JP2020135784A - Smoke detection device and smoke detection method - Google Patents

Abstract

To detect a fire at an early stage by performing class traces with particle filters on images captured in a visible light illumination environment for knowing smoke-specific movement by arranging a plurality of classes indicating smoke regions in chronological order so as to prevent particles from concentrating in one place.SOLUTION: A fire detection unit repeats a process in which a difference image generation unit generates a difference image for each image captured by a monitoring camera, and a class tracking unit scatters particles in the difference image and performs weighting by likelihood indicating smokeiness for each particle to obtain the center of gravity of the smoke region and arranges a class, and then tracks the class to detect a movement trajectory, so that a fire determination unit detects a characteristic change in the smoke from the movement trajectory of the class and determines a fire. The class tracking unit randomly scatters particles to arrange classes, and after arranging the classes, rearranges the classes by scattering particles around a predetermined position above the center of gravity of the class, thus enabling detection of the movement trajectory by arranging a predetermined number of classes without interference.SELECTED DRAWING: Figure 2

Description

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

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

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

For this reason, in the conventional device (Patent Document 1), as phenomena caused by smoke accompanying a fire from an image, a decrease in transmittance or contrast, a convergence of a brightness value to a specific value, and a narrowing of the brightness distribution range to disperse the brightness. The decrease in brightness, the change in the average brightness due to smoke, the decrease in the total amount of edges, and the increase in intensity in the low frequency band are derived, and these are comprehensively judged to enable smoke detection.

However, in such a conventional fire detection system that detects a fire from an image of smoke associated with a fire, the entire image captured by a surveillance camera is processed to detect a characteristic change due to smoke to determine a fire. Therefore, there is a problem that the processing load for determining a fire from the entire image increases and the processing takes time.

In order to solve this problem, in Patent Document 2, there are a dense region and a low-concentration region inside the smoke rising due to a fire, and the dense smoke region gradually appears and moves upward. Since the phenomenon of going on appears, classes are placed in the dense smoke part from the captured image according to the rules of the particle filter, for example, and each class is tracked for each image to obtain the position information, which is characteristic of smoke. By detecting changes that move upward, it is possible to make a reliable fire judgment.

Japanese Unexamined Patent Publication No. 2008-046916 JP-A-2017-191544

By the way, in the method of arranging and tracking the class in the smoke area using such a conventional particle filter, when the particles are scattered on the target image and the class is arranged, the particles are concentrated in one place. Since there is a problem that multiple classes cannot be arranged and tracked, a large number of spot lights are irradiated to the surveillance area, and the images of the areas exposed to the spot lights scattered in the images of the surveillance area captured by the surveillance camera are displayed. By extracting and tracking, it is possible to generate and track multiple classes.

However, a special spot irradiation device is required to irradiate the monitoring area with a large number of spot lights, and an infrared light source is used to obscure the spot light, and the lighting environment is based on normal visible light. There is a problem that it cannot be used in and lacks versatility.

In the present invention, for an image captured in a visible light illumination environment, particles are not concentrated in one place by class tracking by a particle filter, and a plurality of classes indicating a smoke area are arranged in chronological order to be unique to smoke. It is an object of the present invention to provide a smoke detection device and a smoke detection method that can detect various movements and detect a fire at an early stage.

(Smoke detector)
The present invention is a smoke detection device.
An imaging means that sequentially captures images in the monitoring area,
A class tracking means that repeats the process of arranging classes by finding the center of gravity of the smoke region in the image for each image generated by the imaging means, and tracks the classes to detect the movement trajectory.
A fire judgment means for judging a fire by detecting a characteristic predetermined change due to smoke from the movement trajectory of the class detected by the class tracking means,
Is provided,
Further, the fire determination means is characterized in that when the movement trajectory of the class rises with lateral fluctuation, it is judged as smoke, and when it rises linearly, it is judged not as smoke.

