KR20050009135A - Method and device for detecting flames - Google Patents

Abstract

PURPOSE: An apparatus and a method for detecting the flame are provided to reduce the manufacturing cost by without installing the specific flame sensors. CONSTITUTION: An apparatus for detecting the flame includes a video camera(1) and an evaluation stage(2). The evaluation stage evaluates the images supplied from the video camera. The evaluation stage is provided with an algorithm for the localization of the regions having a high optical intensity and a local flicker movement and a processor having a post analysis of the corresponding extracted image. And, the apparatus is characterized in that it analyzes at least one parameter of the radiation generated inside of the monitoring space.

Description

Flame detection method and apparatus {METHOD AND DEVICE FOR DETECTING FLAMES}

The present invention relates to a method and apparatus for detecting a flame by analyzing at least one parameter for radiation occurring in the monitoring space, in a monitoring area called the monitoring space below.

For example, in the prior art as disclosed in US Pat. Nos. 4 866 420 and 4 280 058, it comprises at least one sensor that evaluates the flicker frequency spectrum of radiation, wherein signals located outside a particular frequency are interfering signals. Is evaluated. Thus, typical flicker of sparks in the very low-frequency oscillation range is used as a property to distinguish between the radiation emitted by the spark and the interference radiation. In the simplest example, the frequency band is defined by filters connected upstream of the sensor or by a frequency-selective amplifier connected to another stream, for example a specific pass band of 5 to 25 Hz is obtained in both cases.

These known flame detectors have proved successful, but they are not a negligible cost factor in fire alarm systems. Apart from this, the optimum adjustment of the frequency band to the flicker of sparks, disturbances and inaccurate indications are excluded because it is repeated that any changes in the intensity of ambient radiation are within the passband. These changes in intensity are due to, for example, shading or reflecting vibrations, reflecting sunlight on the water surface, or flickering or varying the light source.

Therefore, an object of the present invention is to solve the above problems.

1 is a diagram illustrating flame detection according to the present invention.

Next, it is an object of the present invention to provide a method for flame detection characterized by having high interference immunity with respect to low cost.

This aim is to produce a video image of the monitoring space and to find a region with high light intensity and local flicker movement in the video image, in which case the region is localized in the first step and the associated excerpt image is in the presence of a flame. Subsequently analyzed.

The most preferred embodiment of the method according to the invention is characterized in that a video image is generated with a certain frequency and intensity images are obtained from it.

A second preferred embodiment of the method according to the invention is that the search for areas with high light intensity and local flicker motion is influenced with the aid of the accumulation matrix, which is obtained from other images of continuous intensity images, the difference The images are characterized by being weighted with weights, with weighting factors specifying information that other images affect the accumulation matrix.

A third preferred embodiment of the method according to the invention is characterized in that the coordinates of the brightest pixels are found with the aid of the accumulation matrix.

A fourth preferred embodiment is characterized by including the brightest pixel or pixels and the desired image area reduced for the original image being defined and analyzed for the presence of the flame.

Further preferred embodiments of the method according to the invention are disclosed in the dependent claims 7 to 10.

The device according to the invention mentioned in the introduction is characterized by a video camera having an evaluation stage for the images supplied by the camera, the evaluation stage being characterized by local flicker motion and regions with high light intensity in the images of the camera. Prosensors with an algorithm for and subsequent analysis of corresponding excerpt images for flame presence.

Preferred embodiments of the device according to the invention are disclosed in the dependent claims 12 to 17.

As CCTV systems and installations are becoming more and more widely used, in many cases video cameras will be present in the monitoring space, with the result that dedicated sensors cannot be installed for flame detection, which will undoubtedly lead to cost reduction. Further cost reduction is achieved by limiting the evaluation only to excerpt images that contain flames, which may significantly reduce computer power. It can also be estimated that the evaluation of these excerpt images is sufficiently strong against disturbances.

