City Walk/WissenZum StadtspaziergangViewer
ImpressumDatenschutz
Leitfaden
  • So funktioniert der Stadtspaziergang
  • Woher die Daten kommen
  • Quellen nach Stadt
  • Wie der Viewer zu einer Angabe kommt
  • Der Weg der Daten
  • Bedienung
  • Glossar
Guide (English)
  • How the city walker works
  • Where the data comes from
  • Sources by city
  • How the viewer comes by what it shows
  • From download to browser
  • Using the viewer
  • Glossary
Entwicklung (English)
  • docs/ — the project knowledge base
  • Rendering — how the data becomes pixels
  • Data pipeline — the bakes, the build step and the artifact contracts
  • Architecture decision records
  • Data → feature provenance
  • Transformation catalog
  • Portability — rendering any location, with whatever data it has
  • Writing a walk
  • Routines: the twin's scheduled agents
  • Implementation plans, backlog and audit history
  • What the data streams carry that the walker does not use yet
Contents
Leitfaden
  • So funktioniert der Stadtspaziergang
  • Woher die Daten kommen
  • Quellen nach Stadt
  • Wie der Viewer zu einer Angabe kommt
  • Der Weg der Daten
  • Bedienung
  • Glossar
Guide (English)
  • How the city walker works
  • Where the data comes from
  • Sources by city
  • How the viewer comes by what it shows
  • From download to browser
  • Using the viewer
  • Glossary
Entwicklung (English)
  • docs/ — the project knowledge base
  • Rendering — how the data becomes pixels
  • Data pipeline — the bakes, the build step and the artifact contracts
  • Architecture decision records
  • Data → feature provenance
  • Transformation catalog
  • Portability — rendering any location, with whatever data it has
  • Writing a walk
  • Routines: the twin's scheduled agents
  • Implementation plans, backlog and audit history
  • What the data streams carry that the walker does not use yet
Guide (English) · 4 / 7

How the viewer comes by what it shows

Deutsch: Wie der Viewer zu einer Angabe kommt

Everything in the viewer comes from open data, but not everything is in the data as it is shown. Some of it is taken over unchanged, some worked out with a formula, some only found by pattern recognition, and some the viewer assumes because nothing at all is known of the thing. This page explains the four badges the viewer marks that with: in the detail view of a building, tree or bridge (under Daten) and here in the knowledge base.

The four badges

BadgeMeaningExample
❝takenAs the source publishes it, shown unchanged.Footprint and eave height from the 3D building model, a tree's species from the street-tree register, an address from OpenStreetMap
=computedBy a fixed formula over measured values. The same data always give the same result.A crown's height as surface minus ground, a meadow's green from the aerial photo's infrared band, a roof's colour as the median of its pixels
◎detectedBy pattern recognition: thresholds, matching two sources or image analysis. It can be wrong.A tree where the surface shows a crown's peak; a shed the laser scan sees; a shop whose point is assigned to a building
≈assumedNothing is known of this thing: a default, an average or a design choice.The shape of a crown, brick or plaster as in the neighbourhood, the traffic at the scene's hour from the typical day

The first two are deterministic and have a solid border: whoever takes the same data comes to the same result, and the result is as good as the measurement. The last two are inferred and have a dashed border: a rule, a match or a design decision has added something that is not measured. The glyph before the word (❝, =, ◎, ≈) tells the four apart without colour.

In the detail view the badges stand at every line under Daten. The facts at the top of the card (height, roof, ground floor …) carry a small ◎ or ≈ after the value only where it is inferred; a value without a mark is taken or computed.

What pattern recognition means here

No part of the viewer is a trained neural network. "Detected" here almost always means: a rule with thresholds decides what something is, or two sources are matched against each other. The rules are open and reproducible, but they can be wrong where the world does not fit the rule.

