KR101490769B1 - Method and apparatus of detecting fire using brightness and area variation - Google Patents

Abstract

Disclosed are a method and an apparatus of detecting fire using brightness and area variation. The method of detecting fire comprises: a conversion step converting an input image to a YCbCr image; a step of generating a brightness variation accumulation image by accumulating brightness variations for the frames of the YCbCr image; a step of generating a primary flame candidate binary image configured only with the pixels corresponding to a flame candidate among the pixels in the brightness variation accumulation image by considering an average brightness of the brightness variation accumulation image; a step of generating a secondary flame candidate binary image by removing non-flame pixels from the primary flame candidate binary image by considering a difference among chrominance components of the primary flame candidate binary image; and a final flame area determining step determining a flame candidate area, whose area included in a block of the secondary flame candidate binary image increases in time, as a final flame area. According to the present invention, an erroneous fire detection rate is reduced, and the fire areas can accurately and quickly be detected.

Description

FIELD OF THE INVENTION [0001] The present invention relates to a method and apparatus for detecting fire using brightness and area change,

FIELD OF THE INVENTION The present invention relates to a flame detection algorithm used in a vision-based fire detection system, and more particularly, to a flame detection algorithm using a luminance change accumulation image and a color feature of a flame in real- And a technique for determining a block in which a change in the area of the flame candidate area continuously occurs as a flame region.

Once a fire occurs, the extent of the damage or the extent of the damage is very large, including damage to property as well as property damage. In addition, the losses caused by the fire can not be recovered or compensated in many cases, such as the disappearance of the same kind in the early Joseon period, which was the treasure No. 479 in Korea due to the fire of Noksansa in 2005. Specifically, the number of fires in Korea in the last five years is about 220,000, and the number of fatalities and injuries has exceeded about 10,000, resulting in property loss of about 1.5 trillion won.

Therefore, it is very important to develop a fire detection system to reduce fire and damage to property. Conventional fire detection systems consist mainly of fire detection sensors (smoke detection, flame detection, temperature detection sensors) and fire surveillance cameras.

However, when a fire is detected by using a fire detection sensor, there is a high possibility of fire detection depending on the surrounding environment. For example, when using a smoke sensor, smoke may not be detected by the sensor due to the diffusion of air. Also, if a thermal sensor is used, it is not possible to quickly detect the exact location of the fire when the ambient temperature is already high. In order to perform fire detection for a wide area, it is necessary to install such sensors at regular intervals. Therefore, the larger the area of the area to be sensed, the exponential increase in the cost of constructing the fire detection system, and the maintenance cost after the system is built up.

In order to solve these problems, a camera capable of photographing a wide area to be monitored has been installed, and a visual fire detection technique using a camera image has been developed. The video fire detection technology can overcome the problems that occur when using the fire detection sensors and has an advantage that it can be applied to the fire detection without using a large cost by using the existing CCTV device. For these reasons, the use of fire detection system using image processing has been increasing recently, and a fire detection method is developed.

Korean Patent Application No. 10-2007-0121221 (hereinafter referred to as reference [1]), entitled " Vision-based Fire Detection System and Method ", discloses a method for detecting motion in a gray- Subtracts the background in the current frame, sets a value larger than a specific threshold value T _n (x) as a motion pixel value, and analyzes the temporal variation amount of the luminance map to distinguish the fire candidate region from the remaining region. However, Reference [1] has a high ratio of the images that are instantly changed in brightness due to changes in the fire.

Korean Patent Application No. 10-2008-0091720 (hereinafter referred to as Reference [2]), entitled " Fire Detection Apparatus and Method ", is used to detect motion pixels using frame differences of images input from a plurality of CCD cameras And detects flame pixels among the pixels having motion. However, reference [2] is disadvantageous in terms of cost because it determines motion using a plurality of cameras, and is weak in noise variation due to illumination because RGB color space is used.

Korean Patent Application No. 10-2009-0014799 (hereinafter referred to as Reference Document [3]), entitled " Image-Based Fire Detection Method and Method and Disaster Prevention System Using the Method ", expresses the characteristics of the flame by color information, The fire is detected using the ratio of the number of pixels included in the area and matching the color distribution characteristic of the fire. However, reference [3] does not consider the continuous characteristics of the image because flame detection is performed using the number of temporary flame pixels, and a lot of calculation resources are required to analyze the color distribution characteristics.

Accordingly, there is a desperate need for an image processing technology that is robust to the illuminance of the camera and the surrounding environment and capable of detecting a fire with a small burden on image processing.

[1] Korean Patent Application No. 10-2007-0121221: "Vision-based Fire Detection System and Method" [2] Korean Patent Application No. 10-2008-0091720: "Fire Detection Device and Method Thereof" [3] Korean Patent Application No. 10-2009-0014799: "Image-Based Fire Detection Method, and Method and Disaster Prevention System Applying the Method"

It is an object of the present invention to provide a fire detection method capable of quickly detecting a fire without being burdened with a large amount of image processing, .