The invention is described in detail below by way of example with reference to a block diagram illustrating an apparatus according to the invention for detecting a flame. Reference 1 denotes a video camera, which provides output video sequences to the evaluation stage 2, which is integrated into or connected to the camera 1. The evaluation stage 2 may be provided at or immediately adjacent to the position where the camera 1 is installed, or may be provided spaced apart from the camera 1, in which case the camera 1 and the evaluation stage 2 are provided. There is a communication link between them.

The evaluation stage 2 comprises a processor (not shown) with an algorithm for the position of the flame found in the camera 1 and subsequent analysis of the corresponding excerpt image. As shown, in the initial process of the algorithm, the video sequences supplied by the camera 1 with the specified image acquisition, intensity and / or chromaticity images x _ij (t) (hereinafter referred to as intensity unknowns) are supplied. Obtained from and i and j are the coordinates of the individual pixels. For example, the frequency of these images is at least 15 images per second, and the image size is 352 × 288 pixels. Intensity images are obtained because the frames represent locations with high light intensity and can further be assumed to have characteristic hue.

In the next process, indicated by preprocessing 4, the flame is obtained from the intensity images X _ij (t) and the flames found are localized in the corresponding excerpt images. This localization is affected by the so-called accumulation matrix, which is formed in the following way:

In a first step, the maximum value of the intensity max [X _ij (t)] and the mean value [X _ij (t)] are determined and the brightness threshold q (t) is determined therefrom, and the following is established:

q (t + 1) = λ ₁ max [X _ij (t)]; If mean [X _ij (t)] <λ ₁ max [X _ij (t)],

q (t + 1) = λ ₂ {max [X _ij (t)] − mean [X _ij (t)]} + mean [X _ij (t)]; In all other cases.

λ ₁ and λ ₂ are constants between 0 and 1, in this case λ ₁ is 0.68 and λ ₂ is 0.05, for example.

The weighting factor w _ij (t) taking into account the flame characteristics is determined from these two conditions:

w _ij (t) = X _ij (t); For X _ij (t)> max [X _ij (t)]-q (t),

w _ij (t) = 0; In all other cases.

This means that all pixels with intensities less than or equal to the value max [X _ij (t)], ie dark objects, are filtered out and no longer considered. As will be apparent, the filtered objects are moving dark objects.

Since it can be assumed that the flame can be identified as a movement with high light intensity, a difference image is formed by comparison of successive images to find this movement; Dark objects are omitted due to the definition of weighting factor w _ij (t), so moving dark objects that cannot be sparks are filtered during formation of the difference image. Thus, an area with high light intensity and local flicker movement is found, which means, for example, that a non-blinking fixed light source is not regarded as a flame and may not be a lamp moving laterally in the monitoring space.

For the difference image Q _ij (t) the following holds:

Q _ij (t) = | X _ij (t)-X _ij (t-1) | w _ij (t)

Next, the difference image Q _ij (t) is used to determine the accumulation matrix A _ij (t):

A _ij (t) = αA _ij (t-1) + (1-α) Q _ij (t)

α is a constant between 0 and 1 indicating the degree to which the difference image Q _ij (t) affects the accumulation matrix A _ij (t). For α = 0, the accumulation matrix becomes equal to the difference image, and for α = 1, the difference image no longer has an effect because A _ij (t) is equal to A _ij (t−1). The accumulation matrix is mainly formed to obtain a smooth image without noise and instantaneous change.

As a final step in the pre-treatment (4), the pixel or pixels with the highest value [i _m, j _m [(t) that is obtained with the aid of the accumulation matrix comprises a so-called pixel or pixel, and desired image area to the flame position ROI is defined:

[i _m , j _m [(t) = {(i, j) | max [A _ij (t)]}

Thus, this determination provides the coordinates of the pixels with local flicker movement and maximum brightness. In general, although a single pixel is involved, it is of course possible to determine a plurality of highest brightness pixels, for which multichannel selection is used and preferably the minimum separation between the individual pixels is defined.

In the next process indicated by analysis (5), first of all the analysis is not affected in the entire initial image of 352 X 288 pixels but in the desired image area ROI with a reduced size of 32 X 32 pixels, for example. Significant reduction in data is affected. This results in a reduction to 1/100. This reduction may, of course, be reduced by 1/15 or even more.

