JP2010238032A - Smoke detection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a smoke detection device for detecting slowly diffused smoke with high accuracy by suppressing an influence of disturbance, in detecting slowly diffused smoke. <P>SOLUTION: The smoke detection device for detecting generation of smoke by performing image processing on an image captured by a monitor camera 1 is provided with: an image memory 10 for storing a plurality of images time-sequentially captured by the monitoring camera 1 and a reference image; and a smoke featured value calculation means 31 for calculating the difference of either the mean value of edge strength or standard deviation as variation for each predetermined cycle between the latest image and a reference image captured by the monitor camera 1 in a smoke detection area specified as a candidate region where smoke is generated and for, when a period for determining generation of a smoke is defined as N cycles (N is an integer of 1 or more), determining that there is high possibility that the smoke has been generated when the calculated variation is continuously kept within a predetermined range for the N cycles. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a smoke detection device that detects the generation of smoke by performing image processing on an image captured by a surveillance camera, and in particular, smoke suitable for detecting slowly diffusing smoke (slow diffusion smoke). The present invention relates to a detection device.

  From the viewpoint of initial fire extinguishment in the event of a fire or prevention of escape delay in a fire accident, early detection of fire or smoke is very important. Thus, in the field of smoke detection devices, research has been conducted on early detection of smoke by performing image processing on an image captured by a surveillance camera.

  As an example, there is a conventional smoke detection device that detects smoke by installing a camera in a tunnel or the like and performing image processing on an image captured by the camera. In image processing for detecting smoke, generally, a reference image (reference image) is stored in advance, a difference image between the latest captured image and the reference image is calculated, and a region where a change has occurred is calculated. Extraction detects smoke. (For example, refer to Patent Document 1).

  In addition, the reference image is regularly updated in order to cope with the temporal change of the reference image due to the influence of sunlight.

Thus, by performing image processing on the image captured by the camera and performing smoke detection, the following two merits can be obtained.
1) By visually confirming the image of the surveillance camera, it is possible to grasp the smoke detection status in a remote place.
2) It is possible to divert already installed surveillance cameras and construct efficient equipment.

Japanese Patent No. 3909665

However, the prior art has the following problems.
In the prior art, in order to detect smoke, a pixel region where a luminance difference from a frame difference image or a background image exceeds a predetermined threshold is extracted. However, for example, when a captured image changes due to an influence of a change in surrounding lighting conditions, there is a problem that a portion where smoke is not generated is erroneously detected as smoke.

  In addition, the smoke itself that is a detection target has a small color tone, and depending on the background color, there may be a small change in luminance (luminance difference) in the captured image. It may be difficult to set and high-sensitivity smoke detection may not be possible.

  The present invention has been made in order to solve the above-described problems. When the slow diffusion smoke is targeted for detection, the influence of the disturbance is suppressed, and the slow diffusion smoke is detected with high sensitivity. An object of the present invention is to obtain a smoke detection device capable of performing the above.

  A smoke detection device according to the present invention is a smoke detection device that detects the generation of smoke by performing image processing on an image captured by a monitoring camera, and a plurality of images captured in time series by the monitoring camera The difference between the brightness value of the latest image captured by the surveillance camera and the brightness value of the reference image in the image memory storing the image and the reference image and the smoke detection area specified as a candidate area where smoke is generated is predetermined. And a smoke feature amount calculating means for determining that the possibility that smoke has been generated is high when the counted pixel exceeds a predetermined allowable value.

  The smoke detection device according to the present invention is a smoke detection device that detects the generation of smoke by performing image processing on an image captured by a surveillance camera, and is captured in time series by the surveillance camera. Average value of edge strength between the latest image captured by the surveillance camera and the reference image in the image memory storing a plurality of images and the reference image, and the smoke detection area identified as a candidate area where smoke is generated Alternatively, when the difference between at least one of the standard deviations is calculated as a change amount for each predetermined cycle, and the period for determining the occurrence of smoke is N cycles (N is an integer of 1 or more), the calculated change amount is Smoke feature amount calculation means for determining that the possibility that smoke has been generated is high when the state within the predetermined range continues for N cycles.

  According to the smoke detection device according to the present invention, when using the calculation result of the smoke characteristic amount suitable for the slow diffusion smoke such as the number of pixels of the difference from the reference image or the edge strength, the slow diffusion smoke is set as the detection target. In addition, it is possible to obtain a smoke detection device capable of detecting a slow diffused smoke with high sensitivity while suppressing the influence of disturbance.

It is a block diagram of the smoke detection apparatus in Embodiment 1 of this invention. It is a figure which shows an example of the brightness | luminance histogram created by the brightness | luminance histogram creation means in Embodiment 1 of this invention. It is explanatory drawing which showed the relationship between the area | region division | segmentation in 1 screen and the mapping result in 1 area | region in Embodiment 1 of this invention. It is a figure which shows the candidate area | region specified by the smoke detection area selection part in Embodiment 1 of this invention. It is explanatory drawing of the smoke detection method based on the correlation value implemented by the smoke feature-value calculation means in Embodiment 1 of this invention. It is explanatory drawing regarding the smoke determination based on the value of the average intensity | strength of an edge strength and the standard deviation implemented by the smoke feature-value calculation means 31 in Embodiment 1 of this invention. It is a figure which shows the mapping result output by the mapping means in Embodiment 1 of this invention. It is a flowchart which shows the flow of the whole process of the smoke detection apparatus in Embodiment 1 of this invention.

