1 jantien stoter kadaster apeldoorn [email protected] & technical university of delft...

1

Jantien Stoter

Kadaster [email protected]

&Technical University of Delft

[email protected]

AUTOMATED GENERALISATION OF TOPOGRAPHIC DATA

Overview

• EuroSDR generalisation research project• TOP10NL-TOP50NL research project• Research questions for follow-up research

automated generalisation for processes Kadaster

EuroSDR Generalisation project

4

Core project team Temporal members Testers VendorsDirk Burghardt (Zurich)Blanca Baella (ICC)Cécile Duchêne (IGNF)Maria Pla (ICC)Nicolas Regnauld (OS GB) Guillaume Touya (IGNF)Connie Blok (ITC)

Karl-Heinrich Anders, Jan Haunert (Hanover)Nico Bakker (Kadaster, NL)Francisco Dávila (IGNS)Peter Rosenstand (KMS, DK)Stefan Schmid (Zurich) Harry Uitermark (Kadaster, NL)

Magali Valdepérez (IGNS)Patrick Revell (OS GB)Stuart Thom (OS GB)Sheng Zhou (OS GB)Willy Kock (ITC, NL)Annemarie Dortland (Kadaster, NL)Maarten Storm (formerly Kadaster, NL)Patrick Taillandier (IGNF)

Axes systems

ESRI

University of Hanover

1Spatial

Objectives

• To study:– capabilities/limitations of commercial software

systems for automated generalisation with respect to NMA requirements

– what different generalisation solutions can be generated for one test case and why do they differ?

Overview of presentation

• Requirement analysis Oct 2006 till June 2007

• Testing June 2007 till Spring 2008

• Evaluation Summer 2008 till Spring 2009

• Finalising the project Autumn-Winter 2009

• Conlusions

Requirement analysisTesting

EvaluationConclusions

Step 1: Selection test cases

Area type

Source dataset

Target dataset

Provided by

Nr input layers

Main layers

Urban area 1:1250 1:25k OS GB 37 buildings, roads, river, relief

Mountainous area 1:10k 1:50k IGN France 23 village, river, land

use

Rural area 1:10k 1:50k Kadaster, NL 29

small town, land use, planar partition

Costal area 1:25k 1:50k ICC Catalonia 74

village, land use (not mosaic), hydrography



ICC, 1:25k IGN France, 1:10K

Kadaster, 1:10k OS GB, 1:1250

Step 2:Defining requirements

• Result: 250 constraints often covering similar situations

Example on one object

Example on two objects

Example on group of objects

Condition to be

respected

Area of buildings >

0.4 map mm2

building must be

parallel to road

target building density should

be equal to initial density ±

20 %



Step 3: harmonised requirements• 45 generic constraints:

– 21 generic constraints on one object– 11 constraints on two objects – 13 constraints on group of objects

• About 300 constraints are defined as specialised specific constraints



Step 4 Analysing test casesTotal number of constraints

Number of objects

Number of constraints for different constraint types

Number of constraints on different feature classes

on one object

on two objects

on group of objects

Model generalisation

Min. dimension and granularity

Position

Orientation

Shape

Topology

Distribution / Stat

Other

Building

Land use

Road

Water

Relief

Coastal features

Any

137 86 23 28 12 80 0 4 19 12 5 5 39 20 16 25 8 19 10

52 27 21 4 11 18 1 0 1 6 0 15 10 13 23 3 0 0 3

61 32 15 14 2 15 2 4 15 12 2 9 33 2 12 9 2 0 3

49 24 13 12 2 16 1 0 0 8 0 22 24 1 8 1 8 0 7

299 169 72 58 16 129 4 8 35 38 7 51 106 36 59 38 18 19 23



Tests

• Were performed:– by project team members on out-of-the-box

versions– by vendors (1Spatial, ESRI, University of Hanover,

Axes systems), possibly on improved and/or customized versions



EuroSDR generalisation state-of-the-art project EuroSDR Meeting 14 May

1:50K, derived from 1:25K, ICC 1:25K, derived from 1:1250, OSGB

1:50K, derived from 1:10K, IGN, France

Evaluation of outputs• Tests performed by project team:

• Tests performed by vendors:

• Total: 35 test outputs were obtained (appr 700 thematic layers). NB: 1 test cost appr 1 week

System ArcGIS CPT Axpand ClarityTest caseIGN 1 3 2 1ICC 2 3 0 2OSGB 1 3 0 1Kadaster 2 3 1 3

System ArcGIS CPT Axpand Clarity

Test caseIGN 0 1 0 1ICC 0 1 0 0OSGB 0 1 0 0TDK 1 1 0 1



Evaluation• Evaluation of:

– System capabilities (based on completed system templates)– Processing (based on actions templates)– Constraint expression (based on constraint expression templates)

• Evaluation of generalised data:– Automated constraint-based evaluation– Evaluation which visually compared different outputs for one test case– Qualitative evaluation by cartographic experts



Evaluation of generalised outputs, three methods

• Automated constraint-based evaluationDirk Burghardt, Stefan Schmidt, University of Zurich

