CN101577033A - Multiband infrared image-type fire detecting system and fire alarm system thereof - Google Patents

Abstract

The invention relates to the field of fire-fighting, in particular to an image identification-based fire detecting system and a fire alarm system thereof. The multiband infrared image-type fire detecting system comprises an image acquisition unit, an image processing unit and an alarming unit, wherein the image acquisition unit sends the collected images to the image processing unit; according to the processing algorithm of the image processing unit, the collected infrared images are judged whether to meet the characteristics of the fire; for the images which meet the fire default values, alarming to the alarming unit is conducted; and the image acquisition unit comprises a half transparent and half reflecting beam splitter, an infrared filter and a camera, wherein the monitored area is synchronously monitored by two cameras added with infrared filters with different wavelengths through the half transparent and half reflecting beam splitter. The invention has the advantages of being sacred from the limitation of environmental conditions such as spatial height, heat barrier, easy explosive, poisonousness and the like, being capable of conducting early stage fire detection in large space and outdoor open space, low cost and the like.

Description

Multiband infrared image-type fire detecting system and fire alarm method thereof

Technical field

The present invention relates to fire-fighting domain, specifically relate to fire detecting system and fire alarm method thereof based on image recognition.

Background technology

Human development is to today, and fire still is the great disaster that threatens human life's safety.Fire is not only engulfed people's property, also can seize people's life, and its significant damage is self-evident.People seek new technology and method always, in the hope of fire is detected control, reach farthest and reduce the disaster that " fire " brings to people.

At present general fire detecting method is the cigarette that detection of fires generates when taking place, and sends fire alarm signal after signal Processing, comparison and judgement, but mostly this type of warning system is the warning carried out after fire has formed.

Yet continuous development along with society, the continuous minimizing of urban land, make City Building develop towards the direction of high stratification, densification, the finishing materials of buildings and mode also get over variation, and along with the increasing of power consumption and rock gas consumption, design has proposed higher, stricter requirement to automatic fire alarm system.The quantity of volumed space building (for example large-scale place of public amusement, theatre, bulk storage plant, large-scale country fair, garage, oil depot, waiting hall and waiting hall for airplanes etc.) and hypogee (as subterranean tunnel, subway station road, underground large parking lot and subterranean commercial area etc.) constantly increased in the last few years, because this type of building interior often height of a house height, span is big, it is bigger that the smog diffusion of fire initial stage is subjected to influences such as air-conditioning that building interior installs and ventilating system, the place personnel that have are intensive, inflammable thing are many, and disaster hidden-trouble is many.And this type of building fire spreads rapidly, and it is big to generate flue gas toxity, and evacuating personnel is taken refuge and the fire attack difficulty, in case breaking out of fire often causes very big economic loss and abominable social influence.Therefore, from passive fire attack develop into initiatively go to survey fire preventing, survey and put out a fire to save life and property parallelly, be the emphasis and the difficult problem of present fire alarm in early days in the hope of it being strangled what do not cause as yet that too havoc takes place.

Make a general survey of present Detection Techniques, as shown in Figure 1, contact is surveyed temperature-sensitive Detection Techniques, smoke detection technology, detection of gas technology, electrostatic detection technology etc.Contactless detection has flame detecting technology, sound-detection technology, image detection technology etc.They all have relative merits separately, such as: the heat detection sensitivity is low, and speed of detection is slow, and time of fire alarming is slow, is subject to temperature or influence of temperature variation, to the smoldering fire Low Response, is not suitable for early warning.The smoke detection technology is that ion smoke detector or optical detector of fire smoke all are that particle is surveyed, and is subject to the interference of particles such as various dusts, water droplet, mist of oil, insect, the rate of false alarm height.The flame detecting technology is the fire detector that a kind of radiation that flame is sent is surveyed, when response wave length is lower than 400nm radiative collision flux, be ultraviolet detection, wavelength is an infrared acquisition when being higher than 700nm radiative collision flux, response speed is fast, but early warning, but be subject to the interference of electric arc welding, thunderbolt, illumination, sunshine.

We can say that all automatic fire alarm technology mainly are based on the detection of sensor.In existing various fire alarms and fire fighting monitoring equipment, in the fire detection in most of places, all adopt the method for conventional detection, as sense cigarette, temperature-sensitive, sensitive detector, they utilize smog, the temperature of flame, the characteristic of light to come fire is surveyed respectively.But in large space occasions such as outdoor warehouse and large-scale indoor warehouse, the sensor signal becomes very faint owing to the huge of space, even high-precision sensor also can can't be worked owing to all interference noises.

Traditional fire detector is used for the real-time change of monitoring on-the-spot responsive phenomenon (as: smokescope, temperature, flame etc.) is detected, and extracts real-time parameter, and its performance quality directly can influence the accuracy and the reliability of automatic fire alarm.Therefore, fire detector plays a part very important in total system.Because the singularity of some large spaces and hypogee, common some type sense cigarette, temperature-sensing fire detecting warning system can't be gathered the cigarette temperature change information that fire sends rapidly, and there is certain defective in these traditional detection methods.

Be because the residing environment of range detector differs bigger each other in the building system of a practical application on the one hand, this different moment in one day or 1 year of each detector differs bigger.

On the other hand, the place of fire hazard often can not the usability cigarette, the temperature-sensitive sensor, therefore because the heat of initial stage fire and cigarette are difficult to arrive very high place, space, as arenas, warehouse etc., this class detector is difficult to the requirement satisfying early detection and forecast this type of building fire.

For a long time, at large space or have the occasion of high velocity air, (as forest fire protection) especially out of doors, the smoke detection of incipient fire worldwide all are difficult problems.Because under this class environment, exist many factors that influence detection, mainly comprise: detection mode, spatial altitude, thermal barrier, coverage, gas velocity, explosive, toxic gas, acceptable rate of false alarm, warning information management and remote signal transmission or the like.Traditional detection means has often lost effect in such environment.In this case, because the image-type fire detection technology has the characteristics of contactless detection for detection, be not subjected to the restriction of spatial altitude, thermal boundary, environmental baseline such as explosive, poisonous, make this technology become the effective means of carrying out detection at large spaces such as integrated mill, warehouse, forest and outdoor open space.

Summary of the invention

One of purpose of the present invention provides a kind of restriction that is not subjected to spatial altitude, thermal boundary, environmental baseline such as explosive, poisonous, can carry out fire early detection at large space and outdoor open space (as forest fire protection), rate of false alarm is low, and cheap multiband infrared image-type fire detecting system.

One of purpose of the present invention provides a kind of multiband infrared image-type fire alarm method of carrying out fire early detection.

One of purpose of the present invention provides a kind of restriction that is not subjected to spatial altitude, thermal boundary, environmental baseline such as explosive, poisonous, can carry out fire early detection at large space and outdoor open space (as forest fire protection), can carry out the multiband infrared image-type fire detecting system that low temperature is surveyed.

In order to realize purpose of the present invention, one of the present invention technical scheme is:

The multiband infrared image-type fire detecting system, comprise image acquisition units, Flame Image Process and control module and alarm unit, wherein, image acquisition units is sent to Flame Image Process and control module with the image of gathering, processing through Flame Image Process and control module, judge whether the infrared image that is collected meets the characteristic of fire, to meeting promptly reporting to the police of fire preset value to alarm unit, wherein, image acquisition units is by semi-transparent semi-reflecting spectroscope, and the image collecting device that installs infrared fileter additional forms, and wherein monitor simultaneously by two image collecting devices that semi-transparent semi-reflecting spectroscope is installed additional the different wave length infrared fileter area to be monitored.

Further; Also comprise the colour imagery shot that does not install the dress optical filter additional in the image acquisition units, described colour imagery shot and described two image collecting devices that install optical filter additional are positioned at same position, be used to detect same conflagration area, when the image by described image collecting device monitoring was met the fire preset value by identification, described colour imagery shot began to capture fire image.

Further; Described image collecting device is the USB digital camera; Or form by CCD simulation camera and SDK image pick-up card.

Further; The multiband infrared image-type fire detecting system also is reserved with communication interface, can link with any fire extinguishing system.

Wherein two optical filter wavelength choose the temperature needs that can survey occasion according to difference, according to Wien's displacement law λ _MaxT=a obtains the peak wavelength of spectral radiant exitance under certain temperature and according to the different images harvester responsiveness that can imaging wavelength is suitably selected optical filter.If the camera (thermal imaging system) that adopts infrared focal plane array to make is then as long as calculate the infrared fileter wavelength of required detecting temperature according to the Wien theorem.

Another technical scheme of the present invention is:

The multiband infrared image-type fire detecting system, comprise image acquisition units, Flame Image Process and control module and alarm unit, wherein, image acquisition units is sent to Flame Image Process and control module with the image of gathering, processing through Flame Image Process and control module, judge whether the infrared image that is collected meets the characteristic of fire, to meeting promptly reporting to the police of fire preset value to alarm unit, wherein, image acquisition units is made up of semi-transparent semi-reflecting spectroscope and infrared focal plane array thermal imaging system, and wherein the area to be monitored is monitored by two focal plane arrays (FPA) thermal imaging systems simultaneously by semi-transparent semi-reflecting spectroscope.