It is another object of the present invention to provide a fire detection device capable of accurately detecting only a fire area while reducing the false detection rate of fire by considering all the features of sequentially changing image frames, .

을 사용하여 상기 RGB 영상을 상기 YCbCr 영상으로 변환하는 단계를 포함하고, R, G, B는 각각 상기 입력 영상의 적색, 녹색, 및 청색 성분을 나타내고, Y는 상기 입력 영상의 휘도를, 그리고 Cb 및 Cr 각각 상기 입력 영상의 색차를 나타낸다. 또한, 상기 휘도 변화 누적 영상을 생성하는 단계는, 시간 t-1 및 t에서의 상기 YCbCr 영상의 화소 (m, n)의 휘도 Y(m,n) _t-1 및 Y(m,n) _t 를 각각 획득하는 단계 및 다음 수학식 C(m,n) = C(m,n) + |Y(m,n) _t - Y(m,n) _t-1 | 을 사용하여, 시간 t-1 및 t 사이의 상기 YCbCr 영상의 화소별 휘도 변화량을 계산하여 상기 휘도 변화 누적 영상 C(m,n)을 생성하는 단계를 포함한다. 더 나아가, 상기 1차 화염 후보 이진 영상을 생성하는 단계는, 상기 휘도 변화 누적 영상 C(m,n)의 전체 화소의 평균 휘도 Y _mean 을 연산하는 단계, 화소 (m, n)의 C(m,n) 값이 상기 평균 휘도 Y _mean 보다 적다면, 해당 화소 (m, n)의 C(m,n)을 0으로 초기화하는 단계, 및 화소 (m, n)에서의 C(m,n) 값을 소정 임계와 비교하고, 다음 수학식

을 사용하여 1차 화염 후보 이진 영상 F ₁ 을 생성하는 단계를 포함한다. 뿐만 아니라, 상기 2차 화염 후보 이진 영상을 생성하는 단계는, 화소 (m, n)에서의 상기 YCbCr 영상의 색차 성분들인 C _b (m,n) 및 C _r (m,n)을 획득하는 단계, 상기 색차 성분들의 차분 |C _b (m,n) - C _r (m,n)|을 연산하는 단계, 및 화소 (m, n)에서의 상기 차분을 소정 임계와 비교하고, 다음 수학식

을 사용하여 2차 화염 후보 이진 영상 F ₂ 를 생성하는 단계를 포함한다. 특히, 상기 최종 화염 영역 결정 단계는, 상기 2차 화염 후보 이진 영상을 소정 개수의 블록으로 분할하는 단계, i 번째 블록에 포함된 화염 후보 영역의 면적 A _i 를 계산하고, 프레임별 A _i 를 누적하여 i 번째 블록의 누적 면적값 D _i 를 연산하는 단계, 시간 t-1 및 t에서의 상기 누적 면적값 D _i 의 차분인 화염 후보 영역 면적 변화량 AV _i 를 다음 수학식 AV _i =|D _i (t) - D _i (t-1)| 을 사용하여 연산하는 단계, 및 상기 화염 후보 영역 면적 변화량 AV _i 가 소정 개수의 프레임 동안 연속적으로 증가하는 블록을 최종 화염 영역으로 결정하는 단계를 포함한다. 더 나아가, 상기 누적 면적값 D _i 를 연산하는 단계는, i 번째 블록에 포함된 화염 후보 영역의 k 번째 중심점 G _k 를 연산하는 단계, 및 상기 k 번째 중심점 G _k 가 상기 i 번째 블록에 포함되는 경우에만 상기 화염 후보 영역의 면적 A _i 에 가산하는 단계를 포함한다. According to an aspect of the present invention, there is provided a fire detection method using brightness and area change. A method for detecting a fire according to the present invention includes a conversion step of converting an input image into a YCbCr image, a step of generating a luminance change accumulation image by accumulating a luminance change amount per frame of the YCbCr image, Generating a first flame candidate binary image composed only of pixels corresponding to flame candidates among the pixels of the luminance change accumulation image, calculating a difference between the first flame candidate binary image and the first flame candidate binary image, Generating a second flame candidate binary image by removing non-flame pixels from the candidate binary image, and generating a second flame candidate binary image by subtracting the non-flame candidate binary image from the second candidate flame candidate image And a final flame area determination step of determining the flame area. In particular, the converting step comprises the steps of: receiving an RGB image;

Wherein R, G and B represent the red, green, and blue components of the input image, Y represents the luminance of the input image, and Cb And Cr represent chrominance of the input image. Further, the step of generating the luminance change accumulated image, the time t-1 and the luminance of the pixel (m, n) of the YCbCr image at t Y (m, n) _t-1, and Y (m, n) _t (M, n) = C (m, n) + < RTI ID = 0.0 & Y (m, n) _t - Y (m, n) _t-1 | The use includes the step of generating the luminance change in the cumulative image C (m, n) by calculating the pixel by the brightness changing amount of the YCbCr image between time t-1 and t. Further, the step of generating the primary flame candidate binary image may include calculating an average luminance Y _mean of all the pixels of the luminance change accumulation image C (m, n), calculating C (m , n) , n) if the value is less than the average luminance Y _mean, the pixel (initializing a C (m, n) of the m, n) to 0, and the pixel (m, n) C (m, n) in the Value is compared with a predetermined threshold, and the following equation

