KR100862409B1 - The fire detection method using video image - Google Patents

Abstract

A fire detection method using video images is provided to detect flames within a detection area automatically and to respond to the initial flame rapidly by obtaining a binary image by a flame candidate pixel, and another binary image by comparing the binary image with the color image. A fire detection method using video images comprises the steps of: outputting an RGB(Red, Green, Blue) color image(Time=t) from a color camera and outputting an RGB color image(Time=t-1); converting two RGB color images into an HSI(High Speed Interface) color model; obtaining a binary image(Ft) by a flame candidate pixel by using the color features of the HIS color space through a following formula; Ft-1(i,j)=1 if (I(i,j)>240 AND S(i,j)<20) OR (100<I(i,j)<200 AND 30<S<70 AND -30<H<60) 0 otherwise, and obtaining a binary image(P) by comparing the binary image(Ft) with the color image(t-1) through a following formula; P(i,j)=Ft-1(i,j)+Ft(i,j); obtaining a binary image(B) by dividing the binary image(P) into areas with proper sizes and comparing the areas with a preset ratio by applying the following formula; B(x,y)=1 if #{Pl P(i,j)=1 AND P(i,j)∈B(i,j)}/#{Pl P(i,j)∈B(x,y)}>r; and comparing the binary image(B) with a threshold which is Theta by using a following formula; if#{BlB=1}>Theta OFF Otherwise to judge the fire finally.

Description

The fire detection method using video image}

1 is a flow diagram illustrating an overall system according to an embodiment of the invention.

2 is a fire image photograph obtained at time t-1

3 is a fire image photograph obtained at any time t moment

4 is a binary image photograph obtained by extracting a pixel determined to correspond to the color of a flame based on the color characteristics of the pixel in the image of FIG.

5 is a binary image photograph obtained by extracting a pixel determined to correspond to the color of a flame based on the color characteristics of the pixel in the image of FIG.

FIG. 6 is an image photograph obtained by performing an exclusive OR of FIGS. 4 and 5 as in Equation (2).

FIG. 7 is an image photograph obtained by determining FIG. 6 based on region 3 by Equation 3;

8 is an image of an omnidirectional camera image of a flame and an electric heater of a similar color coexist

FIG. 9 is an image photograph of the image of FIG. 8 processed by equations (1) and (2).

  The present invention uses a general video camera based on the measurement of visible light to automatically detect and alarm the fire occurred in the monitoring area, enabling the recognition and response early in the fire occurrence to minimize physical and human damage The present invention relates to a fire detection method using a video image.

  In general, a method of detecting fires such as smoke or temperature by using various types of sensors for detecting fires is widely used.

In this case, the device has the advantage of simple and low cost, but there is a possibility of malfunction, there is no way to check the site remotely in case of malfunction, and even if a fire occurs and alarms properly, information such as the size of the fire and the source of the fire You can't get it in real time.

This problem also occurs when using a sensor that directly measures the flame, such as an infrared sensor.

In addition, in the case of a method of detecting a fire by employing a special camera based on the measurement of light having a wavelength other than visible light, such as an infrared camera or a thermal imaging camera, there is a problem that the price is generally expensive.

In order to solve this problem, the fire detection device and method and recording medium robust to the illumination using the color image and the surrounding environment of the pre-registered registration number '10 -0680114 'are inputted as RGB in the video image. It is possible to detect the occurrence of a flame by calculating the egg (R) and the ground (G) signals among the color images, and by identifying the flames caused by the fire and the objects with the color similar to the artificial lighting or the surrounding flames in the video image. It is presenting a way.

However, in the prior art using the RGB color characteristics of the image, since the color value of the flame changes according to the change in the brightness of the surrounding environment, the degree of fire progress, the temperature and the material of the medium, only the color is distinguished like RGB. The method is likely to malfunction. There is a problem that may cause malfunction by detecting the flame of the stove or the flame of the stove as a fire.