Another technical scheme of the present invention is:

The multiband infrared image-type fire detecting method, survey according to the following steps:

1) situation of area to be monitored is monitored by two image collecting devices that install the infrared fileter of different wave length additional simultaneously by semi-transparent semi-reflecting spectroscope; Two two width of cloth infrared target temperature pattern phase place unanimities that image collecting device photographs simultaneously;

2) two width of cloth infrared images that will be photographed simultaneously by two image collecting devices then are through being sent to Flame Image Process and control module, and two width of cloth infrared images that Flame Image Process and control module are at first gathered two image collecting devices carry out the identification of two waveband temperature; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

3) inciting somebody to action wherein again, the multiple image of arbitrary image collecting device continuous acquisition compares, whether the area with infrared image of the out-of-limit doubtful fire target of temperature increases in time, and whether the edge is regular, image area expansion when target object, and when picture shape is irregular, be judged as doubtful fire.

Further; 2) will be sent to Flame Image Process and control module by two width of cloth infrared images that two image collecting devices photograph simultaneously then, continuous two width of cloth images that Flame Image Process and control module are at first gathered same image collecting device subtract each other and take absolute value, when absolute value is non-vanishing; Two width of cloth images that two image collecting devices are gathered simultaneously carry out the identification of two waveband temperature again; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

Further; Carry out the analysis of gray level by the piece image that an image collecting device is wherein gathered, judge the size of the flame envelope of flame, if the pixel value at target image edge greater than internal flame and flame core, then is judged as fire than internal flame and flame core partial pixel value.

Further; Also comprise the colour imagery shot that does not install the dress optical filter additional in the described image acquisition units, described colour imagery shot and described two image collecting devices that install optical filter additional are positioned at same position, be used to detect same conflagration area, when the image by described image collecting device monitoring was met the fire preset value by identification, described colour imagery shot began to capture fire image.

Further; Flame Image Process and control module carry out area dividing to 2 coloured images that monitor (colour imagery shot 3 (HD-HV1302UC) that does not add optical filter), in case certain regional breaking out of fire hidden danger, the corresponding voice broadcast service of storage in advance then calls in system, personnel's emergency escape that notice is on-the-spot and emergent dredging.

Further; Coordinate according to image (coloured image that colour imagery shot 3 monitors) positions and follows the tracks of point of origin.

Semi-transparent semi-reflecting lens described in the present invention is to reflect 50% light intensity to transmittance 50% light intensity of incident, reaches semi-transparent semi-reflecting effect.The semi-transparent semi-reflecting lens applicable wavelengths scope that adopts among the present invention is 400nm-750nm, and this is use always a kind of.It mainly acts on is to provide phase place on all four imaging optical path for two cameras that install infrared fileter additional, this has just guaranteed that two non-overlapping cameras in position can photograph the on all four infrared image of phase place effectively, and the on all four infrared image of this two width of cloth phase place, when carrying out Flame Image Process, can simplify the complexity of fire image recognizer greatly, effectively improve the degree of accuracy that fire is differentiated, rate of false alarm is low.

In general, fire image has following characteristics:

The hot physical phenomenon of incipient fire mainly contains: glow, fire plume, flue gas.

Glowing is a combustion reaction at gas-solid phase interface place, and does not have gas phase flame burning phenomenon.The temperature of glowing is lower, and burning rate is slow, be difficult for finding, thereby danger is very big.Keep and the flames of anger burning representative temperature scope 600K-1000K that glows as the oneself.

Fire plume is the important starting stage that any fire all will experience: promptly enter the shared zone of fresh air by lasting rising of the thermal current of buoyancy-driven above flame.From incipient fire, just exist this combustion phenomena that is called as fire plume.Because it comprises the flame part, so be called fire plume.The temperature of fire plume is to change and different because of incendiary material with position and time.Because the representative temperature of pyrolysis is between 600K-900K, and gas phase flame is 1200K-1700K.Should be to the significant early stage fire plume temperature of detection greater than 1700K.Therefore, typical fire plume temperature range is: 500K-1700K.

Flue gas is the set of the molecule in the products of combustion.Since flue gas in process of flowing with the heat interchange of surrounding environment, its temperature descends gradually, therefore can determine: under the general condition, the upper limit of flue-gas temperature is lower than the temperature of fire plume, and lower limit is higher than environment temperature.Be generally 300K-800K.The temperature range of the various hot physical phenomenons of incipient fire and corresponding peak wavelength scope be as shown in Figure 2:

Above-mentioned hot physical phenomenon all has certain edge and body effect because of existing temperature difference with surrounding environment when fire takes place.Usually, this edge and body effect show in the scope of visible light, are exactly so-called image information.Himself is luminous and show the image information of object body if will take the lead, and its temperature must be more than 600K.After obtaining image information, will it be detected according to the difference of fire radiation and background radiation.

Fire radiation characteristics: according to different character, flame can be divided into two kinds, i.e. premixed flame and diffusion flame.Premixed flame has the feature of blue person's character lights flame, and flame burning is completely.Diffusion flame is more common yellow flame, and this flame is owing to unburnt result forms.The spectral power distribution feature of the flame of these two kinds of different in kinds as shown in Figure 3.

Flame is to comprise the high-temperature gas of various combustion productss and intermediate and constitute based on the high-temp solid particulate of carbon containing matter, inorganics.Therefore, the heat radiation of flame comprises the gas radiation of discrete spectrum and the solid radiation of continuous spectrum.The wavelength of flame in the scope of 0.2-10 μ m, the burning of different material, its radiation intensity is more or less different with Wavelength distribution.The flame spectrum energy distribution of various different materials as shown in Figure 4.

As can be seen, most of materials such as paper, timber, oils, gasoline and other hydrocarbons not only send visible radiation, and also have very strong infrared radiation.

Flicker: flicker is the notable feature that flame is distinguished other radiation.Studies show that the flame under the free combustion state produces random flicker,, can observe about the about 10Hz of its crest frequency if the infrared frequency that flame is sent is analyzed.Certainly, be subjected to the influence of fire scale and wind, its flicker frequency can change in the scope between 2-20Hz to some extent.

Because the radiation of diffusion flame is subjected to this significant modulation voltage effect, and there is not the modulating action of similar this mode under the general situation of background radiation, therefore, can show the improvement very big by the detector of flame flicking principle work, thereby correspondingly reduce rate of false alarm the background resolution characteristic.

Radiance increases: the rate of rise of the radiance of liquid fire is very high, and much lower for the rate of rise of the fire radiance that looks like wood type.Detector is surveyed initial fire disaster, and the rate of rise of the radiance that the check fire is initial is very useful.

Swept area increases: diffusion from taking place just to begin constantly to spread in fire, and its swept area is very high in the initial stage rate increase that fire takes place, and then slowly reduces.Detector is surveyed initial fire disaster, and the initial swept area rate of rise of check fire is very useful.

The background radiation characteristics: in any detection process, all exist certain background signal, therefore, fire signal must come with the background signal difference.For flame detector, the background radiation signal is mainly from the radiation of the natural radiation of the sun (sunlight direct projection, by reflecting surface reflection), other lamp and high-intensity man-made radiation's light source.

Solar radiation: at infrared spectral region, solar radiation is the blackbody radiation that a kind of temperature is about 6000K.Yet the solar spectrum of seeing on earth is subjected to Atmospheric Absorption and changes, and on the wavelength less than 280nm, solar irradiation is absorbed by the ozonosphere of upper atmosphere fully.On 2.7 mum wavelengths, because water vapor and CO ₂Effect, on the wavelength of 4.3 μ m, fully because CO ₂Effect, solar irradiation also is absorbed fully.

The design of flame detector utilizes absorbed these bands of a spectrum of solar irradiation usually.

The background radiation modulating action: although the sun is a stable radiation source, the flicker that the unevenness of atmosphere causes will cause the radiation of arbitrfary point acceptance on the ground that a which amplitude modulation component is arranged.Directly or through the modulating action of the solar radiation of reflection, also may be because cloud and mist block, wind leaf and water surface wave, mechanical rotation etc. cause.The modulating action that is in the background radiation in the flame flicking frequency range must cause enough attention.

The artificial radioactive source: compare with the solar radiation source, artificial radioactive source's radiation ratio is easier to prevention.

The total radiation energy of artificial light source is normally less than the solar irradiation energy.

The variation of artificial light source is moment, radiance and Area Growth speed moment by very big vanishing etc.

