CN102945585A - Method for raising fire alarm through multi-sensor data fusion - Google Patents

Abstract

The invention discloses a method for raising a fire alarm through multi-sensor data fusion. The method is characterized by comprising that (1) sensor nodes conduct sampling within a sampling period to obtain sampling data of an area to be tested, the sampling data are pre-treated to judge whether the sampling data of sensor nodes are to be transmitted to a fusion node or not, and sensor nodes comprise carbon monoxide (CO) sensors, smoke sensors and temperature sensors; (2) the fusion node conducts preliminary data fusion on the sampling data of sensor nodes of the same type to obtain sampling fusion values of sensor nodes of the type within the sampling period; (3) the sampling data of the three types of sensors are fused to obtain a final fusion result, and the final fire probability condition can be obtained through a Dempster synthetic method; and (4) an expert system is used for conducting intelligent decision on the final fusion result and the data produced during fusion, and whether the fire alarm is raised or not is judged according to the result of the intelligent decision. The method improves the ability of a system for fire discrimination.

Description

Carry out the method for fire alarm by Fusion

Technical field

The invention belongs to the microcomputer data processing field, be specifically related to a kind of method of carrying out fire alarm by Fusion.

Background technology

Fire is modal a kind of in the harm humans lives and properties disaster. so the early prediction condition of a fire, and alarm, to prevent fires in combustion not be the task that fire detecting system is finished. and what adopt in the fire detector at present is threshold value comparison method mostly, it also is traditional fire detection data processing method, characteristics are simple and clear and are easy to realize, but environmental suitability and antijamming capability are relatively poor; People to the greatest hope of intelligent fire alarm system are: the total cost of early detection fire, elimination wrong report and reduction system, these factors are mutually restrictions. the one of the main reasons that these situations occur is exactly the data that each sensor do not obtained analysis-by-synthesis in addition under certain criterion, therefore necessary data with multisensor are processed processing, namely adopt sensor Data Fusion. it utilizes the various information of a plurality of sensors acquisitions, draws comprehensive, the correct understanding of environment or characteristics of objects.

The early sign of each flame has the different forms of expression, therefore reflects that the various signals of fire also present different features.Because event of fire is very accidental, observed data is few, so fire signal is prior unknown or unascertainable signal.By the analysis to fire mechanism, can know environment temperature, smokescope, CO content, H ₂Content etc. all can reflect the process of fire, and great many of experiments is observed and shown that the state of these parameters and rate of change thereof and fire exists certain mapping relations.But the above-mentioned physical quantity signal that utilizes sensor to obtain has more than with the condition of a fire and changes, all may cause the variation of signal in the environment such as the installation site of weather, humidity, dust, electronic noise and artificial other activities and sensor, and the feature of this variation often with condition of a fire feature similarity, so fire detection to compare with other typical input be a kind of very difficult input problem.Unit conventional fire detector adopts the unit Detection Techniques more, be single parameter fire detector (comprising threshold triggers formula and nalog quantity type), this detector is to the unevenness of fire characteristic signal response sensitivity. and cause it that the detectivity of actual fire is restricted.For example heat detector is only responsive to the temperature rise that naked light produces, insensitive to smoldering fire, cause that the heat that temperature rises is that fire produces or cigarette or cooking and steam generation nor can distinguish: the optical detector of fire smoke of and for example commonly using at present is a kind of fire sensor that the general condition of a fire is all had higher sensitivity, and smoldering fire is had fabulous Effect on Detecting, but the invisible cigarette (particle diameter is less than 0.4 μ m) that burning is produced or black smoke that naked light occurs are insensitive.Just because of this, still do not have so far a kind of single parameter fire detector can effectively survey all kinds of condition of a fire, cause the fire alarm wrong report to happen occasionally.The present invention therefore.

Summary of the invention

The object of the invention is to provide a kind of and carries out the method for fire alarm by Fusion, has solved usually to depend on the single parameter fire detector in the prior art and survey all kinds of condition of a fire, causes the problems such as the fire alarm wrong report happens occasionally.

