KR101371647B1 - Apparatus for detecting fire flame based image data and method of the same - Google Patents

Abstract

The present invention discloses a fire detection device and a fire detection method. The fire detection apparatus determines a candidate region for determining whether the movement region is a fire candidate region by comparing a camera photographing a predetermined space, a detection unit detecting a movement region in the captured image, and a color of the detected movement region and a predetermined color. If the motion region is a fire candidate region, the feature value extracting unit extracts a plurality of characteristic values for spatial and temporal changes of the fire candidate region, and the fire region in which the fire of the fire candidate region occurs using the extracted plurality of characteristic values. It includes a fire judgment unit for determining the recognition.

Description

Fire detection device based on image data and its method {APPARATUS FOR DETECTING FIRE FLAME BASED IMAGE DATA AND METHOD OF THE SAME}

The present invention relates to an apparatus and method for detecting a flame, and more particularly, to an apparatus and method for extracting a fire candidate region using image data and determining whether a flame is present in a candidate region of fire using a feature parameter vector extracted therefrom. will be.

Fires can cause enormous damage to property and life. Preventing fires is a priority, but in addition to preventing fires, there is a need for a fire alarm system that can detect fires accurately.

Current fire alarm systems are mostly sensor-based detectors such as heat and smoke, and these detectors can only be detected when heat or smoke reaches the sensor after a certain time after the fire has occurred. These methods have the disadvantage of not detecting a fire early and its performance in a large place and open space is poor.

On the other hand, a fire detection system using a camera can be implemented using surveillance cameras installed inside and outside the building, so there is no additional cost, and in case of fire or smoke, there is no need to wait for the spread of heat or smoke through the camera. There is an advantage that can be detected immediately.

Regarding the algorithm for detecting a fire, T. Celik et al. Described the color pixels of fire in the "Fire detection in video sequences using a generic color model" (Fire Safety Journal, vol. 44, pp. 147-158, 2009). We proposed a method that combines statistics and foreground object information (algorithm 1). Ko, BC et al. Proposed a method of detecting fire using support vector machines in "Fire detection based on vision sensor a support vector machines" (Fire safety Journal, vol. 44, pp. 322-329, 2009). (Hereinafter, algorithm 2). Toreyin, BU, et al., In Waveforms obtained by motion domain detection and wavelet transform in "Computer vision-based method for real-time fire and flame detection" (Pattern Recognition Letter, vol. 27, pp. 49-58, 2006). A fire detection method using energy is proposed (Algorithm 3). Borges, P.V.K. Et al., "A probabilistic approach for vision-based fire detection in videos." (IEEE. Trans.on Circuit and Systems for Video Technology, vol. 20, pp. 721-731, 2010) proposed a flame detection method using video data (Algorithm 4).

Although the proposed algorithm shows good results in limited environments, it is pointed out that the accuracy of the general fire environment is low and the false detection rate is high.

An object of the present invention is to provide a flame detection apparatus and method that can extract the spatial and temporal characteristic values of a flame from the image data of the flame, and determine whether or not a fire using the same.

In order to achieve the above object, the flame detection device, in the fire detection device,

A camera for photographing a predetermined space, a detector for detecting a movement region in the captured image, and a candidate region determination unit for determining whether the movement region is a fire candidate region by comparing a color of the detected movement region with a predetermined color. A feature value extractor for extracting a plurality of feature values for spatial and temporal changes of the fire candidate area if the motion area is a fire candidate area; and a fire occurs in the motion area using the extracted feature values It includes a fire judging unit to determine whether or not.

The plurality of characteristic values may be at least two characteristic values among size change characteristics, roughness characteristics, fixed region characteristics, and directional characteristics of the fire candidate region.

The candidate region determining unit may convert the RGB color space of the color of the moving area into a CIE-LAB color space and compare the candidate color in the CIE-Lab color space with a preset flame color to determine the fire candidate area. It may include a comparator for determining the recognition.

The size change characteristic may be calculated a plurality of times a difference value between the candidate area size of a previous frame A _{n -1} and the size of the candidate area of a current frame A _n , so that the size change average value M _A and size change are calculated. The variance value V _A can be calculated.

The roughness characteristic may be calculated as a density average value M _I and a density variance value V _I of the pixels in the candidate region.

The fixed region characteristic may be calculated as a plurality of fixed region characteristic values associated with a region in which there is no motion in the candidate region.

The direction characteristic may be calculated as a motion history and a motion gradient by tracking a flame motion of the candidate region using a motion template.

If it is determined that a fire has occurred in the movement region, the controller may further include an alarm unit generating an alarm signal.