According to Planck's law of radiation, every temperature all can produce infrared radiation greater than the object of absolute zero.The temperature of occurring in nature actual object all is higher than absolute zero.Therefore, all there is infrared radiation in any object of occurring in nature, the infrared radiation of other object around also absorbing simultaneously.According to planck formula, the radiant exitance M of absolute black body _λWith the pass of blackbody temperature T and wavelength X be

$M_{\mathrm{\&lambda;}}={\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">C</figure-callout>}}_1{\mathrm{\&lambda;}}^{-5}\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="e\; c"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">e</figure-callout>}}^{c_{}}-1}---(2-1)$

C wherein ₁=3.7418 * 10 ^-16(Wm ²) be first radiation constant, c ₂=1.4388 * 10 ^-2(mK) be called second radiation constant.The radiant exitance M of black matrix _λImplication be at specified wavelength λ place, the radiation source per surface area is to the radiation power of hemisphere spatial emission, unit is Wm ^-3Planck formula has been described the spectral distribution rule of blackbody radiation, and it has disclosed in radiation and the matter interaction process and the dependence of radiation wavelength and blackbody temperature, is the theoretical foundation of blackbody radiation.

This fence of making a mistake-Boltzmann law: this law be expressed as the black matrix unit area be radiated the hemisphere space at λ ₁～λ ₂General power M (Wcm in the wave band ^-2) be:

$M=\underset{{\mathrm{\&lambda;}}_1}{\overset{{\mathrm{\&lambda;}}_2}{\&Integral;}}{M}_{\mathrm{\&lambda;}}\mathrm{d\&lambda;}---(2-2)$

Differential Planck law formula (2-1) is obtained maximum value, just obtains Wien's displacement law

λ _maxT＝a (2-3)

λ in the formula _MaxThe peak wavelength of-spectral radiant exitance, wherein a is temperature independent constant, and the approximate value of a is 0.2897cmK, and T is a Kelvin temperature.Therefore, the peak wavelength of spectral radiant exitance and absolute temperature are inversely proportional to.Bring (2-2) formula into (2-1) formula, draw another form of Wien's law, provided the peak value of spectral radiant exitance

$M_{{\mathrm{\&lambda;}}_m}=b{T}^5---(2-4)$

In the formula

Peak value (the Wcm of-spectral radiant exitance ^-2μ m ^-1), b is 1.2862 * 10 ^-15Wcm ^-2μ m ^-1K ^-5Provided the relation curve of 300K-1700K absolute black body spectral radiant exitance and wavelength as Fig. 9.

These curves are several characteristics of blackbody radiation as can be seen:

Every curve is non-intersect, and the built-up radiation emittance that is directly proportional with area under a curve increases sharply with the temperature increase, thereby temperature is high more, and the spectral radiant exitance on the wavelength of place is also big more;

The peak wavelength of spectral radiant exitance moves to the shortwave direction with the temperature increase.

Yet, be non-existent at the occurring in nature absolute black body, in order to describe the radiation of non-black-body, introduce the notion of radiant emissivity, use ε _λExpression

${\mathrm{\&epsiv;}}_{\mathrm{\&lambda;}}=\frac{M_{\mathrm{\&lambda;}}}{M_0}---(2-5)$

The implication of radiant emissivity is, under uniform temp, the ratio of the radiant exitance of radiator and the radiant exitance of black matrix is the function of wavelength and temperature, and is also relevant with the surface nature of radiator.According to the object radiation emissivity _λDifference, radiator is divided three classes:

1) black matrix, ε _λ=1;

2) grey body, ε _λ=ε＜1 is with Wavelength-independent;

3) selective body, ε _λ＜1 and change with wavelength and temperature.

As can be seen, the radiation characteristic 1 of black matrix), 2) same being suitable for and general object.

From the infrared radiation of target, the selectivity that was subjected to some gas in the atmosphere before arriving infrared system absorbs, the scattering of suspended particulates in the atmosphere, and therefore, radiation energy is decayed.Infrared radiation can be expressed as by the transmitance of atmosphere:

τ＝e ^-σx (2-6)

σ is an attenuation coefficient in the formula, and x is the length by the atmosphere distance.

Casting process that it should be noted that bigger liquid metals may produce high-intensity man-made radiation's light source.Tungsten lamp is launched 50% power in the 0.75-1.4 mu m waveband, tungsten lamp radiation power maximal value is positioned at 1 mu m waveband, and wavelength is greater than the radiation overwhelming majority of 3.5-4 μ m bulb glass only thoroughly.The radiation that the light of alternating current (especially glow discharge fluorescent tube) emission is modulated, but its modulating frequency is beyond the flame flicking frequency range, so can get rid of by suitable electronic filter.

Combustible burning the time can discharge frequency range from ultraviolet to infrared light wave, at visible light wave range, flame image has the feature of aspects such as unique chromatogram, texture, making it has tangible difference and background on image.Can utilize these features, utilize image process method, fire is discerned.So, utilize the principle of infrared imaging to obtain the infrared image that the burning initial stage sent and carry out Flame Image Process, thereby reach the purpose of monitoring by the image recognition of infrared band.

The selection of optical filter wavelength mainly depends on spectral radiance theoretical foundation.There are intrinsic difference in visible images and infrared image on image-forming mechanism, the feature that feature that the Same Scene visible images provides and infrared image provide is also inequality usually, they have different grey level features, the feature that occurs in visible images might not also occur in infrared image, eliminates the part undesired signal thereby reach.

(1) tests the temperature that records

According to planck formula, the radiant exitance M of absolute black body _λWith the relation of blackbody temperature T and wavelength X suc as formula (2-1):

$M_{\mathrm{\&lambda;}}={\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">C</figure-callout>}}_1{\mathrm{\&lambda;}}^{-5}\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="e\; c"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">e</figure-callout>}}^{c_{}}-1}---(2-1)$

Differential Planck law formula (2-1) is obtained maximum value, just obtains Wien's displacement law (2-3)

λ _maxT＝a (2-3)

λ in the formula _MaxThe peak wavelength of-spectral radiant exitance, wherein a is temperature independent constant, and the approximate value of a is 2897.8 ± 0.4 μ mK, and K is a Kelvin temperature.According to formula (2-3), can obtain the peak wavelength of spectral radiant exitance under certain temperature then, can suitably select optical filter according to this wavelength.

(2) wavelength of infrared radiation

When fire takes place, to our useful temperature range is 600K-1700K, we can report to the police to fire as early as possible in this scope, human eye only has brightness sensation to the electromagnetic wave of 380nm～780nm, and wavelength is called ultraviolet ray less than the electromagnetic wave of 400nm, and wavelength is called infrared ray greater than the electromagnetic wave of 750nm.The infrared radiation of flame combustion mainly concentrates on 950nm～2000nm wave band.

Therefore must adopt the optical filter of certain suitable wave band, make maximum region imaging on camera of surveyed area high-temperature targets radiation, suppress the amplitude of on-the-spot low temperature background imaging, the energy that keeps the fire target of monitoring simultaneously to greatest extent, realize increasing the purpose of contrast, clearly collect the infrared image of fire initial stage generation.

The infrared fileter of different wave length has different effects, and the infrared fileter wavelength of therefore selecting to be suitable on the fire image is necessary.Infrared fileter choose the following two principle of main basis: the infrared wavelength that calculates required detecting temperature according to the Wien theorem with according to different cameras to responsiveness that can imaging wavelength.If the camera (thermal imaging system) that adopts infrared focal plane array to make is then as long as calculate the infrared fileter wavelength of required detecting temperature according to the Wien theorem.Adopt wavelength be 0.8 μ m and 1.0 μ m optical filters transmitance as shown in Figure 7.

From the analysis of Fig. 8 more as can be seen, camera is different in that the image that is collected under the situation of free of light filter is arranged, different infrared fileters are the influence of background interference such as filtering artificial light light effectively, reduce the complexity of software recognizer, thereby can better realize the fire detection in early stage.

At the actual conditions of fire, need supervisory system and to have higher precision to detect disaster hidden-trouble in certain big scope.The single band algorithm is to the collection of fire infrared image, and its pixel can change along with the variation of extraneous factors such as distance and illumination, thereby greatly reduces detection accuracy, has increased the possibility of rate of false alarm.

The two waveband algorithm can be eliminated the interference of extraneous factors such as distance and illumination effectively, and passes through the monitoring of two cameras, greatly reduces rate of false alarm.

Employing adds infrared fileter and detects same width of cloth image before two cameras, detect two wave bands respectively, and then handle, and result who draws and range-independence are as long as the fire within the scope of shooting just can detect.Along with the variation of fire temperature, the infrared light wavelength that it sends is reflected at (as shown in Figure 5) on the image also in continuous variation, be exactly peak value in continuous variation, this just makes detected image pixel value in continuous variation, and can't obtain a definite numerical value.Though wavelength is in continuous variation, but the ratio of the image pixel value (as shown in Figure 5) that sees through long and short two optical filters and obtain is constant greater than 1 before fire takes place, each temperature all has oneself a ratio, through experiment, can obtain the one-to-one relationship of each ratio and temperature.In view of the above, can calculate the relevant temperature of image by detected image, the threshold temperature when taking place with experiment and the resulting corresponding fire of experience is then compared, thereby judgement has or not fire to take place, if calculate the gained temperature greater than threshold value then judge that fire takes place, otherwise do not have fire.

