Quick Read. A single photograph collapses depth. Several overlapping photographs from different positions contain something the single frame lost: parallax. Software can identify common features, estimate where the cameras were, triangulate three-dimensional points and build a spatial model. Add known scale or surveyed control and that model can become measurable.
One-sentence answer: Photogrammetry works by comparing how the same features shift across photographs made from different viewpoints and using those shifts to reconstruct geometry.
Parallax Contains Depth
Hold up a finger and alternate eyes. The finger shifts against the background because each eye sees from a different position. Photogrammetry generalises that principle. If the same corner, texture or mark appears in several photographs, its relative displacement helps constrain where it exists in three-dimensional space.
Overlap Is Evidence
Photographs need substantial shared coverage so algorithms can identify corresponding features. Aerial mapping therefore uses planned flight lines with forward and side overlap; object scanning circles around a subject while keeping neighbouring views connected.
A photograph that sees a completely unrelated part of the object gives the reconstruction system no bridge back to the earlier frames.
First Find the Cameras
Structure-from-motion methods estimate both scene structure and camera pose from image correspondences. The software effectively asks: what arrangement of cameras and 3D points could have produced these observed 2D feature locations?
Once camera positions are estimated, denser matching can produce millions of spatial points, surfaces and textured meshes.
Scale Does Not Appear From Nothing
A reconstructed object may have correct proportions but unknown absolute size unless the system has scale information. A measured reference distance, coded target, GPS/RTK data or survey control can anchor the model to real units.
This is the difference between a convincing 3D shape and a defensible measurement product.
Texture Helps; Shiny Surfaces Hurt
Feature matching likes stable, distinctive visual texture. Blank white walls provide few points to recognise. Glossy or transparent surfaces change appearance with viewpoint, violating the assumption that a visible feature belongs to one stable surface point.
Good photogrammetry capture therefore seeks even illumination, sharp images and enough non-repeating texture for reliable correspondence.
Orthomosaics Correct Perspective for Mapping
A raw aerial photograph contains perspective displacement. Orthorectification uses the reconstructed geometry and camera model to project imagery onto a common map-like surface. Multiple corrected images can then be blended into an orthomosaic suitable for measurement and comparison.
Three Safe Exercises
- Small-object scan. Photograph a textured harmless object from many overlapping angles and reconstruct it with suitable photogrammetry software.
- Missing-overlap test. Deliberately remove connecting views and observe where reconstruction breaks.
- Scale test. Include a measured reference object and compare the model before and after scale is defined.
Common Misconceptions
- “More photographs always improve the model.” Blur, repetition and poor geometry can add data without adding constraint.
- “A realistic model is automatically accurate.” Visual plausibility and measurement accuracy are different standards.
- “One photograph contains full 3D shape.” Ordinary monocular images lose depth; multi-view displacement helps recover it.
The Final Idea
Photography normally flattens the world. Photogrammetry makes that loss reversible enough to measure by giving the camera several chances to look from different places. Each frame is incomplete. Their disagreements contain the depth.
Explore the Culture library.