Fire Detection in the Urban Rural Interface through Fusion Techniques Evangelos Zervas Odysseas Sekkas Stathes Hadjiefthymiades Christos Anagnostopoulos.

Fire Detection in the Urban Rural Interface through Fusion Techniques

Evangelos Zervas

Odysseas Sekkas

Stathes Hadjiefthymiades

Christos Anagnostopoulos T.E.I. Of Athens, Department of Electronics

Pervasive Computing Research Group, Department of Informatics and

TelecommunicationsUniversity of Athens, Greece

MASS-GHS07, 08.10.2007, Pisa, Italy

Fire Detection in Urban Rural Interface (URI)

Early work in the framework of SCIER (FP6-IST) (Sensor & Computing Infrastructure for Environmental Risks)

zone ofinterest

Fire Detection in URI: Architecture

Local Alerting Control Unit (LACU)

•Early fire detection•Fire location estimation•Alerting function

Citizen Owned Sensors Publicly Owned Sensors

Types of sensors:

•Temperature•Humidity•Wind flow•Cameras

Physical Model

Temperature ( T )

Fuel mass function ( F )

after 30sec. from ignition

Fire is s ens ed on ly few

er met ers from

the ignit ion po int

Binary hypothesis problemML Criterion:The “No Fire” Case

sensor measurement for sensor j

Gaussian with mean μ(h)

Mean μ(h) depends on:•time (hours/month),•empirical models,•forecasting,•sensor readings that are more up-to-date

[Walter’s model]

[Drop the D highest and lowest temperature measurements out of K available]

sensor measurement noise (zero mean)

ML Criterion: The “Fire” Case

random variable qj measures the excesstemperature due to fire

Gaussian with mean μq(h)

We consider a heat radiation modelwith mean μq(h) depending on:•ΔH (excess temperature at fire location)•x (distance of the sensor from the fire front)•a (exponent obtained from the physical model)

Receiver Operating Characteristics (ROC)

Parameters:

μ(h) = 300K,σs= 3 K,σn= 0.5 K,σq= 1 K,a= 2.3,ΔH= 700K

Receiver Operating Characteristics (ROC)

R: monitoring area of temperature sensors for creating a dense lattice of sensors for fire early detection

R

Current Work in SCIER

Use of (fuzzy) Neural Nets and/or BN for classification using data from temperature and humidity sensors,

Use of alternative criteria, i.e. CUSUM sequential algorithm,

Use information fusion at a higher level (Computing Subsystem) taking into account the vision sensors.

Thank you

http://p-comp.di.uoa.gr