Map Generalization using Simulated Annealing: Maintaining Feature Alignment

1
Map Generalization using Simulated Annealing
Maintaining Feature Alignment J. Mark Ware and
Nathan J. Thomas School of Computing, University
of Glamorgan, Pontypridd CF37 1DL,
UK Christopher B. Jones Department of Computer
Science, Cardiff University, Cardiff CF24 3XF,
UK Supported by EPSRC/Ordnance Survey CASE Award
2
THE PROBLEM
3
(No Transcript)
4
THE SOLUTION
5
Map generalization

• simplification
• amalgamation
• collapse
• typification
• deletion
• displacement
• resize

6
CONFLICT RESOLUTION BY FEATURE DISPLACEMENT
7
resolve by displacing one or more
features (assume it is permissible to move each
polygon up to a predefined maximum distance from
its origin)
8
POLYGON DISPLACEMENT USING TRIAL POSITIONS
9
Birchgrove
Longford
Bryncoch
Neath
Lonlas
Llansamlet
Llandarcy
Skewen
Swansea
Jersey
Briton
Marine
Ferry
10
Apply PFLP trial position techniques to Polygon
Displacement

• - we have map, n objects, containing conflict

- there will be kn alternative realisations of
the map
- too many to generate and test all (e.g. k8, n
10, gt 1 billion configurations) -need some
strategy for limiting number tested
11
A SIMULATED ANNEALING APPROACH
12
function SimulatedAnnealing input Dinitial,
Schedule, Stop_Conditions DcurrentDinitial Tini
tialT(Schedule) while NotMet(StopConditions) Dnew
RandomSuccessor(Dcurrent) DECOST(Dcurrent)-COST(D
new) if DE gt0 then DcurrentDnew else P e-DE/
T RRandom(0,1) if (RltP) then DcurrentDnew end T
UpdateT(Schedule) end Return(Dcurrent)
13
RandomSuccessor(D) Chooses a polygon pr at
random Chooses a trial position tr at
random Moves pr to tr Cost(D) A x Sum of
all conflict B x Sum of all displacement
14
The Experiment OS MasterMap data and OS Oscar
data
15
OS MasterMap
Ordnance Survey's new large-scale product and
on-line service, providing polygon data and a
unique reference (TOID) for every feature on the
British landscape. Features have been enhanced
and allocated to themes, and map tiles have been
removed to generate a seamless map for the whole
of Great Britain.
16
OS Oscar Data

• OSCAR (Ordnance Survey Centre Alignment Of
  Roads).
• Derived from the Ordnance Survey Roads
  database.
• Constantly revised and updated.

17
MasterMap - filtered (buildings) - pregeneralised
(BuildingSimplify and AreaAggregate) OSCAR -
centre lines
Example 1
18
Symbolize roads
19
Resolve conflict by building displacement
20
Problem not all conflict resolved
21
ADDITIONAL OPERATORS DELETE, REDUCE, ENLARGE
22
How ? create additional trial positions
corresponding to deleted, reduced and enlarged
states of each polygon
Deleted state
Reduced states
Enlarged states
23
RandomSuccessor(D) Chooses a polygon pr at
random Chooses a trial position tr at
random Moves pr to tr Cost(D) A x Sum of
all conflict B x Sum of all displacement
C x Sum of all deletion D x Sum of all
reduction E x Sum of all enlargement
24
Example 2
25
(No Transcript)
26
Resolve conflict by building displacement plus
size reduction
27
(No Transcript)
28
Problem disruption to high order features
Example 3
29
(No Transcript)
30
misalignment
fragmentation
31
Solution 1. - group polygons - apply
modification operators to groups Or 2. - group
polygons - apply modification operators to
polygons - refine cost function so that it takes
group disruption into consideration
32
Grouping polygons
33
Polygon grouping based on - distance -
proximity to roads
34
Application of modification operators to groups
35
(No Transcript)
36

• Advantages
• - High order features
• maintained
• - Remote solutions found more easily
• - Consistency
• - Faster
• Disadvantage
• - Within group
• conflict remains

37
Example 4
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
(No Transcript)
42
Example 5
43
(No Transcript)
44
Problem
too much displacement
too much reduction
45
Cause discrete trial positions Solution ?
continuous search space
RandomSuccessor(D) - Chooses a polygon pr at
random - Chooses a operator and parameters at
random (e.g. displace, R, q), (e.g. reduce, F) -
Apply operator to pr Cost(D) A x Sum of
all conflict B x Sum of all displacement
C x Sum of all deletion D x Sum of all
reduction E x Sum of all enlargement
46
Discrete
47
Continuous
48
(No Transcript)
49
Implementation
50
C / ArcView Implementation
51
Demo
52
Future Work
53

• refine building grouping
• refine cost function
• additional operators
• (e.g. building simplify, area aggregate)
• additional feature types
• (e.g. line modification, map labels)