This principle can simply be expressed as: when radiation energy incided body surface, object was at fixed band λ ₁～λ ₂And λ ' ₁～λ ' ₂On the energy ratio that receives obey an only relevant function and distribute with temperature.

For λ ₁～λ ₂Wave band is got by formula (2-1), (2-2), (2-5), (2-6), and the radiation power that receives in the object unit area is:

$W={\mathrm{\&epsiv;}}_{\mathrm{\&lambda;}}{e}^{-\mathrm{\&sigma;x}}\underset{{\mathrm{\&lambda;}}_1}{\overset{{\mathrm{\&lambda;}}_2}{\&Integral;}}{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">C</figure-callout>}}_1{\mathrm{\&lambda;}}^{-5}\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="e\; c"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">e</figure-callout>}}^{c_{}}-1}\mathrm{d\&lambda;}---(2-7)$

In like manner, for λ ' ₁～λ ' ₂Wave band, the radiation power that receives in its unit area is:

$W^{\&prime;}={\mathrm{\&epsiv;}}_{\mathrm{\&lambda;}}{e}^{-\mathrm{\&sigma;x}}\underset{{\mathrm{\&lambda;}}_1^{\&prime;}}{\overset{{\mathrm{\&lambda;}}_2^{\&prime;}}{\&Integral;}}{\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">C</figure-callout>}}_1{\mathrm{\&lambda;}}^{-5}\frac{1}{{\mathrm{<figure-callout\; id="2"\; label="e\; c"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">e</figure-callout>}}^{c_{}}-1}\mathrm{d\&lambda;}---(2-8)$

Then there is energy ratio D to obey,

(2-9)

(2-9) unique variable T is only arranged in the formula, the two waveband principle must be demonstrate,proved.

In Fig. 5, this principle can be expressed as: the ratio S of two dash area areas ₁/ S ₂Obey the function of temperature.

Fig. 6 is λ under atmosphere ₁～λ ₂Be 3～5 μ m, λ ' ₁～λ ' ₂It is the calculating gained D-T relation curve of 8～12 μ m.

The superiority of two waveband is mainly reflected in following several respects:

The detection means of the two waveband that adopts can be carried out the detection within image pickup scope, and is not subjected to the constraint of distance, improves precision.

By the collection of two cameras, by the computing of two width of cloth image pixel ratios gathered, eliminate the interference of extraneous factors such as distance and illumination effectively, reduce rate of false alarm.

Infrared radiation during flame combustion mainly concentrates on 950～2000nm wave band, can use the camera that has infrared fileter in the time of the Image Acquisition of place, and this optical filter is the filtering visible light fully.

In actual conditions, the most of the time of system is in the state that does not have doubtful burning things which may cause a fire disaster, so can simply have or not the judgement of abnormal conditions to image sequence earlier.Whether native system adopts the state of calculus of differences detected image to change.Operational formula is:

ΔP _i(x，y)＝P _i(x，y)-P(x，y) (3-1)

Wherein, P _i(x y) is pending image, and (x y) is benchmark image to P

And Δ P _i(x, y) value determine according to environment and different, so experimental result adopts a kind of threshold value automatically selecting method further to improve the pretreated judgement dynamics of image.

In fire detection, threshold process can adopt maximum variance between clusters (claiming the Otsu thresholding algorithm again) to obtain.

(i is that (gray level is m to M * N image for i, the j) gray-scale value at some place, might as well suppose f (i, j) value [0, m-1] j) to note f.Note p (k) then has for gray-scale value is the frequency of k:

$p\left(k\right)=\frac{1}{\mathrm{MN}}\underset{f(i,j)=k}{\mathrm{\&Sigma;}}1---(3-2)$

Suppose with gray-scale value t to be that target and the background that Threshold Segmentation goes out is respectively: f (i, j)≤t} and f (i, j)＞t}, so:

Target part ratio: ${\mathrm{\&omega;}}_0\left(t\right)=\underset{0\&le;i\&le;t}{\mathrm{\&Sigma;}}p\left(i\right)$

Target is partly counted: $N_0\left(t\right)=\mathrm{MN}\underset{0\&le;i\&le;t}{\mathrm{\&Sigma;}}p\left(i\right)$

The background parts ratio: ${\mathrm{\&omega;}}_1\left(t\right)=\underset{t<i\&le;m-1}{\mathrm{\&Sigma;}}p\left(i\right)$

Background parts is counted: $N_1\left(t\right)=\mathrm{MN}\underset{t<i\&le;m-1}{\mathrm{\&Sigma;}}p\left(i\right)$

Target mean: ${\mathrm{\&mu;}}_0\left(t\right)=\underset{0\&le;i\&le;t}{\mathrm{\&Sigma;}}\mathrm{ip}\left(i\right)/{\mathrm{\&omega;}}_0\left(t\right)$

Background mean value: ${\mathrm{\&mu;}}_1\left(t\right)=\underset{t<i\&le;m-1}{\mathrm{\&Sigma;}}\mathrm{ip}\left(i\right)/{\mathrm{\&omega;}}_1\left(t\right)$

Grand mean: μ=ω ₀(t) μ ₀(t)+ω ₁(t) μ ₁(t)

Maximum variance between clusters points out to ask the formula of image optimum threshold value g to be:

$g=\mathrm{<figure-callout\; id="0"\; label="Arg\; Max"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">Arg</figure-callout>}\underset{}{\mathrm{Max}}\underset{0\&le;t\&le;m-1}{}[{\mathrm{\&omega;}}_0\left(t\right){({\mathrm{\&mu;}}_0\left(t\right)-\mathrm{\&mu;})}^2+{\mathrm{\&omega;}}_1\left(t\right){({\mathrm{\&mu;}}_1\left(t\right)-\mathrm{\&mu;})}^2]---(3-3)$

In fact in the bracket of formula (3-3) the right is exactly the inter-class variance value, and target and background two parts that threshold value g is partitioned into have constituted entire image, and target value μ ₀(t), probability is ω ₀(t), background value μ ₁(t), probability is ω ₁(t), grand mean is μ, promptly gets this formula according to the definition of variance.Because of variance is the inhomogeneity a kind of tolerance of intensity profile, variance yields is big more, two parts difference that composing images is described is big more, be divided into target and all can cause two parts difference to diminish when part target mistake is divided into background or part background mistake, therefore make to mean the misclassification probability minimum cutting apart of inter-class variance maximum.

μ ₀(t) and μ ₁(t), can represent the mean value gray scale of target and background respectively, μ then represents the center gray scale of entire image, make target and background obtain best cutting apart, and the target that certain hope is partitioned into is tried one's best away from picture centre, i.e. (μ ₀(t)-μ) ²(or | μ ₀(t)-μ |) big as far as possible, background is also tried one's best away from the center, i.e. (μ ₁(t)-μ) ²(or | μ ₁(t)-μ |) big as far as possible, owing to wish that the both is big, so:

1) both weighted sum maximums:

$g=\mathrm{<figure-callout\; id="0"\; label="Arg\; Max"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">Arg</figure-callout>}\underset{}{\mathrm{Max}}\underset{0\&le;t\&le;m-1}{}[{\mathrm{\&omega;}}_0\left(t\right){({\mathrm{\&mu;}}_0\left(t\right)-\mathrm{\&mu;})}^2+{\mathrm{\&omega;}}_1\left(t\right){({\mathrm{\&mu;}}_1\left(t\right)-\mathrm{\&mu;})}^2]---(3-4)$

2) both is long-pending maximum:

$g=\mathrm{<figure-callout\; id="0"\; label="Arg\; Max"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">Arg</figure-callout>}\underset{}{\mathrm{Max}}\underset{0\&le;t\&le;m-1}{}\left[{({\mathrm{\&mu;}}_0\left(t\right)-\mathrm{\&mu;})}^2{({\mathrm{\&mu;}}_1\left(t\right)-\mathrm{\&mu;})}^2\right]---(3-5)$

Note having to μ=ω ₀(t) μ ₀(t)+ω ₁(t) μ ₁And μ (t), ₀(t)≤μ≤μ ₁(t), therefore have:

ω ₀(t)(μ ₀(t)-μ) ²+ω ₁(t)(μ ₁(t)-μ) ²＝(μ-μ ₀(t))(μ ₁(t)-μ)(3-6)

This shows that the two is of equal value.

So system takes the image data stream of continuous 5 frames is calculated g value, the g value taking-up the highest and end forms a threshold range band.If front Δ P _i(x, value y) enters the fire image identification process so in this scope, otherwise just thinks there is not unusual situation.

Face the situation that Otsu takies computer resource, the native system employing is calculated the method for asking as optimal threshold g stage by stage and is realized reducing the consumption to resource.When just beginning system initialization, the Ostu algorithm is necessary.Along with the carrying out that system surveys, adopt a timer to carry out calculating to the Ostu algorithm, its basic thought is as time goes by, and is fewer and feweri to the calculating meeting of Ostu, however in a single day system detects when there is something special the timer calculating that starts anew.

According to the characteristics of infrared fire image, and the view data that combination collects is analyzed.Provided classical fire characteristic model, having comprised: temperature, area, speed, flame profile.According to the numerical value of analyzing, the endless belt line clustering algorithm of dynamic data has been proposed.

