Geographical Map Annotation With Social Metadata In a Surveillance Environment
Post on 16-Apr-2017
287 Views
Preview:
Transcript
GEOGRAPHICAL MAP ANNOTATIONWITH SOCIAL METADATA IN ASURVEILLANCE ENVIRONMENT
Elena RogliaTutor: Prof.ssa Rosa Meo
Università degli Studi di TorinoScuola di Dottorato in Scienza e Alta TecnologiaIndirizzo: Informatica
2
Overview
SMAT-F1 ProjectSecond Level Exploitation of dataObjectives and research questionsMultidimensional data managementMetadata research, management and
visualizationMap annotation with significant tagsConclusions and future works
3
Sistema di Monitoraggio Avanzato del Territorio – SMAT
SMAT Project aims at studying and demonstrating asurveillance system, to support:
prevention and control of a wide range of natural events (fires, floods,landslides)
environment protection against human intervention (traffic, urban planning, pollution and cultivation)
4
5
6
SMAT architecture
SMAT-F1, is the first phase of SMAT project and aims to demonstrate an integrated use of three Unmanned Air Vehicle (UAV) platforms inside of a primary scenario, relevant for the Piedmont Region.
7
SMAT-F1 Architecture
8
SS&C
Before mission: mission planning, UAS tasks allocation.
During mission: mission monitoring, data collection from the CSs, operator support in the interaction with the system
After mission: conclusive report and Second Level Exploitation of data.
9
Second Level Exploitation activity
analyze and organize data collected during missions
prepare mission reportscorrelate dataallow visualization, re-processing and retrieval
of data according to the end-user needsprovide a mechanism to retrieve and search
metadata
10
Metadata Retrieval and Search
Our goal is to add metadata to geo-referenced objects related to missions stored in the SS&C database
Metadata are annotations provided by users of an open, collaborative system (see later!)
The retrieval of annotations occurs by web services exported by the collaborative systems
11
Geo-referenced Spatial Objects
• Target• Airport• Route Waypoints• Executed Route Waypoints (Flown Points).
12
Objectives and Research Questions
How to specify the interesting spatial objects according to the different dimensions involved?
How to search relationships between already stored data?
How to extract significant features in maps?
How to enrich maps?
How to generate a metadata retrieval and search module able to
answer the requirements?
13
Multidimensional approach
Metadata
14
SMAT Multidimensional Data Model
Mission
UAV
Sensor
Target
Airport
Mission Facts
15
Mission Facts
Mission facts are stored in relationship with dimensions:1. Mission in which the fact occurs 2. UAV performing the mission3. Payload sensor 4. Airport 5. Spatial target
Spatial dimensions
16
Metadata Facts
Metadata facts are stored in relationship with spatial objects and involve the dimensions:
1. Spatial objects
2. Metadata creation time
TargetAirportRoute WaypointsFlown Points
17
Query
Abstract Language Specificati
on
Compiler
18
GUI
Metadata Retrieval
COMPILER
PostGis
GeoNames
OpenStreetMap
19
Abstract Specification Language
20
SELECT THE METADATA ASSOCIATED TO THESPECIFIED SPATIAL OBJECT TYPESINVOLVED IN THE MISSIONS SATISFYING ALL THE CONSTRAINTS
21
ASL Compiler: Back – end phase
Optimization •identify mission facts that meet the conditions imposed•identify spatial objects based on these facts•identify metadata associated with these spatial objects
Code Generation
•SQL query statement generation
22
Constraints on dimensions + Spatial Object Types
Compiler
{(ObjectID, MetadataID)} + {(ObjectID, Spatial coodinates)}
24
MDR Tester
The set of constraints the user specifies in her/his query is not available a priori but is known only at run-time.