  Hereinafter, a preferred embodiment of a smoke detection device of the present invention will be described with reference to the drawings.

Embodiment 1 FIG.
FIG. 1 is a configuration diagram of a smoke detection device according to Embodiment 1 of the present invention. The smoke detection apparatus according to the first embodiment includes an image memory 10, a smoke detection area selection unit 20, and a smoke generation detection unit 30. The image memory 10 is configured as an image memory for a plurality of frames so that images captured by the camera 1 can be stored as time-series data for a certain period in the past.

  The smoke detection area selection unit 20 includes a luminance histogram creation unit 21, a luminance change determination unit 22, and a detection candidate area specifying unit 23. The smoke detection area selection unit 20 has a function of specifying an area where smoke detection is to be performed as a smoke detection candidate area based on an image for a predetermined period of time captured by the camera 1 and stored in the image memory 10. Have.

  The smoke generation detection unit 30 includes a smoke feature amount calculation unit 31, a mapping unit 32, and a smoke determination unit 33. Then, the smoke generation detection unit 30 calculates a feature amount for detecting the generation of smoke for the smoke detection candidate region specified by the smoke detection area selection unit 20, and whether smoke is generated based on the calculation result. It has a function to determine whether or not. For example, the characteristic value of slow diffusion, especially thin slow diffusion smoke, is, for example, the correlation value of the luminance distribution between the reference image and the image after the slow diffusion smoke is reduced, but is within a predetermined range, or the slow diffusion smoke Since the image after the occurrence has features such as a decrease in edge strength, the feature amount of smoke is calculated based on these features.

  By providing such a configuration, the smoke detection device according to the first embodiment performs efficient image processing by performing image processing for detecting the presence or absence of smoke on the specified smoke detection candidate region. Sensitive smoke detection can be realized. In particular, it has an image processing function suitable for detecting slow diffused smoke.

  In the first embodiment, smoke detection is not performed on all of the captured images, but the smoke detection candidate region which is a divided region obtained by dividing the image into a plurality of blocks among the captured images. It only features smoke detection. First, the function of the smoke detection area selection unit 20 that realizes this feature will be described.

Step 1: Function of Luminance Histogram Creation Unit 21 The luminance histogram creation unit 21 determines whether or not an intruder or smoke has changed in the captured image for a predetermined period of time for each predetermined pixel. This is done by calculating a histogram of the brightness of the inside.

  Here, the intruder corresponds to a passerby who temporarily passes through the image. The luminance histogram corresponds to a frequency distribution indicating what luminance value distribution the target pixel has in the time-series data of a certain past period.

  FIG. 2 is a diagram showing an example of a luminance histogram created by the luminance histogram creating means 21 in Embodiment 1 of the present invention. Assuming a state where no intruders are generated and no smoke is generated, if the frequency of occurrence of luminance values in a plurality of images within a certain past period is counted for a certain pixel, no change occurs for a certain period. As shown in FIG. 2A, the distribution is concentrated on a narrow range of luminance, and the frequency of occurrence of the narrow range is high.

  Here, the occurrence frequency is the “existence frequency” on the vertical axis in FIG. 2. In other words, the number of images taken in a certain period is the maximum value, and the position coordinates are the same in these images. This is the number of times that a specific pixel has the same luminance value. This histogram shows the brightness value of each pixel in the past. If there is no intruder, the existing brightness is distributed in a narrow range, and the number of times of existence remains high. Keep doing.

  On the other hand, assuming that intruders are generated or smoke is generated, the frequency of occurrence of luminance values in a plurality of images in a certain past period is totaled for a certain pixel. Since a change occurs in the image, as shown in FIG. 2B, a luminance value different from the distribution in the past starts to appear in places other than the narrow range of luminance in FIG. As a matter of course, the number of existing luminances is small. That is, in addition to the high and narrow luminance distribution, the luminance detected as a different distribution can be obtained from the image after the intruder is generated or after the smoke is generated. Alternatively, only a histogram with a low number of existences can be obtained.

Then, the frequency of occurrence of the luminance indicating the newly appearing different distribution is shown in FIG. 2A when the number of images after the occurrence of intruders or smoke is small among the plurality of images in the past fixed period. The result is less than the frequency of occurrence of luminance in a narrow range before occurrence.
In addition, when the luminance difference between the luminance distribution before the occurrence of changes such as intruders and the luminance distribution after the occurrence is small, it is difficult to extract the changed area by detecting the luminance change. By obtaining a luminance histogram for each pixel based on a plurality of images, even if the change in luminance is small, it is easy to identify the change, and it is possible to accurately identify the occurrence of intruders or smoke Become.

