KR101204259B1 - A method for detecting fire or smoke - Google Patents

Abstract

In the present invention, after extracting the background image from the image photographed by the image capturing device to separate the foreground image which is the difference signal image between the captured image and the background image, the characteristics such as brightness and saturation to determine whether the foreground image mobility or motion Analyze the change and extract a predetermined candidate region, and then apply Gaussian mixture modeling method to the extracted candidate to separate and extract the region having the characteristics of smoke or flame to detect the fire occurrence quickly and accurately and inform the manager. It provides a fire detection method using an image processing.

Description

A method for detecting fire or smoke}

The present invention relates to a fire detection method using an image processing, and to a fire detection method for detecting the occurrence of smoke and flames by photographing the monitoring target area, and to quickly notify the fire occurrence according to the smoke and flame generation .

In general, the fire detection system is installed in the monitoring target area using a sensor for detecting smoke or temperature, and there is a method using a camera to photograph the monitoring target area.

The method using the sensor for detecting the smoke or temperature is to determine whether a fire occurs by detecting a certain amount of smoke or a temperature above a certain temperature, or after a fire has occurred, a large amount of smoke after the fire range is widely spread. To detect the occurrence of a fire in the state or the temperature is sufficiently raised, there was a problem that can not be quickly responded to the fire.

The present invention has been proposed to solve the above-mentioned problems, and to provide a fire detection method capable of catching the early signs of smoke, flames.

To this end, in the present invention, the statistical characteristics of the smoke and the flame are determined by using a predetermined probability distribution function, and then the statistical characteristics of the candidate region where the motion is captured in the captured image are extracted based on the data, and the smoke, flame, etc. The purpose of this study is to provide a method of detecting the occurrence of a fire in comparison with the statistical characteristics of.

To this end, in the present invention, the background image is extracted from the image photographed by the image capturing apparatus, and the foreground image, which is the difference signal image between the captured image and the background image, is separated, and the brightness and saturation are determined to determine whether the foreground image is mobile or moved. After extracting a predetermined candidate region by analyzing the characteristic change, etc., the Gaussian mixture modeling method is applied to the extracted candidate to separate and extract the region having smoke or flame characteristics to detect the occurrence of fire quickly and accurately. To provide a fire detection method using an image processing that can inform.

An embodiment of a fire detection chamber according to the present invention includes the steps of: (a) extracting a candidate region with movement from the image photographed by the image photographing apparatus; obtaining statistical characteristics of a predetermined probability distribution function for the candidate region by using color information of corresponding pixels constituting the candidate region; And (c) determining from the statistical characteristic whether the candidate region is a smoke region or a flame region.

In the present invention, in step (b), the color information of the corresponding pixels includes at least one of brightness or chroma information.

In the present invention, a specific example of the predetermined probability distribution function is a Gaussian mixed modeling function.

In the present invention, the determination of step (c) determines whether the candidate area is a smoke area or a flame area by using each average value of a plurality of clusters constituting a statistical characteristic of the Gaussian mixing function for the candidate area.

In the present invention, the determination of step (c) determines whether the candidate area is a smoke area or a flame area by using respective variance values of a plurality of clusters constituting a statistical characteristic of the Gaussian mixing function for the candidate area.

In the present invention, the determination of the step (c) determines whether the candidate region is a smoke region or a flame region using respective average values and variance values of a plurality of clusters constituting statistical characteristics of the Gaussian mixing function for the candidate region. do.

When the fire detection method according to the present invention is carried out, there is an effect of preventing a situation leading to a large fire by capturing signs of smoke or flame in advance.

In addition, the fire detection method of the present invention compares the statistical characteristics of the flame to the smoke through the statistical probability analysis function of the candidate region image in which the motion is captured, thereby accurately specifying the characteristics of the smoke or the probability distribution function of the flame. It has the effect of catching the signs of fire more clearly than usual detection methods.

1 is a schematic block diagram of a system that technically enables a fire detection method according to the invention.
2 is an embodiment for explaining the technical idea of a fire detection method using a video signal processing device proposed in the present invention.
3 is a flowchart for determining whether a candidate region extracted according to the present invention is a postpone region.
4 is a flowchart for determining whether a candidate region extracted according to the present invention is a fire region.
FIG. 5 is an example of a difference signal image (foreground image) extracted from FIG. 2.
6 is an example of a candidate region extracted from FIG. 2.
7 is an example of experiments in the case that the smoke is notified to the manager monitor when the fire detection method according to the present invention is executed.
8 is an example of the experiment when the generation of the flame is notified to the manager monitor during the fire detection method according to the present invention.