• Evaluation which visually compared different outputs for one test caseCecile Duchene, IGN France

• Qualitative evaluation by cartographic experts Connie Blok, Jantien Stoter, ITC



Automated constraint based evaluation of generalised data

Average Constraint Violation of Minimum Distance Constraint between Two Buildings ofthe ICC Data Set

0

0.25

0.5

0.75

1

IGNf (tester) ICC Kadaster Kadaster

Clarity (system) CPT (system) ESRI (system)

Generalisation Systems & Testers

Avera

ge C

onstr

ain

t V

iola

tion



1

2

1. Town centre blocks and streets representation (selection, aggregation)

2. Coastline simplification 3. Conflicts in road interchanges

3

3

3

4. Generalization of suburban buildings (namely: preservation of buildings spatial distribution, buildings alignments)

4

5. Parallelism between roads and buildings

5

Comparison evaluationEvaluation made on "focus zones" defined for each test case after the test stage=> Classical generalisation problems covered=> Feedback from testers/producers



ICC dataset – buildings in suburban areas

(a) Initial (f) CPT, TDK tester (novice)

(g) CPT, ICC tester (expert)

(h) CPT, CPT tester (vendor)

(b) Clarity, IGNF tester (expert)

(c) Clarity, OSGB tester (expert)

(d) ArcGIS, TDK tester (novice)

(e) ArcGIS, ICC tester (novice)



Expert evaluation: methodologyGlobal indicators

Level of manual editions required to meet the constraintsDeviation from initial (ungeneralised) dataPreservation of the geographic characteristics of the test area

LegibilitySeriousness and frequency of main detected errorsNumber of positive aspectsInformation reduction (undergeneralisation / overgeneralisation)

Individual constraints assessed in expert survey

Constraints on one object

Constraints on two objects Constraints on group of objects

minimal dimensions spatial separation between features (distance)

quantity of information (e.g. black/white ration)

granularity (amount of detail)

relative position (e.g. building should remain at the same side of a road)

spatial distribution

shape preservation consistencies between themes (e.g. contour line and river)



Expert evaluation: example results

• Good scores for:

3. Deviation from map of original data

0

2

4

6

8

10

12

14

16

Highlyacceptable

Acceptable Moderatelyacceptable

Notacceptable

Freq

uenc

y of

ans

wer

s

ArcGIS Axpand Clarity CPT

4. Preservation of geographic characteristics

0

2

4

6

8

10

12

14

16

Good Aboveaverage

Average Belowaverage

Poor

Fre

qu

en

cy

of

an

sw

ers




• Lower scores for:

• Future work:– Include interactively generalised– Compare in detail with automated evaluation

1. Legibility

0

2

4

6

8

10

12

14

16

Good Aboveaverage

Average Belowaverage

Poor

Fre

qu

en

cy

of

an

sw

ers


2. Manual editing required

0

2

4

6

8

10

12

14

16

None Exceptionalcases

Some morecases

Many

Fre

qu

en

cy

of

an

sw

ers


5. Main detected errors

0

2

4

6

8

10

12

14

16

Few non-serious

Many non-serious

Fewserious

Manyserious

Noresponse

Fre

qu

en

cy

of

an

sw

ers


6. Number of main positive aspects

0

2

4

6

8

10

12

14

16

Many Fairnumber

Few None Noresponse

Fre

qu

en

cy

of

an

sw

ers


7. Information reduction

0

2

4

6

8

10

12

14

16

Acceptable Undergeneralized Overgeneralized

Fre

qu

en

cy

of

an

sw

ers

ArcGIS Axpand

Clarity CPT

Expert evaluation: example results

Conclusions capabilities of systems (1/4)

Discussed with vendors at IGN, Paris at 22 September 2009

• All systems offer potentials for automated generalisation, especially for single objects


– No generalisation problems are fully solved by the out-of-the-box systems

• Some are close to being solved:– buildings and roads

• Some are far from being solved: e.g.– apply different algorithms/parameters in different contexts

(either not supported and/or detection measures are missing)– operations that concern more than one object (e.g. network

typification)– terrain generalisation– .....


• For other problems solutions do exist (e.g. building simplification), but algorithms are difficult to parameterise– a direct match between parameters and

constraints was often missing


• Satisfying complete NMA requirements requires customisation, progress in automated generalisation should focus on:– Good customisation tools– Generic solutions (includes default parameterisation and

default tools)– User friendly parameterisation: provide insight into effect

• Some shortcomings have been solved by research/NMAs (e.g. detection tools), and also by vendors in parallel tests (e.g. displacement in Clarity and ArcGIS)

Final thoughts

– Disappointing results?• High expectations of the project (constraints, selection

of complex/known problems, high quality paper maps)• Some missing functionalities have been fixed in

vendors’ parallel tests• Not a surprise that out-of-the box versions are not

capable of fulfilling NMAs requirements; customization is definitely required

• Systems are used more satisfactory in practice• Project is well received by vendors to push internal

developments

Future project– Evaluating generalisation software on criteria

beyond constraints– Preservation of topology– Creation of links between initial and output data– Generalisation of incremental updates– Support for parameterisation– Scalability and performance– Customisation!

