From Sensor Data to Triples: Information Flow in Semantic Sensor Networks

March 28th 2011

From Raw Sensor Data to Semantic Web Triples
Information Flow in Semantic Sensor Networks

Nikolaos Konstantinou, Ph.D.
Autonomic and Grid Computing group
Athens Information Technology

March 28th 2011

Lecture Outline
• Introduction
• Sensor data
• Semantic web
• Context-awareness
• The GSN middleware
• Exposing sensor data as triples
• Semantic Integration using SPARQL

March 28th 2011

Introduction
• The Situation
– Decrease in the value of sensors encourage the shift from
Desktop to Ubiquitous Computing
• The Problem
– Extracting meaning from a network of deployed sensors
• Why is this a problem?
– Raw sensor data is useless unless properly annotated
– Limited resources in terms of processing, storage
capabilities and bandwidth
• What is the suggested solution?
– Establish ways to automatically process and manage the
data

March 28th 2011

Concepts involved (1)
– Context-aware systems are able to sense and measure
their environment and include these measurements in
their behavior
• Data fusion
– Combine information and data residing at disparate
sources, in order to achieve improved accuracies and more
specific inferences than could be achieved by the use of a
single data source alone
– Data fusion spans various levels, from signal and object
refinement (low level) to situation and threat assessment
(high level)

March 28th 2011

• Information Integration
– Unify information originating from various sources in order
to allow its processing as a whole
– Obstacles include
• Heterogeneity in source schemas and data
• Various Technical Spaces
• Semantic and syntactic differences
• Semantic Information Integration
– The resulting integration scheme carries its semantics
• Information Merging
– Unification of the information at the implementation level

March 28th 2011

• Information Aggregation
– Report the mean value of a set of measurements
– E.g. the average temperature of a set of temperature
sensors
• Information Annotation
– Inclusion of metadata next or in the actual data
– E.g. ID3 tags in mp3
• Semantic Information Annotation
– Unambiguously define information
– Third parties can understand the information

March 28th 2011

Lecture Outline
• Introduction
• Sensor data
• Semantic web
• The GSN middleware
• Exposing sensor data as triples
• Semantic integration using SPARQL

March 28th 2011

Sensors and Sensor Data (1)
• Sensors are devices that measure physical properties
– Temperature, motion, light, humidity sensors
– Also cameras, microphones, GPS-enabled smartphones
• Sensors provide data that can be
– Streamed data
• Audio/Video content
– Event-based
• Temperature measurement
• RFID tag read
• Light curtain interrupt

March 28th 2011

Sensors and Sensor Data (2)
• Why is sensor data any different than other forms of
data, e.g. multimedia?
– Synchronization issues
• Apply acceptance thresholds
– Erroneous measurements
• Apply aggregation
– Limited resources in the nodes
• Caution when designing where the actual processing takes place
• Keep a sliding window
• Heavy process in the Gateway Nodes
– Streaming may lead to packet losses
• Reconstruct, or take decisions based on what you have

March 28th 2011

Sensor Network Topologies (1)
• Sink nodes collect information
– Higher processing capabilities

March 28th 2011

Sensor Network Topologies (2)

March 28th 2011

Lecture Outline
• Introduction
• Sensor Data
• Semantic Web
• Context-Awareness
• The GSN Middleware
• Exposing Sensor Data as triples
• Semantic Integration using SPARQL

March 28th 2011

Why Semantic Web (1)
• Knowledge in the form of a graph
– (subject, property, object)
• Information is assigned an unambiguously defined
meaning, its semantics
– Queries can be posed by any third parties
– Ontology, a well defined vocabulary
• Numerous ontologies already available and
interconnected on the Web
– Can and should be used when integration is a goal

March 28th 2011

Why Semantic Web (2)
• Enables semantic annotation and integration
• Enables reasoning
– Extract implicit information from the explicitly asserted
– Assure concept consistency and satisfiability
• Open source tools available
– Protégé, Jena, Virtuoso, …
• Allows information to be exposed as Linked Open
Data

March 28th 2011

The Linked Open Data Cloud

2007 2010
Source: linkeddata.org

March 28th 2011

Context-awareness: The big picture (1)

• The sensors capture information from the
environment
• The signal processing components produce
structured information
– Face detector, Body Tracker, Vehicle tracker, Smoke
detection etc.

