Skip to main content

Fire Detection and Management through a Multi-Sensor Network for the Protection of Cultural Heritage Areas from the Risk of Fire and Extreme Weather Conditions

Final Report Summary - FIRESENSE (Fire Detection and Management through a Multi-Sensor Network for the Protection of Cultural Heritage Areas from the Risk of Fire and Extreme Weather Conditions)

Executive Summary:

FIRESENSE “Fire Detection and Management through a Multi-Sensor Network for the Protection of Cultural Heritage Areas from the Risk of Fire and Extreme Weather Conditions” (FP7-ENV-2009-1-244088-FIRESENSE) is a Specific Targeted Research Project of the European Union's 7th Framework Programme Environment (including Climate Change). The project started on December 1, 2009, and finished on February 28, 2013.

One of the main causes of destruction of archaeological and cultural heritage sites, especially in the Mediterranean region, is wildfires. These sites, treasured and tended for long periods of time, are usually surrounded by old and valuable vegetation or situated close to forest regions. The increase in seasonal temperatures has caused an explosion in the number of self-ignited fires in forested areas, which fanned by winds and fuelled by dry vegetation become disastrous. Extreme weather conditions such as storms or floods also pose greats risk for these sites.

Beyond taking precautionary measures to avoid forest fires, early warning and immediate response to a fire break out is the only way to avoid human losses and environmental and cultural heritage damage. Although several technologies based on different sensors have been proposed for wildfire surveillance, the majority of existing fire detection systems does not realize the full potential of state-of-the-art technologies due to the lack of an integrated approach.

In the context of the FIRESENSE project, an automatic early warning system integrating multiple sensors to remotely monitor areas of archaeological and cultural interest for the risk of fire and extreme weather conditions was developed. The system integrates various sensors including optical cameras, infrared cameras at different wavebands, passive infrared (PIR) sensors, a wireless sensor network of temperature and humidity sensors and local weather stations on the deployment sites. The signals and measurements collected from these sensors are transmitted to the control centre, which employs intelligent computer vision and pattern recognition algorithms as well as data fusion techniques to automatically analyze and combine sensor information and detect the presence of fire or smoke.

The control centre is capable of generating automatic warning signals for smoke/flame detection and abrupt temperature rise. Moreover, by reading weather data from official meteorological services as well as from local weather stations, it is can also issue alerts in case of extreme weather conditions. The control centre interface allows monitoring of the site through the cameras, display of maps of the area with multiple layers, manipulation of cameras and sensors and provision of video and statistical data on user demand. Moreover, it can estimate the propagation of the fire based on the fuel model of the area, the local weather conditions and the ground morphology. The estimated fire propagation is visualized on a Google Earth based 3D interface. This information is extremely valuable for efficient fire management by fire fighting forces.

The FIRESENSE control centre adopts a modular architecture, which allows easy integration of different sensors and processing modules. It integrates novel algorithms and techniques for fire and smoke detection based on visible and infrared data, WSN-based fire detection, fusion of multisensory data, vegetation classification and fire propagation estimation. It also adopts a cluster-based WSN architecture implementing novel routing and activity scheduling protocols to enhance network reliability and energy efficiency.

The FIRESENSE system was demonstrated and evaluated in five cultural heritage sites in the Mediterranean area: the sanctuary of Kabeirion in Thebes, Greece, the ancient city of Rhodiapolis in Antalya, Turkey, the Dodge Hall building in Istanbul, Turkey, the Roman Temple of Water in Djebel Zaghouan, Tunisia and Monteferrato-Galceti Park in Prato, Italy. Numerous controlled fire tests were organized in several sites to assess system functionalities and evaluate system performance. The system achieved high detection rates and has successfully detected two real fires in Rhodiapolis in September and October 2012.

Project Context and Objectives:

The challenge

One of the main causes of destruction of archaeological and cultural heritage sites, especially in the Mediterranean region, is wildfires. These sites, treasured and tended for long periods of time, are usually surrounded by old and valuable vegetation or situated close to forest regions. The increase in seasonal temperatures has caused an explosion in the number of self-ignited fires in forested areas, which when fanned by winds and fuelled by dry vegetation become disastrous. In addition, arson events have been repeatedly reported, while common causes of unintentional fires are human carelessness and lightning strikes. Extreme weather conditions such as storms or floods also pose great risks for these sites.

In the summer of 2007, Ancient Olympia, an UNESCO world heritage site and the birthplace of the ancient Olympic Games, was seriously endangered by a fast-moving wildfire. The fire reached the hill overlooking ancient Olympia and it was contained just before entering the archaeological site, but not before reaching a historic pine-covered hilltop above the renowned stadium. Flames licked the edges of the original Olympic stadium and scorched the yard of the museum, home to one of Greece's greatest archaeological collections. The surrounding forest was destroyed. Similar fires have cause significant damages in archaeological areas and treasures all over the Mediterranean, especially during summer months.

Beyond taking precautionary measures to avoid forest fires, early warning and immediate response to a fire break out is the only way to avoid human losses and environmental and cultural heritage damages. Thus, the most important goal in fire surveillance is quick and reliable detection and localization of fire, since it is much easier to suppress a fire when the location of the ignition point is known and while the fire is at an early stage. An automatic fire detection system relying on multi-sensor networks should be able to provide early fire warning and also collect information about the location and spread of fire to facilitate efficient fire management. Based on this information, the firefighting staff can be guided on target to contain the fire before it reaches cultural heritage sites and to suppress it quickly by utilizing the appropriate equipment and vehicles.

The majority of commercial wildfire surveillance systems do not realize the full potential offered by available technologies due to the lack of an integrated approach. Most of the systems use visible range cameras mounted on watch towers to monitor large forested areas. Some systems utilize infrared cameras, which are usually much more expensive compared to regular pan-tilt-zoom (PTZ) cameras and their range may be limited. Few systems employ wireless temperature sensor networks, which can provide real-time feedback for the detection and evolution of fire. All the aforementioned approaches have their own advantages and limitations. What is currently missing is an integrated solution that will combine the outputs of different sensors to increase the detection accuracy and overcome individual sensor limitations.

The goal

FIRESENSE aims to develop an automatic early warning system integrating multiple sensors to remotely monitor areas of archaeological and cultural interest for the risk of fire and extreme weather conditions. FIRESENSE will take advantage of recent advances in multi-sensor surveillance technologies by employing both optical and infrared cameras to monitor the site and the surrounding area as well as a wireless sensor network capable of measuring different environmental parameters (e.g. temperature, humidity). The signals and measurements collected from these sensors will be transmitted to a monitoring center, which will employ intelligent computer vision and pattern recognition algorithms as well as data fusion techniques to automatically analyze and combine sensor information. The control centre will be capable of generating automatic warning signals whenever a dangerous situation arises, i.e. when fire or smoke is detected. Moreover, the system will read weather data from official meteorological services as well as from local weather stations installed at the site and will issue alerts in case of extreme weather conditions. It will also provide real-time information about the evolution of the fire based on wireless sensor network data. Furthermore, it will be able to estimate the propagation of the fire based on the fuel model of the area and other important parameters such as wind speed and direction and ground morphology. Finally, the estimated fire propagation will be visualized on a 3D Geographic Information System (GIS) environment.

The objectives

To main S&T objectives of FIRESENSE are the following:

• Identification of user requirements and system design

o Extensive survey of state-of-the-art algorithms, technologies and systems related to FIRESENSE including EU and national research projects.
o Establishment of an international group of users including people related to fire suppression and cultural heritage preservation and launch of international survey for the identification of user requirements.
o Definition of FIRESENSE system requirements based on the state-of-the-art survey, the user requirements and technical, economical and legal constraints.

• Smoke and fire detection based on cameras and WSN sensors

o Development of novel algorithms for fire and smoke detection based on visible cameras.
o Development of novel techniques for thermal data processing using infrared cameras at different wavebands. Development of low-cost pyroelectric infrared (PIR) sensor based system for indoor fire detection.
o Collection and analysis of weather data from local weather stations and official sources.
o Definition of WSN architecture and related communication protocols; definition of constraints and requirements for the network topology and the network node/gateway hardware/software.
o Hardware and software design, development and integration of sensor nodes, WSN gateways and housing. Design and development of communication protocols, routing algorithms and network topology for maximizing the lifetime of the WSN.
o Establishment of guidelines and test procedures for WSN deployment and maintenance.

• Multi-sensor data fusion

o Development of novel data fusion techniques for the combination of data from multiple sensors to detect fire and smoke.
o Provision of different alarm levels for cases of temperature rise, detection of smoke or fire, extreme weather conditions, etc based on the result of multi-sensor data fusion.

• Estimation and visualization of fire propagation

o Estimation of vegetation distribution and relevant fuel model parameters in monitored areas based on satellite images, pre-existing land cover information or ground surveys.
o Estimation of fire propagation based on the semi-empirical BEHAVE model and examination of physical and hybrid models for fire spread calculations.
o Development of user friendly GIS-based platform for 2D/3D visualization of the estimated fire propagation.

• Development of FIRESENSE Control Centre

o The control centre will provide various functionalities to the end users such as visual and acoustic alarms in case of fire/smoke detection and extreme weather conditions, easy access to camera streams and sensor measurements, manipulation of cameras and sensors, video on demand, maps for location and visualisation, visualization of fire propagation estimation, etc through a user friendly interface.

• System installation, integration and demonstration

o The system will be demonstrated in five cultural heritage sites in the Mediterranean area: the sanctuary of Kabeirion in Thebes, Greece, the ancient city of Rhodiapolis in Antalya, Turkey, the Dodge Hall building in Istanbul, Turkey, the Roman Temple of Water in Djebel Zaghouan, Tunisia and Monteferrato-Galceti Park in Prato, Italy.

• System evaluation

o Development of methodological framework for assessing the performance of the proposed system, in terms of covering the user requirements and expectations.
o Laboratory testing of system components and functionalities.
o Organization of real fire experiments for data collection and system evaluation.
o Evaluation of the final system in terms of technical performance and user acceptance and validation against initial requirements.