(Judgment of the presence or absence of smoke based on the inclination of the class movement trajectory)
The fire judgment means detects the inclination between each class in the movement trajectory of the class, and if the difference between the maximum value and the minimum value of the inclination is equal to or more than a predetermined threshold value or exceeds the threshold value, it is determined as smoke, and is less than the threshold value or the above threshold value. In the following cases, it is judged that it is not smoke.

(Smoke detection by particle filter)
The class tracking means scatters a predetermined number of particles in an image according to a predetermined rule for each image captured by the imaging means, weights each particle by a likelihood indicating smokeiness, and is less than a predetermined value or a predetermined value. Repeat the process of arranging the class while deleting the particles with the likelihood below the value and leaving the particles with the likelihood above or above the predetermined value, tracking the class and detecting the movement trajectory.
In addition, class tracking means
In the case of an image with no classes placed, a predetermined number of particles are randomly scattered in the image to place a new class.
In the case of an image in which classes are placed, a predetermined number of particles are scattered around a predetermined position above the particles left in the previous class placement according to the probability density distribution, the classes are rearranged, and all the classes are rearranged. After that, a predetermined number of particles are randomly scattered in the area excluding the class to place a new class.

(Smoke detection by particle filter using difference image)
Further, a difference image generation means for generating a difference image is provided for each of a plurality of images captured by the imaging means.
The class tracking means tracks the class by arranging and rearranging the class based on particles for each difference image generated by the difference image generating means, and detects the movement locus.

(Class disappearance and reuse)
The class tracking means can arrange up to a predetermined maximum number of new classes, and when it detects the retention of classes, it deletes the accumulated classes and uses it for the allocation of new classes.

(Particle scattering area when relocating the class that has been retained and erased)
When the stagnant class is erased, the class tracking means randomly disperses a predetermined number of particles below the area where the other class exists, or below the erased class when the other class does not exist. Sprinkle and relocate new classes.

(Likelihood of particles)
The class tracking means weights the pixel value of the particles as the likelihood.

(Smoke detection method)
Another embodiment of the present invention is a method for detecting smoke.
Images in the monitoring area are sequentially captured by the imaging means, and
The class tracking means repeats the process of finding the center of gravity of the smoke region in the image and arranging the class for each image generated by the imaging means, and tracks the class to detect the movement trajectory.
The fire judgment means judges a fire by detecting a characteristic predetermined change due to smoke from the movement trajectory of the class detected by the class tracking means.
Further, the fire determination means is characterized in that when the movement trajectory of the class rises with lateral fluctuation, it is judged as smoke, and when it rises linearly, it is judged not as smoke.

The features of the smoke detection method other than this are the same as those of the smoke detection device described above.

(Basic effect)
In the smoke detection device of the present invention, an imaging means that sequentially captures an image of a monitoring region and a process of arranging classes by obtaining the center of gravity of the smoke region in the image for each image generated by the imaging means. A class tracking means that tracks the class and detects the movement trajectory, and a fire judgment means that detects a characteristic predetermined change due to smoke from the movement trajectory of the class detected by the class tracking means and judges a fire. In addition, the fire judgment means judges that if the movement trajectory of the class rises with lateral fluctuation, it is judged as smoke, and if it rises linearly, it is judged as not smoke. For example, since a class generated from an image of the movement of a person or the like has a linear movement, by excluding this from the target of fire judgment, the movement trajectory of the class that fluctuates from side to side, which is peculiar to smoke, is detected. You can definitely judge the fire.

(Judgment of the presence or absence of smoke based on the inclination of the class movement trajectory)
In addition, the fire judgment means detects the inclination between each class in the movement trajectory of the class, and if the difference between the maximum value and the minimum value of the inclination is equal to or more than a predetermined threshold value or exceeds the threshold value, it is determined as smoke and is less than the threshold value or Since it is determined that it is not smoke when it is below the threshold value, it is possible to reliably determine the movement trajectory of the class that moves linearly in an arbitrary direction on the screen and exclude it from the target of fire judgment.