Next, the following image information items are determined for each desired image area:

Average brightness L (t) of the desired image area X _ROI (t):

L (t) = [X _ij (t) | Average of _ROI ]

Color of desired image area X _ROI (t) C (t):

C (t) = [C _ij (t) | Number of _ROIs ] "Flame Saturation Sector" / R (t), where C _ij (t) represents the color pairs V _ij , U _ij of the image X _ij (t) at time t. YUV represents a color space with two color components U and V on the x and y axes, and an intensity Y on the z and z axes, respectively, and the length of the vector from the zero point in the UV plane to the pixel specifies the color saturation of this pixel. The flame saturation sector R (t) is a sector of color space in the typical color range of the flame in the UV plane, which in particular comprises red.

Number of active pixels in accumulation matrix A _ROI (t) R (t):

R (t) = [A _ij (t) | _ROI > number of η ₁ ]; 1 ≤η ₁ <Z For example η ₁ = 30

(Z = total number of pixels in the image area of the desired ROI)

Saturation S (t) of the image area of the desired X _ROI (t)

_{S (t) = [X ij} (t) | ROI> number of _{η 2] · 1 ≤η 2 <} Z For example, η ₂ = 5

In order to keep the result constant, the time integration of the image information items determined during analysis 5 is affected by the process indicated by extraction 6. If the integration is performed for 1 second, for example, it extends over 25 images in PAL format. Thus, the average brightness, color, active pixels and saturation are integrated over time t from t ₀ to t _n to obtain the following properties:

Average value of brightness: F _L = L

Average value of frequency: F _F = F

Average value of amplitude: F _M = M

Average value of spark saturation pixels: F _C = C

Average value of active pixels: F _R = R

Average value of saturation: F _S = S

The average value of the frequency is obtained, for example, by counting pixels with an average brightness L (t). The frequency due to the characteristic flicker of the flame is an important amount for the detection of the flame because it is located in a defined narrow range of 1 Hz to 10 Hz.

In the next process, indicated by pattern recognition (7), the properties obtained during extraction (6) are used to calculate the probability that a flame is present. In this case, the test is influenced for one of the above characteristics, for example, to determine whether the average value is above or below the threshold and correspondingly whether the probability is set equal to one or equal to zero. The overall probability is formed from the probabilities for all n properties.

The same is true for the other characteristics.

Overall probability Π (t):

for w _i, 0≤w _i ≤1 is established, the value w _i is determined empirically. N _F is the sum of w _i for all i.

In the process indicated by decision 8, it is determined whether the alarm is triggered. This includes integration while the overall probability π (t) is integrated over the continuous images. The integration starts at 0 and counts the increment for each Π (t)> k (k is the threshold) and subtracts the increment for each Π (t) <k. If I (t) represents the value of the integral, the following holds:

I (t = 0) = 0

If Π (t)> k, I (t) = I (t-1) + σ + ( if all other cases, the I (t)> S ₊ is saturated with _{S +, I (t) <} S - If S _- (Typically 0)

σ ₊ and σ _- are generally equal to +1.

The determination of whether the alarm is triggered is performed with the help of the integral I (t): If I (t)> β (β is the threshold), the alarm is triggered and in all other cases no alarm is triggered.

The devices described in many applications can use video cameras already installed and do not require the installation of specific flame sensors, which undoubtedly has the advantage of reducing costs. A further reduction in cost is obtained by limiting the evaluation to excerpt images that contain a flame, which can significantly reduce computer power. It can also be assumed that the evaluation of the excerpt images is robust enough to disturb.