March 28th 2011

Context-awareness: The big picture (2)

• The middleware
– Business logic: Program, configure, monitor and control
system behavior
– Storage layer
• Support database: limited historicity, sliding window
• Archive database: enables further processing
• Note the Information/Control duality

March 28th 2011

Signal Processing Components (1)
• Much work carried out in the AGC lab
– Image Processing
• Face Detection/Recognition/Tracking
• Body Tracking
– Audio Processing
• A/V Localization
• Voice activity
detection

March 28th 2011

Signal Processing Components (2)
• Challenges
– Processing is resource-hungry
– Heterogeneous technologies must be combined
• Production Algorithms in C++, prototypes in Matlab
• Multidisciplinary skills required
• Well-defined Interfaces need to be developed using
RMI/Sockets/Web services/JNA
– Video processing differs from streaming
processing
• Processing an avi file differs (greatly!) from processing
an rtp stream

March 28th 2011

Global Sensor Networks (1)
• Open-source, java-based middleware solution
• Available online at sf.net/projects/gsn/
• Adaptability
– Everything is a virtual sensor
– Virtual sensors rely on wrappers
– Every data producer can be integrated into the GSN with a
virtual sensor and wrapper

March 28th 2011

• Simplicity
– Configurable without compiling source code
– Web application for sensor management
• Scalability
– Allows communication
between nodes
– Allows data aggregation
and fusion using an
SQL-like declarative
language

March 28th 2011

• Example
– Integrating a Signal Processing component (e.g. a Smoke
detector) into GSN

March 28th 2011

Example of data aggregation using GSN
<output-structure>
…
<field name="TEMPERATURE" type="int" /> measured
… properties
</output-structure>
…
<storage history-size=“24h" /> sliding
… window
<streams> size
<stream name="input1">
<source alias="source1" sampling-rate="1" storage-size="1">
<address wrapper="temperature">
<predicate key="sampling-rate">10000</predicate> data source
</address>
<query>select TEMPERATURE from wrapper</query>
</source>
<query>select avg(TEMPERATURE) from source1</query> aggregated
</stream> output
</streams>

March 28th 2011

Example of data fusion using GSN
<storage history-size="1h" />
<streams>
<stream name="teststream" rate="1000">
<source name="source1" alias="source1" storage-size="100" slide="0" sampling-rate="1">
<address wrapper="remote-rest">
<predicate key="HOST">localhost</predicate><predicate key="PORT">22001</predicate>
<predicate key="QUERY">select FACE_COUNT from doorwatcher</predicate> source 1
</address>
<query>select FACE_COUNT AS S1 from wrapper</query>
</source>
<source name="source2" alias="source2" storage-size="100" slide="0" sampling-rate="1">
<address wrapper="remote-rest">
<predicate key="HOST">localhost</predicate><predicate key="PORT">22002</predicate>
source 2
<predicate key="QUERY">select TAG from touchatag</predicate>
</address>
<query>select TAG AS S2 from wrapper</query>
</source>
<query>
select source1.S1 as S1OUT, source2.S2 as S2OUT fused
from source1, source2
output
where source1.S1 > 0 AND source2.S2="04dddcb9232580"
</query>
</stream>
</streams>

March 28th 2011

The Storage Layer (1)
• Schema according to the Virtual Sensors
– Virtual Sensor definition example:

<virtual-sensor name="temperature" priority="10">
…
<output-structure>
<field name="ID" type="int" />
<field name="SENSORTIME" type="time" />
<field name="TEMPERATURE" type="double" />
<field name="UNIT" type="varchar(255)" />
</output-structure>

March 28th 2011

• Schema according to the Virtual Sensors
– Schema auto-generated SQL create statement:
CREATE TABLE `gsn1`.`temperature` (
`PK` bigint(20) NOT NULL AUTO_INCREMENT,
added by GSN
`timed` bigint(20) NOT NULL,
`ID` int(11) DEFAULT NULL,
`SENSORTIME` time DEFAULT NULL, depending on the virtual
`TEMPERATURE` double DEFAULT NULL, sensor description
`UNIT` varchar(255) DEFAULT NULL,
PRIMARY KEY (`PK`),
UNIQUE KEY `temperature_INDEX` (`timed`)
) ENGINE=MyISAM AUTO_INCREMENT=122 DEFAULT CHARSET=latin1

March 28th 2011

• Historical data
– According to the Virtual Sensor definition
• Tuple-based: <storage history-size="1" />
• Time-based: <storage history-size="1m" />
• Out-of-the-box support for
– Mysql
– SQL Server
– Oracle
– H2
– Can be extended to support
other RDBMS’s

March 28th 2011

• Storage Layer can be
– Centralized (in the Central Control node)
• Data is pushed to the Central Control node
– Distributed (in the Gateway nodes)
• Interfaces
– Legacy (ODBC/JDBC)
– Web Services (SOAP or RESTful)
– SPARQL Endpoints
• Allow Semantic Information Integration

