JP6096589B2 - Fire detection device and fire detection method - Google Patents

Description

The present invention relates to a fire detection device and a fire detection method for detecting smoke in an early stage of fire from an image of a monitoring area captured by a monitoring camera.

  Conventionally, various devices and systems have been proposed in which a fire is detected by performing image processing on an image of a monitoring area captured by a monitoring camera.

  In such a fire detection device, early detection of a fire is important from the viewpoint of initial extinguishing and evacuation guidance for the occurrence of a fire.

For this reason, in the conventional device (Patent Document 1), as a phenomenon caused by smoke from a fire from an image, a decrease in transmittance or contrast, a convergence of a luminance value to a specific value, a luminance distribution range is narrowed and a luminance distribution is reduced. , A change in average value of brightness due to smoke, a decrease in the total amount of edges, and an increase in intensity in the low frequency band are derived, and these can be comprehensively judged to detect smoke.

JP 2008-046916 A JP-A-7-245757 JP 2010-238028 A

  By the way, in the smoldering combustion which is often in the early stage of fire, a very small amount of smoke rises, the amount of smoke increases with the passage of time, and finally a smoke layer is generated along the ceiling surface. Detects smoke layers generated on the ceiling.

  As described above, it is important to detect a fire at an early stage when a very small amount of smoke rises. However, in a conventional device for detecting smoke by performing image processing on an image, for example, rising smoke The movement (flow) is detected, but it is difficult to detect unless a sufficient amount of smoke rises to detect this flow, and it becomes a very small amount like a thin line. There was a problem that it could not be detected in the early stage of the fire where the smoke of the fire rose.

  In order to solve this problem, the applicant of the present application extracts a linear component of a ridge line generated by smoke from an image of a monitoring area captured by a monitoring camera, and obtains a change in time series with respect to the inclination and the occurrence frequency. Has proposed a fire detection device that detects fires by detecting the very small amount of smoke rising due to smoldering combustion, which occurs frequently in the early stages of fire, by detecting characteristic changes caused by smoke. Application 2013-101883).

  By the way, in a device that detects the rising of a very small amount of smoke that occurs due to smoldering in the early stage of fire from the time-series change of the smoke ridge line component, with a television or projector placed in the monitoring area attached When monitoring, when extracting the ridge line from an image of a video image such as a TV screen and determining the time-series change of the linear component, depending on the video, it may be close to or the same as the characteristic change due to smoke There is a possibility that the image on the TV screen is detected as a characteristic change due to smoke and a fire is misjudged.

  Also, even when there are people or animals in the monitoring area and there is movement in one place, extracting the ridge line from the captured image and determining the time-series change of the linear component may cause a characteristic of smoke depending on the image. May be close to the same or similar, and may be mistaken for a fire. Such a problem may be erroneously determined as a fire for various things that exist in the monitoring area and are moving at that location.

The present invention provides a fire capable of improving reliability by suppressing misjudgment of fire caused by an image other than smoke when smoke is detected from a time-series change in the ridge line linear component of the monitoring image. An object is to provide a detection device and a fire detection method.

(apparatus)
The present invention provides a fire detection device,
An imaging means for capturing an image of the monitoring area;
Ridge line extraction means for performing edge enhancement processing on the image captured by the imaging means and extracting a ridge line;
A linear component extracting unit that divides the image of the ridge line extracted by the ridge line extracting unit into a plurality of image regions, and extracts a linear component of the ridge line for each image region;
For each image region, time-series change detecting means for obtaining a change in time series of the linear component extracted by the linear component extracting means;
Smoke candidate determination means for determining a smoke candidate region when a characteristic predetermined time series change due to smoke is detected from the time series change of the linear component detected by the time series change detection means,
A fire judgment means for judging a fire based on a distribution of smoke candidate areas detected by the smoke candidate judgment means;
It is provided with.

  Here, the fire determination means determines a fire when a distribution extending upward beyond the predetermined number of smoke candidate areas is detected. Further, the fire determination means determines that the fire is not fired when a distribution staying in one or a plurality of smoke candidate areas is detected.

  The smoke candidate determination means determines a smoke candidate region when one or more time-series changes of linear components having a constant slope and different occurrence frequencies are detected, and a time-series of linear components having a constant slope and a constant occurrence frequency When a change is detected, it is determined as a non-smoke candidate area and excluded.