In order to solve these problems of the prior art, technical scheme provided by the invention is:

A kind ofly carry out the method for fire alarm by Fusion, it is characterized in that said method comprising the steps of:

(1) sensor node is sampled within the sampling period and is obtained the sampled data in zone to be measured; Sampled data is carried out pre-service, judge whether the sampled data of sensor node should send to aggregators; Described sensor node comprises CO sensor, smoke transducer, temperature sensor;

(2) aggregators carries out data to the sampled data of sensor node of the same type and tentatively merges, and obtains the sampling fusion value of the sensor node of the type in this sampling period;

(3) sampled data of three types sensor is merged obtain last fusion results, obtain final fire probability situation according to the Dempster synthetic method;

(4) use expert system that the data that produce in final fusion results and the fusion are carried out intelligent decision, judge whether to send fire alarm according to the intelligent decision result.

Preferably, in the described method step (1) sampled data being carried out the method that pre-service adopts is the rate detection method, and described rate detection method comprises that each sensor node of hypothesis obtains n sampled value within a sampling period, be designated as: x (1), x (2) ..., x (n); Suppose that sum function a (m) is the difference sum of the neighbouring sample value x (n) that repeatedly adds up, that is:

$a\left(m\right)=\underset{n=0}{\overset{m}{\mathrm{\Σ}}}[x(n+1)-x\left(n\right)]---\left(I\right);$

Then pre-service result is o _i=a (m)-STD _i; (i=1,2,3) (II);

STD wherein _iRepresent respectively the attribute pre-service threshold value that sensor node detects; Work as o _iGreater than 0 o'clock, the data result that represents this sensor node was that non-stationary changes, and need to be determined further, and gives aggregators these data and carries out fusion treatment; Otherwise the data of this node directly abandon, and are regarded as invalid data.

Preferably, aggregators carries out data to the sampled data of sensor node of the same type and tentatively merges the employing averaging method in the described method step (2), and described averaging method comprises that the sampled data average of the sensor node that hypothesis is of the same type is

Be the fusion value of i class sensor node, that is:

$\stackrel{\‾}{X_i}\left(k\right)=\frac{1}{k}\underset{q=1}{\overset{k}{\mathrm{\Σ}}}{x}_{\mathrm{ik}}---\left(\mathrm{III}\right);$

I=1 wherein, three kinds of sensor nodes of 2,3(), k is the number of such sensor node of not being rejected by step (1).

Preferably, the sampled data to the three types sensor in the described method step (3) merges employing DS evidence theory method, may further comprise the steps:

S1) the preliminary fusion value of the sensor node image data of three types and existing fire model knowledge are shone upon mutually, obtain the basic feasible solution degree according to the interval under the fusion value and distribute;

S2) obtain the probability assignments of three kinds of sensors after, first the data of wherein any two types of sensor nodes merged, ask for first inconsistent degree, i.e. conflict spectrum between each evidence that is provided by knowledge base, the K value:

$K=\underset{A_{i}I{B}_j\≠\mathrm{\φ}}{\mathrm{\Σ}}{m}_1\left(A_i\right)m_2\left(B_j\right)---\left(\mathrm{VI}\right);$

Wherein two evidence ε under the framework Θ are distinguished in supposition ₁And ε ₂, its corresponding basic trust partition function is m ₁And m ₂, burnt unit is respectively A _iAnd B _jΘ represent X might value a domain set, and all elements in Θ all are mutual exclusive, claim that then Θ is the identification framework of X; X represents this proposition (event); When K=0, represent that two kinds of evidences are in full accord, when K=1, represent that two kinds of evidences conflict fully, when 0＜K＜1, represent that two kinds of evidence parts are compatible;

S3) try to achieve two kinds of probable values after the sensor fusion by the Dempster composite formula:

$m\left(A\right)=\left\{\begin{array}{c}\frac{\underset{A_{i}I{B}_j=A}{\mathrm{\Σ}}{m}_1\left(A_i\right)m_2\left(B_j\right)}{1-K},A\≠\mathrm{\φ}\\ 0,A=\mathrm{\φ}\end{array}\right.---\left(V\right);$

For certain the hypothesis A among the identification framework Θ, m (A) is the substantially credible number of this hypothesis, has reacted the reliability size to A itself, and K has reflected the conflict spectrum between each evidence, coefficient 1/ (1-K) becomes the regularization factor, and effect is first standardization with Jiao.Wherein Meaning be independently to obtain a new elementary probability after the elementary probability combination with two;

S4) result after this is merged and the third sensor are according to S2)-S3) step merges again, obtains the probability of last fire generation.

