Cities at full fidelity

Reconstructing a real city in three dimensions, building by building, from LiDAR, third‑party 3D map data, geodata and on‑site photography — one faithful model that the same geometry lets serve entertainment, defense, energy, policing and simulation.

A full‑fidelity city model is a copy of a real place in which every building that exists is present and recognisable — in the right location, at the right height, with its own roof, facade and materials — so that a person who knows the street can name the buildings. No one source describes a city completely: LiDAR fixes shape and height but carries no colour, third‑party 3D map data covers the whole footprint but is uneven up close, photographs carry true materials but no geometry. The model is assembled by letting each source supply what it measures best.

Fig. 1The real place. Aerial reference of the Scheveningen seafront — the pier, the Ferris wheel, the promenade and the line of the boulevard. The reconstruction is measured against survey of the actual location, so the model can be checked building by building against the city it copies.

The sources

The model is built in layers, coarse to fine, each layer constraining the next.

LiDAR survey. Aerial and ground‑level laser scanning return a point cloud: exact heights, roof shapes, street profiles and the relative positions of structures, accurate to the order of a few centimetres. This is the skeleton — geometry without appearance.
Third‑party 3D map data. Photogrammetric mesh and national 3D building datasets give whole‑city coverage of footprints and massing, filling in everything the dedicated survey did not target so that no district is left empty.
Geodata. Open building footprints, the road network and land‑use polygons drive procedural generation of the background fabric — the ordinary streets that must be present and correct but do not each warrant hand work.
On‑site photography. Thousands of ground‑level reference photographs record the true facades, signage, shopfronts and materials. The Scheveningen build alone used 9,674 catalogued photographs; a purpose‑built spatial index ties every photograph to the building and elevation it documents, so a modeller dressing a facade is looking at that exact wall.
Normal maps. Surface relief measured from photographs — brick courses, stone rustication, cornices, ornament — is baked into normal maps. Flat or low‑polygon geometry then catches light as if it carried the relief, giving the close‑range texture of masonry without the polygon cost of modelling every brick.
Manual hero modelling. Landmarks and characterful buildings are modelled by hand against the references. Amsterdam took more than 250 buildings to this level of detail; Scheveningen, more than 200, alongside upward of a hundred bespoke objects — railings, lamps, kiosks, street furniture.

Fidelity is therefore tiered without being uneven. Hero buildings carry full geometry and photographed textures; the surrounding fabric is generated from footprints and heights and dressed with tiled materials and normal maps. Every building is present; effort is spent where the eye spends its attention.

Fig. 2The city at scale. An aerial pass over the assembled Amsterdam model, the waterfront and densely built centre resolved together. The continuous coverage — not a few hero buildings on an empty plate — is what the layered sourcing exists to produce.

Building by building

Under the textures the model is ordinary, explicit geometry: each building is a discrete object with its own mesh, placed at surveyed coordinates. There is no procedural sleight of hand standing in for a real structure. Editing one building does not disturb its neighbours, and any single structure can be inspected, corrected or replaced on its own.

Fig. 3Geometry, before appearance. A cluster of central‑Amsterdam landmarks shown as wireframe in the modelling tool — towers, gables and roofs as explicit per‑building meshes. This is the layer LiDAR and survey fix; texture and relief are added on top of it.

Appearance is then applied from the photographic record. Facades carry their own photographed materials; normal maps give brick and stone their relief; proportions follow the references rather than a template. The result reads correctly not only in a flythrough but at the range of a person standing on the pavement, where stand‑in textures and repeated facades would immediately betray themselves.

Fig. 4Street level. The reconstructed Damrak and Dam square — individual canal houses, the brick and stonework of named buildings, shopfronts and a passing tram, each facade carrying its own photographed material and surface relief rather than a tiled stand‑in.

The city as one model

The buildings are not isolated assets but a single continuous environment: streets connect, scales are consistent, and the model can be entered and traversed without seams. Interiors of major structures are built where they matter, so the reconstruction is navigable rather than only viewable from above. Amsterdam covers roughly 20 km² of the centre as one coherent space.