First, in describing the present invention, when it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted, and the terms used herein are used in the present invention. As terms defined in consideration of functions, they may vary according to the intention or custom of the user or operator, and thus the meaning of each term should be interpreted at the level of ordinary skill in the art based on the contents throughout the specification.

Hereinafter, a specific embodiment for implementing the fire detection method according to the present invention with reference to the drawings will be described.

1 is a schematic block diagram of a system for technically enabling a fire detection method according to the present invention, FIG. 2 is an embodiment for explaining a technical idea of a fire detection method using an image signal processing apparatus proposed by the present invention. 3 is a flowchart for determining whether the candidate region extracted according to the present invention is a smoke region, FIG. 4 is a flowchart for determining whether the candidate region extracted according to the present invention is a fire region, and FIG. 5 is extracted from FIG. 6 shows an example of a difference signal image (foreground image), FIG. 6 is an example of a candidate region extracted from FIG. 2, and FIG. 7 is an experimental example when the occurrence of smoke is notified to the manager monitor when the fire detection method according to the present invention is executed. 8 is an example of the experiment when the generation of the flame is notified to the manager monitor when the fire detection method according to the present invention is executed.

As illustrated in FIG. 1, the fire detection method of the present invention transmits an image photographed through an image capturing apparatus 110 such as a camera to the image signal processing apparatus 112 and then displays smoke or flame characteristics in the captured image. After analyzing whether it is included or not through the various types of alarm signal generating device 114 connected through the wired or wireless network to determine whether to notify the fire alarm.

Hereinafter, a specific embodiment of a fire detection method according to the present invention processed in the image signal processing device 112 of FIG. 1 will be described with reference to FIG. 2.

As shown in FIG. 2, the fire detection method according to the present invention includes a step S201 of receiving a captured image through the image capturing apparatus 110.

The image capturing apparatus 110 may be installed in an area where a fire needs to be prevented in advance, such as indoors of a building such as an office or a building, or an outdoor public place such as a park or an amusement park for fire monitoring and detection. desirable.

The image capturing apparatus 110 to be applied in the present invention includes various types of image capturing apparatuses, such as a pan / tilt camera, a dome camera, an infrared camera, a CCTV, and the like, capable of capturing an external image.

Next, an embodiment of the present invention includes the step S202 of receiving the image captured by the image capturing apparatus 110 as the current image.

Here, the current image means an image including a background image and a foreground image. The background image refers to the image of the fixed surrounding environment captured by the image capturing apparatus 110, and the foreground image is an extraordinary environment image where the position changes with time such as a person, a cloud, a car, a smoke, a flame, and the like. Means an image to be photographed.

Next, the present invention further includes extracting a background image from the current image received through the image capturing apparatus 110 (S203).

Although various methods for extracting a background image from the current image are known in the art related to video communication, in the embodiment of the present invention, the background image is extracted from the current image by using a change amount of color information (for example, YCbCr) of the image. Method was used. However, the case in which the background image is extracted using other commonly known methods is naturally included in the scope of the technical idea of the present invention.

Next, an embodiment of the present invention includes generating a difference signal image by using the difference between the current image and the background image as described above (S204). Here, the difference signal image is not the background image, that is, the same meaning as the foreground image that is commonly used in the industry as an unusual image (for example, automobiles, people, clouds, smoke, flames, etc.).

In more detail, the extraction of the difference signal image is performed by comparing the Y, Cb, Cr image of the background image with the Y, Cb, Cr image of the current image as follows, and determining the difference, and constructing the difference signal image. It is made by extracting the values of the pixels.

D (x, y) = 1, {where │ Iy (x, y) -By (x, y) │> T1 and │Icb (x, y) -Bcb (x, y) │> T2 and │Icr ( x, y) -Bcr (x, y) │> T3}

D (x, y) = 0, otherwise

Where D (x, y) represents the difference signal image, Iy (x, y) represents the Y value of the current image, By (x, y) represents the Y value of the background image, and Icb (x, y) represents the current Cb value of the image, Bcb (x, y) represents the Cb value of the background image, Icr (x, y) represents the Cr value of the current image, Bcr (x, y) represents the Cr value of the background image. T1, T2, and T3 indicate predetermined thresholds for extracting a desired difference signal image.