The number of possible combinations is exponentially large
Automatic procedure to test Compiler
25
Web Search Process
26
Volunteered Geographic Information - VGI
“is the harnessing of tools to create, assemble, and disseminate geographic data provided voluntarily by individuals”
Goodchild, M.F., 2007. Citizens as sensors: the world of volunteered geography. Journal of Geography, 69(4):211-221
27
Geographical Social Metadata
28
OpenstreetMap
29
OSM ElementsNodes (lat/lon-
username-timestamp)
Ways (list of nodes)Relations (nodes, way)
Key = Value
30
GeoNames
over 10 millions of geographical names7.5 millions of unique features: elevation, population, postal codes,administrative division, time zone, etc.
31
Web Services
MDR
http://api.openstreetmap.org/api/0.6/map?bbox=7.639,45.190,7.643,45.192
http://ws.geonames.org/wikipediaBoundingBox?north=45.192&south=45.18&east=7.64&west=7.63
32
Spatial Coordinates
Bounding Box
GeoNames URL preparation
Web Services request
XML File
OpenStreetMap URL preparation
Web Services Request
OSM File
Well-Formed check
CacheStorage
Web Search Process
33
COMPILER
Constraints, Object Types
ObjectID, CoordinatesObjectID, Metadata
Query
File Comparison
34
HistoricalFiles
Files Comparison Process
New Filesfrom the Web
35
Different?
Suggested metadata
37
DEMO
38
Map annotation with significant tags
tags on which the majority of the users agree
tags that annotate really typical features of the given area
39
Method
Hypothesis: all the cells have the same law of features distribution
Central cell tags frequency computation
For each tag, frequency computation in the grid. Central cell excluded!
40
Test
Sample: neighbouring cells
Mean µ and standard deviation σ of feature frequency is computed.
The frequency f of the feature in the central cell is compared with the distribution of frequencies in the sample.
Is f>µ+3σ?
41
Case study: 1
The map of Turin city and its neighbourhood102 distinct tags occurring at least 2 times84 statistical significant tags:
highway: secondary, highway:pedestrian, highway: cycleway
historic:monument, leisure:garden, amenity:fountain
amenity:parking, amenity:atm, amenity:school, amenity:car sharing, amenity:hospitals, railway:station, shop:supermarket.
42
Case study: 2
Very elegant and touristic district of Turin28 distinct tags occurring at least 2 times19 statistical significant tags:
amenity:fountain, amenity:parking, amenity:theatre, historic:monument, tourism:museum, railway:tram, amenity:place of worship, highway: pedestrian, amenity:bicycle rental, amenity:restaurant
amenity:atm, amenity:university,amenity:school, amenity:library, amenity:car sharing, amenity:hospitals, railway:station, amenity:pharmacy, railway:construction, shop: supermarket, shop:bicycle.
Case study 1
43
Case study: 3
Everest Area14 distinct tags occurring at least 2 times9 statistical significant tags:
natural:water, natural:peak, natural:glacier, tourism:camp site,
44
Case study: 4
30 Random Map in Europe:No significant features
45
CASE STUDY 2
High frequency Significant
46
Significance of absent tags
Frequency computation for all
tags in the neighbourhood
Mean µ and standard deviation σ
Frequency computation in the
central cell
Is f<µ-3σ?
47
Absent tagamenity:car wash
µ= 0,1042σ = 0,3713
CASE STUDY 1
48
Method Comparison
M. Tomko and R. Pulves. “Venice, city of canals: Characterizing regions through contentClassification”. Transactions in GIS, 7:295–314,2009.
Object category:over-representation (+)under-representation(-)
49
Empirical Method
Given a tag category we compute:
P1= the ratio between its frequency and the sum of tag frequencies in the central cell.
P2=the ratio between its frequency and the sum of tag frequencies in the neighbourhood cells.
50
P1>P2tag category is significant and it is over-represented in the central cell
P1<P2tag category is significant and it is under-represented in the central cell
ρ=P1/P2ρ<1
ρ>1 over-representation (+)
under-representation (-)
51
Classification problem
• (TP) number of significant tags that are significant for both methods;
• (FN) the number tags that are significant for proposed method but not for the empirical method;
• (FP) the number tags that the empirical method defined to be significant but proposed method finds to be not significant;
• (TN) the number of tags that both methods define to be not significant.