(Method)
The present invention provides a fire detection method,
An image of the monitoring area is captured by the imaging means,
The edge enhancement process is performed on the image of the monitoring area captured by the imaging unit by the ridge line extraction unit, and the ridge line is extracted.
The image of the ridge line extracted by the ridge line extraction unit is divided into a plurality of image regions, and the linear component of the ridge line is extracted by the linear component extraction unit for each image region,
For each image area, the change in the time series of the linear component extracted by the linear component extraction means is obtained by the time series change detection means,
Among the time-series changes of the linear components detected by the time-series change detection means, when the smoke candidate determination means detects a predetermined time-series change characteristic of smoke, it is determined as a smoke candidate area,
Based on the distribution of the smoke candidate area detected by the smoke candidate determination and knowledge means, the fire is determined by the fire determination means.

The other features of the fire detection method of the present invention are basically the same as those of the above-described fire detection device, and thus the description thereof is omitted.

According to the fire detection device and the fire detection method of the present invention, the linear component of the ridge line generated by smoke is extracted from the image of the monitoring area imaged by the imaging means, and the inclination and frequency of occurrence are characterized by smoke in time series. A smoke candidate area is detected and a smoke candidate area is detected. Based on the distribution of the smoke candidate area, for example, when a distribution of smoke candidate areas extending upward beyond a predetermined number of areas is detected, a fire is determined. When the distribution of smoke candidate areas staying in one or more areas is detected, it is judged as non-fire, so the image on the TV screen, the movement of the person's limbs staying in one place, etc., and staying in one place It is possible to reliably prevent misjudgment of fire caused by surveillance images of things with moving parts, etc., and to judge fire from the very small amount of smoke that starts up with a lot of smoldering combustion in the early stage of fire, reliability Improvement possible to be.

Explanatory drawing which showed the monitoring area | region where the fire detection apparatus of this invention was installed An explanatory diagram modeling the state where a very small amount of smoke rises Block diagram showing outline of functional configuration of image processing apparatus Explanatory diagram showing image segmentation Explanatory diagram specifying and showing the area of image processing directly on the fire source, background, and television screen Explanatory drawing showing the time series change of the linear component of the area directly above the fire source Explanatory drawing which showed the time series change of the vector showing the inclination of linear component of FIG. 6, and the occurrence frequency Explanatory drawing which showed the time-sequential change of the linear component of the area | region above the area | region of FIG. Explanatory drawing which showed the time series change of the vector showing the inclination of the linear component of FIG. 8, and the occurrence frequency Explanatory diagram showing the time-series change of the linear component of the background area Explanatory drawing which showed the time-sequential change of the vector showing the inclination of linear component of FIG. 11, and the occurrence frequency. Explanatory drawing showing the time-series change of the linear component of other areas that are the background Explanatory drawing which showed the time series change of the vector showing the inclination of linear component of FIG. 13, and the occurrence frequency. Explanatory drawing which showed distribution of the smoke candidate area | region judged from the area | region specified in FIG. Explanatory drawing which specified the area of image processing when there is a person in the monitoring area, and showed the distribution of the smoke candidate area Flowchart showing the operation of the image processing apparatus of FIG.

[Outline of fire detection device]
FIG. 1 is an explanatory view showing a monitoring area where a fire detection device according to the present invention is installed. FIG. 1 (A) shows a side view, and FIG. 1 (B) shows a front view seen from a monitoring camera.

  As shown in FIG. 1A, a monitoring camera 10 functioning as an imaging unit is installed in the monitoring area 16, and an image is obtained by imaging the state of the monitoring area shown in FIG.

  A combustible material placed in the monitoring area 11 causes a fire for some reason, and in the initial stage of the fire, a very small amount of smoke 24 rises as a fine line from the trash container serving as the fire source 18. The wall surface of the monitoring area 11 appears as straight lines in the vertical and horizontal directions due to the structure and wallpaper. Furthermore, a television 16 is arranged in the monitoring area 11, and when the television 16 is turned on, an image having a motion depending on the received broadcast program is displayed on the television screen 16a.