• FIRESENSE dissemination and exploitation

o Dissemination of project information and results through the project website and project brochure, publications, media releases, meeting with stakeholders, etc.
o Education of inhabitants through a series of lectures aiming at making people living close to test sites conscious about the importance of their surrounding area.
o Organization of Workshop to demonstrate and disseminate project results.
o Market analysis and development of strategy for exploiting project results beyond the life of the project.

Project Results:

In the context of the FIRESENSE project, an automatic early warning system integrating multiple sensors to remotely monitor areas of archaeological and cultural interest for the risk of fire and extreme weather conditions was developed. The system integrates various sensors including optical cameras, infrared cameras at different wavelengths, passive infrared (PIR) sensors, a wireless sensor network of temperature and humidity sensors as well as local weather stations on the deployment sites. The signals and measurements collected from these sensors are transmitted to the control centre, which employs intelligent computer vision and pattern recognition algorithms as well as data fusion techniques to automatically analyze and combine sensor information. The control centre is capable of generating automatic warning signals for smoke/flame detection, abrupt temperature rise and extreme weather conditions. It also allows inspection of the site through the cameras, manipulation of cameras and sensors and provision of statistical data on user demand. Moreover, it estimates the propagation of the fire based on the fuel model of the area, the local weather conditions and the ground morphology. Finally, the estimated fire propagation can be visualized on a Google Earth based 3D interface. In the following, the main science and technology results of the FIRESENSE project are summarized per workpackage.

WP2: Requirements Identification and System Specification

The main S&T results of WP2 are a) the identification of user requirements for an early warning system for the protection of cultural heritage areas from the risk of fire and b) the specification of the FIRESENSE system architecture, components and interfaces.

Analysis of the state of the art

An extensive state-of-the-art survey on technologies related to FIRESENSE was delivered in the early months of the project. Different sensor technologies for surveillance and monitoring were investigated (visible spectrum and IR cameras, PIR and smoke sensors, wireless sensor networks, meteorological sensors). Visible spectrum fire/smoke detection algorithms were examined and their advantages and disadvantages were identified. A survey on the usage of infrared bands for fire detection was also included and possible features in each band were identified. WSN technologies achieving energy efficient communication were also examined and possible network topologies along with factors affecting them were investigated. Methods for fusing data from different sensors were also studied. Different fire spread models and factors affecting the spread were surveyed and existing fire propagation visualization technologies were identified. Finally, existing commercial indoor and forest fire detection systems and standards were studied and related EU projects were reviewed. Advantages and disadvantages of state-of-the-art techniques/systems/projects were indentified and were used for system specification.

Identification of user requirements

The review of state-of-the-art technologies and available systems is very important for system design but equally important are the experience, needs, ideas and concerns of potential users. To identify user requirements, an international group of users was established including stakeholders involved in the preservation of cultural heritage sites (archaeologists, curators, ephorates, local authorities, etc.) and organizations involved in fire prevention/fighting, forest protection and civil protection in Greece, Turkey, Italy and Tunisia. Two user questionnaires were designed aiming at people related to fire suppression/ environmental protection and cultural heritage preservation respectively. Issues concering system design, functionalities, performance, installation and maintenance as well as the status of similar systems in cultural heritage sites in different countries were covered.

User requirements were defined through a two-stage process: first, conclusions/ information were drawn from the state-of-the-art survey and the analysis of user questionnaires. Then, a list of conclusions drawn from interviews and e-mail communications with experts and discussions among partners was generated. The outputs of this process were used to synthesize the final list of requirements including a) technical / operational requirements (cameras, sensors, communication links, power, software modules, interfaces, and maintenance), b) requirements associated with system installation and cost and c) environmental constraints.

The initial system requirements were updated during the final months of the project based on the feedback received by experts. User feedback was received through questionnaires evaluating system performance and functionalities, which were filled during system demonstration activities by users employed in fire service, forest service, cultural heritage organizations, local authorities and research institutions. Feedback from experts in fire prevention, suppression and management was also received through discussions and delivery of brief evaluation reports.

Field tests

Numerous field tests with real controlled fires were conducted in different settings using visible spectrum cameras and infrared cameras on different wavebands. These tests provided valuable information for recognizing potential problems and vulnerabilities of the system and allowed selection of the cameras that are suitable for FIRESENSE. In addition, they resulted in the creation of a large database of video recordings, which were also used for algorithm testing.

System specification

The components and functionalities of the FIRESENSE system were designed taking into consideration: a) user requirements, b) analysis of the state-of-the-art technologies and systems, c) results of field tests and d) various technical, economical and legal constraints. System architecture, specifications for different sensors and communication links, specifications for software modules (video-based fire/smoke detector, IR-based fire detector, WSN sensor and gateway software, data fusion and alarm generation, estimation & visualization of fire propagation) and specifications for control centre components, functionalities and interfaces were defined. Moreover, detailed plans for the deployment of the FIRESENSE system in the five pilot sites were prepared (e.g. sensors and communication links to be used, installation plans, power supply, possible problems, etc) and a set of use cases corresponding to different real world fire detection scenarios was identified.

WP3: Fire-Smoke Detection and Weather Data Collection

The main S&T result of WP3 is the development of algorithms and software for fire and smoke detection based on optical and IR cameras. The developed software enables early detection of fires in large open areas and can be used for the protection of forests and cultural heritage areas, as well as for the detection of fires in landfills, industrial areas (chemical fires, warehouses) and military training areas.

Computer vision based flame detection

Optical cameras and video-based algorithms provide an effective and low cost solution for the detection of flames at an early stage. However, video-based flame detection systems are affected by several limitations that challenge their performance such as the presence of sun reflections, car lights, bad lighting conditions, poor image quality, movements of fire-like coloured objects, etc. To overcome the aforementioned drawbacks, partners focused on improving existing algorithms and developing three new techniques for flame detection. The developed algorithms offer increased detection rates and lower false positive ratios compared to the literature.

BILKENT developed a covariance matrix based fire detection method for video sequences. The algorithm divides the video into spatiotemporal blocks and uses covariance-based features extracted from these blocks to detect fire. Both the spatial and the temporal characteristics of flame colored regions are exploited. Unlike other algorithms used for similar tasks, the proposed method does not use background subtraction, which means that it does not require a stationary camera for the detection of moving flame regions and can, therefore, be also used with moving cameras. This is an important advantage because fixed cameras may sway because of the wind or a PTZ camera can slowly pan an area of interest to detect fire.

CERTH developed a video based flame detection algorithm, which initially applies background subtraction and colour analysis to identify candidate flame regions on the video frames and subsequently distinguishes between fire and non-fire objects based on a set of five extracted features including color probability, spatial variation, temporal variation (flickering), spatiotemporal variance and contour variability of candidate blob regions. Classification is based either on classifiers trained with fire and non-fire video frames or on a rule-based approach.

Finally, SUPCOM developed a real-time flame detection system for video sequences captured by both fixed and moving (PTZ) cameras. First, moving objects are detected in each frame. Then, a set of flame characteristics including color, temporal intensity variance, spatial intensity variance, shape variation and shape complexity are extracted and classified as flame or non-flame using a set of fuzzy Context Independent Variable Behaviour (CIVB) classifiers.

Computer vision based smoke detection

Smoke observed from a long distance and smoke observed from up close have different spatial and temporal characteristics. Wildfire smoke appears to move very slowly after a couple of hundred meters and it does not exhibit turbulent behaviour when monitored by a video camera. Based on this observation, BILKENT developed two different algorithms for close range and long range (wildfire) smoke detection.

The close range smoke detection algorithm first detects regions with smoke-like color (grey/white) to decrease the search area in the frame. Then, the video is divided into spatiotemporal blocks and a set of covariance descriptors is extracted for blocks with smoke-like color. The long-range smoke detection algorithm consists of three main sub-algorithms: (i) slow moving object detection, (ii) smoke-colored region detection, and (iii) correlation based classification. Both algorithms obtain high detection rates while exhibiting increased robustness to false positives due to the presense of clouds, fog or moving objects.

BILKENT also developed an Entropy functional based Online Adaptive Decision Fusion (EADF) framework for image analysis and computer vision applications and improved former smoke detection results. In this framework, sub-algorithms of the smoke detection module are combined with weights, which are adjusted online according to the changing light and environmental conditions. In this way, it is possible to fuse the results of several smoke detection algorithms analyzing the scene in parallel. Improved performance was achieved for wildfire smoke detection at an early stage.

Finally, SUPCOM investigated the possibility to embed an early smoke detector into the camera in order to send a quick alert to the monitoring center immediately after video acquisition. Generally, the two video compression standards MJPEG and MPEG2 are commonly available in most cameras. They both involve a blockwise Discrete Cosine Transform (DCT). SUPCOM’s first contribution for designing such smart camera functionality consists of exploiting the local fractal feature of smoke areas based on the DCT coefficients. The second novelty relies in refining the estimation of the fractal feature by considering larger blocks of coefficients to increase detection accuracy without increasing the complexity. This technique could be very useful in low bit-rate transmission applications.

IR camera based fire detection

Three prototype multispectral cameras were developed by XENICS. Meerkat Fusion consists of three cameras at different wavebands: visible, SWIR and LWIR. Meerkat PTZ consists of two cameras at visible and LWIR spectrums. Finally, Meerkat Fix consists of a LWIR and a SWIR camera. The cameras are capable of recording videos simultaneously and transmit the videos to a computer through Ethernet.

XENICS made an in-depth analysis for infrared data processing. Each aspect of the infrared radiation was studied: (i) physical aspect (physical infrared radiation modelling and effect of environment), (ii) electro-optical system aspect (infrared radiation detection, including image contrast and background noise influenced by lenses, camera resolution and other sensor characteristics) and (iii) image processing aspect (infrared data analysis).

Three approaches were implemented by XENICS for automatic infrared fire detection: the first two approaches are designed for long range fire detection based on SWIR and LWIR cameras. The image processing is performed separately in each waveband and the fusion is performed at the decision level. These two methods take advantage of the high dynamic of these cameras. The third approach is designed for short range fire detection based on both LWIR and visible cameras. The multi-sensor flame detection algorithm first searches for candidate flame objects in both LWIR and visual images based on moving object detection and flame feature analysis. Next, the registration information is used to map the LWIR and visual candidate flame objects on each other. An alarm is issued if the flame probability of the mapped objects is high.