5 is an example of the extracted difference signal image (foreground image). As shown in FIG. 5, pixels constituting the difference signal image (foreground image) are represented by 0 or 1 according to brightness and saturation differences between the background image and the current image, and the pixel region indicated by 1 is an area in which movement is performed. Is recognized.

Next, an exemplary embodiment of the present invention includes a step of calculating a candidate region by clustering adjacent motion pixels with respect to the aforementioned difference signal image (or foreground image) region through a labeling process.

6 shows an example of a candidate region extracted through the above-described labeling process. As shown in FIG. 6, the candidate region obtained through the labeling process for the foreground image forms a rectangular shape surrounding the difference signal image (foreground image) region.

6 illustrates an example in which three candidate regions are extracted, and a target captured by each candidate region includes a moving person, a vehicle, an animal, a flame or smoke.

Next, in the present invention, a predetermined probability distribution function, for example, Gaussian mixture modeling, is applied to extract predetermined statistical characteristics for each candidate region by using color information of pixels constituting each candidate region thus extracted. (S206).

The Gaussian mixture model for flame and smoke detection proposed in the present invention is as follows.

Gaussian mixture modeling is a method of representing input data as a sum of a plurality of Gaussian distribution functions, thereby reducing the amount of computation by modeling complex input data as the mean and variance of each Gaussian distribution function.

The k-dimensional Gaussian mixture model function for the input data can be expressed as follows.

는 각 가우시안 분포의 혼합 가중치를 나타내고,

는 가우시안 분포함수로서 다음 식으로 정의된다.Where k is the number of Gaussian distribution functions,

Represents the mixed weight of each Gaussian distribution,

Is a Gaussian distribution function defined by

는 평균,

는 공분산행렬을 나타낸다.Where d represents the dimension of the input data,

Is average,

Denotes a covariance matrix.

The Gaussian mixture model can be extracted using the EM algorithm to estimate the mean, covariance matrix, and mixed weight of each Gaussian distribution function.

In the present invention, Gaussian mixture modeling is performed using Y, Cb, and Cr values in the candidate region detected as input data.

In the present invention, the three-dimensional Gaussian mixture model is used, but the case of using a higher-order Gaussian mixture model is also included in the technical idea of the present invention. In addition, the Gaussian distribution function is described later as a cluster.

Meanwhile, since the color component of the input image signal may be different according to the performance of the camera selected by the image capturing apparatus 110, it is necessary to change a characteristic value, for example, an average value, for statistical analysis for smoke or flame discrimination.

That is, the average value of the brightness and saturation of the smoke and flame described later (μ _{Ymax ,} μ _{Ymin ,} μ _Cbmax , μ _Crmax , μ _{Cbmid ,} μ _Crmid , μ _Cbmin , μ _C) according to one embodiment of the fire detection method according to the present invention. _{rm in,} etc.) will be adjustable depending on the capabilities and capabilities of the camera selected by the imaging device. Similarly, it is necessary to change the characteristic value (average value or variance value) for statistical characteristic extraction in response to the illumination change (night, daytime, natural light, artificial light, etc.) of the shooting area.

Hereinafter, a process (steps S2071 and S2072 of FIG. 2) for determining whether the candidate region is a smoke or flame region based on the Gaussian mixture modeling result for the candidate region will be described with reference to FIGS. 3 and 4.

In the present invention, the maximum cluster, the intermediate cluster, and the minimum cluster are obtained by sorting the average values of brightness for each cluster for comparison with the flame and smoke characteristics.

As shown in FIG. 3, the present invention compares the statistical characteristics of the candidate regions with those of the smoke through the following conditions.

Condition 1.Minimum Cluster Brightness Average Above 0.4

Condition 2. The difference between the brightness averages of the maximum and minimum clusters is within 0.2

When the above condition is expressed by a formula, it is expressed as the following formula.

Condition 1: μ _Ymin > 0.4

Condition 2: μ _Ymax -μ _Ymin <0.2

Where μ _Ymax and μ _Ymin are the brightness mean values for the maximum and minimum clusters, respectively.

If the determination conditions for the above conditions are not satisfied, it is determined that the characteristic of the smoke is not implied, and the process procedure is terminated.

When the above condition is satisfied, the color difference condition is compared as follows.

Condition 3. The average value of Cb chrominance should be 0.5 or more and 0.7 for all clusters.

Condition 4. The average value of Cr color difference should be 0.45 or more and 0.55 for all clusters.

When the above condition is expressed by a formula, it is as follows.