  An image captured by the monitoring camera 10 is transmitted to an image processing apparatus 12 installed in a manager's room or the like via a transmission path, and a small amount of smoke 24 rising from the fire source 18 is detected by image processing to determine a fire. The fire detection signal is output to the fire alarm equipment to output a fire alarm.

[Detection principle]
The principle of detecting a small amount of smoke according to the present invention will be described as follows. In the present invention, a small amount of smoke 24 rising from the fire source 18 shown in FIG. 2A is detected by image processing. In this case, the initial smoke 24 is translucent, as shown in FIG. It can be considered as a smoke model 24a in which a cylindrical object extends upward from a fire source while fluctuating.

  In the smoke model 24a, as shown in the density distribution of FIG. 2C, the smoke density is higher at the center and relatively thinner at the periphery. It is thought that the center of the line draws the ridgeline that is least transparent.

  Therefore, edge enhancement processing is applied to the image to extract smoke ridge lines, and after the image is divided into fine regions, Hough transform is performed on each region to extract linear components. The linear component due to the smoke extracted in this way extends upward while fluctuating over time. On the other hand, the linear component existing in the background does not change over time and exists constantly. For this reason, if the change of the extracted linear component in the time series is caught, the characteristic time series change due to the smoke can be caught.

  Based on the above results, finding the time-series changes in the direction and cumulative frequency of the linear components of each region extracted from the images taken at predetermined intervals, characteristic time-series changes due to smoke are obtained, It is possible to detect a small amount of smoke 24 that rises as a thin line in the initial stage.

[Fire misjudgment]
As shown in the detection principle of the present invention, edge enhancement processing is applied to an image to extract smoke ridge lines, and after dividing into fine areas, Hough transform is performed on each area to extract linear components When a fire is judged by capturing a characteristic time-series change due to smoke of the extracted linear component, for example, a time-series change of the linear component obtained by processing the video on the TV screen 16a is a characteristic time due to smoke. There is a case where it cannot be distinguished from the series change, and there is a possibility that the moving image of the television screen 16a is erroneously determined as a fire as it is. In the present invention, it is possible to prevent a fire from being erroneously determined from a moving image such as an image on a television screen.

[Fire detection device]
(Functional configuration of fire detection device)
FIG. 3 is a block diagram schematically showing the functional configuration of the fire detection device according to the present invention. As shown in FIG. 3, the fire detection device includes a monitoring camera 10 and an image processing device 12.

  The image processing apparatus 12 includes a CPU, a memory, a computer circuit having various input / output ports and the like as hardware, and a ridge line extraction functioning as a ridge line extraction unit as a function realized by executing a program by the CPU. The unit 28 includes a fire determination unit 30 that functions as a fire determination unit.

  Furthermore, as functions of the fire determination unit 30, a linear component extraction unit 32 that functions as a linear component extraction unit, a time series change detection unit 34 that functions as a time series change detection unit, and a smoke candidate determination unit 36 that functions as a smoke candidate determination unit And a fire determination unit 38 that functions as a fire determination means. The transmission unit 26 uses an appropriate transmission interface that receives image data captured by the monitoring camera 10.

  The monitoring camera 10 functioning as an imaging unit transmits image data in a monitoring area at 30 frames per second, for example, as moving image data by transmission control of the transmission unit 26 and stores the image data in a memory (not shown) provided in the image processing device 12.

  The ridge line extraction unit 28 generates a ridge line image by extracting a ridge line from the frame-unit image stored in the memory. For example, in this case, the ridge line extraction unit 28 applies a Sobel filter, which is one of the edge enhancement processes, to the image, and for example, as illustrated in FIG. 4, the ridge line image 20 including extraction of the smoke ridge line 24b. Generate. Note that the ridge line extraction processing by the ridge line extraction unit 28 may be performed for each frame in which a predetermined number of frames are thinned out because of the processing speed, without targeting all frame images.

  The Sobel filter is present in the image by multiplying the nine pixel values at the upper left and right with a certain target pixel as the center by multiplying a predetermined coefficient by two coefficient matrices in the horizontal and vertical directions to obtain the sum. This is a differential process that makes it possible to detect the boundary (edge) of a certain area, and by applying this, the ridge line extraction unit 28 detects the ridge line of smoke from the image of smoke rising upward from the fire source 18, as shown in FIG. 24b is extracted. Similarly, the ridge line extraction unit 28 extracts each ridge line for the fire source 18, the background, and the television 16 in the monitoring area 11.

