The document discusses CollaGen, a framework for collaborative cartographic generalization. It describes how multiple automatic generalization processes can work together by being orchestrated. Generalization processes are applied to different map spaces. A conductor agent chooses the most appropriate process for a given space based on criteria like pre-conditions and post-conditions. Process application is monitored online and side effects are managed. The goal is to leverage the strengths of different processes to generalize complete maps.

1. 1
CollaGen: Collaboration between Automatic
Cartographic Generalisation Processes
Guillaume Touya, Cécile Duchêne
COGIT Lab
IGN France
International Cartographic Conference 2011

2. 2Background
Initial data Map
Generalisation
After
generalisation
Symbolised for
1:50 000
Cartographic Generalisation
Before
generalisation

3. 3Background
Many automatic cartographic generalisation processes
But…
Adapted to a specific landscape
AGENT [Ruas 99],
[Barrault et al 01]
Adapted to urban areas

4. 4Background
Many automatic cartographic generalisation processes
But…
Adapted to a specific landscape
AGENT [Ruas 99],
[Barrault et al 01]
Adapted to urban areas
Adapted to Land Use
"MIP aggregation" [Haunert 08]
Adapted to a specific theme

5. 5Background
Many automatic cartographic generalisation processes
But…
Adapted to a specific conflict
Adapted to road overlapping
conflicts
Elastic Beams [Bader 01]

6. 6Background
Many automatic cartographic generalisation processes
But…
Adapted to a specific conflict
Adapted to road overlapping
conflicts
Elastic Beams [Bader 01]
Adapted to a mix of landscape, theme and conflict
Adapted to proximities in cities
Simulated annealing [Ware et al 03]

7. 7Background
• Great diversity of generalisation processes
• No process able to solve all generalisation problems
• Very different ways to parameterise a process
• How to benefit from the diversity to generalise complete
maps ?

8. 8Objectives
Process 2
Process 3
Process 1
• Make generalisation processes collaborate
Process 1 on
Process 2 on
Process 3 on
Process 4 on
Process 4

9. 9Objectives
Process 2
Process 3
Process 1
• Make generalisation processes collaborate
Process 1 on
Process 2 on
Process 3 on
Collaborative Generalisation
Process 4 on
Process 4

10. 10
Interoperability Management
Collaborative Generalisation Requirements
(Touya et al, 2010)

11. 11
Interoperability Management
Constraints
model
standard
format
Collaborative Generalisation Requirements
(Touya et al, 2010)

12. 12
Interoperability Management
Constraints
model
Translation
mechanisms
standard
format
Collaborative Generalisation Requirements
(Touya et al, 2010)

13. 13
Interoperability Management
Constraints
model
Ontology
Translation
mechanisms
standard
format
Collaborative Generalisation Requirements
(Touya et al, 2010)

14. 14
Collaborative Generalisation Requirements
(Touya, 2010)
Urban space
Rurban space
Automatically build spaces relevant for generalisation

15. 15Presentation outline
• Background and Objectives
• CollaGen: Collaboration by Orchestration
• Choice of a Space
• Choice of a Process
• Online Observation
• Step-by-step Evaluation
• Side Effects Management
• Results
• Conclusion and Future Plans

16. 16CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
observator
register
process
agents
space
agents
conductor

17. 17CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
• creates spaces
• subscribes to the
register with metadata
observator
register
process
agents
space
agents
conductor

18. 18CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
• chooses a space
observator
register
process
agents
space
agents
conductor

19. 19CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
• evaluates itself
• requests to the
register
observator
register
process
agents
space
agents
conductor

20. 20CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
• chooses the
best process
observator
register
process
agents
space
agents
conductor

21. 21CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
• parameterises
• applies on
space
observator
register
process
agents
space
agents
conductor
• observes if
generalisation is
OK
• reacts if not OK

22. 22CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
• validates state
• re-evaluates
observator
register
process
agents
space
agents
conductor

23. 23CollaGen: Collaboration by Orchestration
AGENT
CartACom
Beams
GAEL
observator
register
process
agents
space
agents
conductor
• detects side effects
• corrects side effects

24. 24CollaGen: Collaboration by Orchestration
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