(Effect of smoke detection by particle filter)
The class tracking means scatters a predetermined number of particles in an image according to a predetermined rule for each image captured by the imaging means, weights each particle by a likelihood indicating smokeiness, and is less than a predetermined value or a predetermined value. The process of erasing particles with a likelihood of less than or equal to the value and arranging the class while leaving particles with a likelihood of more than or equal to a predetermined value is repeated, the class is tracked to detect the movement trajectory, and further, the class tracking means. Is
In the case of an image without classes, a predetermined number of particles are randomly scattered in the image to place a new class, and in the case of an image with classes, the particles left in the previous class placement A predetermined number of particles are scattered around a predetermined position above and the classes are rearranged according to the probability density distribution, and after all the classes are rearranged, a predetermined number of particles are randomly scattered in the area other than the class to make a new one. When relocating the class, the particles are scattered around the upper part of the particles left in the previous class placement to relocate the class, and all the classes are relocated. After that, by arranging new classes by scattering particles in the area other than the class arrangement area, the classes will be arranged without interfering with each other against the smoke mass of constant concentration rising due to the rising smoke vortex. And the rearrangement can be continuously tracked, and the characteristics peculiar to smoke can be captured from the movement trajectories of multiple classes to reliably determine that the fire is occurring.

Also, since smoke is generally difficult to descend, when relocating a class, particles are scattered in the area slightly above the center of gravity of the class and the class is rearranged to duplicate particles between the classes. It can be reliably prevented and the movement trajectory of the class that accurately follows the rise of the smoke mass can be detected.

(Effect of smoke detection by particle filter using difference image)
Further, a difference image generation means for generating a difference image is provided for each of a plurality of images captured by the imaging means, and the class tracking means is a particle-based class for each difference image generated by the difference image generation means. Since the movement trajectory is detected by tracking the class by rearranging and rearranging, it is possible to emphasize only the moving part such as smoke in the image and prevent erroneous detection.

(Effects of class disappearance and reuse)
In addition, the class tracking means can arrange new classes up to a predetermined maximum number, and when a class retention is detected, the accumulated classes are deleted and used for the placement of new classes. Even if the maximum number of classes to be placed is limited, the smoke can be tracked smoothly by arranging the classes in a circular manner.

(Particle scattering area when the retention class is erased)
When the stagnant class is erased, the class tracking means randomly disperses a predetermined number of particles below the area where the other class exists, or below the erased class when the other class does not exist. Since the smoke area rises and disappears from the screen, the class that stays at the top of the screen is erased and the newly appearing smoke area affects other classes because it is sprinkled and a new class is rearranged. It is possible to place classes initially and resume smoke tracking smoothly.

(Likelihood of particles)
Further, since the class tracking means weights the particles with the pixel value as the likelihood, the particles can be weighted by setting the likelihood corresponding to the smoke concentration to appropriately obtain the center of gravity of the smoke region.

The effect of the smoke detection method is the same as the effect of the smoke detection device described above.

Explanatory drawing which showed the monitoring area which installed the smoke detection apparatus of this invention Block diagram showing the outline of the functional configuration of the smoke detection device Explanatory drawing showing a frame image captured by a surveillance camera and its difference image Explanatory diagram showing the procedure to arrange a new class by initial scattering of particles with respect to the smoke area that first appeared. Explanatory drawing showing the scattered area of particles during class rearrangement It is explanatory drawing which showed the arrangement of a new class following the class rearrangement of FIG. Explanatory diagram showing the time-series change of class arrangement and the class movement trajectory for the difference image that changes in time series Explanatory diagram showing the rearrangement of stagnant classes Explanatory diagram showing the movement trajectory of a linear class Flow chart showing smoke monitoring processing by the smoke detection device of FIG. A flowchart showing the details of class tracking by the particle filter in step S2 of FIG. A flowchart showing the details of the fire judgment based on the class tracking trajectory of step S3 of FIG.

