Procedural Modeling of Buildings

1
Procedural Modelingof Buildings

• Pascal Müller ETH Zürich
• Peter Wonka Arizona State University
• Simon Haegler ETH Zürich
• Andreas Ulmer Virtual Entertainment
  Productions
• Luc Van Gool ETH Zürich / K.U. Leuven

2
Presentation Outline

• IntroductionContribution, examples, motivation
  and background
• CGA ShapeA shape grammar for CG architecture
• Mass modelingAssembling solids and shape
  interaction
• ResultsImplementation, rendered animations and
  outlook

3
Contribution

• CGA Shape, a novel shape grammar for the
  procedural modeling of CG architecture.

• Human-readable rules (notation like L-systems)
• Consistent design of mass models and facades
• Not restricted to axis aligned shapes
• Context sensitive rules to specify shape
  interactions

4
Example Semper-Observatory
5
Example Pompeii
6
Example Modern Architecture
7
Example Mayan Architecture
8
Motivation

• Encoding the structural, spatial and functional
  complexity of buildings for CG
• Efficient creation of detailed building models at
  low cost (resulting in billion polygon cities)
• Many applications in entertainment, simulation,
  archaeology and architecture

9
Related Work

• Shape Grammars (Stiny Gips, 1971)
• Analysis / creation of architectural designs
• Derivation usually done manually
• Procedural Modeling of Cities (Parish Müller,
  SG 2001)
• Large-scale city layout tool
• Crude building generation using L-systems
• Instant Architecture (Wonka et. al., SG 2003)
• Split grammar and control grammar
• Shape-based rules restricted in configuration

10
CGA Shape

• Production process
• Rule-driven modification replacement of shapes
• Iteratively evolve a design by creating more and
  more details
• Sequential application (like Chomsky grammars)

11
Shape Rules

• Notation
• A shape consists of
• Symbol (string)
• Geometry (geometric attributes)
• Oriented bounding box called scope (numeric
  attributes)

id predecessor condition ? successor prob
12
Basic Shape Operations

• Insertion I(objId)
• Transformations T(tx,ty,tz), S(sx,sy,sz),
  Rx(a)..
• Branching ...
• Simple example
• 1 A ? T(0,0,6) S(8,10,18) I(cube)
• T(6,0,0) S(7,13,18) I(cube)
• T(0,0,16) S(8,15,8) I(cylinder)

13
The Subdivision Split

• Example
• 1 facade ? Subdiv(Y,3.5,0.3,3,3,3) floor
  ledge floor floor floor

14
The Subdivision Split (Scaled)

• 1 facade ? Subdiv(Y,3.5,0.3,1r,1r,1r)
  floor ledge floor floor floor
• 2 floor ? Subdiv(X,2,1r,1r,2) B A A
  B

15
The Repeat Split

• Example1 floor ? Repeat(X,2) window
• Create as many window elements of
  approximatesize 2 as there is space

16
The Component Split

• Formatid A ? Comp(type,params) B C ...
  Z
• Example1 solid ? Comp(sidefaces) facade

?
17
Mass Modeling

• Building mass models are most naturally
  constructed as an assembling of simple solids
• Basic building blocks L, H, U, T
• Basic roof typesGable, hip, mansard...

18
Assembling Solids Example

• Starting with the building lot as axiom
• Stochastic variationsof mass models
• Generated with four rules only

19
Shape Interaction Motivation

• Problem The volumes are not aware of each
  other ? unwanted intersections

20
Shape Interaction Motivation

• Solution Test the spatial overlap and align
  elements to important lines

21
Shape Interaction Occlusion

• Example1 tile Shape.occ(all) none ?
  door
• Label can be a shape symbol or a flag (all)
• Return values are either none, partial or
  full
• Shape-geometries or scope-boxes can be queried

22
Occlusion Example

• 6 tile Shape.occ(noparent) none ?
  window
• 7 tile Shape.occ(noparent) part ?
  wall
• 8 tile Shape.occ(noparent) full ? e

23
Occlusion Example

• 6 tile Shape.occ(noparent) none ?
  window
• 7 tile Shape.occ(noparent) part ?
  wall
• 8 tile Shape.occ(noparent) full ? e

24
Shape Interaction Snapping

• Snap lines can be generated manually by the user
• .. and are created automatically on intersections

25
Shape Interaction Snapping

• Subdivision split1 floor ?
  Subdiv(X,1r,1r,1r,1r,1r) B B B B B
• Repeat split1 floor ? Repeat(X,0.2r) B

26
Shape Interaction Snapping

• Subdivision split1 floor ?
  Subdiv(XS,1r,1r,1r,1r,1r) B B B B B
• Repeat split1 floor ? Repeat(XS,0.2r) B

27
Snapping Example

• Main application are office and high-rise
  buildings

28
Implementation

• Integrated in the CityEngine framework (C)
• Computation of a model with 50K polygons takes
  about one second
• User interface
• Text editor for rules
• OpenGL with manydifferent view modes
• GIS viewer
• Interactive editing

29
Results Petronas Tower
30
Results Shape Interaction
31
Results Procedural Pompeii
32
Results Suburbia
33
Future Work

• Pushing the CityEngine towards architecture
• Functional shapes for simulation
• Interior design
• Real-time visualization
• Automatic Level-of-Detail generation
• Derivation on demand
• Extracting semantics automatically
• Employ shape grammars for shape understanding
• Methods from computer vision and machine learning

34
Acknowledgments

• Tijl Vereenooghe (Center for Archaeological
  Sciences, KU Leuven)
• Robbie Müller (3D-Freelancer, Zurich)
• EC IST Network of Excellence EPOCH
• EC IST Project CyberWalk
• NGA grant HM1582-05-1-2004
• Pixar
• Greenworks