RGI TOP UP: TOP10NL to TOP50NL

Research team:ITC: Jantien Stoter

ESRI: John van Smaalen, Paul Hardy, Jean Luc MonnotKadaster: Harrie Uitermark , Nico Bakker

Objective

• Define specifications for automated generalisation by enriching existing specifications for interactive generalistaion

Methodology

• Implement current specifications for roads, landuse and buildings in TOP50

• Compare result with current interactively generalised map

• Enrich specification and data if required• Resulting in: improved map generated with

automated generalisation only

Buildings: current specifications

“If density of buildings in urban areas is sufficiently high, buildings can be replaced by built-up area, with the exception of detached houses and buildings in industrial areas. In rural areas built-up areas are never created. Important buildings are never aggregated to built-up area.”

“Remaining buildings should:– never be aggregated– should have a minimum size:

• (20x20m; in exceptional cases 15x15m; protrusions should at least be 15x15 m)

– have minimum distance to other buildings (10m)”

Buildings

• Enrichments:– Information on industrial polygons and on building

types (“detached house”) from other sources– Urban areas were computed based on global

building densities– Sufficiently dense translated to 10%

• Area to consider (partioning)• Presence of ‘too small’ buildings• Locating buildings in limited amount of space• Minimum distance between buildings• Many detached houses in interactively

generalised map are converted into built-up areas (35% in rural areas;65% in urban areas)+ which is not according to specs

Buildings: results compared to interactive data

TOP10NL Interactive generalised TOP50

A

A

A

B

B

A B C

A: original data B: by cartographer C: by ESRI Optimizer

Roads: specifications

• Convert road polygons to lines

• Select important roads

Centrelines in TOP10NL (left) and TOP50 (right), both projected on TOP10NL polygons

Roads: results (1/2)

• Not possible to use TOP10NL road lines:– Road surface in TOP10NL; whole road

constructions in TOP50NL– TOP10NL road lines are not complete

Roads: results (2/2)

• Important roads, dead ends and through roads are not encoded

TOP10NL Interactive Computed

Land Use: trivial problem

a TOP10NL b Collapsing TOP10NL roads

causes gaps in partition c TOP50

Example of an extended land use polygon boundary never meeting a road centreline

Final results

TOP10NL Interactively generalised TOP50 Automatically generalised TOP50

Conclusions

• Satisfying results• Making current specifications suitable for

automated generalisation:• Completing and adjusting specifications• Data enrichment• Formalising specifciations• Include formal description how specifications interact

(i.e. priority, weighting)• Align them with funtionality of systems

Research questions for follow-up researchImplementing automated generalisation in

processes Kadaster

Starting point

• Cartographic multi-scale data set

Cartographic databases not optimal as multi scale information source

Top10NL – Long thin woodlands between road and building (top left), and two similar woods

(bottom and mid right)

Interactively generalised TOP50 – Long thin wood enlarged, and buildings moved for room. Mid-

right wood kept and enlarged, but bottom-right wood deleted

TOP10NL TOP50vector data TOP50 map

• Inaccuracies in cartographic representations• ´Error` propagation in cartographic small scale

databases• No links• TOP10NL is more up to date

Future multi scale information provision

• Explicit separation between cartographic and database representations:– To make users aware of the characteristics of the

data– Different challenges for generalisation– Meet different information demands

Research questions for cartographic representations

• Apply and complement research results in larger context:– Specific attention for road representation at different scales

• Starting from current database, study generalisation of updates:– study characteristics of updates– data matching between different scales

• If data matching not possible: automatically generalise smaller scale data to obtain links (and better coherence)

Research questions for database representation

• Use of multi scale info:– Visualisation requirements– Requirements wrt level of details (different as

traditional map scales?)• Multi ´lod´ information model supporting multi

representations of objects• Generalisation of multi level of detail information:

more dynamic, less demanding?:– vario scale data structure:

• Including issues for client-server, symbolisation, multi scale

Future topographic information provision1. Only cartographic representations at different scales, TOP10NL for GIS analyses

– Advantages:• Less demanding for small scale databases (BRT): only visualisation• Keep current situation

– Disadvantages:• No links between scales• Not meeting multi scale information demand• May be hard for implementing automated generalisation

2. Generate multi scale database representations and keep current cartographic databases at different scales

– Advantage:• Meet multi scale information demand

– Disadvantage:• No links between database rep and carto rep

3. Integrated provision of multi scale information and carto rep– Advantage:

• Optimal starting point for information approach BRT instead of carto, resulting in integrated provision of all small scale topo products

• Ideal for automated update propagation, consistency and vario scale datastructure – Disadvantage:

• Ignore much past efforts• Will users accept (slightly) different products?

Thank you

1 jantien stoter kadaster apeldoorn [email protected] & technical university of delft...

Documents

processes kadaster slide

generic constraints

spatial slide

nl annemarie dortland

objects example

evaluation summer

selection of test cases

defined specifications