According to another embodiment of the present invention, the fire detection method comprises the steps of: imaging a predetermined space; detecting a movement area in the captured image; comparing the color of the detected movement area with a preset color; Determining whether the area is a fire candidate region; extracting a plurality of characteristic values for spatial and temporal changes of the fire candidate region; and using the extracted plurality of characteristic values if the movement region is a fire candidate region. And determining whether a fire has occurred in the fire candidate region.

The plurality of characteristic values may be at least two characteristic values of a size change characteristic, a roughness characteristic, a fixed region characteristic, and a directional characteristic of the fire candidate region.

The determining of the fire candidate area may include converting an RGB color space of the color of the motion area into a CIE-LAB color space and comparing the candidate color in the CIE-Lab color space with a preset flame color. And determining whether the fire candidate region is present.

The size change characteristic may be calculated a plurality of times a difference value between the candidate area size of a previous frame A _{n -1} and the size of the candidate area of a current frame A _n , so that the size change average value M _A and size change are calculated. It may be a dispersion value (V _A ).

The roughness characteristic may be a density average value M _I and a density dispersion value V _I of the pixels in the candidate region.

The fixed region characteristic may be a plurality of fixed region characteristic values associated with a region in which there is no motion in the candidate region.

The direction characteristic may be a motion history and a motion gradient obtained by tracking a flame motion of the candidate area using a motion template.

If it is determined that a fire has occurred in the fire candidate region, the method may further include generating an alarm signal.

The present invention extracts the feature parameter vector from the image data of the flame, and determines whether or not the fire using a plurality of feature parameter vectors extracted by the support vector machine, the accuracy of flame detection is improved, and the false detection is reduced To provide.

1 is a block diagram for explaining a flame detection apparatus according to an embodiment of the present invention;
2 is a flowchart illustrating a flame detection method according to another embodiment of the present invention;
3 is a view showing an image of a flame image in a predetermined area according to another embodiment of the present invention;
4 is a view showing a motion region of a flame in a predetermined region according to an embodiment of the present invention;
FIG. 5 is a view illustrating an image photographing a moving person in a preset area according to an embodiment of the present invention; FIG.
6 is a view showing a motion region of a person moving in a predetermined region according to an embodiment of the present invention;
7 is a view for explaining an RGB color space according to an embodiment of the present invention;
8 is a view for explaining a CIE-LAB color space according to an embodiment of the present invention;
9 is a view showing a flame image captured in a predetermined area according to an embodiment of the present invention;
10 is a view showing contour roughness based on the image shown in FIG. 9 of the present invention;
11 is a view for explaining directional characteristics according to an embodiment of the present invention;
12 is a view for comparing the accuracy of the flame detection method and another flame detection method of the present invention,
13 is a view for comparing the error value of the flame detection method and the other algorithm of the present invention.

Hereinafter, with reference to the accompanying drawings will be described in sequence the configuration of a set-top box according to an embodiment of the present invention.

1 is a block diagram illustrating a fire detection apparatus according to an embodiment of the present invention.

Referring to FIG. 1, a fire detection apparatus according to an embodiment of the present invention may include a camera 100, a detection unit 110, a candidate region determination unit 120, a characteristic value extraction unit 130, and a fire determination unit ( 140 and an alarm unit 150.

The camera 100 may capture a still image or a moving image of a preset space.

In detail, the camera 100 may be configured in various formats of the image data photographed in the predetermined area. That is, image data captured in various formats such as JPG, BMP, GIH, and CNP are constituted.

The detector 110 may receive an image captured by the camera 100 and detect a movement area from the image data.

Specifically, that is, each pixel region of the movement region may be represented by the following equation 1 by a Gaussian function.

는 k번째의 평균이고,

는 k번째의 분산이다. σ는 표준 편차이고 D는 스케일링 성분이며,

이다. In this case, x is video data and θ _k is a component parameter, which includes an average and a variance.

Is the k-th average,

Is the kth variance. σ is the standard deviation and D is the scaling component,

to be.

Position data of each pixel constituting the image captured by Equation 1 described above can be calculated.

Also, the background pixel of the movement area may be determined using Equation 2 below.

Where _{i i , t} are the current pixel values at t time, μ _{i, t-1} are the mean values at t-1 hours, and σ _t-1 are the covariance values at t-1 hours.

Equation 2 determines the background pixel of the motion area. In each frame, the values of ω _k , μ _k , and σ _k can be updated by an expansion likelihood algorithm.

As described above, the moving area can be detected by calculating the position data of each pixel of the captured image and determining the background pixel in each pixel.

The candidate region determiner 120 may determine whether the movement region is a fire candidate region by comparing the detected color of the movement region with a preset color.