March 28th 2011

Exposing sensor data as triples (1)
• OpenLink Virtuoso universal server can be used as
– A web application server
– A relational database repository
– A triplestore
– A web service server
• Open-source version available at
http://virtuoso.openlinksw.com/
• Cluster Configuration
– Parallel and Horizontal scaling

March 28th 2011

• Virtuoso RDF Views
– Export relational data as triples
• SPARQL 1.1 support, plus
– Full Text Queries
– Geo Spatial Queries
– Business Analytics and Intelligence
– SQL Stored Procedure and Built-In Function exploitation from
SPARQL
– Create, Update, and Delete (SPARUL)
• Backward-chaining OWL reasoner

March 28th 2011

• Using Virtuoso, RDF Data can also be accessible via
– ODBC/JDBC
– ADO.NET (Entity Frameworks compatible)
– OLE DB
– XMLA (XML for Analysis) data providers / drivers
• Using the “Sponger” RDF-izer, RDF data can be
extracted from non-RDF sources (e.g. with XSLT)

March 28th 2011

Publishing RDF using Virtuoso (1)
• Conductor: a GUI for server administration
• Virtuoso can be used as a DBMS

March 28th 2011

• Can be combined with GSN to process sensor data
streams and export them as RDF
• Create RDF Views over the relational data

March 28th 2011

• Browseable repository
• A URI for every resource
• Example: Measurement URI

March 28th 2011

Semantic Sensor Network Example (1)
System monitoring and
control node
Mesh
topology

3 node types
Gateway
nodes

Sink
nodes

March 28th 2011

• Sink node
– Operation relies on a relational database
– Limited historical data
– Keep a “sliding window”
• Based on time or tuples
– Do not have semantic capabilities
– One Database per Sink node

March 28th 2011

• Gateway Node
– Operation relies on a semantically-enabled
knowledge base
– Supported by inference procedures
– Maintains historical/archived information
– One Knowledge Base per Gateway node
– Appropriate for Higher Level Fusion (e.g. threats,
events)

March 28th 2011

• Central Control node
– Monitors and controls the network
– Provides system-wide services
• E.g. directory services, secure authentication
– Can store its view over the network for
intelligence extraction

March 28th 2011

Information Flow
• An Information flow example in a decentralized
Semantic Sensor Network

March 28th 2011

Semantic integration using SPARQL (1)
• SPARQL: An SQL-like language for querying RDF
graphs
• SELECT-FROM-WHERE syntax
• WHERE conditions are triple patterns
• SELECT ?x ?y ?z
WHERE
{ ?x ?y ?z }
returns all the triples in the graph

March 28th 2011

• XML over HTTP (RESTful approach)
– http://demo.openlinksw.com/sparql?default-graph-
uri=urn:lsid:ubio.org:namebank:11815&should-
sponge=soft&query=SELECT+*+WHERE+{?s+?p+?o}&form
at=text/html
• No create/update/delete capabilities

March 28th 2011

• SPARQL queries can be named and stored
– A query named sparql-demo listens to:
http://localhost:8890/DAV/sparql-demo
• Can return results over HTTP (XML by default)
• MIME type of the RDF data
– 'rdf+xml' (default) | 'n3' | 'turtle' | 'ttl‘

March 28th 2011

• SPARQL results example in RDF/XML
<ROOT>
<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#” xmlns:rs="http://www.w3.org/2005/sparql-results#"
xmlns:xsd="http://www.w3.org/2001/XMLSchema#">
<rs:results rdf:nodeID="rset">
<rs:result rdf:nodeID="sol193">
<rs:binding rdf:nodeID="sol193-0" rs:name="x“><rs:value rdf:resource="http://localhost:8890/Demo/temperature/PK/4#this"/></rs:binding>
<rs:binding rdf:nodeID="sol193-1" rs:name="y“><rs:value rdf:resource="http://www.w3.org/1999/02/22-rdf-syntax-ns#type"/></rs:binding>
<rs:binding rdf:nodeID="sol193-2" rs:name="z“><rs:value rdf:resource="http://localhost:8890/schemas/Demo/temperature"/></rs:binding>
</rs:result>
…
</rs:results>
</rdf:RDF>
</ROOT>

March 28th 2011

Thank you!

Questions?

From Sensor Data to Triples: Information Flow in Semantic Sensor Networks

Recommended

More Related Content

What's hot (19)

Viewers also liked (11)

Similar to From Sensor Data to Triples: Information Flow in Semantic Sensor Networks (20)

More from Nikolaos Konstantinou (16)

From Sensor Data to Triples: Information Flow in Semantic Sensor Networks