Maybe it’s time to do something with all those 3D city models?

Keynote given at 3DGeoInfo 2021 conference (https://3dgeoinfo2021.github.io/) on 2021-10-12

Recording of the talk is also available:
https://youtu.be/EGf3832F1B0

Transcript

1. Hugo Ledoux
   
   
   Delft University of Technology
   Maybe it’s time to do something
   

   with all those 3D city models?
   3D GeoInfo 2021
   
   
   at home in my pyjamas
   
   
   2021-10-12
   

   View Slide

2. solar potential
   shadow analysis
   noise modelling
   energy estimation
   bomb detonation
   wind turbulence
   pollutant tracking
   spatial planning
   • Goal of 3DGeoInfo conferences: push further 3D data
   modelling
   
   
   • So that 3D datasets are usable and interchangeable
   
   
   • is central as the semantic data model
   

   View Slide

3. More and more 3D (open) data from cities
   3
   New York City
   
   
   (LoD2)
   

   View Slide

4. More and more 3D (open) data from cities
   4
   Switzerland: 70M 3D
   objects are now open
   

   View Slide

5. More and more 3D (open) data from cities
   5
   Helsinki: LoD2
   with textures
   

   View Slide

6. More and more 3D (open) data from cities
   6
   56 cities in
   Japan
   

   View Slide

7. More and more 3D (open) data from cities/countries
   7
   41 cities/countries!
   

   View Slide

8. So, mission accomplished!?
   8
   

   View Slide

9. Yeah, not
   
   
   so sure…
   

   View Slide

10. More and more 3D (open) data from cities
    10
    Whole of the Netherlands:
    
    
    10M+ LoD1 buildings + terrain,
    roads, forest, etc.
    • 3D IMGeo: CityGML ADE
    
    
    • Government could produce the data, but very few could
    read it
    
    
    • Were the datasets really used?
    

    View Slide

11. Where is the File/Open CityGML?
    11
    

    View Slide

12. swisstopo downloads of their 3D model
    12
    -Raphaël Bovier, from his presentation at the 3D city modelling Workshop, Nordic
    Cooperation in the
    fi
    elds of Spatial Data and Land Administration (2021-09-29)
    

    View Slide

13. Scientists in other communities are not really involved
    13
    • CFD/simulation community == OBJ/STL
    
    
    • Computer Graphics == OBJ/STL
    
    
    • Other communities have scientists involved in the
    ecosystem
    
    
    • Why so few here?
    

    View Slide

14. Why are the datasets
    seldom used?
    

    View Slide

15. UML shouldn’t be the main goal
    15
    • Many papers have as result UML
    
    
    • Most OGC standards focus on the
    data model
    
    
    • “Let’s model the full reality” →
    increases complexity
    
    
    • We stay at the theoretical level
    
    
    • Users are not central
    

    View Slide

16. Standards don’t ensure usefulness (quite the contrary one might say…)
    16
    

    View Slide

17. Quiz: you need to store 2D polygons, which one you choose?
    17
    GML 3.2.2 specs
    
    
    (427 pages)
    GeoJSON specs
    
    
    (28 pages )
    

    View Slide

18. ADEs: where are the schemas?
    18
    Biljecki F, Kumar K, and Nagel C (2018). CityGML Application Domain Extension
    (ADE): overview of developments. Open Geospatial Data, Software and
    Standards (2018) 3:13
    • Only 18/44 ADEs have XSD schemas available
    
    
    • Some don’t have UML, just a description
    Files with ADEs are way more difficult
    to read than “vanilla” ones
    
    
    ➡ custom code
    

    View Slide

19. Little focus on usable datasets
    19
    • Assumed encoding (GML) can be automatically derived
    from schemas
    
    
    • Datasets are often hierarchical and have annoying quirks
    

    View Slide

20. “GML madness”: how many ways to store a simple square?
    20
    25 different
    ways
    

    View Slide

21. Complex formats usually have fewer software support
    21
    No known JavaScript
    parser for CityGML files!?
    

    View Slide

22. Real-life reactions of students in my course
    22
    When learning about semantic
    3D city models
    Trying to read a CityGML file
    
    
    with Python
    

    View Slide

23. Many of us promote semantic formats but use… OBJ/STL
    23
    • OBJ and STL are super simple
    
    
    • They are used for decades in computer
    graphics, CFD, etc
    
    
    • Many software support them
    
    
    They. Just. Work.
    