  • Trees without a register: in wood, copse and park the highest point of the surface between 3 and 45 m above the ground becomes a tree, one per cell of about 7 m. A bridge pylon in a park would be as tall as a crown, so bridges are left out.
  • Small buildings: what stands 2–6.5 m high in the laser scan, outside the 3D building model, and sends back a single echo per light pulse (foliage splits a pulse, a roof does not) becomes a shed or a garden house. In the aerial photo's shadow only a clean box counts.
  • Roofs, chimneys, towers, dormers: where the surface model misses the building model's roof by more than 2 m on at least 40 % of the roof, the viewer rebuilds the roof from the measurement — on a fifth of it already where the surface is level there and the model's roof slopes where the surface does not (a low hipped roof drawn over a flat department store). Its slopes are drawn as an ordinary pitched roof: each house of a row and each bend of a wing on its own plane, eaves and ridges level, not dented by its dormers and chimneys, ridges, hips and valleys where two planes cross, and outlines running along the building's walls and meeting at corners rather than curves; only a landmark keeps the measured surface itself, and a slope that fits no roof form stays flat. A tall height above the building model becomes a chimney or tower only where OpenStreetMap names one. The roof's small things come from the surface model alone: every bump over a pitched roof is a dormer, however small and however close to the next; a small one at the ridge, or on a flat roof, is a house chimney — unless it lies along a ridge or valley the building model drew a little off, or stands on a flat roof in a row of others (masts on a platform roof). Where the whole building model lies evenly higher than the measurement (the Taschenbergpalais by 1.2 m), it is set onto it first; a part with steep roof faces drawn off in plan (the palais' north wing by about a metre) is read a second time moved onto the measurement. A steep roof face a shallower upper roof goes on from is a mansard, even where the building model does not call it one; its dormers may stand on up to 80°. A building's dormers facing the same way at one height form a row: set straight, of one size and form, at one height, on one grid of equal spacing along the whole row — over the window axes below where the street photos measured them, else the spacing the dormers agree on best. A single empty place between two takes one like its neighbour, unless a chimney stands there. Last, each dormer takes its height from the building model's exact roof face, not from the 1 m grid, so it sits on the roof that is drawn.
  • Bridge superstructure: what stands at least 3 m above the deck along the bridge's axis, for at least 25 m, becomes an arch or a truss.
  • Assignments: a shop from OpenStreetMap belongs to the building whose footprint holds its point, else to the nearest within a few metres; a landmark from Wikidata to the building parts its outline covers. A bridge is no landmark (it belongs to the bridges) unless Wikidata also knows it as a building.
  • Someone else's recognition: lamps Mapillary found in street photos were detected there by Mapillary's own image recognition — trained networks whose hits the viewer takes over.

How sure is a detection?

Where a rule was checked against an independent source, the result stands with the dataset in Where the data comes from. Two examples of what the numbers do: about a third of the hedges and shrubs only the laser scan finds were the rims of tree crowns, so of them the viewer shows only the long, slim, green ones clear of trees. And Munich's aerial photos have no infrared band; the green from the visible colours agrees only moderately with the real vegetation index (r ≈ 0.7), which is why Munich is yellow there in Sources by city.