[Basic concept of the embodiment]
FIG. 1 is an explanatory diagram showing a monitoring area in which the smoke detection device of the present invention is installed, and FIG. 2 is a block diagram showing an outline of a functional configuration of the smoke detection device.

The basic concept of the present embodiment is to input a moving image taken by the surveillance camera 10 functioning as an imaging means into the smoke detection device 12 to detect a fire at an early stage. From the image input to the smoke detection device 12, the difference image generation unit 24 that functions as the difference image generation means generates a difference image for each of a plurality of images, and the class tracking unit that functions as the class tracking means using the particle filter. According to the method 26, a predetermined number of particles are scattered in the image according to a predetermined rule for each difference image generated by the difference image generation unit 24, and each particle is weighted by a likelihood indicating smokeiness, and is less than a predetermined value. Alternatively, the process of erasing the particles having a likelihood of less than or equal to a predetermined value and arranging the class while leaving the particles having a likelihood of more than or equal to a predetermined value is repeated, the class is tracked to detect the movement locus, and then, The fire determination unit 28, which functions as a fire determination means, detects a characteristic predetermined change due to smoke from the movement locus of the class detected by the class tracking unit 26, and determines a fire.

In particular, the class tracking unit 26 randomly scatters a predetermined number of particles in the image to arrange a new class in the case of a difference image in which no class is arranged, and in the case of a difference image in which a class is arranged, the class tracking unit 26 arranges a new class. , The class is rearranged by scattering according to the predetermined probability density distribution centering on the predetermined position above the particles left in the previous class arrangement, and after rearranging all the classes, a predetermined number of particles are placed in the area other than the class. Randomly scatter and place new classes.

In this way, when the class tracking unit 26 rearranges the classes, the particles are scattered around the position above the particles left in the previous class arrangement to rearrange the classes, and all the classes are rearranged. After that, by arranging new classes by scattering particles in the area other than the class arrangement area, there is no mutual interference with the smoke mass (smoke area) of a certain concentration that rises as a smoke vortex. , Classes can be continuously relocated and tracked, and the movement trajectories of multiple classes can be used to capture the unique characteristics of smoke and reliably determine that it is a fire.

In addition, since smoke is generally difficult to descend, when rearranging a class, particles are placed in a region slightly above the position of each particle when the class was previously placed, and particles belonging to another class. By relocating the classes by scattering the particles in an area that does not include a predetermined range centered on the center of gravity, the particles are surely prevented from overlapping between the classes, and the class that accurately follows the rise of the smoke mass. The movement locus can be detected.

Further, in the present embodiment, the fire determination unit 28 determines that the movement locus of the class fluctuates in the lateral direction, for example, when the movement locus of the class rises with a lateral fluctuation width of a predetermined value or more, it is judged as smoke and rises linearly. If it is, it is judged that it is not smoke. For example, the class generated from the difference image of the movement of people etc. will be a linear movement, so by excluding this from the target of fire judgment, smoke It is possible to reliably judge a fire by detecting the movement trajectory of the class that fluctuates from side to side, which is peculiar to. This will be described in detail below.

[Overview of smoke detector]
As shown in FIG. 1, a surveillance camera 10 is installed in the surveillance area 14, and the surveillance camera 10 sequentially images the surveillance area illuminated by the luminaire at a frame speed of, for example, 30 frames per second or 60 frames per second. It outputs a moving image consisting of a series of color frame images. The surveillance camera 10 is installed in the center of, for example, the upper corner of the surveillance area (surveillance space) 14 partitioned vertically, horizontally, and front and back, and the imaging optical axis is arranged diagonally downward so as to face the opposite wall surface for monitoring. The area 14 can be photographed as a whole.