  Further, the luminance histogram creating means 21 counts the obtained luminance with a certain width when creating the luminance histogram (for example, if the luminance is 100, three luminances including 99 and 101 which are the previous and subsequent luminances) are counted. By calculating the luminance (counting as luminance values 99, 100, and 101), it is possible to reduce the calculation related to the calculation of the histogram, and to obtain a result of reducing the influence of noise. Moreover, the result of reducing the influence of noise can also be obtained by performing a smoothing process after obtaining the luminance histogram.

  In creating the luminance histogram, a histogram may be created for each pixel, but the histogram may be shared by adjacent pixels. For example, a luminance histogram may be created by grouping a total of four pixels in two vertical and horizontal directions, or by grouping a total of nine pixels in three vertical and horizontal directions. Time can be shortened. Preferably, by combining four pixels together, in this case, the processing time is short and the accuracy of intrusion detection is not reduced.

Step 2: Function of the luminance change determination means 22 The luminance change determination means 22 is newly generated by the occurrence of an intruder or smoke based on the luminance histogram calculated (generated) individually for each predetermined pixel by the luminance histogram creation means 21. The presence / absence of a luminance value generated in the above is detected. For example, as shown in FIG. 2B, when a predetermined number of infrequent luminance values occur in the luminance distribution (before generation), the luminance change determination unit 22 determines the portion of the pixel. Therefore, it can be determined that there is a high possibility that an intruder or smoke has occurred.

  The luminance change determination unit 22 performs the above-described determination individually for each predetermined pixel, and outputs the result as a map corresponding to the screen. For example, when the pixels of one screen are p × q (where p and q are integers of 2 or more), the luminance change determination unit 22 selects pixels that are likely to have intruders or smoke “1”. The other pixels are set to “0”, and each pixel of p × q is mapped or binarized to obtain a mapping image (binarized image). In addition, in comparison with the obtained histogram, when the luminance value of the corresponding pixel of the current image is a luminance value that is less than the predetermined number of times, it may be determined that the pixel has an intruder. Good.

Step 3: Function of Detection Candidate Area Identification Unit 23 The data for one screen mapped by the luminance change determination unit 22 is divided into a plurality of predetermined areas. Then, the detection candidate area specifying unit 23 determines whether or not the ratio of the pixels “1” is greater than or equal to a predetermined value for each of the previously divided areas mapped with “1” and “0”.

  FIG. 3 is an explanatory diagram showing the relationship between the area division within one screen and the mapping result within one area in Embodiment 1 of the present invention. FIG. 3A shows a plurality of divided areas set in advance in one screen. As shown in FIG. 3A, the screen is divided into a plurality of rectangular areas in a vertical and horizontal matrix. In front of the door, there is a human being as an intruder, and smoke is present under the right window glass. FIG. 3B shows a state in which each area is divided into pixel units and mapped by the luminance change determination means 22. Thus, one area is composed of a plurality of pixels.

  In FIG. 3B, the portion where the pixel is painted black means a pixel mapped to “1” as a pixel that is highly likely to have an intruder or smoke. Therefore, the detection candidate area specifying unit 23 determines the ratio of the pixels that are “1” (that is, pixels that are blacked out) for each of the divided areas based on the mapping result as illustrated in FIG. It will be determined whether or not there is a predetermined value or more.

In addition, in order to detect the slow diffusion smoke with high accuracy, the detection candidate area specifying unit 23 further performs the following preprocessing 1 to 3 on the mapping result shown in FIG. It can be carried out.
[Pre-processing 1] Removal of isolated pixels Since slow diffused smoke is widely generated over a certain area, pixels that are mapped to "1" exist as isolated pixels that have a high possibility of intrusion or smoke. It's hard to think. Therefore, the pixel mapped to “1” is an isolated pixel (that is, a total of 8 pixels adjacent to the pixel mapped to “1” vertically, horizontally, and diagonally are mapped to “0”. If the pixel is a state pixel), the mapping result is changed from “1” to “0”.

[Preprocessing 2] Difference amount check of luminance value with respect to reference image In a region where slow diffused smoke is generated, the luminance value obtained in the reference image may be decreased by a predetermined amount or more. Therefore, for the pixel mapped to “1”, the difference between the luminance values of the latest image and the reference image is obtained, and those whose absolute value of the difference amount is equal to or larger than a predetermined value remain “1”. If the absolute value of the quantity is less than the predetermined value, it is changed from “1” to “0”. However, an area in which the luminance of the reference image itself is considered to be about the same as the luminance of the slowly diffusing smoke is not suitable for this difference amount check.

[Pre-processing 3] Edge strength check In an area where slow diffused smoke is generated, the image is blurred due to the influence of smoke, so the edge strength may be lowered. Therefore, the edge strength in the neighboring region including the pixel mapped to “1” is calculated. If the edge strength is equal to or less than the predetermined value, “1” is maintained. If the edge strength exceeds the predetermined value, smoke is detected. It is determined that the influence is not, and “1” is changed to “0”.

  Any combination can be used for such preprocessing. For example, if any one of the pre-processing 1 to 3 results in changing from “1” to “0”, this change can be implemented. Alternatively, when a specific combination results in changing the combination from “1” to “0” in common, the change can be performed. Further, a combination of preprocessing can be set in advance for each area of an image to be captured.