Other methods were also investigated such as the detection of the potassium lines or the Time of Flight flame detector. The first method is better fitted for airborne fire detection because the forest could occlude the potassium emission. The potassium lines are seen at the start of the fire because the potassium is released at the beginning of the fire when the temperature is high. These potassium emission lines could be occluded by the dark smoke. The Time of Flight flame detector is best fitted for short-range fire detection.

BILKENT also developed an IR-based flame and smoke detection algorithm. The algorithm starts with segmenting moving hot objects. Then several features including boundary box disorder, histogram roughness, principal orientation disorder, center mass disorder, etc. are extracted from the segmented regions and are used to decide whether the hot moving objects represent fire or not..

PIR sensor based fire detection

A flame detection algorithm using pyro-electric infrared or passive (PIR) sensors was developed by BILKENT. Two versions of this algorithm where implemented. The first can run on a PC. The second is implemented as a standalone system using digital signal microprocessors. A differential PIR sensor is only sensitive to sudden temperature variations within its viewing range and it produces a time-varying signal. This signal is analyzed using Markov models corresponding to the flame flicker process of an uncontrolled fire, ordinary activity of human beings and other objects.

Comparative results show that the system can be used for fire detection in large rooms. Conventional point smoke and fire detectors typically detect the presence of certain particles generated by smoke and fire by ionization or photometry. An important weakness of point detectors is that the smoke has to reach the sensor. This may take significant amount of time to issue an alarm and therefore it is not possible to use them in open spaces or large rooms. The main advantage of the differential PIR based sensor system over the conventional smoke detectors is the ability to monitor large rooms and spaces because it analyzes the infrared light reflected from hot objects or fire flames to reach a decision. The PIR system shows very high detection rate and low false alarm rate. A flood detector, which works by sensing if water is present at its contacts, was also integrated in this board. This standalone system presents an ideal low-cost solution for fire and flood detection in large rooms.

Software platforms for fire/smoke detection

The video-based smoke and flame detection algorithms developed within WP3 have been integrated in two software platforms: a standalone software platform used for algorithm testing using pre-recorded videos and an online software platform that can be connected with a range of different cameras and can detect fire in real-time. The on-line platform can support many cameras at the same time (PTZ cameras also) and provides a user friendly interface that allows the display of live streams from the cameras, issue of visual and audio alarms at real time, addition of new cameras, adjustment of camera parameters, selection of flame/smoke detection algorithm to use, display of camera positions in Google Earth, etc. The online smoke and flame detection platform developed by BILKENT has been installed in many watch towers in Turkey in cooperation with the Turkish General Directorate of Forestry. A software platform for the detection of fire in pre-recorded IR sequences was also developed by XENICS.

Weather data collection

Weather measurements are collected from local weather stations installed at the test sites. Polling of real time weather and forecast data from public weather sites and national meteorological agencies is also supported. Commercial weather stations were installed in most pilot sites, while in Kabeirion a prototype weather station, especially designed for FIRESENSE, was employed.

WP4: Wireless Sensor Network

The main results of the research performed within WP4 is the development of a novel wireless sensor network (WSN) for monitoring environmental parameters, which coupled with a wireless data network (WDN) for video surveillance can be used for the detection of fire.

Wireless Sensor Network

A novel wireless sensor network was designed and developed by CERTH. The design approach primarily targeted at a robust solution for outdoor deployment that can operate on cheap-to-replace batteries for a long period, while maintaining a good response to sensing a wildfire. Moreover, given that the WSN is part of a critical system to which responsible personnel relies for early warnings, any WSN subsystem malfunction should be reported immediately and effectively along with its location. For that reason, a clustered network hierarchy is designed with clusters of approximately 10 sensors governed by a dual-radio clusterhead device with more resources than the sensor platform. Clusterheads uses low range, ultra-low power radio to communicate with sensors that can be placed around them at a distance of 80 meters at most. The clusterheads themselves are collecting local sensor measurements and send them via long range WiFi (upto 200 m) to a WSN aggregation point. In addition, the clusterheads are passively monitoring local sensors and their behavior and report problems immediately. Within FIRESENSE, the clusterheads implemented advanced sleep scheduling and multi aggregation point selection algorithms, which increase robustness and fault tolerance.

The WiFi network formed by the clusterheads is an ad-hoc network that introduces full deployment flexibility since a clusterhead can be placed beyond the 200 m barrier from the aggregator point and use neighbour clusterheads that are closer for measurement propagation in a multi-hop manner.

A considerable amount of effort was invested in integrating the WSN in a controllable and effective way with the Data Fusion module. For that reason, an HTTP-based REST API with human readable XML responses was used to drive the WSN. The API recovers from malfunctions during request processing like network loss or device failure and returns appropriate messages for all error conditions, so that the data fusion can handle them. The integration with data fusion included a long testing period and the behavior of the WSN has been selected to bridge the Data Fusion response demands with the low power requirements. As a result, the Data Fusion engine can effectively interpret temperature and humidity readings to discriminate between fire and direct sun light exposure, which is one of the main problems in WSN-based fire detection.

WSN security has been thoroughly considered and several security extensions have been implemented to both the WiFi adhoc network and to the authentication of REST API messages to avoid man-in-the-middle attacks that wish to deceive the control centre. The WSN has been thoroughly tested and has been deployed in Kabeirion, Rhodiapolis and Dodge Hall test sites.

WSN routing and scheduling

To efficiently utilize WSN energy and to effectively and reliably carry the sensed data to the control center, new WSN routing and activity scheduling protocols were designed and implemented as well. First, efficient routing algorithms and MAC algorithms have been surveyed and two suitable routing protocols and a MAC protocol have been developed by BOGAZICI. Routing algorithms must address the problem of congestion, which is a major source of data loss in wireless sensor networks adversely affecting the reliable data delivery and consequently the success of the fire detection and monitoring applications. Fire detection is an event-triggered application and given the dense deployment of the sensor nodes, many of the sensor nodes detect the same fire event and create a burst of emergency data destined to the sink, which creates transient local congestions in the network. To address these issues, BOGAZICI has developed a Load Balanced Reliable Forwarding (LBRF) algorithm and modified it to its cross-layer geographic forwarding scheme, which aims to increase the reliable data delivery in wireless sensor networks by avoiding congestion using a distributed and a dynamic load balancing approach.

Deployment of multiple sinks (multiple aggregation points) is another solution for the congestion problem in sensor networks, which also provides extra benefits in terms of energy-efficiency and reliability. A multi-sink sensor network is also more robust against the inaccessibility of a sink node due to single point of failures such as node failures (energy exhaustion), node destructions (fire) or communication destructions (fire). Hence, BOGAZICI also developed a routing algorithm for multi-sink sensor networks for fire detection. The proposed algorithm is implemented in the clusterhead platform developed by CERTH. The algorithm has been equipped with a fuzzy decision mechanism for choosing the aggregation point (sink) to be used for each event report. In the case of fire, the target sink may be blocked or destroyed. Multisink routing will allow the WSN to report the emergency events and the temperature readings even while some of the sinks are unreachable.

Considering that the fire risk can be low during some periods of time, or deployment could be dense, not all sensor nodes need to be active all the time. Some of them can be put into sleep mode for a while. By appropriate sleep scheduling (which is also referred to as node activity scheduling or duty cycling), energy consumption can be reduced in wireless sensor networks, especially when the deployment is quite dense, which is the case expected in wireless sensor networks used for fire detection and management. BILKENT has proposed a distributed and energy efficient sleep scheduling and routing scheme that can be used to extend the lifetime of a sensor network while maintaining a user defined coverage and connectivity. The scheme can activate and deactivate the three basic units of a sensor node (sensing, processing, and communication units) independently. A simplified version of this scheduling scheme was specified and incorporated into the sensor network platform developed by CERTH.

Additionally, robust cooperative networking approaches (CWSNs) were developed by SUPCOM to reliably and efficiently carry sensory data over multiple hops and groups of nodes to the center. The main idea consists of selecting, sequentially, a group of relay nodes at each hop, called cooperative nodes. For a group of cooperative nodes, a cost function is attributed to each node. Depending on the value that a cooperative node has compared to its neighbors, it is decided whether it is the best node among its group members to forward the current data packet. The decision depends on the policy chosen for attributing cost values. Two policies were studied by SUPCOM: the first focuses on distances between nodes; the second focuses on link reliability between forwarding nodes (RSSI) and the residual battery energy of cooperative nodes. The cooperative communication approach enhances the network lifetime by reducing the average packet delay and the number of retransmissions.

Wireless Data Network

To carry video and sensory data, a long range wireless data network (WDN) is designed as well. Several different technologies (GSM/GPRS/3G, WiMAX and WiFi) have been investigated and the decision converged to use a WiFi based solution. Point-to-point and point-to-multipoint WiFi links with directional antennas and longer range transmission capabilities are selected and used in pilot sites. Based on the needs and unique characteristics of each site, different WiFi based WDN networks were designed, integrated and installed in each site.

Testing and evaluation of the network

Extensive laboratory and field tests with real experimental fires were performed to evaluate the effectiveness and performance of the WSN and the WDN. Additionally, installations and tests were performed in FIRESENSE pilot sites as well. The experiments were also helpful to tune the performance parameters and update the WSN-based fire detection algorithms. Experimental results show that the developed WSN is capable of robustly carrying temperature and humidity values to the data fusion engine and successfully triggering alarms in case of fire.

WP5: Multi-Sensor Data Fusion and Estimation of Fire Propagation

The main S&T results of WP5 include: a) new algorithms for multi-sensor data fusion, b) new algorithms for vegetation classification based on satellite images, c) ground surveys for the estimation of fuel models, d) development of a Google Earth based application for the estimation and visualization of fire propagation and e) development of the FIRESENSE control centre for the monitoring of culture heritage areas for the risk of fire and extreme weather conditions.