The surveillance camera 10 captures a moving image of the visible image of the surveillance area 14, and the moving image captured by the surveillance camera 10 is transmitted to a smoke detection device 12 installed in a manager's room or the like via a transmission path, and is dusted by image processing. Smoke 16 rising from a fire source 15 such as a fire is detected by image processing using a particle filter to determine a fire, and a fire detection signal is transmitted to a fire alarm facility 18 to output a fire alarm. The size of the moving image taken by the surveillance camera 10 is, for example, 1280 × 720 pixels.

[Functional configuration of smoke detector]
As shown in FIG. 2, the smoke detection device 12 is composed of a computer circuit or the like equipped with a CPU, a memory, various input / output ports, etc. as hardware, and is controlled as a function realized by executing a program by the CPU. A unit 20, an image input unit 22, a difference image generation unit 24, a class tracking unit 26, and a fire determination unit 28 are provided. Further, an alarm display unit 30 and an acoustic alarm unit 32 are provided to notify an alarm when a fire is determined.

The control unit 20 performs overall control of each function provided in the surveillance camera 10 and the smoke detection device 12. The surveillance camera 10 operates in response to an instruction from the control unit 20, transmits image data in a monitoring area of, for example, 30 frames per second as moving image data by transmission control, and the image input unit 22 of the smoke detection device 12 It is received by and stored in a memory (not shown). The image input unit 22 reads the stored moving image in frame units and outputs it to the difference image generation unit 22 according to the instruction of the control unit 20.

[Difference image generator]
3A and 3B are explanatory views showing a frame image captured by a surveillance camera and a difference image thereof, FIG. 3A shows an image of smoke, and FIG. 3B shows a difference image of smoke.

The difference image generation unit 22 generates a difference image according to the following procedure.
(Procedure 1) The frame image is grayscaled and smoothed with a Gaussian filter to reduce noise.
(Procedure 2) The maximum value and the minimum value of each pixel in the latest 30 frames including the current frame image are monitored, and the maximum value image and the minimum value image between the 30 frames are created.
(Procedure 3) A difference image is created from the maximum value image and the minimum value image generated in step 2.

The generation of the difference image is not limited to steps 1 to 3, and an appropriate method for generating the difference image can be applied.

Here, the length of 30 frames for which the difference is taken is the length of processing for an image of smoke with a slow flow velocity, and in the case of smoke with a high flow velocity, it is necessary to set it to, for example, about 6 frames. , It corresponds by parallel processing that creates a difference image with a plurality of frames corresponding to the flow velocity.

As shown in FIG. 3A, in the smoke image 16a of the frame 34 in which the smoke 16 rising from the fire source 15 shown in FIG. 1 is photographed, the smoke has a characteristic of rising while fluctuating, and smoke is present. Focusing on a certain point in the image, it is considered that the presence or absence of smoke changes on the line connecting the viewpoint and the background, and there are a state where the image density is the original background and a state where the smoke density is. Smoke is visualized by difference processing using the characteristics of.

The smoke difference image 16b shown in the difference frame 36 of FIG. 3B is generated by the difference processing of steps 1 to 3, and the smoke difference image 16b has a time series of smoke regions that form a mass of smoke having a constant density. It is visualized that the image rises while appearing multiple times. In the difference frame 36, the fire source image 15a disappears as shown by the dotted line.

[Class Tracking Department]
FIG. 4 is an explanatory diagram showing the procedure up to the placement of a new class by the initial scattering of particles with respect to the smoke region that first appeared, and FIG. 5 is an explanatory diagram showing the scattering region of particles at the time of class rearrangement. FIG. 6 is an explanatory diagram showing the arrangement of new classes following the class rearrangement of FIG.

(Overview of Class Tracking Department)
The class tracking unit 24 arranges a smoke region, which is a mass of smoke, as a class for the difference image by using a particle filter, and tracks this in chronological order to detect the movement trajectory of the class.