In detail, the candidate region determiner 120 may include a converter (not shown) and a comparator (not shown).

The converter may convert the RGB color space of the color of the moving area to the CIE-LAB color space.

The comparator may determine whether the motion area is the fire candidate area by comparing the candidate color in the CIE-Lab color space with a preset flame color. At this time, the color of the flame may be segmented by fuzzy C-mean clustering.

The color space of the moving area is the RGB color space, and when it is analyzed using this, it is not effective, so the RGB color space is converted to the CIE-Lab color space.

If it is determined that the movement region is a fire candidate region, the characteristic value extractor 130 may extract a plurality of characteristic values for spatial and temporal changes of the fire candidate region.

Specifically, the plurality of characteristic values may include two or more of the size change characteristic, the roughness characteristic, the fixed region characteristic, and the directional characteristic of the fire candidate region.

In this case, the area randomness may be obtained by comparing the size change of the fire candidate region of the previous frame with the fire candidate region of the current frame. That is, in general, the image size of the flame is a big difference between the previous frame and the current frame. Therefore, by calculating a difference value between the size of the candidate region of the previous frame and the current frame, and calculating it repeatedly, the average value M _A and the dispersion value V _A of the difference value are calculated as the size change characteristic.

Surface roughness may be obtained by analyzing the intensity of all the pixels. In general, the pixels constituting the image of the flame have a characteristic that the contrast is not uniform. In other words, the image of the flame has a roughness. In determining whether there is a flame, two characteristics may be extracted from the candidate fire region using these characteristics. The two characteristic values are the average value M _I and the variance value V _I of the contrast of all the pixels in the fire candidate region.

Contour roughness is that the contour of the moving area is usually rough and coarse. To extract this fixed region characteristic, an invariant moment can be used. Using the constant moment, seven fixed area characteristic values can be obtained. The characteristic value of the constant moment has a characteristic that does not depend on the position of the screen area.

Motion estimation plays an important role in video flame detection, which provides both spatial and temporal features of the fire candidate region. That is, the motion template tracks the movement of the fire candidate region, thereby obtaining a motion history and a motion gradient in the fire candidate region.

The fire determination unit 140 may detect whether or not a fire occurs as a result of the input by inputting the plurality of characteristic values of the support vector machine.

That is, the support vector machine inputs the plurality of feature value vectors described above as the support vector. We also use a Gaussian radial basis function that automatically generates the weights and thresholds of the support vectors.

A Gaussian radial basis function can be expressed by the following equation.

, n은 트레이닝 세트의 크기이며, δ는 커널 함수의 변량(variance)이다Here, x _y and x _j represent a plurality of characteristic value vectors,

where n is the size of the training set and δ is the variance of the kernel function.

When it is determined that the fire has occurred by the fire determination unit 140, the alarm unit 150 performs a function of notifying the occurrence of fire in various ways such as sound, light, and vibration so that the user can know.

2 is a flowchart illustrating a flame detection method according to an embodiment of the present invention.

2, in the flame detection method according to an embodiment of the present invention, a predetermined spatial imaging step (S200), a motion region detection step (S210), a fire candidate region determination step (S220), a plurality of feature value extraction A step S230, a fire determination step S240, and an alarm generation step S250 are included.

The preset spatial imaging step (S200) is a step of acquiring image data by photographing a preset space.

In this case, the image photographing the preset space is a moving image or a plurality of continuous still images.

The motion region detection step S210 is a step of detecting a motion region in the captured image by using a Gaussian Mixture Model based on the captured image.

Looking at the motion area in more detail as follows.

3 is a view for explaining a color image of the flame according to an embodiment of the present invention.

Referring to FIG. 3, the image captured by the camera is a moving image or a plurality of continuous still images.

3 shows a still-cut of a moving image. 3 shows a color image of the flame.

4 is a view for explaining a motion region of the flame according to an embodiment of the present invention.

Referring to FIG. 4, an area in which a motion occurs in the entire image of the flame may be extracted based on the color image data of the flame illustrated in FIG. 3.

5 is a view for explaining a color image of a moving person according to an embodiment of the present invention.

Referring to FIG. 5, it can be seen that the entire image including the moving person is illustrated. This corresponds to a still cut of a moving image captured by the camera.

FIG. 6 is a diagram illustrating an extraction of a motion region from an entire image including a moving person according to an embodiment of the present invention.

Referring to FIG. 6, it can be seen that the moving region is extracted from the entire image including the moving person illustrated in FIG. 5.

3 and 5 illustrate a color image captured by a camera, and the color image may be implemented in various formats. Moving regions can be detected using a Gaussian mixture model throughout the color image.