Condition 3: 0.5 <μ _Cbmin <0.7, 0.5 <μ _YCbmid <0.7, and 0.5 <μ _Cbmax <0.7

Condition 4: 0.45 <μ _Crmin <0.55, 0.45 <μ _Crmid <0.55, and 0.45 <μ _Crmax <0.55

Where μ _Cbmin, μ _YCbmid, and μ _Cbmax is the average value of the Cb color difference for the largest cluster, the intermediate cluster, a minimum cluster, respectively, μ _Crmin, μ _Crmid, and μ _Crmax is Cr for the maximum cluster, the intermediate cluster, the minimum clusters, each Average value of color difference.

If any one of the color difference conditions is not satisfied, it is determined that the smoke does not contain the characteristics, and the process is terminated.

Next, as shown in FIG. 4, the present invention determines whether the Gaussian mixture modeling result for the candidate region includes the statistical characteristics of the flame through the following conditions.

Condition 5. The minimum cluster brightness average is at least 0.9

Condition 6. The difference between the brightness averages of the maximum clusters and the brightness mean values of the maximum and minimum clusters is 0.75 or more and 0.95 or less.

The above condition is expressed by the following formula.

Condition 5: μ _Ymin > 0.9

Condition 6: 0.75 <(μ _Ymax -μ _Ymin ) / μ _Ymax <0.95

If the result of the determination is not satisfied, the process procedure is ended by determining that the characteristic of the flame is not included. If all of the conditions are satisfied, the following color difference conditions are compared.

Condition 7. The average Cb chrominance is less than 0.55 for all clusters

Condition 8. The average value of Cr chrominance is 0.55 or more and 0.65 or less for all clusters.

The above condition is expressed by the following formula.

Condition 7: μ _Cbmin <0.55 μ _Cbmid <0.55 and μ _Cbmax <0.55

Condition 8: 0.55 <μ _Crmin <0.65, 0.55 <μ _Crmid <0.65, and 0.55 <μ _Crmax <0.65

In the present invention, the process of determining whether the candidate region is a smoke region or a flame region is performed as a step of determining whether the candidate region is a flame region after the determination as to whether the smoke region is a smoke region, but the order thereof is irrelevant and is particularly independent or parallel. Processing is also possible and is included in the technical spirit of the present invention.

Next, the fire detection method according to the present invention further comprises a step (S208) of notifying an administrator through a wired or wireless network when it is determined that smoke or flame has occurred in a predetermined area among the candidate areas. These steps can be built in a variety of ways known in the art so that the additional description will be omitted in the present invention.

7 is an example of the experiment when the smoke is notified to the administrator monitor when the fire detection method according to the present invention, Figure 8 is a case of the flame is notified to the manager monitor when the fire detection method according to the present invention Is an experimental example.

As can be seen from Figures 7 and 8, when performing the fire detection method according to the invention it can be seen that the smoke and flame generation area is displayed correctly on the monitor of the manager.

In the above-described embodiments of the present invention, a method of determining whether the candidate region is a smoke region or a flame region by using an average value of a plurality of clusters derived from the Gaussian mixture modeling of the candidate region has been described. It is only possible to use other statistical variables similar to this as examples only.

For example, by applying Gaussian mixture modeling to the candidate region, a dispersion value of a plurality of clusters is derived and compared with the dispersion value of Gaussian mixture modeling for smoke and flame characteristics from the result, and whether the candidate region is a smoke region. It is also possible to determine whether or not the flame region.

In this case, the variance value derived from the candidate region by applying Gaussian mixture modeling may be used as an independent variable for smoke or flame determination, but it may be preferable to compare the average value with the aforementioned average value.

However, when the variance value is considered, the amount of processing calculation increases, but it may be helpful to more clearly determine whether the candidate area is a smoke area or a flame area.

In addition, the technical idea of the present invention may determine whether the corresponding image of the candidate area is a smoke or a flame by using other statistical characteristics of the smoke or the flame, and the technical features describe the claims for the present invention. It is naturally included within the scope of the present invention within the scope.

As described above, the technical idea of the fire detection method according to the present invention is to extract the background image from the image photographed by the image capturing apparatus and then to separate the foreground image which is the difference signal image between the captured image and the background image, and then the labeling process After extracting the candidate region through the method, the Gaussian mixture modeling method is applied to the extracted candidates, and the average or variance of the clusters is extracted and compared with the statistical characteristics of the flame or smoke to detect the fire quickly and accurately. Suggest a method or system that can notify the user.