  Next, the detection candidate area specifying unit 23 specifies an area in which the ratio of pixels that are “1” with respect to the size of one area is equal to or greater than a predetermined value as a candidate area where smoke detection should be performed in detail. On the other hand, the detection candidate area specifying unit 23 specifies an area in which the ratio of pixels that are “1” is less than a predetermined value as an area that does not perform subsequent smoke detection. Alternatively, an area in which the ratio of pixels that are “0” is equal to or greater than a predetermined value is specified as an area in which subsequent smoke detection is not performed.

  FIG. 4 is a diagram showing candidate areas identified by the smoke detection area selection unit 20 according to Embodiment 1 of the present invention. Here, the blacked-out portion is a region selected by the smoke detection area selecting unit 20, and is a region having a large ratio including the pixel “1” described above. This FIG. 4 corresponds to the state of FIG. 3A, and there are intruders in front of the door and a plurality of black areas as candidate areas under the window glass. .

  Although there are a plurality of regions in the image, it is only necessary to calculate whether or not smoke is generated in the region for only the candidate region specified by the detection candidate region specifying unit 23, It is possible to reduce the overall calculation amount. Through such a series of processes, the smoke detection area selection unit 20 selects candidate areas for which detailed smoke detection is to be performed by the subsequent smoke generation detection unit 30 from a plurality of previously divided areas for a predetermined period of time. From the calculation result of the luminance histogram, it can be specified with high accuracy.

  As a result, it is possible to efficiently detect smoke with high sensitivity by extracting only an area with intruders or smoke from the image while suppressing the influence of disturbance. In other words, if smoke or the like is generated in a certain area, most of the pixels in the area have a different distribution in the histogram and become “1”. Extracted. However, a temporary change such as illumination hardly causes different distributions in the luminance histogram, so there are few pixels that are “1”, and it is difficult to become a candidate region due to the influence of disturbance. Furthermore, by performing the pretreatment as described above, it is possible to select a candidate region suitable for extraction of slow diffused smoke.

  Next, the function of the smoke generation detection unit 30 will be described. The smoke generation detection unit 30 can determine the presence or absence of smoke generation by extracting a feature quantity related to smoke for a candidate area where smoke detection specified by the smoke detection area selection unit 20 should be performed in detail. it can.

Step 1: Function of Smoke Feature Quantity Calculation Unit 31 As typical feature quantity extraction methods, there are the following seven. By applying the following method to the candidate area where smoke detection is to be performed in detail, the smoke feature quantity calculation unit 31 determines whether or not there is a high possibility that smoke has occurred in the area. Can be judged. Note that the calculation of the following smoke feature amount is a method particularly suitable for detecting relatively slow smoke accompanied by a flow. Among them, the extraction methods 5 to 7 are slow diffusion smoke, which is suitable for thin smoke that transmits light, and will be described in detail with reference to the drawings.

[Extraction method 1: Smoke detection based on luminance distribution of pixels]
The smoke feature quantity calculation means 31 calculates the luminance dispersion of the pixels in each area for each candidate area where the smoke detection specified by the smoke detection area selection unit 20 should be performed in detail. In calculating the luminance variance, the smoke feature amount calculating unit 31 does not necessarily need to use all the pixels in the region. The smoke feature amount calculating unit 31 may calculate the luminance variance only for the pixels mapped to “1” as pixels having a high possibility of intrusion or smoke by the luminance change determining unit 22.

  In addition, the smoke feature amount calculation unit 31 basically uses the latest captured image as the image for calculating the luminance dispersion. However, a plurality of captured images can be used retroactively.

  Then, the smoke feature quantity calculation means 31 determines whether or not the calculated luminance dispersion or the standard deviation obtained from the luminance dispersion is within a predetermined range. It can be determined that there is a high possibility that it has occurred.

[Extraction method 2: Smoke detection based on temporal dispersion of average luminance of pixels]
The smoke feature quantity calculation means 31 calculates the average luminance of the pixels in each area for each candidate area where the smoke detection specified by the smoke detection area selection unit 20 should be performed in detail. In calculating the average luminance, the smoke feature amount calculating unit 31 does not necessarily need to use all the pixels in the region. The smoke feature amount calculation unit 31 may calculate the average luminance only for the pixels mapped to “1” as pixels having a high possibility of occurrence of an intruder or smoke by the luminance change determination unit 22.

  Next, the smoke feature amount calculating unit 31 calculates the average luminance in the same region of a plurality of captured images going back in the past for a certain period, and generates time series data of the average luminance for each target region. . Then, the smoke feature quantity calculating means 31 calculates the luminance variance of the generated time series data of average luminance.

  Then, the smoke feature quantity calculation means 31 determines whether or not the luminance variance calculated based on the time series data of the average luminance or the standard deviation obtained from the luminance variance is within a predetermined range. If it is within the range, it can be determined that the possibility that smoke has occurred is high.