What infrared image reflected is the difference of target and background self to extraneous infrared energy, and therefore, what infrared image was mainly described is the heat radiation of target and background, and its fire initial stage infrared image has following characteristics:

The intensity profile of infrared image is in fact corresponding to the distribution of the temperature and the emissivity of fire and background.Infrared imaging can only extract desired signal from infrared distinctive low difference image, available essential information occurs with the picture element intensities form in the digital picture.

Because the number of pictures per second of infrared image picked-up is not very big, the radiation profiles around the target remains unchanged between two frames basically, and this character provides assurance for the pre-service of evaluating objects feature and target image frame by frame.

Because Infrared Image Processing Method is based upon that 2-D data is handled and the basis of random signal analysis on, be characterized in containing much information, thereby calculated amount and memory space are also big, the infrared imaging information processing must be in real time, fast, simple, reliably; High capacity information stores and high speed information are handled, and are the key problem in technology of Flame Image Process all the time.

When wanting breaking out of fire, general combustible can generate a large amount of red-hot particulates.These red-hot particulates make flame emission go out electromagenetic wave radiation, comprise visible light, and these optical characteristics provide possibility for the long-range detection fire.The energy emission and the radiation spectrum that produce all have embodiment at visible light and infrared band.Simultaneously, the set of these red-hot luminous particle can be delineated out flame profile.We just utilize the flame profile that red-hot particulate produced in this time to differentiate the generation of fire by infrared mage pickup.Any flame can be divided into flame envelope, internal flame and flame core three parts.The flame envelope temperature is the highest, secondly is internal flame, and the flame core temperature is minimum.Thereby cause the flame image gray level to be certain regularity of distribution.

Being one group as Figure 20 has image that free of light filter photographs relatively: from the analysis relatively of these images, we as can be seen, infrared fileter is the influence of background interference such as filtering artificial light light effectively, thereby can better realize fire detection in earlier stage.

From Figure 21 and 22, can significantly find out the effect of infrared fileter and the difference that the image information of fire condition when not having arranged.

Its temperature of initial stage of fire generation has the obvious variation process as shown in the above, and the variation of solar temperature and slow, and artificial radioactive source's conversion is usually finished in moment, so can be with variation of temperature speed as separating foundation.By setting suitable rate-valve value, can assist filtering part background information.

If the coordinate of certain point is in the image (m, n), its t pixel value constantly be l (m, n, t), then the rate of change v of its pixel (m n) is:

v(m，n)＝dl(m，n，t)/dt (4-1)

Need during Computer Processing to come the approximate differential computing with difference,

V ' (m, n)=| L (m, n, t ₂)-L (m, n, t ₁) |/t ₂-t ₁(wherein, t ₂＞t ₁) (4-2)

According to test, can get the interval I of variation calculation rate of fire image ₁As v ' (m, n) ∈ I ₁The time, retention point (m, pixel value n); When $v^{\&prime;}(m,n)\&NotElement;I_1$ The time, remove point (m, pixel value n).

This algorithm can be removed the information of the sun and instantaneous variation light source (as opening incandescent lamp), keeps the information of may observe dynamic change.

The value of high more its emittance of the temperature of object is big more, and the image pixel value of corresponding its formation is high more.By setting appropriate threshold, the further information of filtering low temperature background.

If wave band λ ₁～λ ₂The coordinate of image mid point be (m, n), its t pixel value constantly be L (m, n, t); (m is n) at wave band λ ' ₁～λ ' ₂In the image corresponding to point (m ', n '), its t pixel value constantly be L (m ', n ', t).The ratio image pixel value D that generates based on the two waveband principle is then arranged, and (m, n t) are

D(m，n，t)＝L(m，n，t)/L(m′，n′，t) (4-3)

Can get the interval I of ratio image pixel value by test ₂As D (m, n, t) ∈ I ₂The time, retention point (m, pixel value n); When $D(m,n,t)\&NotElement;I_2$ The time, remove point (m, pixel value n).

This algorithm is further removed the information of low temperature background.

Through after the processing of above technology, what image showed is the high temp objects with behavioral characteristics.It both may be a fire, also might be artificially to make high temp objects, as light cigarette candle, lighter etc.Fire is different from other high temp objects notable feature, and its areal extent is in constantly expansion and irregular.By following the tracks of the area change situation of suspected target, can determine whether that fire takes place.

The number of nonzero value picture element is area S (t) in the meter t moment ratio image, then has Area Growth rate α to be:

α＝dS(t)/dt (4-4)

The same difference approximation differential that uses then has

α=S (t ₂)-S (t ₁)/t ₂-t ₁(wherein, t ₂＞t ₁) (4-5)

When α＞1, show that suspected target is a fire; α≤1 o'clock shows that suspected target is not a fire.Show that by experimental test the value that each in detection a period of time gathered is not separate, be a kind of and the distribution fire occurrence tendency.Though the value of each data point is constantly changing, according to the numerical analysis of front as can be known, the set of red-hot luminous particle will be delineated out flame profile in the fire process.After fire took place, along with the increase of the intensity of a fire, flame constantly strengthened, and the image of flame shows as flame area, edge on the one hand in continuous variation; From its change of shape, spatial variations, space distribution the short sequential frame image of certain similarity, the particularly time interval is arranged on the other hand, the similarity of every width of cloth sequential frame image is more obvious.

And according to the characteristic of fire: any flame can be divided into flame envelope, internal flame and flame core three parts.Thereby cause the flame image gray level to be certain regularity of distribution.

4 width of cloth images of continuous acquisition such as Figure 24 from these fire image information and the variation of its corresponding pixel value, can sum up certain law to shown in Figure 30.

Along with the generation of fire, its numerical value change that is presented on the image also can present radiation and gradient form.Around this principle, can select its critical data point (also can be key among a small circle) to simulate this process.

Piece image can be expressed as:

F (m, n) be (m, n) gray-scale value, points all in the image can be expressed as this mode, by the image experimental data of front as can be known, the variation of numerical point is to begin to successively decrease towards periphery from the center, and this is to meet rule that fire takes place, is a kind of ring belt area that has obvious difference and periphery so be presented on the image.

Figure 31 is the pixel value region of variation figure of Figure 24,26,28,30 4 width of cloth images, can draw the flame shake and the image gray levels distribution situation of this four width of cloth image in the contrast clearly.

According to these characteristics, we can set up a model, and the threshold value of the pixel of choosing in the model must be more than or equal to our pre-service the time.Selection comes the center of circle, center of the numerical model that changes as this endless belt greater than key point or zone, calculate its center of circle degree being diffused into less than the endless belt in the scope of threshold value.Consider the feature when fire takes place, an endless belt is set up in point or zone greater than threshold value outside endless belt in addition.Go down by that analogy, have several ring belt areas on the possible piece image data stream.

Extracting the profile that the endless belt border here relates to Flame Image Process extracts.Always there is the edge between the adjacent area that two have different gray-scale values.The edge is the discontinuous result of gray-scale value.Select for use the border tracing to extract the profile of endless belt here.

What the starting point depend primarily on tracking was followed the tracks of on the border chooses and follows the tracks of choosing of criterion.The basic skills that follow the tracks of on the border is: find out the pixel on the target object according to some strictness " detection criterion " earlier, find out other pixels on the target object according to some features of these pixels with certain " tracking principle " again.General tracking principle is: the edge is followed the tracks of from the image upper left corner and is begun then to begin sequential track by picture point scanning when running into marginal point, until the subsequent point of following the tracks of get back to do not have new subsequent point again till.An endless belt is followed the tracks of and is finished, and then continues next point of scanning or zone, and all edges in image are all followed the tracks of and finished.

From the left figure of Figure 32 figure as can be seen, the direction that center pixel can be followed the tracks of has 8, and each direction has been specified direction numbering and side-play amount.So in the time of beginning reading images information, in case find to have information greater than threshold value, then found starting point, this point is noted, defining initial tracking direction is upper left side 0 direction, judge whether this point is impact point, be then this point for being tracked as initial starting point, be rotated counterclockwise 90 degree as new tracking direction, continue to detect the point on this new tracking direction, if not impact point finds impact point then along 45 degree that turn clockwise always.After finding impact point, on the basis of current tracking direction, be rotated counterclockwise 90 degree, use the same method and follow the tracks of the border of next endless belt as new tracking direction.The right figure of Figure 32 is the signal process that follow the tracks of on the border, and stain is represented frontier point, and white point is the internal point of image.The initial point of supposing to follow the tracks of is at the most bottom-right stain, and the initializing set of tracking is upper left side 45 degree.According to the information position that top tracking criterion is constantly preserved stain, know to detect to finish.

On the mark basis that has obtained the image rings region, by each pixel in the image is carried out marking operation, change the pixel value of ring belt area into label, adopt the border tracing again.Follow the tracks of the black pixel label of each ring belt area, the coordinate figure sequence on document image endless belt border.