  The fire determination unit 30 determines a fire by detecting a predetermined predetermined ridge line due to smoke from the ridge lines extracted by the ridge line extraction unit 28. Specifically, the fire determination unit 30 includes a linear component extraction unit 32, a time series, and the like. The change detection unit 34, the smoke candidate determination unit 36, and the fire determination unit 38 detect a fire.

  The straight line component extraction unit 32 divides the image 20 of the ridge line extracted by the ridge line extraction unit 28 into a plurality of regions, for example, 64 × 64 pixel regions, as shown by dotted lines in FIG. Hough transform) is performed to extract the linear component of the ridgeline. The Hough transform is known as a method for extracting a straight line segment in an image. For n points in the image, n curves are obtained on the ρ-θ plane. Of these, m curves are obtained. Are intersected at one point, m points on the image corresponding to the m points are on the same straight line, and thereby a linear component can be extracted.

  The time series change detection unit 34 obtains the time series change of the slope and occurrence frequency of the straight line component extracted by the Hough transform by the straight line component extraction unit 32.

  FIG. 5 is an explanatory diagram showing the areas to be processed for the convenience of explanation with respect to the ridge line image 20 of FIG. 4. The four areas immediately above the fire source 18 are designated as A1 to A4, and two areas are shown. The background areas are A5 and A6, and the area including the television screen 16a is A7 to A18.

  In the ridge line image of FIG. 5, the smoke ridge line 24a rises from the fire source 18 and the tip extends to the area A2. On the other hand, in the background area A5, there are regularly four vertical lines, and in the background area A6, one vertical line and six horizontal edges are present. Yes. The television screen 16a displays a moving image (not shown) depending on the received broadcast program.

(Time series change of straight line components extracted from smoke ridge lines)
FIG. 6 shows time-series changes of the linear components extracted at times t1 to t4 corresponding to four periods of the thinned frame in the area A1 immediately above the fire source 18 in FIG. At time t1, the linear component of the smoke ridge line passing through the area A1 has two occurrence frequencies when the upper direction is θ1 = 0 ° in the two-dimensional coordinates with the center of the lower side of the area as the origin, and the upper right diagonal direction is θ2. The occurrence frequency is two, and if the upper left diagonal is θ3, the occurrence frequency is one. The direction and frequency of occurrence of the linear component of the smoke ridge line passing through the region A1 changes as shown at times t2 to t4 according to the fluctuation of the rising smoke.

  FIG. 7 is an explanatory diagram showing a time-series change of the linear component in the area A1 of FIG. 6, and the linear component is shown by accumulating vectors having a slope θ and the length of occurrence frequency.

As shown in FIG. 7, at time t1, the vector B1 is (θ1, 2).
The vector B2 is (θ2, 2)
The vector B2 is (θ3, 1)
Thus, from time t2 to t4, it increases cumulatively according to the time series change.

  FIG. 8 shows a time-series change of the linear components extracted at times t1 to t4 corresponding to four periods of the thinned-out frame in the area A2 above the area A1 immediately above the fire source 18 in FIG. In the area A2, the rising smoke fluctuates, and therefore, at time t1, the linear component of the smoke ridge line passing through the area A2 has an occurrence frequency of 1 when the upper direction is θ1 = 0 °, and the upper right diagonal If θ2 is θ2, the occurrence frequency is 1, and if the upper left is θ3, the occurrence frequency is 2. Further, if the upper right oblique direction is larger than θ2, the occurrence frequency is one, and the upper left oblique direction is larger than θ3. If θ5, the frequency of occurrence is one.

  Thus, the direction and frequency of occurrence of the linear component of the smoke ridge line passing through the region A2 changes as shown at times t2 to t4 according to the fluctuation of the rising smoke.

  FIG. 9 is an explanatory diagram showing a time-series change of the linear component in the area A2 of FIG. 8, and the linear component is shown by accumulating vectors having the slope θ and the length of occurrence frequency.

As shown in FIG. 9, at time t1, the vector B1 is (θ1, 1),
The vector B2 is (θ2,1)
The vector B3 is (θ3, 2),
The vector B4 is (θ4, 1),
The vector B5 is (θ5, 1)
Thus, from time t2 to t4, it increases cumulatively according to the time series change.