The present invention has been made to solve the above problems, and converts the color, brightness, and saturation from the photographed color image into a color model that can separate hue, brightness, and saturation, such as HSI space, based on the color and dynamic characteristics of the flame. And detecting the pixel corresponding to the boundary, comparing the detected flame candidate pixel with the before and after image, and examining how the flame regions change with time, thereby determining the occurrence of flame, thereby minimizing the possibility of error in flame detection. It is an object of the present invention to provide a fire detection method using a video image having a feature that can be easily applied to a general camera.

The present invention for achieving the above object relates to a fire detection method using a video image, more specifically, an RGB color image (Time = t) of a random time from a color camera, and RGB color image of t-1 time Outputting (Time = t−1); Converting a color space of the two RGB color images into an HSI color model; Obtaining binary image Ft as flame candidate pixel by color characteristics in HSI color space through Equation (1), and comparing binary image Ft with color image of t-1 time through Equation (2) to obtain binary image P ; Dividing the binary image P into a region of an appropriate size and applying the equation (3) to obtain a binary image B by comparing with a predetermined ratio; Finally determining the fire by comparing the binary image B with the threshold value θ at the global level using Equation (4); Characterized in that consists of.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating an entire system according to an embodiment of the present invention. Referring to the entire flow of a fire detection method using a video image according to an embodiment of the present invention, first, an RGB color image output from a color video camera that is generally used is described. It outputs RGB color image acquired at any time t moment and RGB color image at t-1 moment so as to analyze the change of the image caused by fire through it, and accurately detects the flame occurrence from the flame and the object similar to the flame. To convert the RGB color image into HSI color space that can separate brightness, saturation, and color, extract the flame candidate pixel using the color information to obtain binary image Ft, color image and XOR logic operation of t-1 The binary image P is obtained by comparing the result, and the binary image B is obtained by judging it by the equal division and the divided regions of the image. To detect the spread of fire within the area to the entire image is configured to determine the fire.

Hereinafter, an operation example of the fire detection method using the video image of the present invention will be described in more detail step by step, Figure 2 shows a fire image of the instant time t-1, Figure 3 is a fire obtained at any time t moment As shown in the image, as shown, the flame does not have a constant shape, so the next moment, it shows another form as shown in FIG. 3, and as time passes, the shape is enlarged as well as the change in size.

Here, the random time t instant means a time point at which the flame is generated, and the time point of t-1 which compares the instant of flame occurrence with the previous image is determined by the t instant.

Such an image is output as a general RGB color image, which is converted into a HSI color model in which hue, brightness, and saturation are separated, and through this, pixels belonging to an experimentally determined flame color range are first detected.

For example, the following equation (1) is a process of obtaining a binary image Ft from a pixel estimated as a flame pixel by color characteristics in an HSI color space.

Ft (i, j) = 1 if (I (i, j)> 240 AND S (i, j) <20) OR

             (100 <I (i, j) <200 AND 30 <S <70 AND -30 <H <60)

       0 Otherwise expression (1)

In Equation (1), AND and OR mean AND and OR.

In addition, I, S, and H are input color variable values. If the value of the binary image Ft is 1, it is a candidate pixel that can be finally determined as a flame, and if it is 0, it is a non-pixel.

The number used as a threshold in Equation (1) is obtained experimentally, and the specific number or method may be changed depending on factors such as environment and camera used.

4 and 5 show binary images obtained by extracting a pixel determined to correspond to the color of a flame based on the color characteristics of the pixels in the image by applying Equation (1) to FIGS. 2 and 3, respectively. As described in (1), the determination of a flame pixel by color has a problem that an object of a color similar to a flame is also detected.

Therefore, it is correct to call the detected pixel as the flame candidate pixel as shown in equation (1).

To solve this problem, a binary image (t, t-1) composed of two flame candidate pixels obtained from a before and after video image is compared, and a new binary image can be obtained only with a change part.

That is, when the exclusive OR (XOR) operation of FIG. 4 and FIG. 5 is performed as in Equation (2), an image as shown in FIG. 6 can be obtained.

FIG. 6 is an image obtained by exclusive OR of FIGS. 4 and 5 as shown in Equation (2). As shown in FIG. 6, the boundary and peripheral pixels of the flame are extracted as the shape of the flame changes. Even if the equation (2) is performed, it is removed from the binary image.

