The full chain · one area, every step

From a raw pixel to an operational 4D twin.

Follow a single intersection through our entire pipeline — the same patch of Earth, reconstructed step by step from a satellite pass all the way to a drivable, simulation-ready 4D digital twin. The exact workflow we scale to whole cities.

Satellite ortho of the area
Step 01 · Capture

Tri-stereo from space, multi-head from the air

We start with Level-1 satellite imagery — monocular, stereo and tri-stereo, panchromatic and multispectral — then add five-head aerial capture at centimetre resolution, with full camera parameters for every frame.

Tri-stereo satellite · 5-head aerial (8.6 cm) · new tasking or archive
Super-resolution: native vs enhanced
Step 02 · Correct & enhance

TrueHD super-resolution

Atmospheric & geometric correction and pansharpening prepare the imagery; then our proprietary AI super-resolution lifts native 30 cm satellite to 7.5 cm — and 8.6 cm airborne to 2 cm — without a new flight. Left native, right enhanced.

Level-1 → Level-2 · pansharpening · 4–32× super-resolution · SAR–optical cloud removal
Dense point cloud and 3D mesh
Step 03 · Reconstruct

Ortho, DSM/DTM, point cloud & mesh

Bundle adjustment and dense matching produce true-ortho, surface and terrain models, a dense 3D point cloud and a textured mesh — everything measurable and georeferenced.

True-ortho · DSM · DTM · LAS/LAZ point cloud · textured 3D mesh
AI HD vector map
Step 04 · Map

Pixels become an HD vector map

15+ proprietary AI algorithms extract buildings, lanes, sidewalks, furniture and land use as clean 2D / 2.5D / 3D vectors — SD and HD — exported as Shapefile, GeoJSON, KML and lane-accurate ASAM OpenDRIVE (the road-network standard driving simulators read natively).

Semantic · instance · panoptic segmentation → vectors · ASAM OpenDRIVE HD map
Earlier epoch Later epoch Before After
Step 05 · The 4th dimension: time

Detect what changed — automatically

Drag the slider: between two acquisitions, a new development appears. Our AI flags it at building level — so the twin tracks reality instead of freezing at capture time.

Change detection · building & road level · near-real-time update
Textured LoD3 CityGML model
Step 06 · Model

Semantic 3D — LoD1 to LoD3.3

The mesh becomes clean, semantic 3D buildings with textured façades — modelled to OGC CityGML and fused with buildingSMART IFC (BIM), then exported to every open format your tools speak.

OGC CityGML · CityJSON · OBJ · FBX · OpenUSD · glTF · CityGML + IFC fusion
Step 07 · Simulate

A drivable, photoreal twin

The same ground truth becomes a raytraced Unreal-engine twin with Nanite & Lumen — driving, flight, weather, seasons, time of day, traffic, flood and earthquake, all simulated in-engine, with VR/XR.

Unreal Engine · raytracing · Nanite · Lumen · VR / XR · physics & weather
ChatGIS — ask your map in plain language
Step 08 · Ask

An AI agent that understands your twin

ChatGIS answers questions about your maps, CityGML and the twin in plain language; smart NPCs and a MetaHuman guide bring the scene to life — and it all lives in a navigable geospatial metaverse.

ChatGIS / ChatMap · AI agents & smart NPCs · MetaHuman · Web3 metaverse
Now onboarding pilot customers

Let's model your world.

Send us your area of interest and what you need to see. We'll return a sample over your own location and walk you through the platform.

or email hello@twinsphere.example.com