Multi-sensor data fusion

Several techniques for fusing data from different sensors have been examined within Firesense aiming at increasing fire detection rates and reducing the number of false positives.

First, the concept of image fusion at the pixel level was investigated. To this end, CWI focused on fusion of images obtained from visible and IR (i.e. SWIR and LWIR) cameras. One fundamental problem in this respect is that these images need to be co-registered, a task which is complicated by the fact that image properties and characteristics of corresponding points might be quite different, especially when one compares visible or near IR images with images for long wave IR (thermal information). In particular, there are many instances where the statistics of image patches around points of interest are unrelated and therefore difficult to compare (e.g. due to opposite contrast). A feature-based registration algorithm was designed, which uses lines derived from edge pixels (instead of points of interest as the majority of existing algorithms) as edge-correspondences are easier to identify. Experimental results show that this technique outperforms other existing algorithms when registering images captured by an infrared camera and a visual camera.

In the second strand of this research, CWI investigated fusion schemes that can handle the above-mentioned opposite contrast. This is something that occurs quite frequently when a scene is observed in both the visible and MWIR or LWIR spectrum (i.e. predominantly thermal information). Bright patches in visible light might actually be cooler than their surroundings and therefore darker in the IR images. CWI developed an approach in which the fusion rule selects one of the modalities as base image and uses a saliency rule to decide whether or not contrast contributions from the other modalities need to be inverted. A software tool for image registration and fusion providing a graphical user interface has been developed to this end.

CWI concluded its work on image fusion by implementing a saliency-aware multi-modal fusion algorithm in which the wavelet-based fusion is biased to assign a greater prominence to the thermal image wherever in thermal saliency is detected. The result therefore shows an image which strongly resembles the visual input image and is therefore more easily interpreted by a human operator, but at the same time highlights any thermally salient region (e.g. fire) that might be present.

Image fusion at the pixel level requires accurate image registration, which is difficult when the cameras are pointed at natural scenes in which few rigid and fixed keypoints are visible. For that reason, ways of combining the decision information obtained from visible and thermal cameras were also investigated. More specifically, by defining thermal saliency in terms of motion and (thermal) brightness it becomes possible to combine the bounding boxes for flame detection that are obtained in both the thermal and visible images in order to reduce the number of false positives without decreasing the sensitivity of the system. In a related development, CWI has collaborated with partner XENICS to implement a data-driven thresholding algorithm that takes advantage of IR-sensor integration time optimization. The underlying idea is that a judicious choice of the IR sensor integration time allows one to optimize the contrast between hotspots due to fire and the cooler background. It then becomes easier to detect the onset of fire.

CWI also developed a data-driven event-detection algorithm for the data streams emanating from temperature nodes in wireless sensor networks, and tested it for the WSN architecture developed by CERTH. In addition, CWI implemented algorithms for fire risk index computation and extreme weather warnings, based on sudden changes in temperature, barometric pressure and wind speed/direction as measured by the meteorological sensors.

Most of the aforementioned data fusion algorithms are integrated in the data fusion software module, which plays a central role in the generation of alarms and the communication between the cameras, sensors and control centre.

Vegetation classification and fuel modelling

A fuel model is a preliminary representation of vegetation characteristics used in analyzing fire behaviour. One of the objectives of the research performed within FIRESENSE was to propose new techniques to improve the vegetation and fuel model description of the selected pilot sites.

Towards this aim, several algorithms for vegetation classification have been developed by SUPCOM and CNR based on satellite images. Commercial satellite images have reached a fairly high spatial resolution, which allows more powerful textural analysis and more detailed description of soil surface. This improves the capacity to recognize and classify land uses, the amount and typology of vegetation and other potential sources of fuel for wildfires. It also reduces substantially the time and costs for updating vegetation and fuel distribution.

A two-step approach for vegetation classification was proposed by SUPCOM, which applies multi-band supervised SVM (Support Vector Machine) classification followed by temporal analysis. In the first stage, spatial, spectral and textural features characterizing vegetation are extracted from the original data and are combined with the spectral information through the SVM algorithm. The spectral information is introduced through the normalized difference vegetation index (NDVI), the texture features are extracted using Gabor wavelet decomposition and the spatial interaction within each channel is considered by taking into account the ground spectral responses according to the different spectral bands. In the second stage, the temporal behaviour of the land cover is analysed. The type, location and time of the vegetation changes are tracked and the parameters of the fire propagation model are updated accordingly. This spatio-temporal classification approach based on the combination of several features was applied to satellite images of the pilot sites: Galceti Park, Temple of Water, Kabeirion and Rhodiapolis. The results give a good discrimination between classes for all sites, especially for Kabeirion where the vegetation areas are well delimited. The generated vegetation maps were geo-referenced to allow their use by the EFP module. To produce the fuel map for fire propagation estimation, the CORINE nomenclature was used.

CNR also developed a vegetation classification algorithm, which works as follows: first, a pixel-based preliminary classifier is used which explores all available spectral information from multi-temporal satellite images and maps each pixel into a discrete and finite set of labels that identify land-cover-class sets: a) vegetation; b) either bare soil or built-up; c) either water or shadow; d) snow or ice; e) clouds. The second stage includes stratified class-specific context-sensitive feature extraction. In this context, vegetation can be subdivided into evergreen forest, deciduous forest, woody cultivated areas (e.g. olive groves, orchards), grassland and shrub.

CNR evaluated a procedure for estimating the biomass/fuel based on allometric relationships. Allometric models represent a non-destructive and time-efficient alternative for determining biomass indirectly and are based on correlations between biomass and easy-to-measure vegetation parameters, such as basal diameter, shrub height, or vegetation cover. A fuel model was developed in order to obtain a spatial distribution of fine and coarse fuel through remote sensing (Very High Resolution satellite images were used). The model was based on empirical relationships (regression analysis) formulated between the biomass and NDVI. Once the regression parameters were determined, an estimate for the biomass at any pixel in the image (provided vegetation is growing here) could be computed.

Creation of fuel maps can also be carried out by site survey by experienced forestry researchers. In Kabeirion, such a survey was conducted by Dr. Gavriil Xanthopoulos, an expert in fuel modelling and researcher at the Institute of Mediterranean Forest Ecosystems and Forest Products Technology of the National Agricultural Research Foundation of Greece, who was awarded a subcontract by CERTH. This work involved a detailed comparison of satellite images and vegetation at on-site sampling points, judiciously selected to provide accurate and detailed ground truth on fuel-related parameters. The survey has resulted in the identification of 13 existing fuel models (no custom models were used). This made it possible to manually define a ground truth fuel map for a small area around the archaeological site of Kabeirion. This information was then utilized by SUPCOM to producing a more accurate map of the vegetation cover in this area.

A second survey concerned the use of physical and/or hybrid models for the estimation of the probability of fire ignition and spread in the same test site. A subcontract was assigned to OMIKRON LtD (which also participated in the FP7 FireParadox project), which conducted a study for modelling a large rectangular area of approximately 12kmx6km around Kabeirion. The main motivation was to integrate a physical model-based fire spread technique into the EFP (estimation of fire propagation) application in order to be able to compare the results of such models with those provide by the semi-empirical BEHAVE-based models used in the EFP.

Estimation and visualization of fire propagation

Visualization of fire propagation data is important since it enables early intervention of the fire and helps the fire management teams to deploy their forces wisely. To this end, an interactive software application, which provides parameterized simulations of the fire propagation and visualizes the estimated results on a Google-Earth based environment, was developed by CERTH. Fire spread calculations are based on the popular BEHAVE algorithm, while physical models simulations are also supported via the open source VESTA software.

According to the BEHAVE model, fire propagation depends on a number of parameters such as the terrain topology around the ignition points, weather conditions such as wind and moisture, as well as fuel information. The latter factor summarizes vegetation characteristics that have a major impact on fire propagation. These parameters are either provided by the user or are automatically estimated based on available data. Specifically,

• Wind information is obtained in real time from existing weather stations or from internet weather portals.
• Digital Terrain Models (DTMs) are obtained by STRM data (90m resolution) that are freely available for the whole world.
• Fuel moisture information can either be manually provided by the user, directly measured using special sensors or estimated from weather data (i.e. temperature, humidity, etc).
• Fuel information is provided either by CORINE land cover maps, which are converted to fuel maps (containing indices to BEHAVE or custom fuel models) or directly as fuel maps.
• Ignition points can be provided manually or obtained by the Control Centre (the Control Centre provides an estimation of the fire ignition point based on the outputs of the fire/smoke detection module and the data fusion module).

The estimated fire propagation is visualized on a user-friendly 3D environment based on Google Earth. Google Earth was chosen because it is publicly available and widely used by experts and non-experts alike. Besides static views, it also allows the creation of impressive 3D animations of the fire propagation. Google Earth also provides standard access to useful layers that create functionalities and added value for the user such as real time weather information, street view, borders and labels, and Panoramio photos. Additionally, there are layers for important information about the monitored site such as the location of sensors and cameras, vegetation maps and regions of interest. Influential environmental parameters such as the fire ignition point or humidity values can be set interactively, and weather data are automatically acquired, e.g. from onsite or nearby weather stations.

The EFP application produces 2D or 3D visualizations of the fire propagation estimation output: ignition times are displayed as colour coded grids, while flame length animations in KML format are also generated. Support for adaptive resolutions in fire propagation estimation has been implemented, allowing the simulation to run faster, while still attaining the same resolution and performance within the regions of interest. Real-time data from the WSN sensors can be exploited within the EFP simulations by modifying the associated fuel map. In addition, the software offers support for improved spatially varying wind fields using the WINDNINJA client interface. Moreover, moisture estimates can be obtained either from weather parameters using Nelson’s dead moisture model or from specialized sensors (e.g. the 10-h sensor). A model for predicting the probability of transition of a surface fire to a crown fire of the vegetation was also implemented.

Finally, CERTH in cooperation with OMIKRON LtD integrated VESTA (open source Large Scale Fire Simulator software) with FIRESENSE EFP, so that a) a Fire Risk map can be displayed and b) simulations of fire spread using physical or hybrid models can be executed via the EFP GUI.