Preferably, the Using Intelligent Decision-making Method of expert system comprises by production rule structure knowledge base and factbase in the described method step (3), by rule the Object of Knowledge of knowledge base and the facts object of factbase are mated, adopt forward reasoning intelligence to infer whether breaking out of fire.

The present invention combines the method for D-S evidence theory and expert system, multi-sensor data is merged, thereby improve the probability of identifying fire alarm, and adopt corresponding processing scheme.The present invention has proposed concrete scheme in conjunction with expert system, has improved the intelligent of system.For example judge the probability that fire occurs, by the data in the fusion process before, can make further decision-making.

With respect to scheme of the prior art, advantage of the present invention is:

Judge with the fire of single-sensor and to compare that the method for technical solution of the present invention by D-S evidence theory and expert system combines and carry out fire and judge probabilistic probability has been dropped to negligible degree that namely the uncertain of system obviously reduces; And single-sensor originally can not accurately determine whether breaking out of fire, and the correct probability of judging increases after merging, and namely to the data fusion from 3 different sensors, has improved the differentiation fire ability of system.

Description of drawings

The invention will be further described below in conjunction with drawings and Examples:

Fig. 1 is the present invention carries out the method for fire alarm by Fusion schematic diagram.

Fig. 2 is the present invention carries out the method for fire alarm by Fusion detail flowchart.

Fig. 3 is the rule base structural drawing of the inventive method expert system.

Embodiment

Below in conjunction with specific embodiment such scheme is described further.Should be understood that these embodiment are not limited to limit the scope of the invention for explanation the present invention.The implementation condition that adopts among the embodiment can be done further adjustment according to the condition of concrete producer, and not marked implementation condition is generally the condition in the normal experiment.

Embodiment

As depicted in figs. 1 and 2, present embodiment carries out Fusion and carries out the method for fire alarm for high for single-sensor fire alarm rate of false alarm, a kind of multi-sensor data collection that inaccurate problem provides, merge, decision analysis finally obtains more accurately a kind of method of decision information, and its implementation step is as follows:

(1) to CO, smog, the sensor of temperature three types is sampled, and within a sampling period, each sensor node obtains n sampled data.

(2) n data in the sampling period are carried out pre-service, adopt the rate detection method, whether the data of judging this node should send to aggregators is carried out next step fusion.

(3) sensing data of homogeneity merged, adopt averaging method.Obtain the sampling fusion value of the sensor node of the type in the one-period.

(4) data of three types sensor is merged, method is the DS evidence theory method.Probability assignment by expert system acquisition three types sensor so and according to the DS synthetic method obtains final fire probability situation.

(5) utilize expert system that the data that produce in final fusion results and the fusion are made intelligent decision, obtain final solution.

Fig. 1 carries out the sensing system structure that Fusion carries out the method needs of fire alarm.Adopt co, three kinds of sensors of smog and temperature, every kind of sensor is got several environment is monitored.

Below its concrete steps are described in detail:

(1) in a timeslice (sampling period) data are sampled, then the single-sensor node carries out pre-service to the employing data of this timeslice, and pretreated method is the speed sampling method.Concrete grammar is: suppose that each sensor node obtains n sampled value, x (1), x (2) in a timeslice ... x (n), definition---sum function a (m) are repeatedly the difference sum of cumulative neighbouring sample value x (n).

$a\left(m\right)=\underset{n=0}{\overset{m}{\mathrm{\Σ}}}[x(n+1)-x\left(n\right)]---\left(I\right);$

Then pre-service result is o _i=a (m)-STD _i; (i=1,2,3) (II);

STD in the formula _iRepresent respectively CO concentration, smog, the pre-service threshold value of temperature.o _iGreater than 0 o'clock, the data result that represents this sensor node was that non-stationary changes, and need to be determined further, and gives next link these data and processes.Otherwise the data of this node directly abandon, and are regarded as invalid data.Can weed out invalid sensing data like this, save and transmitted the energy that these data need to consume.

(2) through after the pre-service, the data of homogeneity sensor are carried out next step fusion, native system has adopted the algorithm of mean value estimation to through pretreated all types of sensing datas.This algorithm is simple, and complexity is low, and real-time is good.Compare parameter and estimate scheduling algorithm in batches in the data anastomosing algorithm of data level, error burst is in 0.01.Unnecessary fire transaction module, this error do not affect judges decision-making.