To generate a first-flame candidate binary image F ₁ . In addition, the step of generating the second flame candidate binary image, the method comprising: obtaining a pixel _{(m, n) C b (} m, n) and C _r (m, n), which are chrominance components of the YCbCr image in , The difference of the color difference components | _B C (m, n) - C _r (m, n) | comparing the difference in computing a, and the pixel (m, n) with a predetermined threshold, and the following equation

And generating a _second- flame candidate binary image F ₂ using the second-flame candidate binary image F ₂ . In particular, the final flame area determining step, the secondary flame candidates to a predetermined binary image divided into blocks of the number, and calculates the area A _i of the flame candidate region included in the i-th block, accumulate A _i per frame Calculating the cumulative area value D _i of the i- th block, calculating the flame candidate area change amount AV _i , which is a difference between the cumulative area value D _i at times t-1 and t, by the following equation: AV _i = D _i (t) - D _i (t-1) | , And determining a block in which the flame candidate area change amount AV _i continuously increases for a predetermined number of frames as a final flame area. Moreover, the step of calculating the cumulative area value D _i is, i calculating the k-th center point G _k of the flame candidate region included in the second block, and wherein the k-th center point G _k included in the i-th block To the area A _i of the flame candidate area.

을 사용하여 상기 RGB 영상을 상기 YCbCr 영상으로 변환하도록 구성되며, 여기에서 R, G, B는 각각 상기 입력 영상의 적색, 녹색, 및 청색 성분을 나타내고, Y는 상기 입력 영상의 휘도를, 그리고 Cb 및 Cr 각각 상기 입력 영상의 색차를 나타낸다. 또한, 상기 휘도 변화 누적 영상 생성부는, 시간 t-1 및 t에서의 상기 YCbCr 영상의 화소 (m, n)의 휘도 Y(m,n) _t-1 및 Y(m,n) _t 를 각각 획득하고, 그리고 다음 수학식 C(m,n) = C(m,n) + |Y(m,n) _t - Y(m,n) _t-1 | 을 사용하여, 시간 t-1 및 t 사이의 상기 YCbCr 영상의 화소별 휘도 변화량을 계산하여 상기 휘도 변화 누적 영상 C(m,n)을 생성하도록 구성된다. 더 나아가, 상기 1차 화염 후보 이진 영상 생성부는, 상기 휘도 변화 누적 영상 C(m,n)의 전체 화소의 평균 휘도 Y _mean 을 연산하고, 화소 (m, n)의 C(m,n) 값이 상기 평균 휘도 Y _mean 보다 적다면, 해당 화소 (m, n)의 C(m,n)을 0으로 초기화하며, 그리고, 화소 (m, n)에서의 C(m,n) 값을 소정 임계와 비교하고, 다음 수학식

을 사용하여 1차 화염 후보 이진 영상 F ₁ 을 생성하도록 구성된다. 특히, 상기 2차 화염 후보 이진 영상 생성부는, 화소 (m, n)에서의 상기 YCbCr 영상의 색차 성분들인 C _b (m,n) 및 C _r (m,n)을 획득하고, 상기 색차 성분들의 차분 |C _b (m,n) - C _r (m,n)|을 연산하며, 그리고 화소 (m, n)에서의 상기 차분을 소정 임계와 비교하고, 다음 수학식

을 사용하여 2차 화염 후보 이진 영상 F ₂ 를 생성하도록 구성된다. 특히, 상기 최종 화염 영역 결정부는, 상기 2차 화염 후보 이진 영상을 소정 개수의 블록으로 분할하고, i 번째 블록에 포함된 화염 후보 영역의 면적 A _i 를 계산하고, 프레임별 A _i 를 누적하여 i 번째 블록의 누적 면적값 D _i 를 연산하며, 시간 t-1 및 t에서의 상기 누적 면적값 D _i 의 차분인 화염 후보 영역 면적 변화량 AV _i 를 다음 수학식 AV _i =|D _i (t) - D _i (t-1)| 을 사용하여 연산하고, 그리고 상기 화염 후보 영역 면적 변화량 AV _i 가 소정 개수의 프레임 동안 연속적으로 증가하는 블록을 최종 화염 영역으로 결정하도록 구성된다. 바람직하게는, 상기 최종 화염 영역 결정부는, 상기 누적 면적값 D _i 를 연산하기 위하여, i 번째 블록에 포함된 화염 후보 영역의 k 번째 중심점 G _k 를 연산하고, 그리고 상기 k 번째 중심점 G _k 가 상기 i 번째 블록에 포함되는 경우에만 상기 화염 후보 영역의 면적 A _i 에 가산하도록 구성된다. According to another aspect of the present invention, there is provided a fire detection apparatus using brightness and area change. A fire detection apparatus according to the present invention includes a camera for capturing image information of a monitoring object, an image conversion unit for converting an input image received from the camera into a YCbCr image, a cumulative luminance change accumulation unit for accumulating the luminance change amount per frame of the YCbCr image, A first flame generating unit configured to generate a first flame candidate binary image composed of only pixels corresponding to a flame candidate among the pixels of the luminance change accumulation image in consideration of the average luminance of the cumulative luminance change image, A second flame candidate generating unit for generating a second flame candidate binary image by removing a non-flame pixel from the first flame candidate binary image in consideration of a difference between chrominance components of pixels of the first flame candidate binary image, And the area of the flame candidate region included in the block of the second flame candidate binary image is represented by The increased flame candidate region in accordance with the end-parts comprises a flame zone determination unit determining the final flame region. In particular, the image conversion unit receives the RGB image,