Circularity is used for portraying the complexity of object boundary, and they get minimum value when circular boundary., its computing formula is:

Circularity=(girth) 2/ area (4-6)

Girth here is the boundary length of ring belt area, and area is then by calculating the pixel number in this zone.These 2 values can draw in the calculating on endless belt border.

After drawing the circularity of image information stream, along with the carrying out of fire, the variation of image values, this circularity numerical value also can be along with variation.By the situation of the fire of reality, the shape of general fire is complexity constantly, and its circularity value can slowly increase.

From continuous images conflagration area pixel value flows analysis, as can be seen, may have several endless belt in the piece image, and in the continuous images information flow, endless belt also can increase out, both fire spreading trend that come or that spread out.So, the circularity of endless belt of being associated in the consecutive image stream has corresponding variation, both ratio in general greater than 1, but consider the jitter of fire, its respective rings region circularity may can be little than the circularity of a last image, its ratio is less than 1.

At such situation,, problem is converted into decision-making classification problem to AD HOC according to further strengthen judgement based on the Bayesian decision algorithm of two-value.

The Bayesian decision criterion had both been considered the overall probability size that occurs of all kinds of references, had considered the loss size that causes because of erroneous judgement again, and discriminating power is strong.

So, in fact be exactly a kind of discrete feelings recognition category for the estimation of fire alarm at above-mentioned situation.Near critical point value judges, corresponding actual situation is exactly the problem of two classes classification, and the first kind is that non-fire is normal, the second class fire.From the mathematics category, in this critical point, these various situations occurring but is a kind of stochastic distribution.ω ₁The corresponding scene of decision-making fire is arranged, ω ₂The corresponding on-the-spot no fire of decision-making.The pattern feature of system is worth quantification for handling the sampling that conveniently is taken as certain hour, gets n-dimensional space vector x=(x that is: ₁, x ₂..., x _n), to arbitrary characteristics x ₁, x ₂..., x _nValue after the quantification is 0 or 1.Here x is the ratio that consecutive image flows related endless belt, and x is 1 more than or equal to 1 value, otherwise is taken as 0.

Because quantitative mode is judged, consider independently two class problems of each component, this moment, the component of eigenvector was to get 0 or 1 two value.Suppose two discriminant function g ₁, g ₂, the g according to whether then ₁＞g ₂Decide x to range ω ₁, define a discriminant function

g(x)＝P(ω1|x)-P(ω2|x) (4-7)

Differentiating rule is: if g (x)＞0, the x that then makes a strategic decision ranges ω ₁If g (x)＜0 decision-making x ranges ω ₂

If x=is (x ₁, x ₂..., x _d) ^T, and it is independently of one another, component xi value 0 or 1 wherein.

pi＝P(xi＝1|ω1)

qi＝P(xi＝1|ω2) (4-8)

Each feature of pattern is done the classification problem of yes or no.Whether this unknown pattern of final decision incorporates which kind of into, promptly is ω ₁Class or ω ₂Class.Because assumed conditions is independent, so can be write as the product of probability of each component of x to P (x| ω i).

$P\left(x\right|{\mathrm{\&omega;}}_1)=\underset{i=1}{\overset{d}{\mathrm{\&Pi;}}}{p}_i^{x_i}{(1-p_i)}^{1-x_i}---(4-9)$

$P\left(x\right|{\mathrm{\&omega;}}_2)=\underset{i=1}{\overset{d}{\mathrm{\&Pi;}}}{q}_i^{x_i}{(1-q_i)}^{1-x_i}---(4-10)$

Formula (4-9) and formula (4-10) are linear about xi, so can be expressed as

$g\left(x\right)=\underset{i=1}{\overset{d}{\mathrm{\&Sigma;}}}{u}_i{x}_i+u_0---(4-11)$

$u_i=\log \frac{p_i(1-q_i)}{q_i(1-p_i)}(i=1,...,d)---(4-12)$

${\mathrm{<figure-callout\; id="0"\; label="u"\; filenames="A20091005206900361.png,A20091005206900362.png"\; state="\{\{state\}\}">u</figure-callout>}}_0=\underset{i=1}{\overset{d}{\mathrm{\&Sigma;}}}\log \frac{1-q_i}{1-p_i}+\log \frac{P\left({\mathrm{\&omega;}}_1\right)}{P\left({\mathrm{\&omega;}}_2\right)}---(4-13)$

Decision function is the linear function about xi, like this for the classification that measures pattern under the actual conditions for the decision-making detector, the just value of the decision function that can try to achieve.

When the decision-making of g (x)＞0 unknown pattern is ω ₁Class is when the decision-making of g (x)＜0 unknown pattern is ω ₁

For Flame Image Process and control module: through the image that infrared fileter is taken, be not to be pure image gray in fact,, then gray level image carried out binary conversion treatment so be gray level image with image transitions earlier.Carry out the edge at zone bright in the binary image again and detect and edge treated, can also be converted to histogram, observe the intensity profile situation.

Simply introduce the principle of changing and handling in each process below:

The mutual conversion of gray scale image (Y) and non-pure gray scale image (R)

Conversion formula is as follows:

$\left\{\begin{array}{c}Y=0.299R+0.587G+0.114B\\ U=-0.147R-0.289G+0.436B\\ V=0.615R-0.515G-0.100B\end{array}\right.$

$\left\{\begin{array}{c}R=Y+1.14V\\ G=Y-0.39U-0.58V\\ B=Y+0.23U\end{array}\right.$

Be converted to bianry image, need thresholding.The principle of thresholding is mainly as follows:

Thresholding can be regarded as a special case of slicing, as shown in figure 12: be not difficult to find out, as long as make g in the slicing _1old=g _2oldJust realized thresholding.Threshold value is just as individual threshold, and bigger than it is exactly white, and littler than it is exactly black.Become the black and white binary map through the image after the thresholding processing.Carry out thresholding and only need provide threshold point g _1oldGet final product.

When we need know intensity profile situation among the width of cloth figure, can adopt grey level histogram to represent, (histogram of the present invention's gained image in once testing) as shown in figure 13, the horizontal ordinate among the figure is represented gray-scale value, ordinate is represented the number of times (frequency) that this gray-scale value occurs.The pixel of low gray scale has accounted for the overwhelming majority.Also adopted histogrammic function in the present invention.

The edge detects and profile extracts

Have following advantage:

A) can obtain the girth of object more accurately

B) describe the shape of object with the compacter mode of this image itself, can be used for calculating the flame circular diagram, with one of this judgment criterion that whether takes place as fire.

C) extract by profile, can reduce counting of processing, improve operation efficiency, calculate the variation of bright region area simultaneously easily.

Find out marginal point by order and follow the tracks of out the border.The result who after profile is followed the tracks of, obtains, as shown in figure 14.

The profile track algorithm on a closed border of simple binary picture is very simple: at first presses from top to bottom, and sequential search from left to right, first stain that finds must be the most upper left frontier point, is designated as A.Its right side, the bottom right down, has at least one to be frontier point in four adjoint points in lower-left, is designated as B.B looks for from beginning, by right, and the bottom right, down, and the lower-left, a left side, upper left, on, upper right order is looked for the frontier point C in the consecutive point.If C is exactly the A point, then show to make a circle EOP (end of program); Otherwise continue to look for from the C point, till finding A.Judge whether frontier point is easy to: if its four neighbours up and down all be stain then be not frontier point, otherwise be frontier point.

Because the computing machine that is sent to of every two field picture all has certain time interval,, realize calculus of differences so directly front and back two width of cloth image subtractions that same camera is monitored take absolute value.As shown in figure 16; If surpass threshold value then enter the temperature algorithm computation, otherwise return, and deposit image in memory block.

As shown in figure 17, image being carried out temperature threshold handles.Less than the pixel of threshold value gray-scale value being set directly for gray-scale value is 0; Gray-scale value keeps initial value and begins to carry out area algorithm greater than the pixel of threshold value, as shown in figure 18.

For image collecting device of the present invention, methods such as digital camera or simulation camera and image pick-up card all can be used.Wherein the USB digital camera comprises CMOS camera and focal plane arrays (FPA) thermal imaging system.

Most of image pick-up cards all are the multichannel inputs on the market, are convenient to a computer like this and can detect a plurality of zones simultaneously.The analog video signal that video capture device can produce the simulation camera converts digital signal to, and then it is stored in the computing machine.The vision signal that the simulation camera photographs must become digital signal with analog signal conversion through video frequency collection card, just can be transformed on the computing machine and use.，

Digital camera can will photograph to such an extent that image directly passes in the computing machine by USB interface.Now the camera on the computer market is substantially based on digital camera, and in the digital camera to use the USB digital camera of new types of data transmission interface, visible in the market major part all is this product.

The CCD camera: when selecting camera, camera lens is very important.Divide by the sensor devices classification, now the camera lens that camera uses on the market mostly is two kinds of CCD and CMOS greatly.CCD (ChargeCoupled Device wherein, Charged Coupled Device) is the high-end technology assembly that is applied in shooting, image scanning aspect, CMOS (Complementary Metal-Oxide Semiconductor, additional metal oxide semiconductor assembly) then is applied in some low side video products mostly.When but such location was not illustrated in concrete camera use, both had very big difference.In fact pass through technological transformation, the gap of the actual effect of CCD and CMOS has reduced greatly at present.