$\stackrel{\‾}{X_i}\left(k\right)=\frac{1}{k}\underset{q=1}{\overset{k}{\mathrm{\Σ}}}{x}_{\mathrm{ik}};i=\mathrm{1,2,3}---\left(\mathrm{III}\right);$

The average of trying to achieve in the formula

Be the fusion value of i class sensor node, this value sent into next link merge.

(3) dissimilar sensing datas is merged, fusion method is DS evidence theory method (Dempster synthetic method), and ultimate principle is as follows:

If Θ represent X might value a domain set, and all elements in Θ all are mutual exclusive, claim that then Θ is the identification framework of X, arbitrary proposition of being concerned about so is all corresponding to the subset of Θ.If Θ={ θ ₁, θ ₂, the power set of Θ is then arranged: 2 ^Θ={ φ, Θ, { θ ₁, { θ ₂.

Basic reliability distribution (Basic Probability Assignment) is called for short BPA.BPA on identification framework Θ is a function m:2 ^Θ→ [0,1] is called the mass function.And satisfy following condition:

$\left\{\begin{array}{c}m\left(\mathrm{\φ}\right)=0\\ \underset{A\⊆\mathrm{\Θ}}{\mathrm{\Σ}}m\left(A\right)=1\end{array}\right.---\left(\mathrm{IV}\right);$

M is the basic reliability distribution on the framework Θ, and m (A) is the substantially credible number of A.Substantially credible number has reflected the reliability size to A itself.

The Dempster composition rule also claims the evidence composite formula, and it is defined as follows:

Suppose two evidence ε under the identification framework Θ ₁And ε ₂, its corresponding basic trust partition function is m ₁And m ₂, burnt unit is respectively A _iAnd B _j, establish

Then the Dempster composition rule is:

$m\left(A\right)=\left\{\begin{array}{c}\frac{\underset{A_{i}I{B}_j=A}{\mathrm{\Σ}}{m}_1\left(A_i\right)m_2\left(B_j\right)}{1-K},A\≠\mathrm{\φ}\\ 0,A=\mathrm{\φ}\end{array}\right.---\left(V\right);$

In the formula

$K=\underset{A_{i}I{B}_j\≠\mathrm{\φ}}{\mathrm{\Σ}}{m}_1\left(A_i\right)m_2\left(B_j\right)---\left(\mathrm{VI}\right);$

K has reflected the conflict spectrum between each evidence, and coefficient 1/ (1-K) becomes the regularization factor.The belief function given by m becomes m ₁And m ₂Quadrature and or direct sum, be designated as

If

Be false, then m ₁And m ₂Quadrature and

Do not exist.

The Dempster compositional rule is the rule of reflection evidence combined effect.Provide on the same identification framework based on the belief function of different evidences, if these several evidences do not conflict fully, so just can calculate a belief function as the belief function that produces under those several evidence combined effects with the Dempster compositional rule.

Basic step is as follows:

1) value of sensor is shone upon mutually with existing fire model knowledge, obtains the basic feasible solution degree according to the interval under the fusion value and distributes (this fire model knowledge is usefulness for example only, and expert system knowledge base provides again). and for example the CO measured value is 18; With reference to table 1, the probability assignments situation of CO is that m1{ has fire as can be known, and is without fire, uncertain }={ 0.6,0.1,0.3}.

Table 1: the confidence value under the different parameters scope

2) obtain the probability assignments of three kinds of sensors after, first wherein two kinds are merged.

Suppose that sensor is CO:18, smog 6.5, temperature 55 through the value of gained after single type node fusion steps.Here get first two kinds of sensor fusion of CO and temperature, at first ask for the inconsistent degree K of CO and smoke transducer, the probability assignments situation of two kinds of sensor nodes is that m1{ has fire as shown in Table 1, without fire, uncertain }={ 0.6,0.1,0.3}, m2{ has fire, without fire, uncertain }={ 0.7,0.1,0.2}

According to formula (VI): K=0.6*0.1+0.1*0.7=0.13

3) formula (V) is tried to achieve two kinds of probable values after the sensor fusion.

