Linking objects of different spatial data sets by
integration and Aggregation
An article by Monika Sester, Karl-Heinrich Andres and
Volker Walter
Lecture by Gil Zellner
What is a map?
wikipedia: A map is a visual representation of an area—a symbolic depiction highlighting relationships between elements of that space such as objects, regions, and themes.
What is a map (cont’d) A map is not just a 2d image:
• List of objects• Partitions of areas• Linking information• Different versions of the same area• Aerial Photo• Satellite Image
Outline The article discusses ways of
integrating different maps onto a single easily accessible database, without losing information.
What is the problem with unification ?
Satellite images are not always available, often outdated, and more expensive.
What is the problem with unification? (cont’d)
Aerial photo limits• Aerial reconnaissance photos are taken
as “strips” of a larger whole. • Even the slightest (and with current
technology, unavoidable) shift in angle, connecting them is difficult
What is the problem with unification? (cont’d)
Even if we still had all the data:• Inaccuracies prevent us from matching
objects• Terrain is not flat, angle of
photography…• Information is not Absolute
Motivation Many maps today exist in many
different formats, each containing :• some correlating information• some different information
The TRUE potential of this information is when it is integrated and we can see all of it at once…
Motivation- examples Multi-national forces in IRAQ\
Afghanistan use non-stanag equipment, which uses arcane map formats, maps are essential for efficient cooperation!
- STANAG is a family of NATO standards for military equipment.
Motivation- examples (cont’d) Information from freely available
maps on web sites can be used to see trends in demographics, economy etc…
What is the closest chinese restaurant ?
Motivation- examples (cont’d)
Motivation- examples (cont’d)
Motivation- examples (cont’d)
Problem Many formats exist, integrating them
can be quite difficult without losing information
DLM = digital landscape modelCadastre = bordered maps
Solution? Conversion into a single format ?
Not a viable option, since data can become bloated and hard to decipher, also – some data STILL will be lost!
Solution – take 2 We keep all the original data, and
simply link the objects together, choosing when to use one format or another.
This article focuses on the linking aspects.
Our formats GDF – specifically designed for road
network data – vehicle navigation
Our formats (cont’d) ATKIS – Topographic data system
Our formats (cont’d)
Since the common data between system is roads, they are the matching primitives
Matching at object level The usual system for matching
information
This is not possible here!
What is geometric matching?
Matching at geometry level This we CAN do!
The different Approaches
Examples of geometric matching
Matching examples (cont’d)
Matching examples (cont’d)
How do we efficiently match these objects?
Cardinality of the matching pairs
Efficient matching (cont’d)
Normal Machine vision is clunky and difficult Solution: use noise margins, and Map the matching problem onto a communication system!
Noise margins
Series10
1
2
3
4
5
6
upper boundlower boundsample
Matching problem mapped onto a communication system
Matching function
2
| .
|; log
i i
i j j i
i ji j
i
P a is the probability that a is sent fromthetransmitter
P a b is theconditional probability that b was received whena was sent
P a bI a b
P a
Matching function (cont’d) In order to calculate the mutual
information I(D1,D2), the 2 data sets are seen as
messages which consist of symbols represented by our match primitives – the centerlines of streets.
Matching function (cont’d) For the matching of GDF and ATKIS
data we take account the length, shape, and position of start and end points
Matching function (cont’d) Our final function:
Results
Medium scale object from large scale data through Aggregation
Now that we know how to establish connections between objects of the same scale, we have another problem:
Multi-scale data objects
Multi scale data objects How do we match objects of different
scale ?• First we transform them to a similar
scale (data aggregation problem)
Scaling
Our formats: German ALK (1:500) ATKIS DLM25 (1:25000)
The process Classification
• Based on usage• Relations are check by combination
Aggregation• Adjoining parcels are aggregated• Separated areas are merged accordingly
Learning Aggregation rules Usage of “typical” machine learning
can be used here• What to group• Why group• When to group
Learning Objects and Semantic relations
1) Object Types2) Classification is derived from the
data set3) Classes created
Learning Objects and Semantic relations (cont’d)
Learning Objects and Semantic relations (cont’d)
Learning Objects and Semantic relations (cont’d)
1st phase Classification
Final Classification
Structural Description of knowledge acquired
Summary Linkage of objects based on
geometry Linkage of different scaled objects
Article Criticism Lack of proper explanation
Not self contained
Addresses problems without proper explanation of “Train of thought”
Questions?