180 likes | 286 Views
A Multi-level Data Fusion Approach for Early Fire Detection. Odysseas Sekkas Stathes Hadjiefthymiades Evangelos Zervas. Pervasive Computing Research Group, Department of Informatics and Telecommunications, University of Athens, Greece Department of Electronics, T.E.I. Of Athens, Greece.
E N D
A Multi-level Data Fusion Approach for Early Fire Detection Odysseas Sekkas Stathes Hadjiefthymiades Evangelos Zervas Pervasive Computing Research Group, Department of Informatics and Telecommunications, University of Athens, Greece Department of Electronics, T.E.I. Of Athens, Greece CIDM-2010, 25.11.2010, Thessaloniki, Greece
Fire Detection in Urban Rural Interface (URI or WUI) zone of interest Work in the framework of SCIER (FP6-IST) (Sensor & Computing Infrastructure for Environmental Risks)
Computing Subsystem Public infrastructure Local Alerting Control Unit LACU LACU LACU private infrastructure LACU LACU LACU LACU control SCIER architecture
Sensing Subsystem • Sensor Infrastructure • In-field sensor nodes (humidity, temp, wind speed/direction) • Out-of-field vision sensors (vision sensor) • Sensor Data Fusion
Localized Alerting Subsystem-LACU • Receives sensor data and executes fusion algorithms. • Generates fused data with degree of reliability. • Fused data fed to the Computing Subsystem. • The false alarm rate (fire detection in case of no fire) is parameterized • user requirements • season of the year (e.g. summer) • risk factor of the monitored area
Computing Subsystem (CS) • Computation and Storage • Environmental models • Main functionalities of CS • Collect and store sensor-measurements from the area of interest • Perform fusion-algorithms to assess the level of risk • Trigger a simulation in case of an alarm, i.e. retrieve geographical data from the GIS Database on the terrain layout of the area of interest. • Predictive Modeling (simulations of fire propagation using GRID Computing)
DB Computing Subsystem Architecture C.S. C.S. GRID GRID Storage Subsystem Storage Subsystem Simulation Subsystem Simulation Subsystem FF FF Sim Sim Data Storage Data Storage DB DB DS Manager DS Manager Simulation IF Simulation IF LACU Manager LACU Manager Fusion Subsystem Fusion Subsystem GIS GIS User Interface User Interface From/To From/To LACUs LACUs
Multilevel fusion scheme • Monitors the distribution of sensor data (e.g. ambient temperature) • Assigns in each sensor a probability on “fire” case • Collects probabilities on “fire” case from in-field sensors and cameras • Probabilities combined through DS theory in order to make a final decision about fire occurrence
First level fusion Sequence of random variables (e.g. values of temperature sensor) density in “no fire” case, μ0 denotes the mean temperature value density in “fire” case, μF denotes the mean temperature value superscripts e, h, f and m denote empirical,historical, forecasting and measured estimates respectively. empirical estimation of temperature Walters’ model [Walter ‘67]
First level fusion • Change detection [Gombay ’05] • Cumulative Sum (CUSUM) test • conclude that a change from the initialμ0 mean value to μFoccurs at time τ. • Basic probability assignments (BPA) for each sensor or use anincreasing function g(·) that maps the interval [μ0,μF] to the interval [0,1]. The same techniques of change detection can beapplied also for humidity sensors. In this case μ0 denotes theambient relative humidity which decreases in the “fire” case
Second level fusion • Collection of probabilities on the “fire” case • camera: significant change in the contrast orthe luminance of a scene is translated to a probability of “fire” • Cases where a camera tile does not oversee any sensor(s), or a/any sensor(s) is/are not overseen by a camera • fusion process will be carried out taking into account the probabilities of a single camera tile or any sensor(s) respectively. • Combination of probabilities through DS-theory [Shafer ‘76] • decision of experts Si and Sj
Second level fusion • For eachsensor we need the BPAs • m(F), “fire” case • m(no - F),“nofire” case • m(F U no - F),the uncertainty of thesensor. • With 3 or more sensors we calculate m123…M(F), m123…M(noF) and m123…M(F U no - F) For the fire detection we use the result m123…M(F) and compare it to a threshold t
Fire front evolution • The fusion result indicates “fire” in a specific location • SCIER CS initiates asimulation of several runs in the GRID infrastructure • each run computes the expected evolution of the fire front line for up to 180 minutes after fire detection • The model is fed with information about • the topography, • moisturecontent, • type of the surface fuel • dynamic environmentalparameters such as the wind
Conlusions • Adoption of a layered fusion scheme • cope with different type of sensors • use in-field and out-of-field sensors • increase the reliability of the system • reduce false alarm rates • satisfy the early detection requirement • Future work: • use alternative combination rules other than DS • adoption of the Fuzzy Set theory to deal with uncertainty, imprecision and incompleteness of the underlying data
System Validation & Evaluation • Gestosa, Portugal (experimental and controlled burns)
System Validation & Evaluation • Stamata, Attica, Greece (system deployment)
Thank you http://p-comp.di.uoa.gr