[Extraction method 3: Smoke detection based on low frequency intensity of average luminance of pixels]
The smoke feature quantity calculation means 31 generates time series data of average luminance for each target area in the same manner as the extraction method 2 described above. Then, the smoke feature amount calculation means 31 performs Fourier transform on the generated time series data of the average luminance to calculate a power spectrum.

  Next, the smoke feature quantity calculation means 31 extracts a predetermined low frequency component from the power spectrum calculated based on the time series data of the average luminance, calculates the intensity for the mode, and the intensity is a predetermined value. When it is below, it can be judged that the possibility that smoke was generated is high.

[Extraction method 4: Smoke detection based on difference average with reference image]
The smoke feature quantity calculation means 31 has, for each candidate area where smoke detection specified by the smoke detection area selection unit 20 is to be performed in detail, each pixel in the candidate area and a reference stored in the image memory 10 in advance. A luminance difference value with a corresponding pixel of the image is obtained. Further, the smoke feature quantity calculation means 31 obtains an average value of the luminance difference values for each candidate area, and the possibility that smoke has occurred when the average value is larger than a predetermined value or within a predetermined range. Can be determined to be high.

[Extraction method 5: Smoke detection based on correlation value with reference image]
The smoke feature quantity calculation means 31 has, for each candidate area where the smoke detection specified by the smoke detection area selection unit 20 is to be performed in detail, each pixel in the candidate area and a standard stored in the image memory 10 in advance. A correlation value between luminance values with corresponding pixels of the image is obtained. Further, the smoke feature quantity calculation means 31 obtains a correlation value for each candidate area as a feature quantity related to smoke, and determines that there is a high possibility that smoke has occurred when the value of the correlation value is within a predetermined range. can do.

  FIG. 5 is an explanatory diagram of the smoke detection method based on the correlation value performed by the smoke feature amount calculation unit 31 according to Embodiment 1 of the present invention. 5A is a reference image, FIG. 5B is the latest image captured when no smoke is generated, FIG. 5C is the latest image captured when a human invades, and FIG. 5D shows the latest images captured when smoke is generated. In each figure, a portion surrounded by a square on the lower left side of the window glass in the figure corresponds to one candidate area.

  When the image of FIG. 5B (image in which smoke is not generated) is obtained as the latest captured image, the correlation value in the candidate area with the reference image of FIG. 5A is calculated. Then, since both images are similar, a high value will be shown. Further, for example, assuming that there is a change in illumination, the image in FIG. 5 (b) at that time has a large correlation value because the overall brightness does not change but the overall brightness does not change. A high value can be obtained without any influence.

  On the other hand, when the image shown in FIG. 5C (an image in which a human is present in the candidate area) is obtained as the latest captured image, the correlation value is greatly disturbed and becomes a low value.

  On the other hand, when the image of FIG. 5D (image in which smoke is generated) is obtained as the latest captured image, the correlation in the candidate area with the reference image of FIG. When the value is calculated, the correlation value between the two images is lower than the correlation value when smoke is not generated due to the influence of smoke, but is higher than the correlation value when a human is present. This is because smoke is more transparent than humans, and background image information remains. Therefore, when the calculated correlation value falls within the correlation value within a predetermined range, it can be determined that there is a high possibility that smoke has occurred in the candidate area.

  In calculating the correlation value, the smoke feature amount calculating unit 31 does not necessarily need to use all the pixels in the candidate area. The smoke feature amount calculation unit 31 may calculate the correlation value only for pixels mapped to “1” by the luminance change determination unit 22 as pixels having a high possibility of intrusion or smoke.

[Extraction method 6: Smoke detection based on edge intensity value]
When smoke is generated in the candidate area, it is conceivable that the average density and standard deviation of the edge intensity in the image gradually decrease due to the influence of the smoke. On the other hand, when a moving object enters the candidate area, it is conceivable that the average intensity and standard deviation of the edge intensity in the image change rapidly. Therefore, it is possible to determine whether or not there is a high possibility that smoke has occurred in the candidate area based on the average intensity and standard deviation of the edge intensity. The edge strength used in this extraction method 6 is a value that can be calculated for each pixel using, for example, a well-known sobel filter.

  FIG. 6 is an explanatory diagram related to smoke determination based on the average intensity and standard deviation of the edge intensity, which is performed by the smoke feature amount calculation unit 31 according to Embodiment 1 of the present invention. FIG. 6A shows a reference image stored in advance in the image memory 10 as a normal state where no smoke is generated. In contrast, FIG. 6B is the latest image captured when smoke is generated as an intruder. In each figure, a portion surrounded by a square corresponds to one candidate region.

  As shown in FIG. 6A or FIG. 6B, the smoke feature amount calculating unit 31 calculates the edge strength of the reference image and the edge strength of the latest image. Note that, as an evaluation index of the edge strength in the candidate region, the smoke feature amount calculating means 31 can use the average value (average intensity) of the edge strength obtained in the candidate region and its standard deviation.