CCD is a kind of semiconductor imaging device, thereby has highly sensitive, anti-high light, advantages such as distortion is little, volume is little, the life-span is long, anti-vibration.To the CCD chip, CCD is according to the electric charge of the power accumulation corresponding proportion of light through lens focus for the image of subject, and the electric charge of each pixel accumulation is under the video time sequence control, and pointwise moves outward, after filtering, processing and amplifying, forms vision signal output.Vision signal is connected to the video inputs of monitor or television machine just can see the video image identical with original image.

USB camera: the real plug and play of camera support of using USB interface, in case your insertion equipment, system also can report immediately, and be that it seeks suitable driver, and, the employed power supply of USB camera can directly obtain from the mainboard USB interface, no longer needs clumsy independent current source converter; USB (V2.0) interface provides the 480Mbps transmission bandwidth in theory, and transmission speed is much higher than the existing peripheral port of computer.From being not difficult to find out here, adopted the computer camera of USB interface, on speed, gather around and have great advantage.

The hot video camera of infrared focal plane array: the thermoviewer technology has become the popular industrial technology of whole world common concern, CCD solid-state image pickup technology as the mid-90, because the CCD manufacturing technology of develop rapidly is ripe, the batch process that has in the relevant producer of each of global range, the wait that has is gone into operation, the application market sharp increase of CCD solid state camera, develop into the new industry of global range very soon, produced deep and wide influence the every field that comprises sphere of life (as digital video camera etc.).Undoubtedly, development trend from present global Infrared Focal plane Array Technologies, because the infrared eye technology as the thermoviewer guardian technique obtains considerable progress, the particularly breakthrough that obtains of uncooled infrared focal plane array technology, realized the miniaturization infrared video camera of high-performance cheapness, solve the high for a long time application obstacle that causes of price, opened up wide civil and military market.

Image pick-up card: but the capture card that the present invention adopts is day SDK2000 series capture card of quick company band secondary development.

10MOONS SDK2000 is a special high-quality PCI video card at system development merchant and computer DIY fan.SDK2000 has high-quality video acquisition performance, possesses the high-speed PCI bus, and compatible plug and play (PNP) is supported one-telephone multi-card.

The secondary development bag (hereinafter to be referred as SDK) of complete function is provided.Can select multiple programming languages such as VisualBasic, visual c++, Delphi to develop, comprise DLL dynamic base (VC use) among the SDK, OCX control (VB, Delphi uses) and detailed description thereof by SDK.Input port that can be by SDK control chart picture, brightness of image, contrast, colourity, input signals such as gray scale, whether dynamic cut-away view picture has moving target or the like with the AVI form detecting image of recording a video.

Description of drawings

Fig. 1 fire detector classification synoptic diagram

Fig. 2 incipient fire temperature and peak wavelength distribution

The spectral power distribution of Fig. 3 flame

The flame spectrum energy distribution of the various different materials of Fig. 4

Fig. 5 two waveband schematic diagram

Fig. 6 D-T relation curve

The optical filter transmittance curve synoptic diagram that Fig. 7 the present invention is selected

Fig. 8 has the image of free of light filter to compare

The relation of Figure 93 00K-1700K absolute black body spectral radiant exitance and wavelength

Figure 10 multiband infrared image-type fire detecting of the present invention system

Figure 11 is the bright multiband infrared image-type fire detecting system works of a present invention flow process block diagram

The schematic diagram of Figure 12 thresholding

Figure 13 grey level histogram

Figure 14 original image

Result after Figure 15 profile is followed the tracks of

Figure 16 rate algorithms process flow diagram

Figure 17 temperature algorithm flow chart

Figure 18 temperature algorithm flow chart

Figure 19 multiband infrared image-type fire detecting system chart of the present invention

Figure 20 has the comparison of free of light filter institute images acquired

Image that collects when Figure 21 has situation and numerical information

Image that collects when Figure 22 does not have situation and numerical information

Figure 23 fire image information 1

Figure 24 and fire image information 1 corresponding pixel value

Figure 25 fire image information 2

Figure 26 and fire image information 2 corresponding pixel value

Figure 27 fire image information 3

Figure 28 and fire image information 3 corresponding pixel value

Figure 29 fire image information 4

Figure 30 and fire image information 4 corresponding pixel value

Figure 31 continuous images conflagration area pixel value flows and analyzes

Synoptic diagram is followed the tracks of on Figure 32 border

Embodiment

The present invention is described further below in conjunction with accompanying drawing:

Embodiment 1: 10 as seen in conjunction with the accompanying drawings, the multiband infrared image-type fire detecting system, image acquisition units is made up of semi-transparent semi-reflecting spectroscope, infrared fileter and USB camera 1 and USB camera 2, the camera 1 that installs 0.8 mum wavelength optical filter additional is fixing towards two vertical light paths with the camera 2 that installs 1.0 mum wavelength optical filters additional, monitors the area to be monitored simultaneously by semi-transparent semi-reflecting spectroscope.Image acquisition units is sent to Flame Image Process and control module with the image of gathering, and through the processing of Flame Image Process and control module, judges whether the infrared image that is collected meets the characteristic of fire, to meeting promptly reporting to the police to alarm unit of preset value.

USB camera 1 is the HD-HV1302UM type with USB camera 2; USB camera used in the present invention is not limited thereto, and the various models of USB camera of market sale all can be used.

Image acquisition units adopts model to be: EC5-1719CLDNA (star of embedding) is the high-performance single-borad computer of a employing Intel notebook computer chipset 945GM design.The star of embedding used in the present invention is not limited thereto, and any model PC all can use on the market.

Wherein, the camera that installs optical filter additional can be replaced with the infrared focus plane thermal imaging system, its output signal is directly changed into digital signal, can save image pick-up card and directly send into Flame Image Process and control module, the infrared focus plane thermal imaging system is a low temperature imaging wide ranges as the main advantage of image acquisition units, can be used for the detection occasion of cryogenic object combustible.And adopt the CCD camera to install that optical filter mainly acts on additional is to substitute expensive infrared focus plane thermal imaging system, finish infrared temperature image acquisition to monitoring objective, the CCD camera reduces the cost of whole fire detecting system with advantages such as its cheap and high temperature imaging wide ranges significantly.

Embodiment 2: among the embodiment 1, the selection of the wavelength of optical filter can be according to the needs of concrete fire detection occasion, the infrared wavelength that calculates required detecting temperature according to the Wien theorem with according to the different images harvester responsiveness that can imaging wavelength is chosen different wavelength.If the camera (thermal imaging system) that adopts infrared focus plane to make is then as long as calculate the infrared fileter wavelength of required detecting temperature according to the Wien theorem.As the timber burning-point is 250 ℃～300 ℃, then the wavelength of infrared fileter should be chosen between 5.54 μ m～5.06 μ m, the cotton burning-point is 210 ℃～255 ℃, then the wavelength of infrared fileter should be chosen between 6.0 μ m～5.49 μ m, 700 ℃～800 ℃ of cigarette end surface temperatures, then the wavelength of infrared fileter should be chosen between 2.98 μ m～2.7 μ m, the paper burning-point is 130 ℃, then the wavelength of infrared fileter should be chosen between 7.19 μ m～7.0 μ m, the wheat straw burning-point is 200 ℃, then the wavelength of infrared fileter should be chosen between 6.12 μ m～5.9 μ m, and the polyster fibre burning-point is 390 ℃, and then the wavelength of infrared fileter should be chosen between 4.37 μ m～4.1 μ m etc.

Embodiment 3: among embodiment 2 or the embodiment 1, image acquisition units also comprises a colour imagery shot 3 that does not install optical filter additional, when the image of camera 1 and camera 2 monitorings was met preset value by identification, the colour imagery shot 3 that does not install optical filter additional began to capture fire image.

Colour imagery shot 3 adopts the HD-HV1302UC model.

Embodiment 4: as shown in figure 10, the present invention adopts the infrared fileter of semi-transparent semi-reflecting lens and different wave length to come by 2 (HD-HV1302UM) USB camera collection dual-band infrared images in system, surveys camera 1 and whether camera 2 surveillance areas have fire by the algorithm of software.In case there is suspicious fire to take place, then send warning message to far-end server by the gsm module that is connected with serial ports, simultaneously also by the network interface of system platform, warning message is transferred on the far-end server.

Wherein, that gsm module adopts is the TC35 of Siemens company, TC35 is the radio communication gsm module that Siemens company releases, and module mainly is made up of GSM baseband processor, GSM radio-frequency module, supply module (ASIC), flash memory, ZIF connector, antennal interface six parts.

Communication interface adopts CAN bus or RS-485 serial ports, be mainly used in all kinds of on-the-spot audio alert after fire takes place, remote I NTERNET warning, the warning of GSM short message telephone etc., and and existing fire-fighting fire-extinguishing equipment interlock control, communication interface can provide the startup of instruction control fire-protection equipment according to the needs of Flame Image Process and the default condition of a fire grade of control module.

