3D Documentation of Heritage Sites Before They Change
Photographs and condition reports describe a site as it looked on one visit. 3D documentation of heritage sites keeps a dated, walkable record from before it changes, without an in-house 3D team.
A heritage site rarely announces the day it starts to change. A roof gives out during a routine repair, a flood reaches a basement archive, or a restoration grant finally arrives and scaffolding goes up long before anyone thought to record what was underneath it. 3D documentation of heritage sites is the record made before that day, not after: a dated, walkable scene of the place as it actually stood, not a handful of photos taken on the last site visit anyone happens to remember.
Conservation teams already know their buildings and objects are ageing. What most do not have is a pipeline that turns a single site visit into a record precise enough to matter later, and that does not mean a survey firm on retainer or a 3D department built from nothing.
What photographs alone cannot answer later
When the Great Buddha of Bamiyan was destroyed in March 2001, a team at ETH Zurich tried to rebuild the larger statue from whatever images existed. Of the fifteen photographs they could source, only four were usable for reconstruction, the rest too inconsistent in angle and lighting to triangulate a surface from (Grun, Remondino and Zhang, The Photogrammetric Record, 2004). That is the real failure of photograph-only documentation. Not that the photos were bad, but that nobody shot them with reconstruction in mind, so the record could not answer a question nobody had thought to ask yet.
Most heritage documentation still works this way: a condition report, a folder of images from the last inspection, measurements taken by hand for one specific purpose and never repeated the same way twice. That is enough to describe a site. It is rarely enough to rebuild it, compare it against itself a year later, or settle a dispute about what changed and when. Heritage documentation before restoration matters most of all here, because once scaffolding goes up, the "before" a reviewer or a donor wants to see is gone for good.
What does 3D documentation of heritage sites actually involve?
It means capturing a site with photos, ordinary video, 360 footage or a LiDAR scan, then reconstructing it in the cloud as a Gaussian Splat: a continuous, photoreal 3D scene rather than a sparse point cloud or a flat set of photos. The result opens in a browser, on a phone or in a headset, and it can be revisited and compared years later.
A laser scan on its own produces a point cloud: accurate, but a cloud of measured points that reads as a diagram to anyone outside a survey team. Reconstructing that same capture as a Gaussian Splat turns it into continuous surfaces a curator, a donor or a board member can actually walk through rather than decode. For the mechanics of how that reconstruction is built, see what a 3D Gaussian Splat actually is.
The value of doing this before a site changes is not hypothetical. Art historian Andrew Tallon laser scanned Notre-Dame de Paris in 2015, producing more than a billion data points accurate to five millimetres. When fire tore through the cathedral's roof in April 2019, that scan became one of the primary references architects used to rebuild what had burned (MIT Technology Review, 2019). Nobody scanned Notre-Dame because they expected a fire. They scanned it because the record was worth having anyway, and climate stress is making that habit harder to postpone: a 2025 study found 80% of UNESCO World Heritage cultural sites already burdened by climate stress, with 19% built of materials, stone and wood among them, that are directly threatened (Chen et al., Communications Earth & Environment, 2025).
Turning a site visit into a 3D record
3D documentation of heritage sites breaks into four stages here, and none of them require a 3D specialist added to the conservation team.
Plan the capture
Walk the site once before recording, and decide what has to be unambiguous later: a facade under threat from subsidence, a collection room before an HVAC retrofit, an interior before a repaint. Overlap matters more than resolution here. A phone held at chest height, walked slowly with each frame overlapping the last by half, covers a room better than a handful of carefully composed, widely spaced photos ever will.
Capture on site
Most heritage interiors and facades need nothing more specialised than a phone or a 360 camera (Insta360, Ricoh Theta and GoPro Max footage all work). A single room takes ten to fifteen minutes to walk properly. Where a site already carries LiDAR data from a prior structural or planning survey, in E57, LAS, LAZ or PLY, that scan can be turned into the same kind of scene through lidar to 3D reconstruction rather than recaptured from scratch, since it holds millimetre-level geometry a phone walk cannot match. For an interior too large or too dark to shoot well handheld, a 360 camera walk converts the same way through 360 video to 3D.