  Here, one frame, which is the timing for sequentially capturing images, is set to about 33 ms, and one cycle is set to an interval for capturing images of 128 frames. For example, the latest image is updated every cycle (about 4.2 s). think of. In this case, the smoke feature quantity calculation means 31 calculates the average gray level and standard deviation of the edge intensity in the latest image every cycle. Edge strengths in the candidate regions are schematically shown on the right side of FIGS. Thus, when slow diffusion smoke is generated, the average intensity and standard deviation of the edge intensity gradually decrease.

  Therefore, the smoke feature quantity calculation means 31 has the respective change amounts (differences) between the average density and standard deviation of the latest image calculated for each cycle and the average density and standard deviation of the reference image calculated in advance. Is calculated. The edge change amount is the difference between the current image and the reference image, but the edge change rate may be used.

  Then, the smoke feature amount calculation means 31 determines whether each change amount is within a predetermined smoke determination range for each cycle. The smoke feature amount calculation means 31 is configured such that the state in which the calculated change amount is within a predetermined range continues for N cycles when the period for determining the occurrence of smoke is N cycles (N is an integer of 1 or more). It is judged that there is a high possibility that smoke has occurred.

  Note that the smoke feature quantity calculation means 31 counts up the judgment value when each change amount is within the predetermined smoke judgment range every cycle, and is outside the predetermined smoke judgment range. Alternatively, smoke may be determined by counting down the determination value and calculating the determination value. Specifically, it is conceivable that the initial value of the determination value is zero, +1 when counting up, and -1 as a lower limit value when counting down.

  Thus, by counting every cycle, it is possible to determine whether there is an edge change amount continuously. In the case of smoke, as will be described later, since there is a continuous edge change unlike humans, smoke and humans can be identified and detected.

  Finally, when the determination value updated for each cycle reaches a smoke determination threshold value (for example, 10 counts), the smoke feature amount calculation unit 31 generates slow diffusion smoke. Judge that the possibility is high. In the above description, the determination value is counted up and down by looking at the change state of both the average gray level and the standard deviation. However, a separate counter can be used according to each change state. In this case, it can be determined that there is a high possibility that slow diffused smoke is generated when one of the count values reaches the smoke determination threshold value. The threshold itself can also be kept separate.

  Here, mention is made of obtaining a feature quantity using an edge. In the case of slow smoke, particularly dark slow diffuse smoke, the edge inside the moving object is weak and the edge strength becomes weak. However, in many cases, a person has a relatively strong edge inside the moving object. Slightly thin smoke has the characteristic of weakening the edge of the background because it is translucent, but the human part often has a stronger edge than the background. Furthermore, in terms of time, in the case of slow smoke, the continuous edge strength decreases due to the property of stagnating near the ceiling, but in the case of a person, the continuous edge strength decreases due to fluctuations. Hateful. From the above, it is possible to discriminate between human and smoke by calculating the average value of the edge intensity for each cycle and comparing it with that of the reference image.

[Extraction method 7: Smoke detection based on the number of difference pixels]
When smoke is generated in the candidate area, it is conceivable that the luminance value in the image gradually changes compared to the reference image due to the influence of the smoke. Therefore, the smoke feature amount calculation means 31 calculates the difference in luminance value between the reference image and the latest image for each pixel in the candidate area, and counts pixels for which the calculated difference value is equal to or greater than a predetermined difference threshold value. To do.

  Note that the predetermined difference threshold value need not be uniform for all candidate areas, and may be variably set according to the average contrast of the reference image in each candidate area. Specifically, when the contrast in the candidate region is low, it is considered that the change in contrast due to the generation of smoke is small, so the predetermined difference threshold is set to a low value according to the average contrast. On the other hand, when the contrast in the candidate region is high, it is considered that the change in contrast due to the generation of smoke is large, so the predetermined difference threshold is set as a high value according to the average contrast. As a result, an appropriate threshold value can be set according to the characteristics of the image of the candidate area, and the smoke detection accuracy within one screen can be stabilized.

  Finally, the smoke feature quantity calculation means 31 generates slow diffuse smoke when the number of pixels equal to or greater than a predetermined difference threshold counted in the candidate area exceeds an allowable value. Judge that the possibility is high. In this determination, it may be determined that there is a high possibility that slow diffused smoke is generated when such a state continues for a plurality of predetermined frames. This is because when a moving person appears in the monitoring image, pixels exceeding the difference threshold are generated temporarily, but the number of pixels continuously exceeds the allowable value.

  In addition, it can prevent detecting the change of illumination erroneously by providing the detection of this difference pixel number. In the case of a change in illumination, it is a temporary change (for a moment) and does not continue. That is, once the number of pixels equal to or greater than the difference threshold is counted, when the number of pixels is counted over several cycles, the number of pixels equal to or greater than the predetermined threshold is almost eliminated. On the other hand, in the case of smoke, if the number of pixels is continuously counted, the number of pixels equal to or greater than a predetermined threshold is counted for a predetermined time. can do.

  In the above description, calculation and determination processing are performed for each cycle corresponding to 128 frames. For example, it is possible to divide one cycle and perform calculation and determination processing for every 32 frames. It is.