Using this Gaussian mixture model, it is effective for the flame detection algorithm. That is, it can successfully handle lighting changes and weaken the effects of finely repeating motions such as moving vegetation such as leaves.

The input data of this Gaussian mixture model is color image data, and the output data is a binary image.

Determining whether the motion area is a fire candidate area (S220) includes converting an RGB color space into a CIE-Lab color space and determining whether the motion area is a fire candidate area.

The conversion of the RGB color space to the CIE-Lab color space is described in more detail below.

7 is a diagram for describing an RGB color space according to an embodiment of the present invention.

Referring to FIG. 7, the RGB color space is a color space consisting of a RED axis, a GREEN axis, and a BLUE axis. Fuzzy C-mean clustering can segment the color of the motion region. A fuzzy C-mean clustering scheme may be used to segment the color of the moving area.

The color space of the color of the moving area is the RGB color space. RGB color space is not efficient. Therefore, in an embodiment of the present invention, the RGB color space is converted to the CIE LAB color space and used.

 8 is a view for explaining a CIE-Lab color space according to an embodiment of the present invention.

Referring to FIG. 8, the CIE-Lab color space is a color space defined by the CIE based on the opposite colors of yellow-blue and green-red for detecting color. The CIE-Lab color space makes it easy to guess the range and direction of color errors when toning.

Here, L * represents a reflectance (brightness equal to human luminous sensitivity), and units below the decimal point may be expressed in steps from 0 to 100. a * is a chromaticity diagram. + a * represents red, and -a * represents green. b * is a chromaticity diagram, with + b * representing yellow and -b * representing blue.

The fire candidate region determination step (S220) determines whether the motion region is a fire candidate region by comparing the candidate color in the CIE-Lab color space with a preset flame color.

Two elements (A and B) are used in the CIE-Lab color space. By spatial fuzzy C-mean clustering, a cluster of motion regions is formed, and the center value of each cluster of the motion regions is calculated.

That is, by comparing the center value of each cluster and the center value of the flame with each other, it is possible to select a fire candidate region from the movement region.

In the extracting of the plurality of characteristic values (S230), when the motion region is determined to be a fire candidate region in step S220, the plurality of characteristic values are extracted for spatial and temporal changes of the fire candidate region.

In the extracting of the plurality of characteristic values (S230), the plurality of characteristic values of the fire candidate region may be extracted. That is, the plurality of characteristic values of the fire candidate region include values of at least two characteristics of the size change characteristic, the roughness characteristic, the fixed region characteristic, and the directional characteristic.

First, the size change characteristic is an average value M _A obtained by using a plurality of difference values calculated by comparing the candidate region size of the previous frame A _{n -1 with} the candidate region size of the current frame A _{n a} plurality of times. ) And the variance value (V _A ).

Next, the roughness characteristics are the contrast average value M _I and the contrast dispersion value V _I of the pixels in the fire candidate region. This means that the gray image of the flame uses a characteristic in which the contrast of the pixels is uneven, so that the roughness characteristic of the fire candidate region becomes the contrast average value M _I and the intensity dispersion value V _I of the pixels of the candidate region.

9 is a diagram illustrating a flame image captured in a preset area according to an embodiment of the present invention.

Referring to FIG. 9, it may be confirmed that the flame is an image generated in the preset region.

10 is a diagram illustrating contour roughness based on the image shown in FIG. 9 of the present invention.

Referring to FIG. 10, the contour of the moving area is usually rough and coarse. This contour roughness characteristic can be extracted using invariant moments. In other words, the contour characteristics of the roughness of the fixed plurality of regions associated with the non-moving region in the fire area candidate values _{(H 1, H 2 ... H} 7). The value of the invariant moment is not dependent on the area position of the screen.

11 is a view for explaining a direction characteristic according to an embodiment of the present invention.

Referring to FIG. 11, the left view of FIG. 11 is an image of a moving person and a situation in which a flame has occurred. FIG. 11 is a diagram illustrating an image of a fire candidate region extracted from moving regions in the left view. 11 is a diagram representing the motion history and the motion gradient of the intermediate view.

That is, motion estimation is related to the calculated motion history and motion gradient by tracking the motion of the fire candidate region using a motion template.

In the fire determination step (S240), when it is determined that a fire has occurred in the fire candidate region (Y in S240), an alarm is generated to notify the fact that the fire has occurred (S250).

12 is a view for comparing the accuracy of the flame detection method and another flame detection method of the present invention.

According to the flame detection method and other flame detection method according to the present invention, the accuracy of fire detection is calculated by the following equation (4).