52
Case study: 1
The map of Turin city and its neighbourhood137 significant tagsThreshold 0,125 0,167 0,333 0,5 1
# tags 161 161 156 151 133Correlation 0,823 0,823 0,862 0,897 0,893Precision 0,851 0,851 0,878 0,907 0,962Recall 1 1 1 1 0,934
ρ<1
Threshold 1,2 1,4 1,6 1,8 2 2,2 2,5
# tags 125 119 118 115 113 105 100
Correlation 0,897 0,871 0,864 0,845 0,830 0,780 0,749
Precision 0,992 1 1 1 1 1 1
Recall 0,905 0,869 0,861 0,839 0,825 0,766 0,73
ρ>1
FP
FN
53
Case study: 2
Very elegant and touristic district of Turin38 significant tagsThreshold 0,125 0,167 0,333 0,5 1
# tags 70 70 68 68 52Correlation 0,667 0,667 0,682 0,682 0,821Precision 0,543 0,543 0,558 0,558 0,731Recall 1 1 1 1 1
ρ<1
Threshold 1,2 1,4 1,6 1,8 2 2,2 2,5
# tags 48 44 43 40 40 39 37
Correlation 0,806 0,823 0,835 0,844 0,844 0,858 0,823
Precision 0,75 0,795 0,814 0,85 0,85 0,872 0,865
Recall 0,947 0,921 0,921 0,895 0,895 0,895 0,842
ρ>1
FP
FN
54
Other
• Hills of Turin
• Industrial area of Turin
• Everest
• Random Maps
55
1. when statistical method does not identify significant characteristics the classifier still extracts significant tags, producing many false positives as characteristics of the area.
2. when proposed method identifies significant features:
if their number is low, the classifier continues to produce an high number of false positives
if their number is high, the classifier improves in performance, reducing the number of false positives, but can make some mistakes producing false negatives.
56
When the area to is stronglycharacterized, the empiricalmethod tends to produce more tags than those produced by proposed method, which acts, in general, as a morerestrictive filter for features
57
Conclusions
Metadata Retrieval and Search ModuleAllow the SS&C operator to show historical
metadataSuggest new metadata as annotation of the
geo-referenced spatial objects
Map annotation with significant tags
58
Future Work• Spatial object annotation according to a
unique tagging system: adopting the tag ontology provided by a unique system as a referential knowledge base and then trying to learn the correspondences between tags in the different systems
• Recognition of related annotations which appear to be different (different nouns or synonymous referred to the same concept).
59
• The study of Data Mining methods for the elaboration and the integration of Web resources in order to make communicate the world of ”Internet of Things” with the world of ”Semantic Web”.
• The study and the application of an algorithm that suggests the area most characterized in order to apply the proposed statistical method.
60
QUESTIONS?
61
My pubblications• E. Roglia, R.Meo, E.Ponassi, Geographical map annotation with significant tags
available from social networks, Chapter in XML Data Mining: Models, Methods, and Applications, A.Tagarelli (ed.), 26 pp, Idea Group Inc., to appear in February 2011.
• E. Roglia, R.Meo, A SOA-Based System for Territory Monitoring, Chapter in Geospatial Web services:Advances in Information Interoperability, Peisheng Zhao and Liping Di (eds.), 27 pp, Idea Group Inc., October 2010. ISBN: 978-1609601928.
• E.Roglia, R.Meo, A Composite Wrapper for Feature Selection, in Proceedings of Workshop on Data Mining and Bioinformatics in AI*IA - Intelligenza Artificiale e Scienza della Vita (DMBIO08) Cagliari (Italy), 13 September, 2008.
• E.Roglia, R.Cancelliere, R.Meo, Classification of Chestnuts with Feature Selection by Noise Resilient Classifiers, in Proceedings of the 16th European Symposium on Artificial Neural Networks - Advances in Computational Intelligence and Learning (ESANN08) Bruges (Belgium), 23-25 April, 2008.
top related