Embodiment 5

In conjunction with Figure 10, Figure 11 and shown in Figure 19, the present invention will be further described:

The multiband infrared image-type fire detecting method, according to the following steps

1) situation in detected zone is monitored by two (the image HD-HV1302UM of Daheng type) camera 1 and (the image HD-HV1302UM of Daheng) cameras 2 that install the infrared fileter of (wavelength 0.8 μ m and 1.0 μ m) additional simultaneously by semi-transparent semi-reflecting spectroscope; Two two width of cloth infrared target temperature pattern phase place unanimities that camera photographs;

2) then this two width of cloth image is sent into Flame Image Process and control module (EC5-1719CLDNA (star of embedding)) through image acquisition units ((HD-HV1302UM)), two width of cloth images that Flame Image Process and control module are at first gathered two cameras carry out temperature identification; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

3) inciting somebody to action wherein again, the multiple image of arbitrary camera continuous acquisition compares, whether the area of looking to have the target infrared image of the out-of-limit doubtful fire of temperature increases in time, and whether the edge is regular, image area expansion when target object, and when picture shape is irregular, be judged as fire.

Embodiment 6

The present invention is described further in conjunction with Figure 10, Figure 11, Figure 17, Figure 18, Figure 19

The multiband infrared image-type fire detecting method, according to the following steps

1) situation in detected zone is monitored by two (the HD-HV1302UM model) USB cameras that install the infrared fileter of difference (wavelength 0.8 μ m and 1.0 μ m) wavelength additional simultaneously by semi-transparent semi-reflecting spectroscope; Camera 1 is consistent with two width of cloth infrared target temperature pattern phase places that camera 2 photographs;

2) then this two width of cloth image is sent into Flame Image Process and control module (EC5-1719CLDNA (star of embedding)) through image acquisition units, two width of cloth images that Flame Image Process and control module are at first gathered simultaneously two cameras carry out temperature identification; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

3) again will be wherein the multiple image of (HD-HV1302UM model) USB camera 1 or (HD-HV1302UM model) USB camera 2 continuous acquisition compare, whether look its area with target infrared image of the out-of-limit doubtful fire of temperature increases in time, and whether the edge is regular, image area expansion when target object, and when picture shape is irregular, be judged as fire.

In the said method, in step 2) then this two width of cloth image is sent into Flame Image Process and control module through image acquisition units, Flame Image Process and control module at first to (HD-HV1302UM) USB camera 1 or (HD-HV1302UM) continuous two width of cloth images gathered of USB camera 2 subtract each other and take absolute value, when absolute value is non-vanishing; Two width of cloth images that camera 1 or camera 2 gathered simultaneously carry out temperature identification again; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

Further, by to (HD-HV1302UM) USB camera 1 wherein or (HD-HV1302UM) piece image gathered of USB camera 2 carry out the analysis of gray level, judge the size of the flame envelope of flame than internal flame and flame core partial pixel value, if the pixel value at target image edge greater than internal flame and flame core, then is judged as fire.

Further, image acquisition units comprises that also (HD-HV1302UC) colour imagery shot 3 that does not install optical filter additional detects detected zone simultaneously, when the image that is detected by two cameras that install infrared fileter additional was met preset value by identification, the camera that does not install optical filter additional began to capture fire image.

Also the coloured image that monitors can be carried out area dividing, in case certain regional breaking out of fire hidden danger, the corresponding voice broadcast service of storage in advance then calls in system, personnel's emergency escape that notice is on-the-spot and emergent dredging.

Also can point of origin be positioned and follow the tracks of according to the coordinate of image.

Voice are dredged: the image that will monitor occasion is divided into a plurality of area coordinates in " well " font mode, in Flame Image Process and control module, by the software recognizer, judge it is which area coordinate breaking out of fire,, then pass through the program of this zone correspondence in case judge certain regional breaking out of fire, sound card in Flame Image Process and control module, play the voice of recording in advance and dredge broadcasting, carry out orderly personnel and dredge, even more serious disaster takes place in order to avoid the scene of fire is chaotic.

Claims (11)

1, the multiband infrared image-type fire detecting system, comprise image acquisition units, Flame Image Process and control module and alarm unit, wherein, image acquisition units is sent to Flame Image Process and control module with the image of gathering, processing through graphics processing unit, judge whether the infrared image that is collected meets the characteristic of fire, to meeting promptly reporting to the police of fire preset value to alarm unit, it is characterized in that: image acquisition units is by semi-transparent semi-reflecting spectroscope, and the image collecting device that installs infrared fileter additional forms, and wherein monitor simultaneously by two image collecting devices that semi-transparent semi-reflecting spectroscope is installed additional the different wave length infrared fileter area to be monitored.

2, the multiband infrared image-type fire detecting system, comprise image acquisition units, graphics processing unit and alarm unit, wherein, image acquisition units is sent to graphics processing unit with the image of gathering, processing through graphics processing unit, judge whether the infrared image that is collected meets the characteristic of fire, to meeting promptly reporting to the police of fire preset value to alarm unit, it is characterized in that: image acquisition units is made up of semi-transparent semi-reflecting spectroscope and infrared focal plane array thermal imaging system, and wherein the area to be monitored is monitored by two infrared focal plane array thermal imaging systems simultaneously by semi-transparent semi-reflecting spectroscope.

3, multiband infrared image-type fire detecting according to claim 1 system, it is characterized in that: also comprise the colour imagery shot that does not install the dress optical filter additional in the image acquisition units, described colour imagery shot and described two image collecting devices that install optical filter additional are positioned at same position, be used to detect same conflagration area, when the image by described image collecting device monitoring was met the fire preset value by identification, described colour imagery shot began to capture fire image.

4, according to claim 1 or 3 described multiband infrared image-type fire detecting systems, it is characterized in that: described image collecting device is the USB digital camera.

5, according to claim 1 or 3 described multiband infrared image-type fire detecting systems, it is characterized in that: described image collecting device is made up of CCD simulation camera and SDK image pick-up card.

6, according to claim 1 or 3 described multiband infrared image-type fire detecting systems, it is characterized in that: the multiband infrared image-type fire detecting system is reserved with communication interface, can with any fire extinguishing system interlock.

7, multiband infrared image-type fire detecting method is characterized in that: according to the following steps

1) situation of area to be monitored is monitored by two image collecting devices that install the infrared fileter of different wave length additional simultaneously by semi-transparent semi-reflecting spectroscope; Two two width of cloth infrared target temperature pattern phase place unanimities that image collecting device photographs simultaneously;

2) two width of cloth infrared images that will be photographed simultaneously by two image collecting devices then are through being sent to graphics processing unit, and two width of cloth infrared images that graphics processing unit is at first gathered two image collecting devices carry out the identification of two waveband temperature; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

3) inciting somebody to action wherein again, the multiple image of arbitrary image collecting device continuous acquisition compares, whether the area with infrared image of the out-of-limit doubtful fire target of temperature increases in time, and whether the edge is regular, image area expansion when target object, and when picture shape is irregular, be judged as doubtful fire.

8, multiband infrared image-type fire detecting method according to claim 7, it is characterized in that: 2) will be sent to graphics processing unit by two width of cloth infrared images that two image collecting devices photograph simultaneously then, continuous two width of cloth images that graphics processing unit is at first gathered same image collecting device subtract each other and take absolute value, when absolute value is non-vanishing; Two width of cloth images that two image collecting devices are gathered simultaneously carry out the identification of two waveband temperature again; The infrared image of long and short two wave band collections is carried out pixel relatively, if ratio is greater than 1, illustrate that target temperature is lower than set alarm temperature threshold value, in case area to be monitored breaking out of fire, the pixel ratio of the infrared image that then long and short two wave bands are gathered is less than 1, illustrate that then the target temperature is higher than set alarm temperature threshold value, is judged as doubtful fire;

9, according to claim 7 or 8 described multiband infrared image-type fire detecting methods, it is characterized in that: carry out the analysis of gray level by the piece image that an image collecting device is wherein gathered, judge the size of the flame envelope of flame than internal flame and flame core partial pixel value, if the pixel value at target image edge greater than internal flame and flame core, then is judged as fire.

10, multiband infrared image-type fire detecting method according to claim 9, it is characterized in that: also comprise the colour imagery shot that does not install the dress optical filter additional in the described image acquisition units, described colour imagery shot and described two image collecting devices that install optical filter additional are positioned at same position, be used to detect same conflagration area, when the image by described image collecting device monitoring was met the fire preset value by identification, described colour imagery shot began to capture fire image.

11, multiband infrared image-type fire detecting method according to claim 10, it is characterized in that: graphics processing unit carries out area dividing to 2 coloured images that monitor, in case certain regional breaking out of fire hidden danger, the corresponding voice broadcast service of storage in advance then calls in system, personnel's emergency escape that notice is on-the-spot and emergent dredging.

Images