Obtain first the regularization factor 1/ (1-K):

1/ (1-K)=1/ (1-0.13)=1.15, according to formula (V):

The probability m(A1 of breaking out of fire)=1.15*(0.6*0.7+0.6*0.2+0.7*0.3)=0.8625

The probability m (A2)=1.15*(0.1*0.1+0.1*0.2+0.3*0.1)=0.069 of not breaking out of fire

The probability of unknown situation is m (A3)=1.15* (0.3*0.2)=0.069

The fusion probability that therefore can obtain CO and two kinds of sensors of smoke transducer is { fire being arranged, without fire, uncertain }={ 0.86,0.07,0.07}.

4) in like manner, result and the third sensor after this fusion are merged again, obtain the probability that last fire occurs.

The value that the temperature sensor that not yet participates in merging is surveyed is 55, and can get the probability assignments situation according to table 1 is that m3{ has fire, without fire, uncertain }={ 0.7,0.15 0.15} is with the result { 0.86 of (3) step fusion, 0.07,0.07} the step according to 2,3 liang of steps merges, finally passable is { 0.95 to the result, 0.038,0.013}.

According to above-mentioned for example can table 2:

Table 2

As can be seen from Table 2, probabilistic probability has dropped to negligible degree, and namely the uncertain of system obviously reduces; And single-sensor originally can not accurately determine whether breaking out of fire, and the correct probability of judging increases after merging, and namely the data fusion from 3 different sensors has been improved the differentiation fire ability of system.

As can be seen from Table 2, probabilistic probability has dropped to negligible degree, and namely the uncertain of system obviously reduces; And single-sensor originally can not accurately determine whether breaking out of fire, and the correct probability of judging increases after merging, and namely the data fusion from 3 different sensors has been improved the differentiation fire ability of system.

(4) expert system merges decision-making to final fire probability information.

Expert system adopts the mode of forward reasoning, utilizes the CO that obtains in merging early stage, temperature, and smog, the probability of fire can be known following state, 1. high 3. high temperature of dense smoke 2.CO concentration 4. fire 5. few cigarette 6. low CO concentration 7 low temperature 8 are without fire.These states are the optional precondition of expert system.

Expert system represents knowledge by production rule, the data of knowledge base as shown:

struct?rule{

Int used; Whether // sign rule is used

Int result; The conclusion that // sign rule is released;

Int flag; Whether // sign is the termination rule

Int count; The number of the regular contained prerequisite of // sign

Cause*causep; The fact of // rule is the prerequisite part

Rule*next; Next rule of // sensing

}

Rule objects is the entity of knowledge, in this system rule objects and inference mechanism is packaged together.Form an independently blocks of knowledge.The same user interactions of rule objects and facts object.Be illustrated in figure 3 as the structural drawing of rule base, the user carries out the intelligent decision request by known key element (CO content, smog, temperature) to expert system.

Factbase in the project comprises: 9. solid-state burning class fire 10 liquid-phase combustion class fire 11 gas bursts 12 CO reveal 13 cigarettes 14 and cook.Facts object has recorded current state, and gives directions the user to implement corresponding scheme.The fact and knowledge Unified number have also been set up factbase when setting up knowledge base.Factbase and knowledge base are entities, and true number (namely 91011121314) is true unique key word, and the rule in the knowledge base contacts with the fact by true number.For example

Certain rule is 4157813 in the rule base, i.e. expression has 4 preconditions, is respectively dense CO, and low temperature is smokeless, without fire.The conclusion that obtains according to rule is that 13CO reveals.

Above-mentioned example only is explanation technical conceive of the present invention and characteristics, and its purpose is to allow the people who is familiar with technique can understand content of the present invention and according to this enforcement, can not limit protection scope of the present invention with this.All equivalent transformations that Spirit Essence is done according to the present invention or modification all should be encompassed within protection scope of the present invention.

Claims (5)

1. one kind is carried out the method for fire alarm by Fusion, it is characterized in that said method comprising the steps of:

(1) sensor node is sampled within the sampling period and is obtained the sampled data in zone to be measured; Sampled data is carried out pre-service, judge whether the sampled data of sensor node should send to aggregators; Described sensor node comprises CO sensor, smoke transducer, temperature sensor;

(2) aggregators carries out data to the sampled data of sensor node of the same type and tentatively merges, and obtains the sampling fusion value of the sensor node of the type in this sampling period;

(3) sampled data of three types sensor is merged obtain last fusion results, obtain final fire probability situation according to the Dempster synthetic method;

(4) use expert system that the data that produce in final fusion results and the fusion are carried out intelligent decision, judge whether to send fire alarm according to the intelligent decision result.