Wherein R, G, and B represent the red, green, and blue components of the input image, Y represents the luminance of the input image, and Cb And Cr represent chrominance of the input image. (M, n) _t-1 and Y (m, n) _t of the pixel (m, n) of the YCbCr image at times t-1 and t (M, n) = C (m, n) + | Y (m, n) _t - Y (m, n) _t-1 | (M, n) by calculating the luminance variation of each pixel of the YCbCr image between times t-1 and t . Further, the primary flame candidate binary image generating unit, wherein the luminance change in the cumulative image C (m, n) C (m, n) for calculating the average luminance Y _mean of all the pixels, and the pixel (m, n) value of If less than the average luminance Y _mean, a C (m, n) value in the pixels to initialize the C (m, n) of the (m, n) to zero, and, to a pixel (m, n) a predetermined threshold And the following equation

To generate a first-order flame candidate binary image F ₁ . In particular, the secondary flame candidate binary image generating unit obtains C _b (m, n) and C _r (m, n) , which are chrominance components of the YCbCr image in the pixel (m, n) Difference | _B C (m, n) - C _r (m, n) |, and the calculation, and compares the difference in the pixel (m, n) with a predetermined threshold, the following equation

To generate a secondary flame candidate binary image F ₂ . In particular, the final flame area determination section, wherein said secondary flame candidate binary images the predetermined divided into blocks of the number, and calculates the area A _i of the flame candidate region included in the i-th block, and accumulating the A _i per frame i the cumulative area value of the _i-th block D calculated, and the time t-1 and the cumulative area value difference D _i is the amount of change in the flame area of the candidate region in the AV t _i the following equation: AV _i = | D _i (t) - D _i (t-1) | And a block in which the flame candidate area change amount AV _i continuously increases for a predetermined number of frames is determined as a final flame area. Preferably, the final flame area determination section, wherein the cumulative area value D _i a to operation, i computing a k-th center point G _k of the flame candidate region included in the second block, and wherein the k-th center point G _k the only if contained in the i-th block is configured to be added to the area a _i of the flame candidate region.

According to the present invention, a flame candidate binary image is generated to detect a flame using the luminance change accumulation image and color information of an image, and the flame candidate binary image is labeled to separate into flame candidates. Then, Since the block in which the area change of the area continuously occurs is determined as a flame block, a fire detection method capable of quickly detecting a fire without being burdened with many image processing robust to changes in the illuminance of the shot image or the surrounding environment is provided.

In addition, according to the present invention, a flame motion region is detected by accumulating luminance differences between adjacent images, a flame candidate region is generated using the color of the flame, and finally, a temporal change The fire area is determined by using the fire detection device. Therefore, a fire detection device capable of detecting only the fire area accurately and quickly is provided.

1 is a flowchart schematically illustrating a fire detection method according to an embodiment of the present invention.
2 is a flowchart illustrating a fire detection method according to the present invention in detail.
3 is a view for explaining a process of determining a center point and an area of a flame candidate region in the fire detection method according to the present invention.
FIG. 4 is a graph showing cumulative area change amounts for the flame region and the non-flame region in the fire detection method according to the present invention.
5 is a block diagram schematically showing a fire detection apparatus according to another aspect of the present invention.
Figure 6 is an image illustrating the result of the flame detection measured using the present invention.

In order to fully understand the present invention, operational advantages of the present invention, and objects achieved by the practice of the present invention, reference should be made to the accompanying drawings and the accompanying drawings which illustrate preferred embodiments of the present invention.

Hereinafter, the present invention will be described in detail with reference to the preferred embodiments of the present invention with reference to the accompanying drawings. However, the present invention can be implemented in various different forms, and is not limited to the embodiments described. In order to clearly describe the present invention, parts that are not related to the description are omitted, and the same reference numerals in the drawings denote the same members.

Throughout the specification, when an element is referred to as "including" an element, it does not exclude other elements unless specifically stated to the contrary. The terms "part", "unit", "module", "block", and the like described in the specification mean units for processing at least one function or operation, And a combination of software.

1 is a flowchart schematically illustrating a fire detection method according to an embodiment of the present invention.

