Hierarchical Simplification of City Models to Maintain Urban Legibility

1
Hierarchical Simplification of City Models to
Maintain Urban Legibility

• Remco Chang
• Thomas Butkiewicz
• Caroline Ziemkiewicz

Zachary Wartell Nancy Pollard William Ribarsky
University of North Carolina Charlotte
Carnegie Mellon University
2
Research DirectionKnowledge Visualization
Minimize Resources, Maximize Information

• Rendering Effective Route Maps Improving
  Usability Through Generalization Agrawala and
  Stolte 2001
• Wire-transfer fraud detection and visualization
  project with Bank of America by clustering
  millions of accounts

3
Knowledge Visualization
?
City of Xinxiang, China 30k buildings 280k
polygons
4
Urban Legibility

• Kevin Lynch in The Image of the City (1960, MIT
  Press)
• categorized Urban Legibility into
• Paths highways, railroads, canals
• Edges shorelines, boundaries
• Districts industrial, residential
• Nodes Time Square in NYC
• Landmarks Empire State building
• ...the ease with which its parts may be
  recognized and can be organized into a coherent
  pattern.

5
Using Urban Legibility in Computer
Science

• Dalton 2002 history of the use of Urban
  Legibility in computer science
• Empirical justification of Urban Legibility
• Paths, Edges, and Districts are very important to
  human navigation (Darken and Sibert 1996,
  Magliano, Cohen et al. 1995)
• Landmarks do not always improve navigation
  (Tlauka and Wilson 1994 Magliano, Cohen et al.
  1995 Steck and Mallot 2000)
• Using Urban Legibility in Graphics and
  Visualization
• Ingram and Benford navigating abstract data
  spaces

6
Why Urban Legibility?
Visually different, but quantitatively similar
7
Our Goal

• Create simplified urban models that retain the
  image of the city from any view angles and
  distances.

Demo!
Original Model
45 polygons
18 polygons
8
Related Works in
Urban Flythrough

• Visibility and Occlusion
• Wonka et al. 2000 and Schaufler et al. 2000
• Imposters
• Marciel and Shirley 1995, Sillion et al.
  1997, and Shalabi 1998
• Procedurally Generated Buildings
• Wonka et al. 2003 Mueller et al. 2006
• Popping
• Google Earth 2005

9
Algorithms to Preserve Legibility

• Identify and preserve Paths and Edges
• Create logical Districts and Nodes
• Simplify model while preserving Paths, Edges,
  Districts, and Nodes
• Hierarchically apply appropriate amount of
  texture
• Highlight Landmarks and choose models to render

10
Identifying and Preserving Paths
and Edges (1)
bc
de
def
abc

• Single-Link Clustering
• Iteratively groups the closest clusters
  together based on Euclidean distance
• produces a binary tree (dendrogram)
• Penalizes large clusters to create a more
  balanced tree

bcdef
abcdef
11
Identifying and Preserving Paths
and Edges (2)
12
Creating logical
Districts and Nodes (1)
13
Creating logical
Districts and Nodes (2)

• Merge two clusters by combining foot prints

• (c) Shows the resulting Merged Hull
• (d) Shows the Negative Space created
  from the merger

14
Simplification while preserving Paths,
Edges, Nodes, and Districts (1)
6000 edges
1000 edges
Demo!
15
Simplification while preserving Paths,
Edges, Nodes, and Districts (2)

• After the polylines have been simplified
• Create Cluster Meshes
• The height of the Cluster Mesh is the median
  height of all buildings in the cluster

16
Hierarchical Textures (1)

• Each Cluster Mesh contains 6 textures
• 1 Side Texture
• 1 top-down view of the roof texture
• 4 roof textures from 4 angles
  (south, west,
  east, north)

Side texture
17
Hierarchical Textures (2)

• Clusters are divided into bins
• Each bin creates a texture atlas of the same
  dimension

n/2
n/4
n/8

.
18
Runtime Levels of Detail

• Starting with the root node of the dendrogram
• Approximate the Negative Space as a 3D box
  shown as the red box
• Project the visible sides of the box onto screen
  space
• Reject if the number of pixel is above a
  user-defined tolerance

19
Landmark and Skyline Preservation (1)
Original Skyline
20
Landmark and Skyline Preservation (2)

• Project a user-defined pixel tolerance (a) onto
  the top of each cluster
• If any building within that cluster is taller
  than the projected tolerance (shown in green), it
  is drawn separately from the cluster mesh.

21
Results
22
Conclusion

• Per-pixel error is not indicative of the visual
  quality of simplified urban models
• Expert knowledge from city planning helps extract
  visually salient features
• Urban Legibility allows efficient and intuitive
  simplification of urban models

23
Limitations and Future Work

• The rendering engine
• Currently not using display lists, vertex arrays,
  frustum culling etc.
• The pre-processing steps
• Clustering, merging, and simplification are all
  O(n3) processes
• Incorporating detailed models
• Manually created
• Procedurally created
• Allow user interactions
• Let experts manually select Districts
• Deep hierarchy tree
• Binary trees are deeper than quad trees

24
Special Thanks

• This work is supported by the Department of
  Defense's MURI program, administered by the Army
  Research Office
• Eric Sauda and Jose Gomez from the Architecture
  Department for lessons on Urban Legibility
• Sonia Leach for inspiration on the clustering
  algorithm
• Evan Suma for making this software run in stereo
• Danny Fregosi and Hunter Hale for data preparation

25
Questions and Comments?

• http//www.viscenter.uncc.edu