FIRESENSE Control Centre

SUPCOM designed and implemented the FIRESENSE Control Centre (CC) and its related Graphical User Interface (GUI). The architecture of the CC application and its connections to the other modules is outlined in Figure 19. The Control Centre allows:

• display of layered maps of the monitored site;
• monitoring of the site through different sensors and cameras;
• collection and visualization of measurements from the sensors and videos from the cameras;
• configuration of sensors and manipulation of cameras;
• display of sensor statistics;
• generation of different types and levels of alarms when a situation is judged as suspect or dangerous by the data fusion module.

For efficient visualisation, the GUI comprises of three graphical interfaces: the main screen, the video screen and the maintenance screen. The main screen shows the map of the supervised area and the location of installed equipment (WSN sensor, cameras, PIR sensors, weather stations, communication links are presented as icons) and displays the alarms. The user can click on a WSN sensor icon to get information about its status, incoming data flow and parameters that are reconfigurable. In the same way, clicking on a camera icon allows the display of a small window showing the video acquired by this camera in real time.

The alarms generated by the Data Fusion module are displayed on the Main Screen as shown in Figure 20: a visual and sound alarm is displayed on the top along with a text box indicating the alarm type (e.g. smoke). A bounding box and a flag indicating the alarm location (defined by the GPS coordinates of the ignition point) is also shown on the Google map. The main screen also allows the user to trigger the EFP application, which visualizes the estimated fire propagation.

The video screen is dedicated to the cameras and their manipulation. This interface is composed of two sub-windows. The biggest one is dedicated to the display of the video stream of a given camera selected by the user. The second sub-window displays a mosaic of all the video streams from the remaining optical and IR cameras. Furthermore, the user can adjust camera parameters (e.g. pan-tilt-zoom of PTZ cameras). In case of a fire alarm, the cameras can automatically turn to monitor the area surrounding the estimated ignition point on user demand. Finally, on user demand, the video screen offers the possibility to display the result of fusion of visible and infrared images acquired by the optical and IR cameras connected to the CC based on the image fusion algorithm described above.

The maintenance screen allows controlling the status of the sensors/cameras/communication links. For each sensor, its energy consumption can be displayed, while an alert is issued in case of dysfunction or breakdown (e.g. vandalism or battery out of order).

WP6: System Integration and Demonstration

In the context of WP6, the FIRESENSE system was installed and demonstrated in five cultural heritage sites: Kabeirion (Thebes, Greece), Rhodiapolis (Antalya, Turkey), Dodge Hall (Istanbul, Turkey), Temple of Water (Djebel Zaghouan, Tunisia) and Monteferrato-Galceti Park (Prato, Italy). The five pilot sites have quite different characteristics: Dodge Hall is a building; Kabeirion is an archaeological site mainly surrounded by cropland, where one of the major fire risks comes from burning straws and crop residues; the archaeological site of Rhodiapolis is surrounded by greenhouses and, closer to the ruins, by poor vegetation of brushes and pine trees; the Temple of Water is surrounded by a relatively dense forest and by some cropland; Galceti is at the border between a pine wood, some cropland and town outskirts. The local orographic situation is also quite different among sites, with Galceti and the Temple of Water exemplifying relatively steep, uneven landscape and Kabeirion situated in a relatively flat area.

System installation at pilot sites

The first prototype was installed in the archaeological site of Kabeirion, 8km west of Thebes, Greece. The sanctuary of Kabeirion was dedicated to gods Kabeiros, Demeter and Pais. It is located at an area which was isolated at ancient times deep in the quiet countryside as the rituals that were performed there had to be kept secret and only the initiated could attend the ceremonies. The sanctuary complex consists of the temple, the shrine, a theatre, a gallery and other buildings. It covers approximately 7,000 square meters. Nowadays the remains are surrounded by olive trees, pine trees and cereal cultivation.

The system installed in Kabeirion includes the following components: 2 PTZ cameras and 1 SWIR camera installed inside the archaeological site, a Wireless Sensor Network comprising of 50 sensor nodes (including temperature and humidity sensors) and 5 clusterheads, a prototype weather station installed at the museum of Thebes and the control centre PC installed also at the museum. A wireless IP network link with a reflection point has been deployed to connect the archaeological site with the museum of Thebes.

The second prototype was installed in the ancient city of Rhodiapolis in Antalya, Turkey. The most important remains of the city include the theatre, bathhouse, agora/stoa, sebasteion, temples, church, cisterns, cenotaph, necropolises and houses. The site is surrounded by forestall areas. An excavation house used by archaeologists during the excavation period is located near the site.

The system installed in Rhodiapolis includes the following components: 4 PTZ and two fixed optical cameras, a multispectral image sensing platform (including visible, SWIR and LWIR cameras), a WSN comprising of approximately 20 sensor nodes (including temperature and humidity sensors), a commercial weather station and the control centre PC, which is installed at the excavation house.

The third prototype was installed in Dodge Hall in Istanbul, Turkey. Dodge Hall is located on a wooded hill overlooking the Bosporus, in the campus of Bogazici University. It is listed as a second degree heritage building by the Turkish Ministry of Culture. Dodge Hall has a large interior space similar to most buildings in Istanbul (e.g. Agia Sophia, Blue Mosque). It is demonstrative of building type monuments and has to be protected from both external and internal fires. It is an actively used indoor historical site for sport activities, physical education courses and university administration activities.

The system installed in Dodge Hall includes the following components: 2 optical cameras - one installed inside the Hall and the other monitoring the building from the outside, a WSN comprising of 5 sensor nodes (including temperature and humidity sensors) and a gateway, a PIR sensor, 3 commercial smoke sensors and the control centre PC.

The fourth prototype is installed in the Temple of Water in Tunisia, which is encompassed by the national park of Djebel Zaghouan. This temple marks the site of an aqueduct and a canal network built in the 2nd century BC under the Roman Emperor Hadrian, which used to carry water from the city of Zaghouan to the city of Carthage over 130 km away. The pilot site corresponds to an area of 3,000 ha and it is surrounded by a forest. The vegetation cover is typical Mediterranean including pine trees, mastic trees, olive trees, etc. An Eco-museum is located within the archaeological site, while the surrounding area is a leisure activities centre.

The system installed in the Temple of Water includes the following components: 2 PTZ cameras - one installed on the roof of the Eco-museum and the other on the roof of a nearby hotel, a commercial weather station and the control centre PC installed in the museum. A WiFi link connects the second camera with the museum.

The fifth prototype is installed in Galceti Park in Prato, Italy. Galceti Park is an open air multi-purpose site located in a Natura 2000 area, which includes a renaissance chapel, an archaeological Mousterian site, a Natural Science museum and a small zoo. It occupies a surface of 1,760 ha mainly including pinewoods.

The system installed in Galceti includes the following components: 5 optical cameras – four installed inside the site and the fifth on the roof of a nearby school, a commercial weather station that belongs to a network of weather stations in the area of Prato and the control centre PC installed in the Natural Science Centre. WiFi links connect the cameras with the control centre.

Prior to equipment installation, special permissions were received from archaeological, cultural or other public authorities for the installation of the system in the five pilot sites. Also, significant effort was required for the preparation of test sites for system installation, which included establishment of Internet connection for open air sites, supply of power, cleaning of vegetation, repairing of existing infrastructure or old buildings, etc.

System demonstration

Several activities were organized in all pilot sites to validate system functionalities and demonstrate the system to interested parties and stakeholders.

Initially, “internal” demonstration activities without user involvement were organized, including controlled fires and artificial smoke tests to validate system functionalities, verify the correct operation of sensors and test system performance. In Galceti, artificial smoke generated by smoke candles, small and controlled fires as well as smoke from bonfires lighted by farmers in the countryside were used to test the fire/smoke detection capabilities of the system. Moreover, a set of experiments with controlled bonfires were also organized to test the IR cameras and WSN sensors. Several real fire experiments with cameras and WSN sensors have been organized in Volos and Kabeirion. Another real fire test was organized in the parking lot of Bogazici University in cooperation with the local fire department. Dry ice and boiling water were used for smoke generation in an indoor test in Dodge Hall. Two field tests with visible and IR cameras were conducted in Rhodiapolis. Several tests with small controlled fires were also organized in the Temple of Water in cooperation with local authorities. All the aforementioned tests also served for collecting sensor data (videos, signals, measurements) for system evaluation.

"Open" demonstration activities involving local users were also organized in all test sites. These events aimed at demonstrating FIRESENSE technologies to potential stakeholders, the scientific community and the general public through presentations, talks, videos and real or video demonstrations and tests of the FIRESENSE system. Interaction with the audience gave valuable feedback for system evaluation and update of user requirements.

In Kabeirion, an open day was organized by the IX EPCA in collaboration with CERTH. The event was attended by people from local authorities, the Ministry of Culture, the fire department, security enterprises, interested inhabitants, etc. The workshop included several presentations and videos as well as a live demonstration of the system. In Rhodiapolis, an open demonstration was organized during the FIRESENSE workshop, including a real fire simulated by the fire brigade, which was successfully detected by the system. An open demonstration for stakeholders and University staff and students was organized in Dodge Hall. The event included presentations and live system demonstration with artificial smoke generated by combination of dry ice and boiling water and artificial flame coming from a lighter. Two open demonstration events were organized in the Temple of Water: people from fire fighting departments, the Tunisian General Directorate of Forestry and the University were invited. A live demonstration of the Control Center operation was performed by setting real controlled fires using tree leaves and branches to generate smoke. In Galceti, an open day including presentations and videos was attended by local authorities and stakeholders. In all these events, system demonstration was followed by discussion and filling of questionnaires by the attendees.

The FIRESENSE successfully detected two real wildfires near Kumluca, Antalya. The first wildfire started on 02/09/2012 near the ancient town of Olympos (Cirali) next to Rhodiapolis and the second on 18/10/2012 near the archaeological site of Rhodiapolis.