25. 25Choice of a space
Orchestration infered on partial knowledge

26. 26Choice of a space
Orchestration infered on partial knowledge
Schema
Generalisation
Geometry
Collapses
Selection
Cartographic
Generalisation
Graphic
Generalisation
5 steps and rules to chain them

27. 27Choice of a space
Orchestration infered on partial knowledge
« road network rural spaces »
Schema
Generalisation
Geometry
Collapses
Selection
Cartographic
Generalisation
Graphic
Generalisation
5 steps and rules to chain them
Inside a step: some rules to
choose spaces

28. 28CollaGen: Collaboration by Orchestration
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

29. 29Choice of a process
Monitor for building min area
Monitor for road shape
Monitor for proximity
localise
Constraints monitors related to each constraint
•building min area
• maintain road
shape
• min proximity b/w
roads and buildings

30. 30Choice of a process
Monitor for building min area
Monitor for road shape
Monitor for proximity
localise
Perfectly satisfied monitor
Moderately satisfied monitor
Non satisfied monitor
Generalisation
evaluate
Constraints monitors related to each constraint
•building min area
• maintain road
shape
• min proximity b/w
roads and buildings
Monitors are used in all the next steps of CollaGen

31. 31Choice of a process
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000

32. 32Choice of a process
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000
Process 1
Pre-conditions:
• rural (4/5)
• urban (2/5)
Post-conditions:
• ‘maintain building alignments’
(4)
• ‘proximity b/w buildings’ (5)
• ‘building minimum area’ (5)
Scale Range
1:15000 à 1:50000

33. 33Choice of a process
Process 2
Pre-conditions:
• roads (4/5)
• rivers (2/5)
Post-conditions:
• ‘maintain initial shape’
• ‘avoid coaslescence’
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000
Process 4
Pre-conditions:
• rural (5/5)
Post-conditions:
• ‘building minimum area’
• ‘proximity b/w buildings’
• ‘proximity building/road’
• ‘parallelism building/road’
Scale Range:
Up to : 1:25000
Process 1
Pre-conditions:
• rural (4/5)
• urban (2/5)
Post-conditions:
• ‘maintain building alignments’
(4)
• ‘proximity b/w buildings’ (5)
• ‘building minimum area’ (5)
Scale Range
1:15000 à 1:50000
Process 3
Pre-conditions:
• rural (4/5)
• urban (3/5)
• mountain (2/5)
Post-conditions:
• ‘proximity building/road’ (4)
• ‘proximity b/w buildings’ (5)
• ‘parallelism building/road’ (3)
• ‘building minimum area’ (5)
• ‘relative positioning b/w
buildings’ (3)
Scale Range
1:15000 à 1:35000

34. 34Choice of a process
Process 2
Pre-conditions:
• roads (4/5)
• rivers (2/5)
Post-conditions:
• ‘maintain initial shape’
• ‘avoid coaslescence’
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000
Process 4
Pre-conditions:
• rural (5/5)
Post-conditions:
• ‘building minimum area’
• ‘proximity b/w buildings’
• ‘proximity building/road’
• ‘parallelism building/road’
Scale Range:
Up to : 1:25000
Process 1
Pre-conditions:
• rural (4/5)
• urban (2/5)
Post-conditions:
• ‘maintain building alignments’
(4)
• ‘proximity b/w buildings’ (5)
• ‘building minimum area’ (5)
Scale Range
1:15000 à 1:50000
Process 3
Pre-conditions:
• rural (4/5)
• urban (3/5)
• mountain (2/5)
Post-conditions:
• ‘proximity building/road’ (4)
• ‘proximity b/w buildings’ (5)
• ‘parallelism building/road’ (3)
• ‘building minimum area’ (5)
• ‘relative positioning b/w
buildings’ (3)
Scale Range
1:15000 à 1:35000
Filtering by pre-conditions