First, the fire detection method 100 according to the present invention converts an input image into a YCbCr image (S110). The reason for converting the input image into the YCbCr image in the present invention is that the luminance variation is larger in the case of the flame than in the surrounding pixels. Therefore, the flame can be easily detected by extracting the luminance component of the image. Furthermore, as will be described later, in the fire detection method 100 according to the present invention, it is advantageous to convert the RGB image into the YCbCr image because the flame candidates are separated in consideration of not only the luminance but also chrominance components of the flame.

When the YCbCr image is acquired, the luminance variation of each frame of the YCbCr image is accumulated to generate an accumulated luminance change image (S130). It is accurate to detect the flame based on the luminance change amount with the previous frame rather than detecting the flame in one frame of the acquired image itself. Further, according to the present invention, since the luminance change amount with respect to the previous frame is accumulated, the flame can be detected more accurately. That is, the spreading rate of fire is not constant. If the progress of the fire is slow, the image size of the flame generally does not differ greatly between the previous frame and the current frame. In the case of such a fire image, it is difficult to obtain a good performance with the conventional technology, but the reliability of the present invention increases because the amount of change in the progress area of the flame continues to accumulate.

When the luminance change accumulation image is generated, a first-order flame candidate binary image composed of only pixels corresponding to the flame candidates among the pixels of the luminance change accumulation image is generated in step S150 in consideration of the average luminance of the generated cumulative luminance change accumulation image. In the present invention, instead of directly using the luminance change accumulation image to measure the flame, a pixel having a luminance lower than the average luminance is initialized to improve the fire detection precision. The process of generating the first flame candidate binary image will be described in detail below with reference to FIG.

When the first flame candidate binary image is generated, the second flame candidate binary image is generated by removing the non-flame pixel among the first flame candidate binary image considering the difference of chrominance components of the pixel of the generated first flame candidate binary image (S170). The secondary flame candidate binary image thus generated has a flame candidate region. If the area of the flame candidate region continuously changes, the final flame region is determined (S190).

As described above, the fire detection method 100 according to the present invention detects a flame using the luminance change accumulation image and the color information of the image. More specifically, the generated flame candidate binary image is labeled and divided into blocks, and a block in which the area of the flame candidates included in each block continuously changes is determined as a flame block. Therefore, it is possible to accurately detect only the fire area without being required to perform complicated image processing, which is robust against ambient illumination and environmental changes.

2 is a flowchart illustrating a fire detection method according to the present invention in detail.

When a CCTV image is input (S205), the input RGB image is converted into a YCbCr color space to be separated into luminance Y and chrominance components Cb and Cr (S210). The conversion equation that can be used at this time is as shown in the following equation (1).

(M, n) _t-1 and Y (m, n) _t of the pixel (m, n) of the YCbCr image are obtained by using the following equation (2) .

In Equation 2, the left side of C (m, n) is the time t of the pixel (m, n) C (m, n) of intensity variation represents the accumulated value, the right side at time t of the pixel (m, n) -1 . ≪ / RTI > Also, | Y (m, n) _t - Y (m, n) _t-1 | Represents the absolute value for the pixel by the luminance difference between the current brightness of the image Y (m, n) and _t before the luminance of the image _{Y (m, n) t-} 1. That is, it is a sum of the pixel by the brightness changing amount between time t and t-1 to the luminance change amount an accumulated value of the time t-1 is the luminance change amount an accumulated value of the time t.

When the obtained luminance change accumulation image is obtained, the obtained luminance change accumulation image C (m, n) is used to separate the non-flame area and the flame area. This is because the variation of the temporal luminance of the flame region is larger than that of the other regions in the fire video.

In other words, an increase in C (m, n) due to noise or movement of objects due to changes in illumination or a sudden change in the surrounding environment increases the false detection rate of the flame. Therefore , C (m, n), which is smaller than the average luminance mean value Y _mean of the luminance change accumulation image, is initialized to 0 by using the property that the luminance change amount continuously increases in the flame region (S220). This process can be expressed by the following equation (3).

The noise-canceled C (m, n) can be used to generate a first-flame candidate binary image. That is, the first flame candidate binary image F ₁ is generated by leaving only the pixels whose luminance change amounts are equal to or greater than the predetermined threshold value δ by using the following equation (4), and initializing all the values of other pixels to zero.

When the first flame candidate binary image is generated, the case where the difference between the chrominance components C _b and C _r among the pixels of the first flame candidate binary image is not equal to or greater than the threshold value is determined as the non-flame region, And generates a binary image F ₂ . To do this, it is first determined whether the difference between the chrominance components C _b and C _r is equal to or greater than a threshold value (S230). Otherwise, the value of the corresponding pixel of the primary flame candidate binary image is initialized to 0 (S235). The initialized image is returned to the image inputting step (S205) and used in the image processing process for the next frame.

If the difference between C _b and C _r is equal to or greater than the threshold value and the value of the first flame candidate binary image of the pixel is 1, a binary flame candidate binary image is generated by setting the pixel value to 1 as shown in the following Equation 5 (S240).