Here, the accuracy is the percentage of positive video (number of fire frames) and total number of the detected frames.

Referring to FIG. 12, it can be seen that the method according to the present invention has higher accuracy than the method according to the other method.

13 is a view for comparing the error value of the flame detection method and the other algorithm of the present invention.

The error rate (error) of the fire detection is calculated by the following equation (5).

Here, the error rate is a percentage of negative video (number of non-fire frames) and total number of fire detected frames.

Table 1 below is a table comparing the performance of the fire detection method according to the present invention and the fire detection method according to another algorithm (1 to 4).

Referring to Table 1, it can be seen that the method of the present invention exhibits a remarkable effect with high accuracy and low error rate compared to other algorithms.

This extracts a moving region from the image of the preset region, extracts a fire candidate region from the extracted moving region, and calculates a unique characteristic value of the extracted fire candidate region. By inputting the calculated characteristic value into the support vector machine, it is determined whether a fire has occurred in the candidate fire area, and thus the accuracy of judging whether a fire has occurred is increased and the error rate is low.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.

Claims (16)

In the fire detection device,
A camera for photographing a predetermined space;
A detector configured to detect a movement area in the captured image;
A candidate region determination unit comparing the detected color of the movement region with a preset color to determine whether the movement region is a fire candidate region;
A feature value extracting unit extracting a plurality of feature values for spatial and temporal changes of the fire candidate region if the movement region is a fire candidate region; And
And a fire determination unit determining whether the fire of the fire candidate region is a fire region using the extracted plurality of characteristic values.
The candidate region determination unit,
A converter for converting an RGB color space of the color of the moving area into a CIE-LAB color space; And
And a comparator for determining whether the motion area is the fire candidate area by comparing the candidate color in the CIE-Lab color space with a preset flame color. The method of claim 1,
The plurality of characteristic values,
And at least two characteristic values of a size change characteristic value, a roughness characteristic value, a fixed region characteristic value, and a direction characteristic value of the fire candidate region. delete 3. The method of claim 2,
The magnitude change characteristic value is,
A plurality of difference values calculated by comparing the candidate region size of the previous frame A _{n -1} with the candidate region size of the current frame A _{n a} plurality of times, and the average value change size of the plurality of difference values M _A And a dispersion value V _A. 3. The method of claim 2,
The roughness characteristic value is
And a contrast mean value (M _I ) and a contrast variance value (V _I ) of pixels in the candidate region. 3. The method of claim 2,
The fixed region characteristic value is
And a plurality of fixed region characteristic values calculated in relation to the non-movement region among the fire candidate regions. 3. The method of claim 2,
The direction characteristic value is
Flame detection apparatus characterized in that the motion history and motion gradient extracted by tracking the flame movement of the candidate region using a motion template (Motion Template). The method of claim 1,
And a fire unit for generating an alarm signal when it is determined that a fire has occurred in the fire candidate region. In the fire detection method,
Imaging the predetermined space;
Detecting a movement area in the captured image;
Comparing a color of the detected motion area with a preset color to determine whether the motion area is a fire candidate area;
Extracting a plurality of characteristic values for spatial and temporal changes of the fire candidate region if the movement region is a fire candidate region; And
And determining whether the fire candidate region is a fire region in which a fire occurs using the extracted plurality of characteristic values.
The determining of the fire candidate area may include:
Converting an RGB color space of the color of the motion area into a CIE-LAB color space; And
And determining whether the motion area is the fire candidate area by comparing the candidate color in the CIE-Lab color space with a preset flame color. 10. The method of claim 9,
The plurality of characteristic values,
And a value of at least two of a size change characteristic, a roughness characteristic, a fixed region characteristic, and a directional characteristic of the fire candidate region. delete 11. The method of claim 10,
The magnitude change characteristic value is,
A plurality of difference values calculated by comparing the candidate region size of the previous frame A _{n -1} with the candidate region size of the current frame A _{n a} plurality of times, and the average value change size of the plurality of difference values M _A And a dispersion value V _A. 11. The method of claim 10,
The roughness characteristic value is
And a contrast average value (M _I ) and a contrast value dispersion value (V _I ) of pixels of the candidate region. 11. The method of claim 10,
The fixed region characteristic value is
And a plurality of fixed region characteristic values associated with a region in which there is no motion in the candidate region. 11. The method of claim 10,
The direction characteristic value is
Flame detection method characterized in that the motion history and motion gradient extracted by tracking the flame movement of the candidate region using a motion template (Motion Template). 16. The method of claim 15,
And if it is determined that a fire has occurred in the fire candidate region, generating an alarm signal.

Images