35. 35Choice of a process
Process 2
Pre-conditions:
• roads (4/5)
• rivers (2/5)
Post-conditions:
• ‘maintain initial shape’
• ‘avoid coaslescence’
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000
Process 4
Pre-conditions:
• rural (5/5)
Post-conditions:
• ‘building minimum area’
• ‘proximity b/w buildings’
• ‘proximity building/road’
• ‘parallelism building/road’
Scale Range:
Up to : 1:25000
Process 1
Pre-conditions:
• rural (4/5)
• urban (2/5)
Post-conditions:
• ‘maintain building alignments’
(4)
• ‘proximity b/w buildings’ (5)
• ‘building minimum area’ (5)
Scale Range
1:15000 à 1:50000
Process 3
Pre-conditions:
• rural (4/5)
• urban (3/5)
• mountain (2/5)
Post-conditions:
• ‘proximity building/road’ (4)
• ‘proximity b/w buildings’ (5)
• ‘parallelism building/road’ (3)
• ‘building minimum area’ (5)
• ‘relative positioning b/w
buildings’ (3)
Scale Range
1:15000 à 1:35000
Filtering by scale

36. 36Choice of a process
Process 2
Pre-conditions:
• roads (4/5)
• rivers (2/5)
Post-conditions:
• ‘maintain initial shape’
• ‘avoid coaslescence’
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000
Process 4
Pre-conditions:
• rural (5/5)
Post-conditions:
• ‘building minimum area’
• ‘proximity b/w buildings’
• ‘proximity building/road’
• ‘parallelism building/road’
Scale Range:
Up to : 1:25000
Process 1
Pre-conditions:
• rural (4/5)
• urban (2/5)
Post-conditions:
• ‘maintain building alignments’
(4)
• ‘proximity b/w buildings’ (5)
• ‘building minimum area’ (5)
Scale Range
1:15000 à 1:50000
Process 3
Pre-conditions:
• rural (4/5)
• urban (3/5)
• mountain (2/5)
Post-conditions:
• ‘proximity building/road’ (4)
• ‘proximity b/w buildings’ (5)
• ‘parallelism building/road’ (3)
• ‘building minimum area’ (5)
• ‘relative positioning b/w
buildings’ (3)
Scale Range
1:15000 à 1:35000
Ranking by relevance

37. 37Choice of a process
Process 2
Pre-conditions:
• roads (4/5)
• rivers (2/5)
Post-conditions:
• ‘maintain initial shape’
• ‘avoid coaslescence’
20 monitors in rural space (importance):
• 1 ‘maintain building alignments’ (4)
• 5 ‘proximity building/road’ (4)
• 4 ‘proximity b/w buildings’ (5)
• 5 ‘parallelism building/road’ (3)
• 5 ‘building minimum area’ (5)
Rural Space
Scale: 1: 15000 -> 1: 50000
Process 4
Pre-conditions:
• rural (5/5)
Post-conditions:
• ‘building minimum area’
• ‘proximity b/w buildings’
• ‘proximity building/road’
• ‘parallelism building/road’
Scale Range:
Up to : 1:25000
Process 1
Pre-conditions:
• rural (4/5)
• urban (2/5)
Post-conditions:
• ‘maintain building alignments’
(4)
• ‘proximity b/w buildings’ (5)
• ‘building minimum area’ (5)
Scale Range
1:15000 à 1:50000
Process 3
Pre-conditions:
• rural (4/5)
• urban (3/5)
• mountain (2/5)
Post-conditions:
• ‘proximity building/road’ (4)
• ‘proximity b/w buildings’ (5)
• ‘parallelism building/road’ (3)
• ‘building minimum area’ (5)
• ‘relative positioning b/w
buildings’ (3)
Scale Range
1:15000 à 1:35000
1
2
Ranking by relevance

38. 38CollaGen: Collaboration by Orchestration
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

39. 39Online Observation
Observator Agent
Space Agent
Which monitor sample?
Monitor (Satisfied)
During the generalisation of a space agent

40. 40Online Observation
Observator Agent
Space Agent
Which monitor sample?
Monitor (Satisfied)
During the generalisation of a space agent

41. 41Online Observation
Observator Agent
Space Agent
Which monitor sample?
Monitor
Unsatisfied Monitor
(Satisfied)
During the generalisation of a space agent
During generalisation

42. 42Online Observation
Observator Agent
Analysis
Space Agent
Which monitor sample?
Monitor
Triangulation
Unsatisfied Monitor
(Satisfied)
During the generalisation of a space agent

43. 43Online Observation
Observator Agent
Analysis
Space Agent
Which monitor sample?
Monitor
Triangulation
Unsatisfied Monitor
Observed conflict area
(Satisfied)
During the generalisation of a space agent
new subspace agent