2. method according to claim 1, it is characterized in that in the described method step (1) that it is the rate detection method that sampled data is carried out the method that pre-service adopts, described rate detection method comprises that each sensor node of hypothesis obtains n sampled value within a sampling period, be designated as: x (1), x (2),, x (n); Suppose that sum function a (m) is the difference sum of the neighbouring sample value x (n) that repeatedly adds up, that is:

$a\left(m\right)=\underset{n=0}{\overset{m}{\mathrm{\Σ}}}[x(n+1)-x\left(n\right)]---\left(I\right);$

Then pre-service result is o _i=a (m)-STD _i; (i=1,2,3) (II);

STD wherein _iRepresent respectively the attribute pre-service threshold value that sensor node detects; Work as o _iGreater than 0 o'clock, the data result that represents this sensor node was that non-stationary changes, and need to be determined further, and gives aggregators these data and carries out fusion treatment; Otherwise the data of this node directly abandon, and are regarded as invalid data.

3. method according to claim 1, it is characterized in that the middle aggregators of described method step (2) carries out data to the sampled data of sensor node of the same type and tentatively merges the employing averaging method, described averaging method comprises that the sampled data average of the sensor node that hypothesis is of the same type is Be the fusion value of i class sensor node, that is:

$\stackrel{\‾}{X_i}\left(k\right)=\frac{1}{k}\underset{q=1}{\overset{k}{\mathrm{\Σ}}}{x}_{\mathrm{ik}}---\left(\mathrm{III}\right);$

I=1 wherein, 2,3, be used for representing three kinds of sensor nodes, k is the number of such sensor node of not being rejected by step (1).

4. method according to claim 1 is characterized in that the sampled data to the three types sensor merges employing DS evidence theory method in the described method step (3), may further comprise the steps:

S1) the preliminary fusion value of the sensor node image data of three types and existing fire model knowledge are shone upon mutually, obtain the basic feasible solution degree according to the interval under the fusion value and distribute;

S2) obtain the probability assignments of three kinds of sensors after, first the data of wherein any two types of sensor nodes merged, ask for first inconsistent degree, i.e. conflict spectrum between each evidence that is provided by knowledge base, the K value:

$K=\underset{A_{i}I{B}_j\≠\mathrm{\φ}}{\mathrm{\Σ}}{m}_1\left(A_i\right)m_2\left(B_j\right)---\left(\mathrm{VI}\right);$

Wherein two evidence ε under the framework Θ are distinguished in supposition ₁And ε ₂, its corresponding basic trust partition function is m ₁And m ₂, burnt unit is respectively A _iAnd B _jΘ represent X might value a domain set, and all elements in Θ all are mutual exclusive, claim that then Θ is the identification framework of X; X represents this proposition; When K=0, represent that two kinds of evidences are in full accord, when K=1, represent that two kinds of evidences conflict fully, when 0＜K＜1, represent that two kinds of evidence parts are compatible;

S3) try to achieve two kinds of probable values after the sensor fusion by the Dempster composite formula:

$m\left(A\right)=\left\{\begin{array}{c}\frac{\underset{A_{i}I{B}_j=A}{\mathrm{\Σ}}{m}_1\left(A_i\right)m_2\left(B_j\right)}{1-K},A\≠\mathrm{\φ}\\ 0,A=\mathrm{\φ}\end{array}\right.---\left(V\right);$

For certain the hypothesis A among the identification framework Θ, m (A) is the substantially credible number of this hypothesis, has reacted the reliability size to A itself, and K has reflected the conflict spectrum between each evidence, and coefficient 1/ (1-K) is with the first standardized regularization factor of Jiao,

For independently obtaining a new elementary probability after the elementary probability combination with two;

S4) result after this is merged and the third sensor are according to S2)-S3) step merges again, obtains the probability of last fire generation.

5. method according to claim 1, the Using Intelligent Decision-making Method that it is characterized in that expert system in the described method step (3) comprises by production rule structure knowledge base and factbase, by rule the Object of Knowledge of knowledge base and the facts object of factbase are mated, adopt forward reasoning intelligence to infer whether breaking out of fire.

Images