Everything at a glance

What is drawnFromHow
Ground heightsTerrain model DGM1❝taken
The ground as a triangle mesh, to 15 cmTerrain model DGM1=computed
Walls and steps, sharpened in the groundDGM1 and OpenStreetMap◎detected
Wall heights where none is mappedDOM1=computed
What an area is (road, water, meadow …)Basis-DLM (Hamburg, Berlin: OpenStreetMap)❝taken
The colour per areaa palette of the viewer's≈assumed
Green of meadows and foliageaerial photo, infrared (Munich: visible colours)=computed
Footprint, height and roof shape of buildings3D building model LoD2❝taken
Roofs where the model misses themSurface model DOM1◎detected
Chimneys, towers, newer buildingsDOM1, confirmed by OpenStreetMap◎detected
Dormers and house chimneysDOM1 over the building model's roof◎detected
Platform roofs, carports, sheltersDOM1, confirmed by OpenStreetMap◎detected
Sheds and garden housesLaser scan◎detected
Roof colourAerial photo=computed
Facade colour and material, where mappedOpenStreetMap❝taken
Glass facades where nothing is mappedstreet photos (Mapillary), read by an image model◎detected
Brick or plaster where nothing is mappedOpenStreetMap, the neighbourhood≈assumed
Tint per building, storey bandsthe viewer≈assumed
Front doorsOpenStreetMap, the entrances❝taken
Passages through buildings, their widthOpenStreetMap, the passages (tunnel=building_passage)❝taken
A passage's height where none is mapped, its rectangular openingthe viewer, from the ground storey≈assumed
A mural's pictureWikimedia Commons, a photo that shows the whole mural straight on❝taken
A mural's drawing: its pen lines, its wash and its inkcomputed from that photo (its light flattened, lines where its lightness turns, its colours sorted into five fields and lifted into the pastel)=computed
A mural's size and its place on the wallOpenStreetMap (else Wikidata), on the nearest wall of the building model=computed
How high a mural starts where OpenStreetMap does not saythe viewer, a storey up≈assumed
Shop or restaurant on the ground floorOpenStreetMap, assigned to the building◎detected
LandmarksWikidata, assigned to the building parts◎detected
An asked landmark's facts under its name in the scene (built, architect, rebuilt, height)Wikidata❝taken
Several buildings of the city model as one (the Residenzschloss with the Hausmannsturm)OpenStreetMap, one named building outline over several buildings, across tile edges too◎detected
Street trees: place, species, sizeStreet-tree register❝taken
Trees without a register: placeDOM1 or laser scan◎detected
Trees without a register: heightDOM1 or laser scan=computed
Crown shape, leaf-out, autumn colourthe viewer, per genus≈assumed
Hedges: courseBasis-DLM, OpenStreetMap or the laser scan❝taken ◎detected
Hedges: heightLaser scan (without one: the surface model DOM1; a hedge it cannot see is left out; else a default)=computed ≈assumed
Bridges: outline and kindBasis-DLM (Hamburg, Berlin: OpenStreetMap)❝taken
Bridges: height of the roadwayDOM1=computed
Bridges: arch or truss above the deckDOM1◎detected
Bridges: piersthe viewer, clear of the fairway≈assumed
Tracks and their courseBasis-DLM or OpenStreetMap❝taken
Which level a track runs on (ground, deck, cut)from ground and decks along the whole line◎detected
A monorail's beam: its height and supportsfrom the ground along the whole line, a car's clearance; the supports the viewer's=computed ≈assumed
Lamps, benches, street furniture: placeOpenStreetMap❝taken
Lamps from street photosMapillary◎detected
Their form; where a bench without a direction facesthe viewer≈assumed
Lamps: post height; which way a tall mast's arm reachesDOM1, per street; OpenStreetMap, towards the nearest carriageway=computed ≈assumed
Monuments: place and nameBasis-DLM, OpenStreetMap❝taken
Monuments: form, where measuredDOM1=computed
Playground pieces where none is mappedDOM1◎detected
Sports grounds: outline, sport, surfaceOpenStreetMap❝taken
Their linesthe sport's standard dimensions≈assumed
Orchard trees: place and heightDOM1◎detected =computed
Allotment plots; orchard trees without DOM1 or an entrythe viewer≈assumed
Water outlineBasis-DLM❝taken
Water level, one per water bodyDGM1=computed
Water colourDOP=computed
Beaches and bars inside the waterDGM1 + DOP◎detected
Foam at weirsOpenStreetMap, DGM1❝taken ≈assumed
Waves, fog, haze, paper grainthe viewer≈assumed
Sky light and horizonTerrain and building model; beyond the site Copernicus GLO-30=computed
Sun position and shadows at the scene's timethe viewer=computed
Data layer motor traffic: daily countsTraffic counts❝taken
Data layer motor traffic: value at the hourthe typical day≈assumed
Data layers cycling, tramslive counters, timetable❝taken

The developer documentation carries the same badges on every entry of the transformation catalogue.

What the badges do not say

A badge describes what the viewer does with the data, not how the source itself was made. The 3D building model is partly modelled automatically by the survey office from laser scan and footprints; OpenStreetMap is mapped by volunteers and is as complete as the street someone mapped. How good a source itself is, Where the data comes from and Sources by city describe. And what is taken can still be out of date: the datasets have different editions, which the detail view names at every line.

Previous
Sources by city
Next
From download to browser
Source on GitHub: docs/guide/en/methods.mdDeutsche Fassung →
On this page
  • The four badges
  • What pattern recognition means here
  • How sure is a detection?
  • Everything at a glance
  • What the badges do not say