Direct sun through stained glass or across a stone facade blows out highlights the same way it would in a photograph, and a blown-out surface gives a reconstruction the least to work with. Flat, even light produces a cleaner scene than a bright one.
Process and receive the scene
Processing happens in the cloud: 10 to 120 minutes depending on the size and density of the capture, with no local GPU and nothing to install on a museum laptop that was never meant to run reconstruction software. The output is a Gaussian Splat that opens as .spz, .ply, .splat or .ksplat, shareable as a link, an embed on the institution's own site, or a QR code printed beside the object it shows.
Add context and archive it
Hotspots, added in the Splat Editor, carry the text, image or video a curator wants attached: an inscription's translation, a comparison photo from a prior restoration, a link to the accession record. Multiple captures of the same site, made across separate visits before and after a repair or a season of weather, can be aligned and compared directly through Multi Splat, which renders the difference between two visits as a 4D timelapse rather than two archive entries a researcher has to compare by eye.
| Photographs and condition report | Laser scan alone | Captures Studio | |
|---|---|---|---|
| Time on site | Under an hour, repeated on every follow-up question | Half a day to a full day with survey equipment | 10 to 15 minutes to capture |
| Deliverable | A folder of images and a written report | A dense point cloud, accurate but hard to read unassisted | A walkable, photoreal scene, ready in 10 to 120 minutes |
| Who can open it | Anyone, but it answers only what was photographed | Someone with point cloud software and training | Anyone with a browser, phone or headset |
| Cost to redo | A full repeat visit each time | A survey firm booked again | A short recapture, comparable directly to the last one |
What a conservation team actually asks before trying this
Do we need to buy anything? A phone, and for a large or dark interior, a 360 camera, are usually enough. There is no rendering rig to buy and no local GPU required, since reconstruction runs on Captures Studio's servers rather than the museum's own hardware. 3D scanning for museums and archives does not have to start with a laser scanner. For most rooms, a phone walk is the actual starting point.
Is it as accurate as our survey instruments? No, and it is not sold as a substitute for the sub-millimetre readings a structural engineer takes with a total station for active monitoring. What it gives a conservation team is something a point cloud does not: a scene a board member, a grant reviewer or a member of the public can actually walk through and understand, built from the same site visit.
Who has to learn something new? Capturing the video or photo walk takes roughly the skill already used to shoot condition photographs. The reconstruction step needs nobody on staff to learn photogrammetry software, since it runs from the footage already collected.
What happens to the data afterwards? The scene opens in open formats, .spz, .ply, .splat and .ksplat, so an institution is not committing a decade of institutional memory to one vendor's proprietary viewer. Enterprise accounts can also run a white-labelled, offline viewer that opens scenes without an internet connection, useful for an archive that cannot rely on a link surviving as long as the record needs to.
Turning an existing LiDAR scan into a walkable scene through Captures Studio is not self-serve software: a firm sends the scan data it already holds and receives back a hosted, embeddable scene. It suits an institution with survey data already on file more than one starting from a blank archive.
FAQ
Do we need our own 3D team to start?
No. The site visit itself, walked with a phone, ordinary video camera or 360 camera, is close to the same skill as shooting condition photographs. Reconstruction happens in the cloud, so nobody on staff needs to learn photogrammetry or 3D software to get a finished scene back.
Can an existing photo archive be reconstructed retroactively?
Sometimes, if the photos overlap enough and were shot from varied angles, though a purpose-built capture walk nearly always reconstructs better than photos taken for a different reason at the time. An archive holding only a handful of images per room, the way Bamiyan's did, usually cannot be rebuilt with any confidence.
How often should a site be recaptured?
That depends on what is changing. A facade under active weathering or a structure mid-restoration is worth recapturing every season or after each intervention, while a stable interior might only need recapturing before and after a planned change. Aligning two captures through Multi Splat shows exactly what moved between visits, rather than leaving that comparison to memory.
A folder of photos from a 2021 site visit cannot answer a question raised in 2026 about a wall that no longer exists. A dated 3D scene can, because whoever is answering is not relying on having taken the right picture of the right detail five years earlier. They are standing back inside the room. That is the actual shift here: not sharper pictures, but digital preservation of historic sites built to be revisited by someone who was not in the room the day it was made, asking a question nobody had thought to ask yet.