Claims (17)

In a monitoring area called a monitoring space, a method of detecting a flame through the analysis of at least one parameter of radiation occurring in the monitoring space, A video image of the monitoring space is generated, an area with high light intensity and local flicker movement is found in the video image, in which case the area is localized in the first step and the associated excerpts are imaged for the presence of the flame. Subsequently analyzed. The method of claim 1, wherein the video image is generated with a specific frequency, and an intensity image [X _ij (t)] is obtained from the video image. The method of claim 2, wherein the search for regions with high light intensity and local flicker movement is influenced by an accumulation matrix [A _ij (t)] obtained from difference images of continuous intensity images [X _ij (t)]. Wherein the difference image is weighted with a weighting factor, the weighting factor specifying information for which the difference images affect the accumulation matrix [A _ij (t)]. 4. The method of claim 3, wherein all pixels with brightness are located below a predetermined threshold such that all moving dark objects are filtered during the formation of the difference image and during formation of the accumulation matrix [A _ij (t)]. . 5. The method of claim 4, wherein the coordinates of the brightest pixels are found with the aid of the accumulation matrix [A _ij (t)]. 6. A method according to claim 5, wherein the desired image area [ROI], which is comprised of the brightest pixel or pixels and is reduced for the original image, is defined and analyzed for the presence of the flame. 7. The method of claim 6, wherein the size of the desired image area [ROI] is at most 1/15 of the size of the original image. 8. The method of claim 7, wherein image information items of brightness [L (t)], color [C (t)], number of active pixels [R (t)] and saturation [S (t)] above a certain intensity threshold are described above. The method determined in the desired image area [ROI]. 10. The method of claim 8, wherein the image information items are integrated over a specific time and number of images, the average value of which is determined as additional parameters during the integration of the mean value of frequency [F] and the mean value of amplitude [M]. The probability of the presence of is calculated for each of said average values. 10. The method of claim 9, wherein the probabilities of the mean values are used to calculate an overall probability of the presence of a flame in the reduced image area [ROI], wherein the overall probability is integrated over a plurality of images, the integrated value being How the alarm is triggered when the threshold is exceeded. An apparatus for detecting a flame by analyzing at least one parameter of radiation occurring in the monitoring space in a monitoring area called a monitoring space, A video camera [1] having an evaluation stage [2] for the images supplied by the video camera [1], wherein the evaluation stage [2] is adapted to localization of regions with high light intensity and local flicker movement. And a processor having an algorithm for and subsequent analysis of the corresponding excerpt image regarding the presence of the flame. 12. The apparatus of claim 11, wherein the algorithm comprises a process called image acquisition [3], wherein during the image acquisition an intensity image [X _ij (t)] is obtained from a video image at a particular frequency. 13. The method of claim 12, wherein the algorithm includes a process called preprocessing [4], wherein during this preprocessing the accumulation matrix [A _ij (t)] is determined for searching for regions with high light intensity and local flicker movement. The accumulation matrix is obtained from difference images of continuous intensity images [X _ij (t)], the difference images are weighted with a weighting factor, and the weighting factor is the difference images with the accumulation matrix [A _ij (t). A device that specifies the extent to which it affects]. 14. The desired image area according to claim 13, wherein during the preprocessing [4], with the aid of the accumulation matrix [A _ij (t)], the coordinates of the brightest pixels include the brightest pixel or pixels and are reduced for the original image. The device for which [ROI] is defined. 15. The method of claim 14, wherein the algorithm comprises a process called analysis [5], for the analysis of the desired image area [ROI] during this preprocessing, brightness [L (t)], color [C (t)]. And the number [R (t)] of active pixels above a certain intensity threshold and the image information items of saturation [S (t)] are determined. 16. The method of claim 15, wherein the algorithm comprises a process called extraction [6], wherein during the extraction the image information items are integrated over a particular time and number of images, and average values of the image information items are determined, The average value [F] of frequency and the average value [M] of amplitude are determined as additional parameters during the integration, and the probability of the presence of a flame is calculated for each of the average values. 17. The method of claim 16, wherein the algorithm comprises pattern recognition [7] and a decision process [8], wherein during these processes the mean values are used to calculate the overall probability of the presence of a flame in the image region [ROI] where the mean values are reduced. Wherein the overall probability is integrated over a plurality of images, wherein an alarm is triggered when the integrated value exceeds a threshold.