That is, in the present invention, both the chrominance value of the first flame candidate binary image and the luminance of the first flame candidate binary image are considered to generate the second flame candidate binary image. Therefore, since the non-flame pixels can be removed more accurately, the false detection rate of fire is reduced.

When the second flame candidate binary image is generated, the second flame candidate binary image is divided into a predetermined number of blocks, and the flame region is finally detected in consideration of the area change amount of the fire candidate region in each block. For example, the image is divided into 32x32 blocks, and the continuity of the area change of the flame candidates in the block is determined as the final flame judgment reference.

3 is a view for explaining a process of determining a center point and an area of a flame candidate region in the fire detection method according to the present invention.

As can be seen from FIG. 3, the labeling is performed on the secondary flame candidate binary image F ₂ , and the center point G _k and the area A _k of the regions having the kth label included in the divided block are calculated. Then, the area of the flame candidate area included in the i- th block is stored in the block-by-block cumulative area value D _i as shown in Equation (6). That is, the area A _i of the flame candidate area included in the divided i- th block is calculated, and the cumulative area value D _i of the i- th block is calculated using the following equation 6 by accumulating A _i for each frame.

Referring to the example of FIG. 3, it can be seen that a portion of the area having G ₂ of the two blocks exists over the two blocks. In this case, the area of the area having G ₂ is not included in the cumulative area of the second block, but is calculated as the cumulative area of the first block including the center point.

2, the flame candidate area change amount AV _i , which is a difference between the cumulative area value D _i at time t-1 and t , is calculated using the following equation (7) (S250).

That is, the AV _i flame candidate area from the area change amount of the i-th block is defined as the absolute value of the difference between the cumulative area value D _i of the cumulative area value D _i and the time t at the time t-1.

Then, it is determined whether the calculated flame candidate area change amount AV _i is greater than 0 (S260). If the flame candidate area change amount AV _i is not larger than 0, it indicates that there is no accumulated area change in the corresponding block. Therefore, it is determined that the flame image does not exist in the corresponding block (S270).

FIG. 4 is a graph showing cumulative area change amounts for the flame region and the non-flame region in the fire detection method according to the present invention.

Continuously as shown in FIG. 4, the accumulated value of the AV _i continues to increase on the passage of time if the flame region, and transient non-flame zone noise or due to motion of an object is missing or changes in AV _i accumulated value as It does not increase. In other words, when there is a flame in a certain block, the value of D _i continuously increases. Therefore, the region where there is no change or only a temporary change is judged as a noise region, thereby preventing false detection of the flame region.

2, when the flame candidate area change amount AV _i of the i- th block is larger than 0, it means that a flame is present in the corresponding block. Therefore, the count value Cnt _i for the block is incremented by 1 (S280). That is, the count value Cnt _i represents the number of frames in which the flame candidate area change amount AV _i continuously increases in the i- th block. Then, it is determined whether the area change of the flame candidate area occurs continuously for a predetermined number of frames (S285). As a result of the determination, if the area change of the flame candidate area occurs continuously for a predetermined number of frames, the block is finally determined as a block in which the flame is generated as shown in Equation (8).

In Equation (8), Flame _i is a value indicating whether the i- th block includes a flame. If Flame _i is 1, it means that the block contains a flame. If Flame _i is 0, it means that it is a block without flame.

As described above, the present invention accurately detects a flame portion in a fire image, and shows a stronger effect on noise according to the amount of sunshine than other methods. Especially, even if there is an object similar to the color of the flame, it can be precisely filtered because it is filtered using the luminance change and the flame motion. Therefore, it is possible to detect the accurate flame area, thereby improving the reliability of the fire detection system.

5 is a block diagram schematically showing a fire detection apparatus 500 according to another aspect of the present invention.

5, the fire sensing apparatus 500 includes a camera 510, an image conversion unit 520, a luminance change accumulation image generation unit 540, a primary flame candidate binary image generation unit 560, A secondary flame candidate binary image generating unit 570, and a final flame region determining unit 590.

The camera 510 captures the image information of the object of monitoring and transmits the image information to the image converter 520. The photographed image may be an RGB image.

The image converting unit 520 converts the input image received from the camera 510 into a YCbCr image. Here, Y represents the luminance of the input image, and Cb and Cr represent the chrominance of the input image, respectively, as described above.

When the input image is converted into the YCbCr image, the luminance change accumulated image generating unit 540 accumulates the brightness change amount of each frame of the received YCbCr image to generate the luminance change accumulated image. For this purpose, the change in luminance of the cumulative image generation unit 540 is time t-1 and the luminance Y (m, n) _t-1, and Y (m, n) of the pixel (m, n) of the YCbCr image at t _t (M, n) = C (m, n) + | Y (m, n) _t - Y (m, n) _t-1 | Calculating pixel by the brightness changing amount of the YCbCr image between time t-1 and t according to the equation to produce the cumulative change in luminance image C (m, n).

