferred when you need an editable polygon mesh.
vs LiDAR
LiDAR (used by Matterport's Pro2 and the iPhone Pro sensor) measures depth directly with laser pulses. It is faster and more accurate for interiors. Photogrammetry is cheaper because it only needs a camera. See the 3D virtual tour overview for how these capture paths compare.
vs 360 panorama
A 360 panorama captures imagery from one point; photogrammetry reconstructs full 3D geometry from many points. Different jobs. For walkable interior tours, see the vendor comparison hub.
How to actually get a photogrammetry scan — and when not to bother
Free / DIY. Open-source COLMAP plus Meshroom or OpenMVS on your own GPU. Steep learning curve, full control.
Mobile. Polycam photo mode, KIRI Engine, or RealityScan run on a phone. Good for objects, marginal for whole rooms.
Pro. RealityCapture or Agisoft Metashape. Hundreds to thousands of dollars in licensing, plus a workstation GPU.
Honest take: if the goal is a walkable interior tour for your business, photogrammetry is the long way around. TourReady's $99 short-video walkable tour uses a Gaussian-splatting pipeline that delivers a hosted scene in about two minutes — without you running COLMAP yourself.