  As described above, the time-series change of the linear component in each area of the smoke ridge line detected by the time-series change detection unit 34 is, as shown at time t4 in FIGS. A plurality of so-called daisy patterns exist, and this is a characteristic time-series change due to smoke.

(Time series change of straight line components extracted from the background ridgeline)
FIG. 10 shows a time-series change of the straight line components extracted at times t1 to t4 for four cycles of the thinned-out frame in the area A5 as the background of FIG. In the area A5, four linear components are constantly present in the upper and lower sides in the background. Therefore, at all times t1 to t4, the linear components existing in the area A5 are generated when θ1 = 0 ° upward. The frequency is four.

  FIG. 11 is an explanatory diagram showing a time-series change of the linear component in the area A5 of FIG. 10, and the linear component is shown by accumulating vectors having the slope θ and the length of occurrence frequency.

  As shown in FIG. 11, at time 1, the vector B1 becomes (θ1, 1), and from time t2 to t4, the vector B1 increases cumulatively with a constant occurrence frequency = 4 according to the time series change.

FIG. 12 shows a time-series change of the straight line components extracted at times t1 to t4 for four cycles of the thinned frame in the area A6 as the background of FIG. In the area A6, there are one straight line component in the vertical direction and six straight line components in the background in the background, and in the two-dimensional coordinates with the origin at the center of the lower side of the area, the right side is θ5 = + 90 ° and the left side Is θ6 = −90 ° and the diagonally upper left is θ7, at all times t1 to t4,
The frequency of occurrence of θ1 is one, the frequency of occurrence of θ5 is two, the frequency of occurrence of θ6 is two, the frequency of occurrence of θ7 is two, and the inclination and the frequency of occurrence are constant and constant.

  FIG. 13 is an explanatory diagram showing a time-series change of the linear component in the area A6 of FIG. 12, and shows the linear component accumulated by a vector having a slope θ and a length of occurrence frequency.

As shown in FIG. 13, at time t1, the vector B1 is (θ1, 1).
Vector B5 is (θ5,2)
The vector B6 is (θ6, 2)
The vector B7 is (θ7,2)
From time t2 to time t4, it increases cumulatively at a constant frequency according to the time series change.

  As described above, the time-series change of the linear component of each region of the background ridge line detected by the time-series change detection unit 34 has a constant slope and a constant frequency of occurrence as shown at time t4 in FIGS. It becomes a steady pattern and can be clearly distinguished from a daisy pattern that shows characteristic time-series changes due to smoke.

(Time series change of linear components extracted from TV screen)
A video ridge line image extracted from the video of the broadcast program is generated on the TV screen 16a in FIG. 4, and this video ridge line image is an image that moves in various ways in time series. The time series change of the ridge line linear component for each of the regions A7 to A18 by the series change detection unit 34 is a so-called daisy pattern in which a plurality of cumulative vectors having different slopes and occurrence frequencies exist in the same manner as shown in FIGS. The change is similar to the characteristic time-series change caused by smoke and cannot be distinguished as it is.

[Smoke area candidates and fire judgment]
The smoke candidate determination unit 36 in FIG. 3 determines a smoke candidate region when a characteristic predetermined time series change due to smoke is detected from the time series change of the ridge line linear component detected by the time series change detection unit 34. . For example, the smoke candidate determination unit 36 determines a smoke candidate region when detecting a plurality of time-series changes of linear components having a constant inclination and different occurrence frequencies, when detecting a so-called daisy pattern, while determining that the inclination is constant. When a time-series change of a linear component having a constant occurrence frequency is detected, when a so-called steady pattern is detected, it is determined as a non-smoke candidate region and excluded.

  The fire determination unit 38 of FIG. 3 determines a fire based on the distribution of smoke candidate areas determined by the smoke candidate determination unit 36. For example, the fire determination unit 38 detects a distribution extending upward beyond a predetermined number of smoke candidate areas because a very small amount of smoke rises in the smoldering combustion that is often performed at the early stage of the fire. Judge. On the other hand, the fire determination unit 38 determines that there is no fire when it detects a distribution in which the smoke candidate area remains within one or a plurality of areas, such as the smoke candidate area determined from the video on the television screen.