Fig. 5Inside the model. The interior of a reconstructed Amsterdam station — vaulted roof, platforms and rolling stock — showing that the city is built as an enterable space, not a hollow set of exterior shells.

What the model is for

Once a city has been reconstructed to this standard, the model is the asset, and it is neutral as to use. The same geometry, heights, materials and surface relief that make a convincing ride also make a usable simulation substrate. Building the city to full fidelity once, rather than to the minimum a single product needs, is what allows it to be reused across domains that otherwise each commission their own coarse model:

Entertainment. VR rides and location‑based attractions, virtual production and film backdrops, and games set in a recognisable real city. Both builds here were delivered as ride attractions.
Defense and security. Mission rehearsal in a faithful urban environment; line‑of‑sight, cover and sightline analysis against true building geometry; radio‑frequency and sensor occlusion studies; route and approach planning; training that takes place in the actual streets a unit will operate in rather than a generic stand‑in.
Carbon and energy. Per‑roof solar‑irradiance and photovoltaic‑yield estimation; overshadowing and daylight studies; building‑stock energy and retrofit modelling across a district; urban heat‑island analysis; first‑order embodied‑material estimates read from building volumes.
Policing and public safety. Incident and scene reconstruction with correct sightlines and distances; camera‑placement and coverage optimisation; crowd‑flow and evacuation simulation for events; security planning along a parade or race route before the day.
Urban planning. Visual‑impact and viewshed assessment of a proposed building; right‑to‑light and overshadowing disputes settled against real neighbours; pedestrian‑level wind and microclimate studies around new towers; zoning and massing shown in context.
Telecommunications. Antenna siting and 5G or millimetre‑wave propagation modelling, where coverage depends on diffraction and reflection off the true shapes of real facades.
Water and climate resilience. Pluvial and fluvial flood routing over the actual ground surface and building footprints, and storm‑surge modelling on an exposed coast — directly relevant to a seafront such as Scheveningen.
Autonomous systems. A digital twin in which self‑driving vehicles and delivery drones are tested against real geometry, and from which labelled synthetic sensor data — camera, LiDAR — is generated with exact ground truth for training perception models.

The common thread is that none of these uses tolerates a stand‑in. A sightline, a shadow, a radio path or a flood front is wrong the moment the building that casts it is wrong. Fidelity is not a finish applied at the end for the entertainment case; it is the property that makes every other use possible.

Two cities, delivered as rides

The method is described above in the abstract; in practice it has been carried out twice, and in both cases the first delivery was a VR ride — the application that most demands close‑range fidelity at speed, and so the most exacting first customer for the model.

Central Amsterdam. Roughly 20 km² of the centre, with more than 250 buildings modelled in detail and the surrounding districts generated from open geodata, dressed from a scouted photographic survey and indexed by a purpose‑built location service. Delivered as a roller‑coaster ride some 16 km long through the city’s landmarks.

Scheveningen. The seafront and pier, with more than 200 real buildings and over a hundred bespoke objects, dressed from 9,674 catalogued on‑site photographs managed through a dedicated geographic information system. Delivered as a roughly 5 km ride along the coast on a motion platform driven by the cart’s telemetry, synchronised to the virtual movement so the body feels what the eyes see.

Fig. 6The first delivery. The Amsterdam reconstruction in motion as a roller‑coaster ride, the cart passing landmark facades at speed. The ride is one use of the model; the same reconstruction stands behind every other application listed above.

In short

A real city can be rebuilt in three dimensions to a standard where every building is its own, recognisable to someone who knows the street, by combining LiDAR for shape, third‑party 3D map data and geodata for coverage, scouted photography for material, and normal maps for relief — hero buildings by hand, the rest generated, all present. The two cities here reached the public as VR rides, but the ride is only the most demanding early use of the asset. The asset is the city: a faithful, navigable, building‑by‑building model that the same geometry lets serve defense, energy, policing, planning and simulation as readily as entertainment.

AAIC built the city reconstruction; the VR ride attractions were built on it by JetXR. Footage is from the two completed city builds as delivered for those attractions; Fig. 1 is aerial reference of the actual Scheveningen seafront. Figures are silent; each caption is the clip’s full description.