Ft(i,j) 식 (2)P (i, j) = Ft-1 (i, j)

Ft (i, j) Formula (2)

는 배타적 논리합 연산자를 의미한다. However, in formula (2), Ft is a binary image obtained by processing the color image obtained at an arbitrary time t by the formula (1),

Means the exclusive OR operator.

The next step of the invention is region level image processing.

That is, the image is divided into regions of a suitable size, and if the pixels determined to be related to the flame in each region are more than a predetermined ratio, the region is determined as the flame region.

Equation (3) represents an area level determination method, and if the number of pixels determined as 1 by Equation (2) out of the total number of pixels P in any area B exceeds a predetermined ratio, the area is determined as a flame area. do.

  B (x, y) = 1 if # {P | P (i, j) = 1 AND P (i, j) ∈B (x, y)} / # {P | P (i, j) ∈B (x, y)}> r

         0 Otherwise. Formula (3)

Where B (x, y), where 1≤x≤N, 1≤x≤M, is an arbitrary area when the image is divided into N × M areas, and # (·) is an operation symbol indicating the number of elements It is used to mean “Number of” and r has a value between 0 and 1 at a randomly designated rate by the user.

FIG. 7 shows an image obtained by region-based determination by Equation (3) in FIG. 6. As can be seen from the figure, an area-level process is to analyze the original image by lowering the resolution.

The next step of the invention is the determination by observation at the global level of the image.

That is, if the flame region obtained in the previous step in the whole image is greater than or equal to the arbitrary threshold value θ, it is finally determined as a fire, and specifically, according to equation (4).

At this stage, we will look at how the flame zones change over time, and if the number of flame zones in the whole image increases by more than a certain amount, it can be interpreted as a fire that starts and a fire increases.

This phenomenon is an important feature in comparison with the small and almost constant size of flame generated from a cooking appliance such as a gas burner or a hot air balloon such as a heater.

  Fire judgment = ON if # {B | B = 1}> θ

           OFF Otherwise expression (4)

  The method of the present invention according to the above procedure is more accurate than the method of detecting a fire using only the color of the pixel, and is relatively independent of the type of the camera, and is generally used for omnidirectional cameras, web cameras, or closed circuit cameras. It can be easily applied to the camera image being used.

FIG. 8 is an image of an omnidirectional camera, in which a flame of a similar color and an electric heater coexist, and FIG. 9 shows a result of processing the image of FIG. 8 by equations (1) and (2). For example, it can be seen from the image of the omnidirectional camera that the electric fan-type electric stove is hardly detected as shown in FIG. 9 even though its color is similar to the flame.

That is, the structure of the present invention is to process two consecutive color images output from the camera at the pixel level and the area level, and to make a decision at the global level.

According to the fire detection method using the video image of the present invention described in the above configuration, in detecting the flame, not only prevents malfunction due to the similar color of the surroundings, but also affects the fire situation without being affected by the flames such as the stove and the stove. It can be detected and can be applied to video cameras that are commonly used.

Claims (1)

Outputting an RGB color image (Time = t) at an arbitrary time and an RGB color image (Time = t-1) at a time t-1 from a color camera; Converting a color space of the two RGB color images into an HSI color model; Expression, Ft (i, j) = 1 if (I (i, j)> 240 AND S (i, j) <20) OR              (100 <I (i, j) <200 AND 30 <S <70 AND -30 <H <60) The binary image Ft is obtained from the candidate candidate flame by the color characteristics in the HSI color space through 0 Otherwise, and the binary image Ft is expressed as the color image of t-1 time, and P (i, j) = Ft-1 (i, j )

Obtaining a binary image P by comparing through Ft (i, j); Divide binary image P into appropriately sized regions and use the equation, B (x, y) = 1 if # {P | P (i, j) = 1 AND P (i, j) ∈B (x, y)} / # {P | P (i, j) ∈B (x, y)}> r          0 Otherwise. Applying a to obtain a binary image B by comparing with a predetermined ratio; Binary image B is formulated, fire judgment = ON if # {B | B = 1}> θ            Finally determining a fire by comparing the threshold θ at a global level using OFF Otherwise; Fire detection method using a video image, characterized in that consisting of

Images