Step 2: About the function of the mapping means 32 The mapping means 32 performs mapping for each candidate region where smoke determination should be performed in detail based on the extraction results of the seven types of extraction methods 1 to 7 by the smoke feature amount calculation means 31. . For example, the mapping unit 32 sets “1” for an area where smoke is highly likely to be generated by any one or more of the seven types of extraction methods 1 to 7 and “0” for other areas. Can be set.

  As another mapping method, the mapping means 32 sets “1” for an area where it is determined that smoke is likely to be generated by two or more types of extraction methods, and “0” for other areas. Can be set. Alternatively, an area where it is determined that there is a high possibility that smoke is commonly generated by the combination of the seven types of extraction methods 1 to 7 is set to “1”, and the other areas are set to “0”. Can be set. In particular, when detecting the slow diffusion smoke, it is conceivable that mapping is performed with an emphasis on the results of extraction methods 5 to 7 suitable for the detection.

  FIG. 7 is a diagram illustrating a mapping result output by the mapping unit 32 according to the first embodiment of the present invention. By calculating the feature value relating to the smoke for the candidate area shown in FIG. 4, as shown in FIG. 7, it is possible to select an area where smoke is likely to be generated from the candidate areas. it can. In FIG. 7, the candidate area of the door portion (see FIG. 4) is an area extracted by the presence of an intruder such as a passerby that passes through the image, and it is determined that the possibility that smoke has occurred is low. , Three of the three candidate areas under the window (see FIG. 4) indicate a case where it is determined that the possibility that smoke has occurred is high.

Step 3: Function of Smoke Determining Unit 33 Based on the data of each area for one screen mapped by the mapping unit 32, the smoke determining unit 33 has a (candidate) area mapped as “1”. It is determined whether or not detection is continued for a predetermined period across the number of regions. For example, the smoke determination means 33 is a time series in which a region of n (n is an integer of 2 or more) in the vertical direction or the horizontal direction is mapped as “1” by the mapping means 32 and sequentially captured from the past to the present. When the connected area is detected continuously on the screen m (m is an integer of 2 or more) times or more, it can be finally determined that smoke is generated.

  Next, the flow of the entire process of the smoke detection device of the present invention having the configuration of FIG. 1 will be described based on the flowchart. FIG. 8 is a flowchart showing a flow of overall processing of the smoke detection device according to Embodiment 1 of the present invention.

  First, in step S801, the luminance histogram creating unit 21 generates a luminance histogram for each predetermined pixel based on a plurality of time-series image data stored in the image memory 10 (FIGS. 2A and 2B). )reference). Next, in step S802, the luminance change determination unit 22 detects the presence or absence of a luminance value newly generated by the generation of an intruder or smoke based on the generated luminance histogram, and the occurrence of the intruder or smoke is generated. Pixels that are likely to have been mapped are output (see FIG. 3B).

  Next, in step S803, the detection candidate area specifying unit 23 determines whether the ratio of pixels mapped as having a high possibility of occurrence of intruders or smoke is greater than or equal to a predetermined value for each area divided in advance. Judgment is made and a candidate area for smoke detection is specified (see FIG. 4). In step S803, the preprocessing 1 to 3 as described above can be performed.

  The above steps S801 to S803 are a series of processes by the smoke detection area selection unit 20. As described above, the smoke detection processing by the smoke generation detection unit 30 is performed only on the candidate area where the smoke detection specified by the smoke detection area selection unit 20 is to be performed in step S804 and subsequent steps. For this reason, the amount of calculation can be reduced compared with the case where the whole image is processed.

  In step S804, the smoke feature quantity calculation means 31 calculates the feature quantity related to smoke using only the extraction methods 1 to 7 described above for only the areas identified as candidate areas for which smoke detection is to be performed. Next, in step S805, based on the results of the extraction methods 1 to 6, the mapping unit 32 performs mapping output of a region where there is a high possibility that smoke has occurred (see FIG. 7).

  Next, in step S806, the smoke determination unit 33 specifies an area where smoke finally occurs based on the temporal distribution and spatial distribution of the mapping result of the divided areas by the mapping unit 32.

  By performing such a series of processes, the smoke generation detection unit 30 performs the smoke detection specified by the smoke detection area selection unit 20 from the calculation result of the luminance histogram for a certain period in the past with respect to the candidate area where the smoke detection should be performed in detail. Thus, by extracting the feature quantity related to smoke, it is possible to efficiently determine whether or not smoke is generated. As a result, it is possible to detect smoke with high sensitivity while suppressing the influence of disturbance. In particular, by combining seven extraction methods as the smoke feature amount extraction means 31, it is possible to detect only smoke from an image by eliminating what can be a false alarm source such as a human.

  As described above, according to the first embodiment, a configuration is provided in which a luminance histogram within a certain past period is calculated for each predetermined pixel from a plurality of images captured in time series. As a result, it is possible to easily detect the presence or absence of a brightness value newly generated by the generation of an intruder or smoke, and to detect smoke with high sensitivity while suppressing the influence of disturbance. A detection device can be obtained.

  In particular, even when the difference in luminance between the luminance distribution before the occurrence and the luminance distribution after the occurrence is slight, it is newly generated because the luminance histogram is obtained for each pixel based on a plurality of images within a certain past period. The presence of a luminance value can be identified with high accuracy.

