Photogrammetry and 3D Models

Photogrammetry is the process of reconstructing three-dimensional shape from ordinary photographs. Take enough overlapping pictures of an object from enough angles, and software can work backwards from them to determine where every surface sits in space.

It is not a scanner and it is not LiDAR. Understanding the difference is most of what a buyer needs to know, because it determines which jobs this technique answers well and which ones it cannot answer at all.

How the reconstruction works

The process has three stages, and each one has a failure mode worth knowing about.

Matching. Software finds the same physical feature across many photographs — the corner of a parapet, a crack in the paving, a bolt head. This is why texture matters so much. A surface with visible detail produces thousands of matched points. A smooth white wall, a sheet of glass, or still water produces almost none.

Triangulation. Knowing where each photo was taken and where a feature appears in each frame, the software solves for that feature's position in space. Repeated across millions of features, this produces a dense cloud of measured points.

Surfacing. The point cloud is turned into a continuous mesh, and the original photographs are draped over it as texture. The result is a model that both measures correctly and looks like the real thing.

What you receive

The point cloud is usually the file that matters to a technical team, because it is the one that carries measurable geometry into Autodesk products. The mesh is what non-technical stakeholders actually look at.

What it is good for

Existing-conditions capture. Before a renovation or an addition, a current model of what is physically there. Far faster than measuring by hand, and it captures things nobody thought to measure.

Facade and structure documentation. Elevations of a building, bridge, or tower without scaffolding or a lift. Particularly useful where the surface is high, awkward, or unsafe to reach.

Coordination inputs. A model of the site as built, brought into the same environment as the design model, so the two can be compared.

Volume and area work. Material stockpiles, excavation, and surface areas calculated from captured geometry rather than estimated.

Visualization. Marketing, entitlement hearings, and community presentations, where a model people can rotate communicates more than a set of drawings.

Where photogrammetry fails

This is the section most providers leave out, and it is the one that prevents wasted money.

It cannot see through vegetation. Photogrammetry models the visible surface. If tree canopy covers the ground, the model shows canopy. There is no processing step that recovers the ground beneath it, because no camera ever saw the ground. Projects that need bare-earth terrain under trees need LiDAR, which uses laser pulses that can find gaps in foliage. We will say so rather than sell a model that cannot answer the question.

Reflective and transparent surfaces reconstruct badly. Glass curtain wall, standing water, polished metal, and wet asphalt all fail to produce stable matches. These areas come out warped, holed, or smeared. They are flagged in delivery rather than quietly smoothed over.

Featureless surfaces are unreliable. A large blank wall, a fresh concrete slab, or a snow field gives the matcher nothing to lock onto.

Interiors are harder than exteriors. Enclosed spaces have limited camera positions, uneven light, and repeating features that confuse matching. It is doable, but it is a different flight and a different scope.

Moving objects create artifacts. Vehicles, equipment, and people appear in some frames and not others, and end up as ghosts or smears.

Photogrammetry compared to LiDAR

Both produce point clouds. They are not interchangeable.

For structures, open sites, stockpiles, and anything where you want the model to look like the real thing, photogrammetry is usually the better value. For ground elevation under trees, LiDAR is the answer and it is worth paying for.

Scoping a model correctly

The single biggest driver of cost and usefulness is level of detail, and it is decided before the flight.

A model for a community presentation needs to look right from a hundred feet. A model supporting renovation coordination needs to be correct at the scale of a wall opening. They are different flights, different altitudes, different image counts, and different processing times. Asking for "a 3D model" without settling this is how projects end up with a file that is either too coarse to use or so large nobody can open it.

Tell us three things and the scope resolves itself: what the model feeds into, who opens the file, and what the smallest thing is that has to be visible in it. Scope and starting prices for this work are on the 3D modeling service page.

Frequently asked questions

Is a photogrammetric model accurate enough for construction?

It depends on what the model is for. These models are widely used for existing-conditions capture, quantity work, documentation, and coordination. Where a legal boundary or a survey of record is involved, a licensed land surveyor governs and the model supports that work rather than replacing it.

Can you model the inside of a building?

Yes, on a smaller aircraft or with a handheld capture, but interiors are scoped separately. Enclosed spaces limit camera angles and lighting, and both affect the result.

What if the site has heavy tree cover?

Then the ground under those trees will not be in the model. If that ground is what the project needs, this is a LiDAR job. We would rather tell you that up front than deliver a surface model that shows treetops where you expected terrain.

How long does processing take?

Longer than the flight, often considerably. Dense reconstruction is computationally heavy and larger or higher-detail models take longer. Expect delivery several business days after capture.

Can the model be simplified for our machines?

Yes. Full-resolution meshes are large. Decimated versions are produced on request so the file opens on ordinary hardware without a specialist workstation.