When the cumulative luminance change image C (m, n) is generated, the first-order flame candidate binary image generator 560 calculates the average luminance of the cumulative image C And generates a first-flame candidate binary image composed of only the first flame candidates. To this end, the primary flame candidate binary image generation unit 560 first luminance change accumulated image C (m, n) calculating an average luminance Y _mean of all the pixels, C (m, a pixel (m, n) and the n ) Is smaller than the average luminance Y _mean , C (m, n) of the pixel (m, n) is initialized to zero. Also, the first-order flame candidate binary image generator 560 compares the value of C (m, n) in the pixel (m, n) with a predetermined threshold,

를 사용하여 1차 화염 후보 이진 영상 F ₁ 을 생성한다.

To generate a first-flame candidate binary image F ₁ .

When the primary flame candidate binary image F ₁ is generated, the secondary flame candidate binary image generating unit 570 generates the secondary flame candidate binary image F _{1 based} on only the pixels corresponding to the flame candidates among the pixels of the luminance change accumulation image, And generates a first flame candidate binary image. To this end, the secondary flame candidate obtained a binary image generation unit 570, first the pixel (m, n) C _b (m, n) and C _r (m, n), which are color difference components of the YCbCr image in the color difference Difference of ingredients | C _b (m, n) - C _r (m, n) | Then, the difference in the pixel (m, n) is compared with a predetermined threshold,

To generate a second flame candidate binary image F ₂ .

If the secondary flame candidate binary image F ₂ is generated, the final flame region determining unit 590 is the second of the area A _i of the flame candidate region included in the flame candidate dividing the binary image into blocks of a predetermined number, and the i-th block . The final flame region determining unit 590 calculates the cumulative area value D _i of the i- th block by accumulating A _i for each frame, and then calculates the cumulative area value D _i of the flame candidates, which is the difference between the cumulative area values D _i at time t- the region area variation AV AV _i = _i | D _i (t) - D _i (t-1) | . In order to calculate the cumulative area value D _i , the final flame region determining unit 590 calculates the k- th center point G _k of the flame candidate region included in the i- th block, and the k- th center point G _k is included in the i- Is added to the area A _i of the flame candidate area only.

When the flame candidate area change amount AV _i is calculated, the final flame area determination unit 590 determines a block in which the flame candidate area change amount AV _i continuously increases for a predetermined number of frames as a final flame region.

Figure 6 is an image illustrating the result of the flame detection measured using the present invention.

As shown in FIG. 6, it can be seen that the block including the flame image can be accurately discriminated from the result of dividing the image generated by the fire into the respective blocks.

As described above, the fire is detected using the color and motion specification of the flame in the video image. First, in order to efficiently separate the luminance from the color difference, the RGB color space is converted into the YCbCr color space. In addition, the fire detection technique according to the present invention accumulates the difference in luminance between adjacent images to detect a motion region of the flame, and generates a flame candidate region using the color of the flame. Finally, the flame region is determined using the temporal change in the area of the flame candidate region. As shown in FIG. 6, the fire detection technology according to the present invention can separate the flame region quickly while reducing the burden on the image processing.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art.

In addition, the method according to the present invention can be embodied as computer-readable code on a computer-readable recording medium. A computer-readable recording medium may include any type of recording device that stores data that can be read by a computer system. Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and may be implemented in the form of a carrier wave (for example, transmission via the Internet) . The computer readable recording medium may also store computer readable code that may be executed in a distributed manner by a distributed computer system connected to the network.

Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

The present invention can be applied to a fire detection system.

Claims (14)

In a fire detection method using brightness and area change,
A conversion step of converting an input image into a YCbCr image;
Accumulating brightness change amounts of each frame of the YCbCr image to generate a luminance change accumulated image;
Generating a first-order flame candidate binary image including only pixels corresponding to a flame candidate among the pixels of the luminance change accumulation image, taking into account an average luminance of the cumulative luminance change image;
Generating a second flame candidate binary image by removing a non-flame pixel in the first flame candidate binary image, taking into account the difference in chrominance components of the pixel of the first flame candidate binary image; And
And a final flame region determining step of determining a flame candidate region in which the area of the flame candidate region included in the block of the secondary flame candidate binary image increases with time, as a final flame region,
Wherein,
Receiving an RGB image; And
The following equation

And converting the RGB image into the YCbCr image using the RGB image,
Here, R, G, and B represent the red, green, and blue components of the input image, Y represents the luminance of the input image, Cb and Cr represent the chrominance of the input image,
Wherein the step of generating the luminance change accumulation image comprises:
Obtaining luminance Y (m, n) _t-1 and Y (m, n) _t of the pixel (m, n) of the YCbCr image at time t-1 and t , respectively; And
The following equation
C (m, n) = C (m, n) + | Y (m, n) _t - Y (m, n) _t-1 |
(M, n) by calculating a luminance change amount of each pixel of the YCbCr image between times t-1 and t ,
Wherein the step of generating the primary flame candidate binary image comprises:
Calculating an average luminance Y _mean of all the pixels of the luminance change accumulation image C (m, n);
The method comprising: C (m, n) value of a pixel (m, n) if less than the _mean average luminance Y, initialize C (m, n) of the pixel (m, n) to zero; And
The value of C (m, n) in the pixel (m, n) is compared with a predetermined threshold,

