ACI Map On Demand 92000 A. Ruas 1

1
Evaluation of generalisation process

• Synthesis

2
Evaluating what ? What for ?

• Some generalised data
• A process
• A system
• Some algorithms

3
Evaluating the data
- the data (changes) - according to user needs
4
Comparing Data, to what?
How well the generalised data represent the real
world ?
?
5
User needs
DATA SCHEMA SELECTION RULES
...
necessary to enrich the description of the data
6
Evaluating a process

• To Evaluate a system (soft)
• convergence, efficiency
• missing tools
• bugs
• To Learn about the process
• How to generalise?

7
Data Process
the system
the output
8
Evaluating a system what ?
User needs
Generalised Data
Initial Data
9
Comparing what to what ?
Manual
User needs
Generalised Data
Dynamic
Generalised Data
principles
Initial Data
Generalisation
Sequences used
Ergonomy
algorithms
10
The OEEPE Test on generalisation

• Aim-1 learn about the process
• Aim-2 identify deficiency
• Method
• perform each generalisation
• recording the action and decision
• analyse each process
• compare processes

11
OEEPE test what has been evaluated?
User needs
Procedural
geographical
Ergonomy
Generalised Data
knowledge
Initial Data
Dynamic
Generalisation
Sequences used
algorithms
principles
12
Tests organisation

• 3 data bases
• Road network IGN BDCarto (10m)
• Medium scale IGN BDTopo (1m)
• Large scale ICC Data
• User need like existing maps
• gt 1250K
• gt 1 50K
• gt 1 15K
• On existing systems
• with generalisation packages

13
Difficulties

• Differences btw systems
• level of automation batch, interactive, 1/2
  interactive
• algorithms
• missing algorithms
• Terminology to describe
• action, motivation, assessment
• understanding of user needs

14
Consequences

• Hard to analyse tests performed on CHANGE
  (Hanover)
• Batch processes
• Lack of algorithms
• Processes adapted to possibilities
• Specifications slightly different btw tests

15
Process description

• Which objects?
• Why?
• How?
• Operation
• Algorithm
• Parameter(s)
• Evaluation
• 0/1/2
• remaining pbm

• Common n objects
• list of conflicts
• list of
• Operations
• Algorithms
• list of conflicts

16
Conflicts code
17
Example urban block
18
The analysis of the results

• Global results
• nb operations, assessments
• Operations algorithms
• qty used
• Process
• sequence used
• Conflicts
• conflicts lt-gt solution ?
• Remaining conflicts
• Algorithms
• assessment, parameter, bugs

19
Full results test 1 2OEEPE
PublicationEarly 2001
20
Test 1 BDCarto
21
Test 1 Road Network

• MGE/MG Intergraph
• M. Jordan Munich
• M. Pla, B. Blanca ICC
• Lamps Laser-Scan
• L. Harrie Lund
• PlaGe IGN-COGIT
• S. Mustière COGIT
• C. Roux COGIT
• CHANGE Hanover B. Husen

22
Principal results

• Strong correlation btw conflicts and algorithms
• symbolisation necessary
• algo
• caricature displacement are missing
• filtering are not excellent
• line segmentation is useful
• line shape is badly preserved

23
Test2 BDTopo
24
Test 2 Medium scale

• MGE CHANGE ICC
• M. Pla, B. Blanca ICC
• MGE/MG Intergraph
• M. Jordan Munich
• Lamps Laser-Scan
• L. Harrie Lund
• A. Boffet COGIT
• Stratège IGN-COGIT
• A. Ruas COGIT
• CHANGE Hanover B. Husen

25
Principal results

• Interactive operations 23,7
• 86,3 are object removal
• lack of selection process
• Generalisation performed by algorithms
• Object removal 18,1 (inc. Typifica.)
• Aggregation 8,2
• Simplification 17,2
• Dilation 16,1
• Squaring 13,1
• Displacement 5,4
• Orientation 6,6
• Collapse 8,5
• Data Enrichment 12,2

26
Assessment

• Remaining conflicts
• over proximity
• loss of relative position
• build. structure
• loss of shape character
• squaring, simplification

27
Process (1)

• Object removal (size, type)
• operations on urban block
• building removal (typification 2)
• geometry improvement
• dilation, simplification
• object removal (to correct pbm)
• generalisation of vegetation
• aggregation, fusion

28
Process (2)

• Street removal
• at the beginning 2 tests
• within urban block, before buildings 2 tests
• at the end, before vegetation 1

29
Process (3)

• Building generalisation
• building before houses
• typifications damage spatial organisation
• always dilation / change to symbol
• generate overlapping, solved by deletion
• dilation before simplification
• squaring the last operation

30
Visual results
31
Test 3 ICC data
32
Test 3 Large scale

• MGE/MG Intergraph
• M. Jordan Munich
• CHANGE Uni. of Hanover
• B. Husen Hanover
• CHANGE MGE ICC
• M. Pla, B. Blanca ICC
• Lamps Laser-Scan
• L. Harrie Lund
• PlaGe IGN-COGIT
• F. Lecordix COGIT

33
Visual results
CHANGE MGE ICC
CHANGE
LAMPS
PlaGe
34
Level of automation

• CHANGE 1 macro operation
• 21 remaining conflicts
• CHANGE MGE ICC 40
• Lamps 143
• Mge/MG 543
• PlaGe 660

35
Process

• Hanover ANGI CHANGE
• ICC ANGI CHANGE
• Ortho-aggre Clarification (MGE)
• Munich MGE
• clarification (aggregation squaring)
• IGN PlaGe
• build-simp Nickerson (Douglas)
• Lund LS
• corner-flipping Douglas-dispike

36
Remaining conflicts

• Road and building too close
• roads do not exist...
• buildings too close
• lost of building shape

37
Repetitive small shapes - Inner proximity
38
Test 3 Conclusions

• Bad test organisation (my fault)
• algorithms comparison
• not process sequences analysis
• Use of samples would have been better

39
General conclusion (1)

• Time consuming
• test
• cleaning analysis
• Process analysis
• adapted to step by step process
• need log in with controls
• Large scale
• samples to detect deficiency

40
General conclusion (2)

• Correlation btw conflicts and solutions
• automation is possible BUT
• lack of spatial analysis tools
• Lack of
• selection / deletion (contextual)
• displacement
• geometric change
• pbm / shape preservation

41
FUTURE

• Evaluation of
• generalised data (robust measures)
• algorithms (with samples)
• process (interactive automated)
• Samples of data on web
• different levels of granularity
• Log in tools for measuring
• use of machine learning

42
I am leaving the WGNeed for new chairmanto
carry on