  FIG. 14 is an explanatory diagram showing the distribution of smoke candidate areas determined at times t1 to t4 for four cycles of a predetermined thinned frame from the area specified in FIG. 5, and the smoke candidate areas are indicated by hatching.

  In FIG. 14, the smoke candidate areas A1 to A4 show a distribution that extends upward with the passage of time, and this is a distribution specific to smoke in which a very small amount of smoke rises due to smoldering combustion. Judged as a fire. In this case, it is determined that there is a fire at the timing of time t4 when the number of smoke candidate areas extending upward reaches a predetermined number of areas, for example, 3 areas and becomes 4 areas.

  On the other hand, the smoke candidate areas A7 to A18 are determined to be smoke candidate areas from the video on the TV screen, and the distribution remains within a rectangular range corresponding to the size of the TV screen 16a. The distribution does not extend upward beyond the number, and it is not judged as a fire.

  Further, FIG. 14 illustrates a case where all the areas A7 to A18 are determined as smoke candidate areas. However, in a program video with little movement, some areas in the areas A7 to A18 are discretely smoke candidate areas. It is assumed that a distribution extending upward within a rectangular range corresponding to the screen size is assumed, but since it does not have a distribution extending upward beyond the predetermined number of areas, a fire is determined. There is no.

  FIG. 15 is an explanatory diagram showing the image processing area when there is a person in the monitoring area and the distribution of the smoke candidate area. As shown in FIG. 15A, when a person 15 is sitting on a sofa placed in the monitoring area, the limbs of the person 15 show various movements in this state, and the time series of the ridge line component of the moved part is shown. From the change, a smoke candidate region indicated by hatching in FIG.

  In this case as well, as in the case of the smoke candidate area corresponding to the television screen shown in FIG. 14, the distribution remains within the range corresponding to the posture of the person 15 and exceeds the predetermined number of areas like smoke. Therefore, the distribution does not extend upward, and it is not judged as a fire.

  In the case where the person 15 moves in the monitoring area, a daisy pattern having a short vector length is temporarily generated in each area through which the person 15 passes, for example, as shown at time t1 in FIG. 9 or FIG. The cumulative vector that becomes the time-series cumulative occurrence frequency does not increase if it passes, and a predetermined threshold is set for the cumulative occurrence frequency, and an area having an accumulated occurrence frequency equal to or greater than the threshold is determined as a smoke candidate area Therefore, a fire is not judged from the time-series change of the ridge line component accompanying the movement of the person.

  In addition, when a person stays in the monitoring area and does not move, it becomes a time-series change of the same steady pattern as the background ridgeline, and does not affect the fire judgment.

[Fire detection operation]
FIG. 16 is a flowchart showing a fire detection operation by the image processing apparatus of FIG.

  In FIG. 16, the image processing apparatus 12 acquires an image of a monitoring area captured at, for example, 30 frames / second as a moving image by the monitoring camera 10 in step S1 (hereinafter “step” is omitted), and stores the acquired image in a memory. Then, the ridge line is extracted from the image by applying the Sobel filter by the ridge line detection unit 28.

  Subsequently, in S3, the image of the ridge line is divided into a plurality of regions by the straight line component extraction unit 32, and the Hough transform is performed for each region in S4 to extract the straight line component of the ridge line. To obtain a time-series change based on the slope of the linear component extracted in S3 and the cumulative frequency of occurrence. In S4, the fire detection unit 36 determines the characteristic linear component due to smoke from the time-series change of the slope of the linear component and the occurrence frequency. A smoke candidate region is determined by detecting a predetermined time-series change that is an inclination and a cumulative frequency of occurrence, for example, a daisy pattern.

  Subsequently, in S7, it is determined from the distribution of the smoke candidate area whether or not there is a distribution peculiar to smoke that extends upward with the passage of time. As a result, when a fire is determined in S8, a fire detection signal is output to the fire alarm facility in S9 to output a fire alarm. On the other hand, if the fire is not determined in S8, the process returns to S1 and the same processing is repeated.