WP7: Technical Assessment and Evaluation

In WP7, an assessment plan was established for evaluating the performance of the FIRESENSE system and assessing its compliance with user requirements and expectations. The assessment process was performed in two stages: first party assessment of system and system components was carried out by the partners (laboratory testing during the development phase and field testing during the demonstration phase) and second party assessment was carried out directly by system users providing feedback via questionnaires.

Laboratory testing of system modules

During system development, the main components of the FIRESENSE system (video-based smoke and fire detection, infrared based fire detection, wireless sensor network based fire detection, generation of fuel maps by vegetation classification, fire propagation estimation, and data fusion) were evaluated in laboratory conditions.

Video-based smoke and fire detection algorithms were evaluated using a test video database consisting of several sequences of fire and smoke scenes at various environments, which is available at the project website. This database includes videos downloaded from the Internet as well as numerous videos recorded during field tests and experiments organized within the project. LWIR, SWIR, MWIR and PIR recordings are also included.

Experimental results show that the developed flame detection techniques obtain high recognition rates and low false positives and are robust to problems created by sun reflections and fire-like coloured moving objects (e.g. car lights). The long range smoke detection algorithm capabilities were successfully tested using sequences of real forest wildfires. The algorithm has increased performance compared to other algorithms in the literature and can successfully discriminate between smoke and similar phenomena such as clouds or fog, which usually produce many false alarms. The PIR-based fire detector was compared against conventional smoke sensors, which it outperformed both in terms of detection accuracy and time of response.

Several tests with IR cameras at different wave-bands (LWIR, MWIR and SWIR) and fires of different sizes at various distances from the camera were organized for the evaluation of IR-based fire detection techniques. False alarm generation due to the sun, sun glint and car lamps was also examined and useful conclusions were drawn for IR-based detection performance at different bands and conditions.

Several tests were also performed for assessing different components and parameters of the WSN such as temperature reading sensitivity, connectivity range between the Zigbee dongle and the sensor, battery life, worst observed delay, response to fire, etc. The proposed implementation was shown to have good performance in typical outdoor deployments and good response for fire detection.

Vegetation classification techniques were evaluated using ground truth data from site surveys and available vegetation maps. The algorithm for the estimation of fire propagation (EFP) was tested with historical data of past fires in Greece and Turkey where none or negligible human intervention occurred.

Finally, the data fusion algorithms were evaluated using data recorded during field tests with different cameras and WSN sensors. The algorithm for the registration and fusion of multimodal images (visible and IR images) was tested on pairs of IR and ViS images/videos describing indoor and outdoor scenarios. The proposed line-based registration technique showed clear advantages over the point-based registration. Observing the IR and visible sequences, only the fused image showed both the smoke and the fire clearly, deducing that the image fusion can be helpful for human visualization and inspection. Moreover, the fusion of visible and thermal data at the decision level showed that it can significantly reduce the number of false positives without reducing the sensitivity of the fire detection.

Assessment and optimization of WSN performance

The evaluation and optimization of the fire detection performance of the WSN was performed separately via OPNET simulations. To this end, BOGAZICI and CERTH investigated the event reporting capabilities of the WSN under realistic fire simulations. BOGAZICI implemented flat and Zigbee+WiFi based WSN models in OPNET and CERTH developed a realistic fire propagation estimation software module (EFP). Various temperature models that rely on heat transfer by thermal radiation were studied and integrated into the EFP software.

The integrated evaluation system works as follows: Given the fire ignition point, area morphology and fuel model and local weather conditions, the EFP module outputs the temperature changes at each sensor location at given time steps. The OPNET simulator uses this information for data generation. The sensor data generation model supports two modes of operation: normal and alert. The period for reporting temperature data is longer in the normal mode and shorter in the alert mode. Two temperature thresholds are defined: the first threshold determines when the sensors switch from normal operation mode to alert mode. The second threshold defines the temperature value over which the sensors are destroyed, affecting the routing decisions and accessibility to sinks and, hence, the fire detection performance of the WSN. Three OPNET node models were designed and implemented: Zigbee capable sensor device, Zigbee+WiFi capable hybrid local gateway device and WiFi capable central gateway device.

The OPNET models were tested in simulations driven by realistic fire scenarios produced by the EFP. The aim was to determine the effects of sensor and WSN related factors (number of deployed sensors, local gateway WiFi communication range, temperature threshold values, reporting frequency in alarm mode, etc.) on the fire detection performance of the WSN. The effect of weather and environment related factors (wind speed/direction, number/position of ignition points, fuel and moisture model, etc.) was also investigated. The metrics to be gathered from the simulations were determined as: a) freshness of temperature map, b) reporting delays, c) percentage of report losses, d) time between fire ignition and first alarm report received. Simulation results indicate that the freshness of the temperature map and the percentage of report losses is greatly affected by the alarm mode reporting frequency (higher reporting frequencies leading to more fresh temperature maps while increasing the report losses).; The reporting delays are influenced mostly by the local gateway WiFi communication range (longer ranges resulting in less delays). The time between the fire ignition and the first alarm report is affected by the number of sensors deployed (higher number of sensors deployed in the same area results in faster fire detection). These results provide an insight towards the optimization of the tunable WSN parameters to achieve the best fire detection and monitoring performance.

Technical assessment of the FIRESENSE system

The test scenarios to evaluate the FIRESENSE system basically involved building an artificial fire which generates heat, smoke and visible flames, and the system components detecting this fire, hence allowing the evaluation of the system’s fire detection performance. The control center software and the graphical user interface were utilized during the tests, and various visual and textual data gathered from the sensors were displayed and assessed.

The video-based smoke and fire detection performance was assessed in field tests organized in Rhodiapolis, Galceti, Temple of Water and Ankara. Fires of different sizes and at different distances from the cameras were set outdoors. Indoor tests with artificial smoke were also organized in Dodge Hall. In all cases, the system successfully detected smoke and flame and issued an early fire warning. The IR-based fire detection capabilities of the system were also evaluated in several tests organized in Belgium, Rhodiapolis and Galceti. The detection performance of MWIR, LWIR and SWIR cameras was assessed for fires at distances of upto 1 km in different environmental settings. The developed techniques have shown high detection accuracy and increased robustness to false positives caused by sunlight, moving objects, car lights, etc

The WSN-based fire detection system was evaluated in outdoor real fire tests conducted in Rhodiapolis, Dodge Hall, Ankara and Volos. The WSN sensors successfully detected temperature rise, while the proposed CUSUM-based methodology based on measurements obtained by clusters of sensors has shown increased robustness to false alarms caused by direct sunlight or ambient temperature rises.

The user-friendliness of the system and the control center user interface was assessed through questionnaires filled by system users. The users thought that the system is easy to use even by inexperienced personnel, its functionalities are clear and its interface is satisfactory.

A security evaluation of the system was also performed and conformity of subsystems to ISO/IEC 15408 was assessed. The FIRESENSE system uses an acceptable security scheme that uses subsystems which conform to the Common Criteria for data transmissions, and thus the eavesdropping from an adversary is very difficult. Nevertheless, the major concern of the FIRESENSE security scheme is to immediately detect system tampering or false data transmissions that can result in fire detection failures. This challenge has been addressed to both networking security and physical deployment level where practices that conform to ISO/IEC 15408 have been followed.

Finally, the system’s performance, functionalities and compliance with user requirements and expectations were thoroughly evaluated by 54 people employed in fire service, forest service, cultural heritage organizations, local authorities and research institutions through detailed questionnaires filled during demonstration activities. In general, the users found the smoke and fire detection performance of the system mostly satisfactory. Another important conclusion is that most of the users gave high ratings for the quickness of the fire detection performance. From the user friendliness perspective, the system GUI is found to be mostly average, but also a significant portion of the users think that the system GUI is easy to use. On the overall, most of the participants think that the FIRESENSE system is a novel approach in cultural heritage and environmental monitoring and can be applied to other areas too. The most important result is that, a large majority of the participants are eager to apply the FIRESENSE system in their organizations if accompanied by professional support by system developers and researchers.

Publications & patents

Results from the research carried out within the FIRESENSE project have been published in international conferences and scientific journals. More specifically, 23 journal papers and 35 conference papers have been published or have been accepted for publication. These papers cover several research areas related to FIRESENSE technologies such as fire and smoke detection based on visible video, fire detection based on IR data processing, PIR-based flame detection, WSN architecture and design, WSN routing protocols, WSN activity scheduling, cooperative WSNs, vegetation classification based on satellite images, estimation and visualization of fire propagation, image fusion, multi-sensor data fusion, OPNET simulations, etc. A detailed list of these publications is presented in Table A1 of Section 4.2. Moreover, two patents for a) the PIR-based fire detection system and b) a smoke detection system using nonlinear video analysis were applied. These are presented in Table B1 of Section 4.2.

Potential Impact:

Natural hazards do not respect national boundaries; therefore, coordinated and collaborative research is required at the European level to reduce the uncertainty, the unpredictability and the consequences of natural hazards. Since the loss of a cultural heritage site is irreversible, there is great significance in integrating the technological components required for the protection of these sites. Archaeological sites located not only in the Mediterranean region but across Europe can greatly benefit from the FIRESENSE system.

The innovation of the FIRESENSE project capitalizes both on basic and emerging research directions that demonstrate strong potential, but are not established yet as an integrated solution. A significant outcome of the project is the development of new algorithms for outdoor fire and smoke detection using regular video, IR cameras and PIR sensors. Various algorithms were developed for real-time fire and smoke detection. Experimental results have shown that the developed algorithms are robust enough to false alarms and yet capable of detecting fire in its early stages. In addition, a novel multispectral image sensing platform including a visible, a SWIR and a LWIR camera was designed and implemented within FIRESENSE. Another significant outcome of the project is the design and development of a novel WSN for outdoor deployment that can operate on cheap-to-replace batteries for a long period, while maintaining a good response to sensing a wild fire. To this end, new WSN routing and activity scheduling protocols were designed and implemented as well.