44. 44CollaGen: Collaboration by Orchestration
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

45. 45CollaGen: Collaboration by Orchestration
Perfect
generalisation
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

46. 46CollaGen: Collaboration by Orchestration
Good
generalisation
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

47. 47CollaGen: Collaboration by Orchestration
Bad
generalisation
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

48. 48Step-by-step Evaluation
0
5
10
15
20
25
Satisfied Medium Non
satisfied
before
after
Improvement in comparison to previous state
• 3 criteria for evaluating state validity :
Monitors number
Monitors satisfaction

49. 49Step-by-step Evaluation
Post-conditions:
• ‘building minimum area’
• ‘building proximity’
• ‘road/building proximity’‘road/building orientation’
satisfied medium non satisfied
‘building proximity’
‘road/building proximity’
‘building minimum area’
‘forest granularity’
Improvement in comparison to previous state
Internal Quality
• 3 criteria for evaluating state validity :

50. 50Step-by-step Evaluation
Global Quality
satisfied medium non satisfied
Use the mean and the standard deviation of monitor satisfactions
Improvement in comparison to previous state
Internal Quality
• 3 criteria for evaluating state validity :

51. 51CollaGen: Collaboration by Orchestration
Choice of
a space
Choice of
a process
Online
observation
generalisation
Evaluation
Side Effects
management

52. 52Plan
• Background and Objectives
• CollaGen: Collaboration by Orchestration
• Results
• Conclusion and Future Plans

53. 53Results
1: 15k to 1:50k map (9 processes and ~50 spaces)

54. 54Results
1: 15k to 1:50k map (9 processes and ~50 spaces)

55. 55
Comparison with benchmark results (Stöter et al, 2010)
initial CollaGen
Results

56. 56
Comparison with benchmark results (Stöter et al, 2010)
initial CollaGen
Clarity1
2
Results

57. 57
Comparison with benchmark results (Stöter et al, 2010)
initial CollaGen
CPT1
2
Results

58. 58Conclusion
CollaGen allows automatic orchestration of
generalisation processes with good results

59. 59Conclusion
CollaGen allows automatic orchestration of
generalisation processes with good results
• Orchestration inference with partial knowledge

60. 60Conclusion
CollaGen allows automatic orchestration of
generalisation processes with good results
• Orchestration inference with partial knowledge
• Generalisation process register

61. 61Conclusion
CollaGen allows automatic orchestration of
generalisation processes with good results
• Orchestration inference with partial knowledge
• Generalisation process register
• Online observation and dynamic correction

62. 62Conclusion
CollaGen allows automatic orchestration of
generalisation processes with good results
• Orchestration inference with partial knowledge
• Generalisation process register
• Global and generic evaluation methods
• Online observation and dynamic correction

63. 63Conclusion
CollaGen allows automatic orchestration of
generalisation processes with good results
• Orchestration inference with partial knowledge
• Generalisation process register
• Side effects management system
• Global and generic evaluation methods
• Online observation and dynamic correction

64. 64Future Plans
• Improve the side effects management
• Improve the global evaluation

65. 65Future Plans
• Improve the side effects management
• Improve the global evaluation
• Test CollaGen for various applications
• Extend CollaGen to Web service processes

66. 66
CollaGen: Collaboration between Automatic
Cartographic Generalisation Processes
Thanks for your attention ! Questions ?
Guillaume Touya
guillaume.touya@ign.fr

67. 67Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

68. 68Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

69. 69Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

70. 70Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

71. 71Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

72. 72Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

73. 73Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

74. 74Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start

75. 75Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start
End

76. 76Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start
End

77. 77Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start
End

78. 78Choice of a space
Orchestration infered on partial knowledge
(R1). start urban spaces
(R2). urban spaces road network
(R3). road network rural spaces
(R4). mountain spaces end
Space Agents
urban
rural
mountain
roads
rurban
Start
End

79. 79Side Effects management
• Problems of global harmony
(a) (b)
before after
(a) (b)
Generalisation
• Conflicts at the space neighbourhood
River network
generalisation
same building density (b) more dense than (a)

80. 80Side Effects management
Side effects are…
• identified by specific monitors
• corrected by diffusion, deformation or arbitration