    View Slide

24. What can we do
    about it?
    

    View Slide

25. Shouldn’t we apply the Rule of Least Power?
    25
    

    View Slide

26. OK, but concretely here?
    26
    1. Make one usable profile of CityGML that supports most important
    parts (similar to Simple Features)
    
    
    2. Please no GML for the encoding, no developers want this
    
    
    3. Write parsers for the format (Python, JS, C++) so that others can
    just reuse/copy. Maybe an API to manipulate the data?
    
    
    4. Think “like a designer” (how will the product be used?)
    
    
    5. “Eat your own dog food”
    

    View Slide

27. Two projects where we
    applied those principles
    

    View Slide

28. CityJSON v1.0
    28
    1. "Building"
    
    
    2. "BuildingPart"
    
    
    3. "BuildingInstallation"
    
    
    4. "Road"
    
    
    5. "Railway"
    
    
    6. "TransportSquare"
    
    
    7. "TINRelief"
    
    
    8. "WaterBody"
    
    
    9. "PlantCover"
    
    
    10. "SolitaryVegetationObject"
    
    
    11. "LandUse"
    
    
    12. "CityFurniture"
    
    
    13. "GenericCityObject"
    
    
    14. "Bridge"
    
    
    15. "BridgePart"
    
    
    16. "BridgeInstallation"
    
    
    17. "BridgeConstructionElement"
    
    
    18. "Tunnel"
    
    
    19. "TunnelPart"
    
    
    20. “TunnelInstallation”
    
    
    21. “CityObjectGroup”
    
    
    • community standard
    
    
    • all modules of CityGML v2.0 are mapped 💪
    
    
    • Subset of CityGML (~97% of features)
    
    
    • software for full conversion CityGML <-> CityJSON
    
    
    • several software support it
    

    View Slide

29. Same information as CityGML, but in JSON format
    29
    {
    
    
    "type": “CityJSON",
    
    
    "version": “1.0”,
    
    
    "metadata": {
    
    
    "referenceSystem": "urn:ogc:def:crs:EPSG::7415",
    
    
    },
    
    
    "CityObjects": {
    
    
    "id-1": {
    
    
    "type": "Building",
    
    
    "attributes": {
    
    
    "measuredHeight": 22.3,
    
    
    "roofType": "gable",
    
    
    "owner": “Elvis Presley"
    
    
    },
    
    
    "geometry": [
    
    
    {
    
    
    "type": "MultiSurface",
    
    
    "boundaries": [
    
    
    [[0, 3, 2, 1]], [[4, 5, 6, 7]], [[0, 1, 5, 4]]
    
    
    ]
    
    
    }
    
    
    ]
    
    
    }
    
    
    },
    
    
    "vertices": [
    
    
    [23.1, 2321.2, 11.0],
    
    
    [111.1, 321.1, 12.0],
    
    
    ...
    
    
    ],
    
    
    "appearance": {
    
    
    "materials": [],
    
    
    "textures":[],
    
    
    "vertices-texture": []
    
    
    }
    
    
    }
    
    
    human-readable file
    
    
    computers prefer this over XML
    
    
    ~6X compacter than CityGML
    

    View Slide

30. Python parsing is very easy
    30
    

    View Slide

31. CityJSON can be easily parsed with JavaScript
    31
    Developed by MSc
    students
    

    View Slide

32. CityJSON is “ready” for machine learning
    32
    new!
    

    View Slide

33. 3D BAG: all 10M+ buildings in the Netherlands
    33
    https://3dbag.nl
    

    View Slide

34. LoD 2.2
    

    reconstruction
    3D BAG is based on open datasets
    34
    Point cloud
    (AHN3)
    Footprint (BAG)
    +
    X 10,000,000
    
    
    100% automatic
    

    View Slide

35. 3 different LoDs
    35
    

    View Slide

36. 3D BAG is tiled into *small* and *manageable* datasets
    36
    BAG polygons
    Quadtree
    
    
    Max 3500 buildings per tile
    

    View Slide

37. Download is simple and fast
    37
    Stats last 2 weeks
    
    
    (~32000 downloads)
    
    
    CityJSON 56%
    
    
    OBJ 23%
    
    
    GPKG 21%
    
    
    PostgreSQL <1%
    

    View Slide

38. Very careful about geometric validity
    38
    ~98% valid
    

    View Slide

39. Amsterdam put our data in their Unity online platform within days
    39
    

    View Slide

40. People outside our field used the data
    40
    Historical cadastre Shadow analysis
    

    View Slide

41. We reached non-geo people
    41
    

    View Slide

42. thank you.
    h.ledoux@tudelft.nl
    
    
    3d.bk.tudelft.nl/hledoux
    Hugo Ledoux
    https://cityjson.org
    https://3dbag.nl
    

    View Slide