Since the developed system is based on multi-sensor technology (optical cameras, IR (short and long wave) cameras, PIR sensors, wireless temperature and humidity sensors and meteorological sensors), novel data fusion techniques were proposed and developed to increase the reliability of the system. Furthermore, a new software platform for estimating fire propagation (EFP) based on weather data, Digital Terrain Models and fuel information was developed. Novel techniques based on multispectral satellite images were also developed for automatic vegetation classification. The EFP platform offers a user friendly interface, which allows 3D visualization of fire propagation on Google Earth maps. Finally, the FIRESENSE control centre provides various functionalities to the end users such as: video on demand from cameras, weather data, a local map of the supervised area along with the sensors’ location, different levels of alarm etc.

The FIRESENSE project developed a powerful cost-efficient approach that can be used for the protection of cultural heritage providing:

• High reliability: The system utilizes different sensing technologies (CCTV cameras, PTZ cameras, IR cameras, PIR sensors, temperature and humidity sensors, and meteorological sensors). The different types of sensors operate independently
• Early detection of fire: Automatic detection of flame/smoke/rise in temperature.
• Forest fire management: The system estimates and visualizes the fire propagation based on the area’s fuel model (vegetation), the local weather conditions and ground morphology.
• Automation of the fire fighting: The output of the FIRESENSE system can activate water pipe networks for watering, like the fire sprinkler in buildings. Such water pipe networks are usually organized in sectors, which can be timely and separately activated in the areas threatened by the fire.
• Early warning for extreme weather conditions: Local weather stations provide useful sensor readings like temperature, wind direction and speed, relative humidity, barometric pressure, rain gauge etc. External weather forecasting is made available to the system as well, which makes it straightforward to use it as an early warning system for extreme weather conditions.

Furthermore, two significant features of the FIRESENSE system, which make it applicable to numerous archaeological sites across Europe, are:

• Modular architecture that allows for easy system upgrades and extensions depending on the particular needs of different archaeological sites.
• Protection of archaeological sites through non-destructive and non-intrusive intervention.

Taking into account the technological benefits of the developed technology, the FIRESENSE project is expected to significantly contribute to the protection of forested areas and the safeguarding of cultural heritage, particularly monuments and open archaeological sites. Forest fires cause adverse ecological, economic and social impacts such as:

• Life casualties and loss of properties;
• Loss of valuable timber resources;
• Degradation of water catchment areas resulting in loss of water;
• Loss of biodiversity and extinction of plants and animals;
• Loss of wild life habitat and depletion of wild life;
• Loss of natural regeneration and reduction in forest cover and production;
• Global warming resulting in rising temperature;
• Loss of carbon sink resource and increase in percentage of CO2 in the atmosphere;
• Change in the micro climate of the area resulting in unhealthy living conditions;
• Soil erosion affecting productivity of soils and agricultural production;
• Ozone layer depletion;
• Indirect effects on agricultural production;

Moreover, the proposed technology can be used in other sensitive areas and/or villages and towns located next to forests. As a result, the proposed system is expected to provide significant societal and economic benefits. More specifically, the FIRESENSE system will:

• locate high risk areas before the outbreak of fires and prevent human casualties and property losses;
• specify appropriate actions when facing forest fires, which can result in better management of resources and reduced loss of forested area;.
• protect forested areas of extreme cultural importance, which constitute a significant portion of the historical heritage in many European countries;
• have a positive contribution to environmental issues, as the forest fires are significant causes of air pollution, harmful carbon emissions, biodiversity loss through elimination of animal and plant species and water supply problems;
• reduce losses for natural hazards and prevent man-made hazards (forest arsons) from happening;
• contribute to the protection of cultural heritage, the basic asset on which tourism is built. Tourism, closely related to Cultural Heritage, is, at the moment the main industry in the world, with an increasing ratio of 12% of the Gross Domestic Product (GDP). This sector employs 8 million people in Europe and accounts for nearly 5.5 % of European GDP.

The dissemination activities included a series of actions that provided third parties outside the consortium with information relevant to the project aiming to increase interest from stakeholders and support the exploitability of the FIRESENSE system. The dissemination activities of FIRESENSE include the following:

• Project logo
• Project brochure
• Web-site
• Posters
• FIRESENSE dissemination videos
• Video demos and the FIRESENSE video database for fire/smoke detection
• Press releases and TV interviews
• Events / meetings
• FIRESENSE Workshop
• Clustering activities and events
• User group establishment
• Education of inhabitants in pilot sites
• Publications in international journals and conferences

The main goal of these activities was to raise awareness about the protection of cultural heritage from natural disasters such as wildfires, disseminate the project’s results, educate the inhabitants and exploit the FIRESENSE product.

A full list of FIRESENSE dissemination activities is presented in Table A2 of Section 4.2.

Project logo

The project logo presents the project scope and technology in the simplest way. The aim of the design is twofold: a) it emphasizes the use of sensor technology for early detection of fire, while b) the meander symbolises the protection of cultural heritage.


One of the first priorities in terms of dissemination was the preparation of the project brochure. The brochure was used to support project presentation in different events / meetings and has also served as a communication tool for potential mailings to groups of users. Except for the English version, the brochure was also translated and printed in Italian, French and Turkish.

Apart from providing an overview of the project consortium, and some contact information, the four-page brochure mainly focused on the project concept and aims, offering an overview of the overall architecture and modules. The pilot applications are presented as well. Having both end-user-oriented as well as technical content, the brochure is attractive for all target groups identified by the project. Several brochures were distributed to a large number of potential stakeholders. An electronic version of the brochure was also made available for download on the project website, and was provided in an accessible format.


The FIRESENSE website is a tool for general dissemination activities presenting the project to the outside community and facilitating on-line collaboration between partners. The FIRESENSE website is reachable via the domain It contains a public and a private part. The website is updated on a continuous basis by ITI-CERTH.


Posters were used at events such as exhibitions and international fairs in which partners participated. All posters followed a similar design approach. At the top of all posters there was the EU flag and the logo of both the FP7 and the Environment programmes, as well as the banner of the FIRESENSE website. An electronic version of all posters produced for dissemination purposes is available for download on the project website. The subjects of the designed posters include: a) general description of the FIRESENSE project, b) technical description of the FIRESENSE system and c) FIRESENSE technologies puzzle.

FIRESENSE dissemination videos

During the project, three videos (long, middle and short version) that demonstrate the objectives and results of FIRESENSE were created. These videos are available in a streaming format to the general public through the project’s website and popular video-sharing websites. Videos were also delivered in CD or DVD format to all interested parties (EU, User Group, other potential users/buyers, etc.) and were also displayed in exhibitions, conferences and/or other dissemination activities. The main focus of this audiovisual material was to inform the broad public about the FIRESENSE overall benefits with respect to efficient fire detection and management for the protection of cultural heritage areas and demonstrate the advantages of the proposed multi-sensor system over existing technologies.

Press releases and TV Interviews

The dissemination activity of FIRESENSE includes the publishing of numerous press articles, interviews and press releases. These can be downloaded from the project website. Briefly, TV interviews, newspaper articles and articles in the web are presented below:

• TV interviews in Spanish, Italian, Greek and Turkish media

o Interview of Nikos Grammalidis on CANAL PATRIMONIO, Spain (12/11/2010).
o Press conference in Galceti park, Prato, Italy (10/6/2011)

 Prato TV,
 Toscana TV and
 RTV38

o Interview on ET3 - Greek National TV Channel (23/6/2011)
o Interview on TRT Channel - Turkish Radio and Television (8/7/2011)
o Documentary about Rhodiapolis on Turkish television channel NTV (23/7/2011)
o News report on Turkish Channel 24 (5/7/2012)

• Articles in Greek, Turkish and Italian newspapers

o Hurriyet (27 July 2010)
o Il Tirreno Livorno (19 August 2010)
o Aggelioforos (6 May 2011- Open Day CERTH-ITI)
o Il Tirreno Prato (11 June 2011)
o Antalya (5 July 2011)
o Article in PARLIAMENT Magazine, Issue 335, pp. 57 October 2011
o Milliyet Ankara (4 February 2013)


• Articles and interviews that were released in Greek, Turkish and Italian web-sites



Events / meetings

FIRESENSE project and research achievements were presented in the following events / meetings:

• AR & PA Innovation (Valladolid, Spain, 11-14 November 2010 and 24-27 May 2012)
• Infosystem 2010 (Thessaloniki, Greece, 8 October, 2010)
• Open Day organized by ITI-CERTH (Thessaloniki, Greece, June 2010 & May 2011)
• Meeting of Prof. Enis Cetin with officers from the Turkish Directorate of Forestry (30 October 2011)
• Workshop organized by Prof. Enis Cetin and Prof. Ibrahim Korpeoglu during the URSI General Assembly and Scientific Symposium of International Union of Radio Science – (URSIGASS2011) (Istanbul, Turkey, 13-20 August 2011)
• Special session on "Signal processing for disaster management and prevention" organized by Prof. Enis Cetin and Dr. Nikos Grammalidis during the 19th European Signal Processing Conference (EUSIPCO2011) (Barcelona, Spain, 31 August 2011)
• Lecture on "Flame detection and fire propagation estimation" by Dr Kosmas Dimitropoulos (CERTH, Thessaloniki, Greece, 5 October 2011)
• Presentation about "Supervised vegetation Classification for Fire Propagation Estimation" to the United Nations/Vietnam Workshop on Space Technology Applications for Socio-Economic Benefits by Prof. Ferdaous Chaabane from SUPCOM (Hanoi, Vietnam,11 October 2011)
• 4th Archaeological Meeting of Thessaly and Central Greece (Volos, Greece, 15-18 March 2012)
• In the cultural association “Laios” by the director of THEPKA Mrs Alexandra Harami who talked about the archaeological site of Kabeirion (Thebes, Greece, 24 June 2012)
• 4th International Euro-Mediterranean Conference on Cultural Heritage (EuroMed 2012). A stand was used for presentation and dissemination of FIRESENSE. In this conference the paper "Flame detection for video-based early fire warning for the protection of cultural heritage" authored by K. Dimitropoulos,O. Gunay, K. Kose,F. Erden, F. Chaabene, F. Tsalakanidou, N. Grammalidis and E. Cetin was awarded with Best Full Paper Award (29 October - 3 November 2012, Lemesos, Cyprus).