[Modification of the present invention]
(Thing with movement)
In the above embodiment, the image of the TV screen as an object to be determined as a smoke candidate area other than smoke, and the movement of a limb when a person stays in one place is taken as an example. All movements such as the movement of the clock pendulum, the movement of the pet animal, the air volume adjustment of the air conditioner, the swinging movement of the fan, etc. may be mistakenly judged to be a fire from the distribution even if it is judged as a smoke candidate area. It can be prevented.

(Ridge line extraction)
In the above embodiment, the ridge of smoke is extracted by applying a Sobel filter to the image, but an appropriate filter used for edge enhancement processing such as a Previtt filter is applied. Also good.

(Linear component extraction)
In the above embodiment, the smoke ridge line is extracted by applying the Hough transform. However, a processing method for extracting a linear component from an image such as Line Segment Detector (LSD) may be applied.

(Image processing device)
In the above embodiment, the surveillance camera and the image processing apparatus are separately arranged and connected by a transmission path, but an apparatus in which both are integrated may be used.

The present invention is not limited to the above-described embodiment, includes appropriate modifications without impairing the object and advantages thereof, and is not limited by the numerical values shown in the above-described embodiment.

10: surveillance camera 12: image processing device 14: fire alarm equipment 16: monitoring area 18: fire source 20: image 24: smoke 24a: smoke model 24b: smoke ridge 26: transmission unit 28: ridge extraction unit 30: fire determination unit 32: Linear component extraction unit 34: Time-series change detection unit 36: Smoke candidate determination unit 38: Fire determination unit

Claims (8)

An imaging means for capturing an image of the monitoring area;
A ridge line extracting unit that performs edge enhancement processing on the image captured by the imaging unit and extracts a ridge line;
A linear component extraction unit that divides the image of the ridge line extracted by the ridge line extraction unit into a plurality of image regions, and extracts a linear component of the ridge line for each image region;
Time series change detection means for obtaining a change in time series of the linear component extracted by the linear component extraction means for each image region;
Smoke candidate determination means for determining a smoke candidate area when a characteristic predetermined time series change due to smoke is detected from the time series change of the linear component detected by the time series change detection means;
Fire determination means for determining a fire based on the distribution of smoke candidate areas detected by the smoke candidate determination means;
A fire detection device comprising:
In the fire detection device according to claim 1,
The fire detection device determines that a fire is detected when a distribution extending upward beyond a predetermined number of the smoke candidate areas is detected.
In the fire detection device according to claim 1,
The fire detection device determines that the fire is not fire when a distribution staying in one or a plurality of smoke candidate areas is detected.
The fire detection device according to claim 1, wherein the smoke candidate determination means determines a smoke candidate region when one or more time-series changes of linear components having a constant inclination and different occurrence frequencies are detected, and the inclination is determined. A fire detection device characterized in that a non-smoke candidate region is excluded when a time-series change of a linear component having a constant occurrence frequency is detected.
An image of the monitoring area is captured by the imaging means,
Applying edge emphasis processing by the ridge line extraction means to the image of the monitoring area imaged by the imaging means to extract the ridge line,
The image of the ridge line extracted by the ridge line extraction unit is divided into a plurality of image regions, and the linear component of the ridge line is extracted by the linear component extraction unit for each image region,
For each image area, the time-series change detecting means obtains the time-series change of the linear component extracted by the linear component extracting means,
Among the time-series changes of the linear components detected by the time-series change detection means, when the smoke candidate determination means detects a characteristic time-series change characteristic of smoke, it is determined as a smoke candidate area,
Based on the distribution of the smoke candidate area detected by the smoke candidate determination means, determine a fire by the fire determination means,
A fire detection method characterized by that.
In the fire detection method according to claim 5,
The fire determination means determines that a fire has occurred when a distribution extending upward beyond a predetermined number of smoke candidate areas is detected.
In the fire detection method according to claim 5,
The fire detection method according to claim 1, wherein the fire determination unit determines a non-fire when a distribution staying in one or a plurality of smoke candidate areas is detected.
  6. The fire detection method according to claim 5, wherein the smoke candidate determination means determines a smoke candidate region when one or more time-series changes of linear components having a constant inclination and different occurrence frequencies are detected, and the inclination is determined. A fire detection method characterized in that a non-smoke candidate region is determined and excluded when a time-series change of a linear component having a constant occurrence frequency is detected.

Images