The technical idea of the present invention may include a case where a certain step proposed by the present invention is replaced or changed with a separate technology known in the art. In particular, in addition to the proposed Gaussian mixture model, smoke or flame statistical characteristics may be used. The same applies to the case of using various types of probability distribution functions that can be compared with statistical features of the candidate region, and such changes and substitutions are naturally included in the scope of the present invention within the scope of the claims.

For example, although YCbCr color information is used in the present invention, it is also possible to substitute and substitute RGB color information. In addition to Gaussian mixture modeling, a method of capturing the signs of a fire from its statistical characteristics by using other probability distribution functions for the candidate region is described. Naturally, it is included in the technical idea of this invention.

Claims (6)

(a) receiving the image captured by the image capturing apparatus as a current image:
(b) extracting a background image from the current image using a change amount of color information;
(c) generating a difference signal image by using the difference between the current image and the background image;
(d) extracting a candidate region by grouping adjacent motion pixels through a labeling process on the difference signal image region;
(e) obtaining statistical characteristics of a Gaussian mixture modeling function for the candidate region by using color information of corresponding pixels constituting the candidate region; And
(f) determining from the statistical characteristic whether the candidate region is a smoke region or a flame region,
In the extracting of the difference signal image in step (c), the Y, Cb, Cr image of the background image and the Y, Cb, Cr image of the current image are extracted by the following equation.
D (x, y) = 1, {where │ Iy (x, y) -By (x, y) │> T1 and │Icb (x, y) -Bcb (x, y) │> T2 and │Icr ( x, y) -Bcr (x, y) │> T3}
D (x, y) = 0, other
Where D (x, y) represents the difference signal image, Iy (x, y) represents the Y value of the current image, By (x, y) represents the Y value of the background image, and Icb (x, y) represents the current Cb value of the image, Bcb (x, y) is the Cb value of the background image, Icr (x, y) is the Cr value of the current image, Bcr (x, y) is the Cr value of the background image, T1, T2 and T3 are predetermined threshold values for extracting a desired difference signal image. The method of claim 1,
In the step (b), the color information of the corresponding pixel comprises at least one or more of brightness or chroma information. delete The method of claim 2,
The determination of the step (f) is used to determine whether the candidate region is a postpone region using each average value of a plurality of clusters constituting the statistical characteristics of the Gaussian mixed modeling function for the candidate region,
As a result of calculating the statistical characteristics of the Gaussian mixed modeling function in step (e),
The minimum brightness value of the minimum cluster is 0.4 or more, and the difference between the maximum and minimum cluster brightness averages is less than 0.2,
If the following Equations 1 and 2 are satisfied,
Equation 1: 0.5 <μ _Cbmin <0.7, 0.5 <μ _YCbmid <0.7, and 0.5 <μ _Cbmax <0.7
Equation 2: 0.45 <μ _Crmin <0.55, 0.45 <μ _Crmid <0.55, and 0.45 <μ _Crmax <0.55
(Where μ _Cbmin, μ _YCbmid, and μ _Cbmax is the average value of the Cb color difference for the largest cluster, the intermediate cluster, a minimum cluster, respectively, μ _Crmin, μ _Crmid, and μ _Crmax is for up to a cluster, the intermediate cluster, the minimum clusters, each Average value of Cr color difference)
Fire detection method, characterized in that determined in the step (f) smoke area. The method of claim 2,
The determination of the step (f) is to determine whether the candidate region is a flame region using each average value of a plurality of clusters constituting the statistical characteristics of the Gaussian mixed modeling function for the candidate region,
As a result of calculating the statistical characteristics of the Gaussian mixed modeling function in step (e), the following Equations 1 and 2 are satisfied.
Equation 1: μ _Ymin > 0.9
Equation 2: 0.75 〈(μ _Ymax − μ _Ymin ) / μ _Ymax 〈0.95
Where μ _Ymin is the brightness mean of the smallest cluster and μ _Ymax is the brightness mean of the largest cluster.
If the following Equations 3 and 4 are satisfied
Equation 3: μ _Cbmin <0.55 μ _Cbmid <0.55 and μ _Cbmax <0.55
Equation 4: 0.55 <μ _Crmin <0.65, 0.55 <μ _Crmid <0.65, and 0.55 <μ _Crmax <0.65
Where μ _Cbmin is the average Cb color difference for all clusters and μ _Crmin is the Cr color difference average for all clusters.
Fire detection method, characterized in that it is determined as the flame zone in the step (f). delete

Images