  Furthermore, it is possible to perform high-accuracy smoke detection that prevents erroneous detection while reducing calculation load by performing smoke detection by narrowing down to a candidate area where the presence of a newly generated luminance value is detected. .

  Furthermore, whether or not there is a high possibility that smoke is generated in the candidate area by obtaining a plurality of feature quantities as feature quantities for performing smoke detection and by a specific combination of individual judgment results based on the plurality of feature quantities Therefore, the detection accuracy can be further improved. In addition, when calculating the feature quantity regarding smoke, the extraction methods 1 to 3 described above do not need to store a reference image in advance, unlike the extraction methods 4 to 7. Therefore, when the feature amount is obtained without using the extraction methods 4 to 7, a merit that the reference image is unnecessary is obtained.

  Further, if the area mapped as having a high possibility that smoke has been generated using the characteristic that smoke spreads over a specific area is detected continuously for a predetermined period across a certain number of areas, the smoke Smoke judging means for judging that the occurrence of smoke is provided. As a result, erroneous detection can be suppressed, and stable and highly accurate smoke detection can be realized. In particular, by using the extraction methods 5 to 7, slow diffusion smoke can be detected with high accuracy.

  Since the present invention is configured as described above, a luminance histogram is calculated from a plurality of past images, and a luminance with a low probability is detected as an intruder, so that a region with high sensitivity can be extracted even with smoke with a small luminance difference. It is also possible to absorb influences such as changes in illumination. In addition, since the calculation of the histogram for each pixel is enormous in both the storage amount and the calculation amount, the calculation amount is reduced by sharing the histogram between adjacent pixels.

  In the present invention, the smoke candidate region is extracted from the image by the luminance histogram creation unit and the luminance change determination unit. For example, after the image is divided into a plurality of small detection regions in a matrix, each detection is performed. The contrast of the area may be calculated, and the area where the contrast value has decreased may be determined as the smoke candidate area. In this case, a means for detecting a decrease in contrast in this region is a means for configuring the smoke detection area selection unit.

  In the present invention, one image is divided into 7 * 10 vertical areas and the smoke feature amount in each area is calculated. However, the areas are divided more finely and the respective areas are overlapped. You may do it. For example, the size of one area itself is the same as the area shown in FIG. 3, and the setting method of the area on the screen is changed as follows. That is, the areas are set so that adjacent areas overlap each other in half in both the vertical and horizontal directions. That is, one area in FIG. 3 is substantially covered by nine areas, so that one image can be set in a more detailed area, and the accuracy of smoke detection can be further increased. It becomes.

  By performing such region division, it is possible to prevent the detection of joints between regions from becoming unstable. In this case, the determination of the presence or absence of the final smoke may be performed by determining how many squares having a size of 1/4 of the area of FIG. 3 are mapped.

  DESCRIPTION OF SYMBOLS 1 Camera, 10 Image memory, 20 Smoke detection area selection part, 21 Luminance histogram preparation means, 22 Luminance change determination means, 23 Detection candidate area | region identification means, 30 Smoke generation | occurrence | production detection part, 31 Smoke feature-value calculation means, 32 Mapping means, 33 Smoke judging means.

Claims (4)

A smoke detection device that detects the generation of smoke by performing image processing on an image captured by a surveillance camera,
An image memory for storing a plurality of images and reference images captured in time series by the monitoring camera;
Pixels in which the difference between the luminance value of the latest image captured by the monitoring camera and the luminance value of the reference image is equal to or greater than a predetermined difference threshold in the smoke detection area identified as a candidate area where smoke is generated And a smoke feature amount calculating means for determining that the possibility that smoke has been generated is high when the counted pixel exceeds a predetermined allowable value. The smoke detection device according to claim 1,
The smoke feature amount calculating means obtains an average value of luminance values of pixels included in the smoke detection area in the latest image, and smoke is generated using the difference threshold value according to the average value. It is judged whether or not there is a high possibility that the smoke detection device is characterized. A smoke detection device that detects the generation of smoke by performing image processing on an image captured by a surveillance camera,
An image memory for storing a plurality of images and reference images captured in time series by the monitoring camera;
In the smoke detection area identified as a candidate area where smoke is generated, the difference between at least one of the average value and the standard deviation of the edge intensity is calculated between the latest image captured by the monitoring camera and the reference image. The change amount is calculated for each predetermined cycle, and when the period for determining the occurrence of smoke is N cycles (N is an integer of 1 or more), the calculated change amount is in the predetermined range. A smoke feature amount calculating means for determining that the possibility that smoke has been generated is high when the smoke continues. The smoke detection device according to claim 3,
The smoke feature amount calculating means calculates, for each predetermined cycle, a difference in average value of edge strengths as a first change amount and a difference in standard deviation as a second change amount between the latest image and the reference image. When the state where the first change amount is within the first predetermined range continues for the N cycles, or when the state where the second change amount is within the second predetermined range continues for the N cycles. It is judged that there is a high possibility of occurrence of smoke.

Images