A two-day Workshop was organized in Antalya, Turkey on 8-9 November 2012 to disseminate and validate the FIRESENSE outcomes. The aim of the International Workshop on Multi-Sensor Systems and Networks for Fire Detection and Management was to bring together researchers from all over the world, who deal with fire detection and management using multi-media and multi-sensor devices and networks. A number of key stakeholders were invited; the Workshop was also open to all interested parties. The objective of this workshop was to widely disseminate the FIRESENSE concept and diffuse results achieved in pilot sites. Discussion about the validation of the FIRESENSE system and its exploitation strategy were also made.

A call for papers covering different areas related to fire detection and management was published on 31/07/2012. 12 papers were submitted and were presented in the Workshop in three regular sessions covering the following topics: i) “Video-based wildfire detection”, ii)” Wireless Sensor Networks” and iii) “Wildfire prediction and readiness”.

The Workshop also included presentations given by invited speakers from the Turkish General Directorate of Forestry, the Tunisian General Directorate of Forestry and the Greek Fire Service, as well as a presentation by an Italian expert in wildfire prediction.

Figure 37: FIRESENSE Workshop, Antalya, Turkey, 8-9 November 2012.

Clustering activities

The FIRESENSE consortium organized a clustering activity with representatives from BIOSOS, SCIER and FIREPARADOX EU projects on Friday 9 December 2011 in CERTH premises. The BIOSOS project was represented by Prof. Maria Petrou and Dr. Vasiliki Kosmidou. Prof. Maria Petrou was the Director of ITI-CERTH, while Dr. Kosmidou is a postdoctoral research fellow in ITI-CERTH. The SCIER project was represented by Dr. Gavriil Xanthopoulos. Dr Xanthopoulos is a forest fire researcher at the Institute of Mediterranean Forest Ecosystems and Forest Products Technology of the National Agricultural Research Foundation in Athens, Greece. Finally, the FIREPARADOX project was represented by Dr Antonis Mantzavelas. Dr Mantzavelas is the manager of Omikron Ltd and President of the Permanent Committee for the Management of Forest Fires in Greece. The main topic of this event was the use of remote sensing technology for vegetation classification and fuel modeling. However, issues concerning fire detection technology and fire suppression techniques were discussed as well.

Furthermore, on 25th May 2012, the FIRESENSE project was presented at INTERVALUE (Inter-regional cooperation for valorization of R&D) project’s Workshop. Finally, on 26th February 2013, a meeting was held between representatives of CERTH and representatives of the Greek national project “Forest Fire Prevention with INCA methodology” in order to investigate common interests and possible future collaboration.

User group

Potential users of FIRESENSE system and technologies can be classified in the following categories based on the type of organization they work for:

• Cultural heritage preservation/protection
• Civil protection
• Fire fighting
• Forest protection
• Environmental protection
• Volunteers
• Researchers

After establishing the user requirement group, the consortium focused on the establishment of a wider network of potential users. This network consists of users who can register through the project’s web-site and participate in an on-going, two-way process. Within this process, the partners can publicize the results of their research, but they can also obtain feedback from potential users. At the end of the project the total number of registered users is 261 users from 49 different countries.

Education of inhabitants

Lectures were organized in test sites (Rhodiapolis, Kabeirion, Galceti Park and Temple of Water) to present the history of the sites, point out the importance of the conservation of the archaeological heritage and inform local inhabitants about the main objectives and results of the FIRESENSE project.

Prof. Enis Cetin visited Kumluca Vefa Hill High school in 2011 and 2012 to talk to students about the FIRESENSE project. This week was called “The Forest Week”. Two concerts were organized in Rhodiapolis in the summer of 2011 and 2012, respectively. The chief archaeologist of the archaeological site Dr. Isa Kizgut gave talks to the concert attendees before the concerts and pointed out the he fact that wildfires are a major threat for cultural heritage treasures.

IX EPCA (HMC) organised educational programmes for three different categories of inhabitants: adults, school classes 9-12 years old and families with children. Specifically, three organised archaeological tours at the Temple of Kabeirion were conducted by Dr Vassilis Aravantinos, Director of the IX EPCA, at October 2010, and by Alexandra Charami, Directorin of the IX EPCA at April 2012 and June 2012 respectively.

Furthermore, special educational programmes for school classes (for children 9-12 years old) about the Temple of Kabeirion were conducted by IX EPCA on 14-21 May 2012 (International Museum Day), including informative presentations with power point, arts and crafts, and theatrical play.

Open weekend educational programmes for families were organized by IX EPCA on 21 May 2012 (International Museum Day) and on 21 October 2012, including informative presentation with power point, arts and crafts, and theatrical play. Finally, three lectures about the Temple of Kabeirion were presented by IX EPCA in May 2012, June 2012, and January 2013.

Lectures to local inhabitants at Galceti test site took place in December 2012. A parallel session was also organized by CNR in order to keep children busy in activities linked to the lecture theme, i.e. something linked to fire. In Tunisia, lectures were organized in January 2013 to inform Sup’Com undergraduate students about the Firesense project.

Figure 38: Photos from educational programmes and archaeological tours organized in Kabeirion, Thebes.


Results from the research carried out within the FIRESENSE project have been published in international conferences and scientific journals. FIRESENSE partners published 22 articles in peer-reviewed journals and 35 articles in conferences proceedings.


To facilitate the successful introduction of the proposed early fire warning system for the protection of cultural heritage and related applications into the market, detailed exploitation strategies were elaborated. The exploitation strategy analysis was initially based on internet data, products knowledge, system analysis and contacts with the distributor networks of Titan and Xenics. Around twenty different early fire detection applications were identified and studied. These helped the Consortium to define the market segments and identify the potential users of FIRESENSE technologies.

Moreover, to generate a more comprehensive picture of the market, its composition with respect to different user groups reflecting their particular expectations and needs, its takeover potential, market opportunities and barriers to adoption and related trends as well as the competitive environment were highlighted. More specifically, the following market segments were identified for possible application of the FIRESENSE technology:

• Archaeological sites
• Museums fire protection
• Historic Buildings
• Art galleries fire protection
• Shops
• Parking fire Protection
• Tunnel fire protection
• Railways
• Metros
• City and Metros
• Nuclear Industry Plant
• Chemical Industry Plant
• Petrochemical
• Toxic material storage
• Logistics
• Offshore oil fire protection
• Gas infrastructure
• Pipe line infrastructure
• High voltage grids in forest areas
• Forest fire
• Waste fire
• Coal mine fire
• Waste treatment facilities
• Waste bunkers • Highways and railways going through forests
• Satellite monitoring (image fusion – large areas)
• Aviation monitoring (image fusion – large areas)
• Helicopter monitoring (image fusion – large areas)
• UAV monitoring (image fusion – large areas)
• Pole and mast monitoring
• Integrated security and fire detection
• Critical Infrastructure
• Airports
• Ports
• Portable Firefighting
• Monitoring critical vessel temperature
• Torpedo Car Refractory Thermal Monitoring
• Fire detection for Traffic Monitoring
• Area of cement works

Furthermore, the security market was divided in two parts: the external security or defense (military) and the private security sector (private).

The integrators are also considered as potential customers, since they could integrate different parts of the FIRESENSE system (control centre, cameras, WSN, image processing, data fusion, fire propagation estimation software). In any case, the fire detection system architecture must be tuned in function of the application. The system includes basic blocks that should be integrated. Even if all the blocks are available to build a fire detection system, we need at the technical level the following competences: system engineering, project management, system integration, maintenance and training. More specifically, the system engineer will realize a step by step approach with iterative loops of the following phases: he will realize a thorough analysis with the end users of the mission taking into account the local context and organization. Before performing a survey analysis, he will realize an initial simulation and model based on geographic information system (GIS). After this first analysis, a detailed site survey will be conducted with the customer. A fine selection of the components such as the sensors, communication links, data bandwidth, power system and others will be defined and proposed. He will write the technical specification, the integration plan, the test plan and maintain the configuration management. A pilot network implementation for field testing with future operators of the system will be conducted for a first evaluation and a first tuning of a partial system. This will be followed by a full deployment of the overall system. The system engineer will also provide support to the end users during the preliminary phases.

The project manager will be responsible for the timely delivery of a high quality product and system. On the other hand, the system integrator will develop the large scale project in modular solutions by using the most appropriate sensors, communication system and computer systems. Finally, an Integrated Logistic Support (ILS) ensures the maintainability of the system during its design and development. The ILS aims to address all the aspects of maintainability throughout the acquisition and whole life cycle of the equipment. Training should be provided to the customer.

Since cost of the equipment is one of the main criteria for successful market penetration, a cost analysis of sensors was also performed with emphasis on the IR cameras, which are the most expensive equipment of the system.

List of Websites:

The website of the FIRESENSE project is


The FIRESENSE consortium consists of 10 partners (6 academic and research institutes, 3 SMEs and 1 state authority) from 6 countries:

• Centre for Research and Technology Hellas, Information Technologies Institute (Greece) - Coordinator
• Bilkent Universitesi (Turkey)
• Ecole Supérieure des Communication de Tunis (Tunisia)
• Xenics nv (Belgium)
• Stichting Centrum voor Wiskunde en Informatica (Netherlands)
• Marac Electronics S.A. (Greece)
• Bogazici Universitesi (Turkey)
• Hellenic Ministry of Culture, IX Ephorate for Prehistoric and Classical Antiquities (Greece)
• Titan Building Systems Technology, Industry and Trade Limited Company (Turkey)
• Consiglio Nazionale delle Ricerche (Italy)

Contact information

For more information about FIRESENSE please contact the Project Coordinator:

Dr Nikos Grammalidis
FIRESENSE Coordinator
Researcher Grade B

Information Technologies Institute
Centre for Research and Technology Hellas
1st km Thermi - Panorama Road, 57001, Thessaloniki, Greece
Tel.: +30 2310 464160 (ext. 112)
Fax: +30 2310 464164