And generating a first-flame candidate binary image F ₁ using the luminance and area changes.
The method according to claim 1,
Wherein the step of generating the secondary flame candidate binary image comprises:
Obtaining C _b (m, n) and C _r (m, n) which are chrominance components of the YCbCr image in the pixel (m, n) ;
The difference of the chrominance components | Calculating C _b (m, n) - C _r (m, n) | And
The above difference in the pixel (m, n) is compared with a predetermined threshold, and the following equation

And generating a _second flame candidate binary image F ₂ by considering both the chrominance value of the primary flame candidate binary image and the luminance of the corresponding primary flame candidate binary image using the luminance and area change Fire detection method used.
3. The method of claim 2,
Wherein the final flame zone determination step comprises:
Dividing the secondary flame candidate binary image into a predetermined number of blocks;
i cumulative area value of the i-th block by calculating the area A _i of the flame candidate region included in the second block, and accumulating the frame-by-frame _i A method comprising: calculating a D _i;
The flame candidate area change amount AV _i , which is the difference between the cumulative area value D _i at time t-1 and t,
AV _i = | D _i (t) - D _i (t-1) |
; And
And determining a block in which the flame candidate area change amount AV _i continuously increases for a predetermined number of frames as a final flame area.
The method of claim 3,
The step of calculating the cumulative area value D _i comprises:
calculating a k- th center point G _k of the flame candidate region included in the i- th block; And
The k-th center point G _k is the i-th block only method, the fire detection by the brightness change and the area comprising the step of adding to the area A _i of the flame candidate region if it is included in.
In a fire detection apparatus using brightness and area change,
A camera for capturing video information of a monitoring target;
An image converter for converting an input image received from the camera into a YCbCr image;
A luminance change accumulation image generation unit for accumulating luminance change amounts of each frame of the YCbCr image to generate a luminance change accumulation image;
A first flame candidate binary image generation unit for generating a first flame candidate binary image composed of only pixels corresponding to a flame candidate among the pixels of the luminance change accumulation image in consideration of the average luminance of the luminance change accumulation image;
A second flame candidate binary image generating unit for generating a second flame candidate binary image by removing non-flame pixels from the first flame candidate binary image, taking into account differences of chrominance components of pixels of the first flame candidate binary image; And
And a final flame region determining unit for determining a flame candidate region in which the area of the flame candidate region included in the block of the secondary flame candidate binary image increases with time, as a final flame region,
The image converter may include:
Receives an RGB image, and
The following equation

To convert the RGB image into the YCbCr image,
Here, R, G, and B represent the red, green, and blue components of the input image, Y represents the luminance of the input image, Cb and Cr represent the chrominance of the input image,
The luminance change accumulation image generation unit may include:
Obtain the luminance Y (m, n) _t-1 and Y (m, n) _t of the pixel (m, n) of the YCbCr image at time t-1 and t , respectively
The following equation
C (m, n) = C (m, n) + | Y (m, n) _t - Y (m, n) _t-1 |
The use, by calculating the pixel by the brightness changing amount of the YCbCr image between time t-1 and t is configured to generate the brightness change in the cumulative image C (m, n),
Wherein the primary flame candidate binary image generating unit comprises:
Calculates an average luminance Y _mean of all the pixels of the luminance change accumulated image C (m, n)
If the C (m, n) value of a pixel (m, n) is less than the _mean average luminance Y, and initialize the C (m, n) of the pixel (m, n) to zero, and,
The value of C (m, n) in the pixel (m, n) is compared with a predetermined threshold,

Is used to generate a _first flame candidate binary image ( F ₁₎ using a luminance and area change.
6. The method of claim 5,
Wherein the secondary flame candidate binary image generating unit comprises:
Obtaining a pixel _{(m, n) C b (} m, n) and C _r (m, n), which are chrominance components of the YCbCr image in, and
The difference of the chrominance components | C _b (m, n) - C _r (m, n)
The above difference in the pixel (m, n) is compared with a predetermined threshold, and the following equation

To generate a second flame candidate binary image F ₂ using the luminance and area changes.
The method according to claim 6,
The final flame area determination unit
Dividing the secondary flame candidate binary image into a predetermined number of blocks,
calculating the area A _i of the flame candidate region included in the i-th block, and calculating the cumulative area value D _i of the i-th block by accumulating the frame-by-frame, and A _i,
The flame candidate area change amount AV _i , which is the difference between the cumulative area value D _i at time t-1 and t,
AV _i = | D _i (t) - D _i (t-1) |
, ≪ / RTI > and
Wherein a block in which the flame candidate area change amount AV _i continuously increases for a predetermined number of frames is determined as a final flame area.
8. The method of claim 7,
The final flame region determining unit may calculate the final flame area D _i ,
calculating a second center point k G _k of the flame candidate region included in the i-th block, and
And adding to the area A _i of the flame candidate area only when the k- th center point G _k is included in the i- th block. delete delete delete delete delete delete

Images