The class tracking processing unit 26 repeats the processing of sampling, prediction, and observation, which are the processing procedures of the particle filter. That is, the class tracking processing unit 26 scatters a plurality of particles (particles) in the image according to a predetermined rule for each difference image generated by the difference image generation unit 24, and the likelihood of showing smokeiness for each particle. The process of arranging the classes is repeated to obtain the center of gravity of the smoke region by weighting the class, and the movement trajectory is detected by tracking the class. Here, one particle corresponds to, for example, one pixel, the density value of the particles is used as the likelihood, and weighting is performed according to the pixel value.

The size of the particles may be a predetermined number of a plurality of pixels. In this case, the total pixel values of the pixels included in the particles are weighted as the likelihood.

(Initial placement of class)
When the class is not arranged in the processing up to that point, the class tracking unit 26 randomly scatters a predetermined number of particles, for example, 40 particles in the read difference image to create a new class. Performs the process of arranging.

For example, as shown in FIG. 4, for the processing frame 38-1 in which the smoke region 41 that first appears as a smoke mass surrounded by a dotted line, 40 particles 40 are randomly scattered over the entire region of the image.

Subsequently, as shown in the processing frame 38-2, when the pixel value of each particle is weighted as the likelihood, the particles having a low likelihood are designated as white circles, and the particles having a high likelihood are designated as black circles, as shown in the processing frame 38-3. When 34 low-likelihood particles are extinguished to leave highly-likelihood smoke estimation particles 42 and the center of gravity is calculated for the weight values of the remaining smoke estimation particles 42, as shown in processing frame 38-4. , Class 44-1 having a center of gravity 46-1 is arranged. The center of gravity 46-1 means the position of each particle scattered in the class rearrangement.

Here, class 44-1 shows a smoke region that becomes a smoke mass, but it is actually a set of smoke estimation particles 42, and is shown as a circular region for the sake of simplicity.

(Relocation of classes)
As shown in FIG. 4, the class tracking unit 26 arranges the class in the smoke area that first appears, and then when the next difference image is read, the class tracking unit 26 refers to the six particles remaining in the previous class arrangement. , A number such that the total number of rearranged particles is 40, determined by the likelihood of the particles that are the reference of the position, centered on the predetermined position above the position (center of gravity) 46 of each particle when it was placed last time. , For example, the particles are scattered at the coordinate positions generated according to the random numbers of the normal distribution, and the class is rearranged in the state where 40 particles are scattered in total. At this time, the higher the likelihood of the particles at the time of the previous placement, the larger the number of particles to be rearranged based on the position.

Here, the distance L from the position (center of gravity) 46 of each particle placed last time to the scattering center 48 is changed according to the moving speed corresponding to the color of smoke or the moving speed of the past class. This is because the movement of smoke between images changes depending on the speed at which the smoke rises, so if the speed of movement of smoke is high, the distance from the center of gravity of the class 46 to the center of scattering of particles 50 when the class is rearranged. By lengthening L and shortening the distance L if the smoke moving speed is low, particles can be scattered in the optimum area for class rearrangement according to the smoke moving speed.

For example, black smoke from a combustion fire has a high ascending speed, so the distance L to the particle scattering center is increased, and white smoke from a smoked fire has a low ascending speed, so it reaches the particle scattering center L. By shortening the distance, it is possible to accurately rearrange the class according to the movement of smoke.

Further, the class tracking unit 26 is a region below the horizontal line 52 passing through the position (center of gravity) 46 of the previously placed particles among the particles scattered for class rearrangement in the scattered region 50 shown in FIG. Is set as a non-scattering area 54 as shown by a dotted line, and particles existing in the non-scattering area 54 are excluded from the calculation of the center of gravity position of the class to be rearranged, and exist in the scattering area 50 existing above the horizontal line 52. The center of gravity is obtained from the likelihood of the particles and rearranged.

The exclusion of particles existing in the non-scattering region 54 can be eliminated by setting the weight of the particles to zero. Further, the center of gravity of the class may be calculated as the particles in the scattered area 50 by moving all the particles existing in the non-scattered area 54 on the horizontal line 52.

When rearranging the class in this way, the smoke is usually downward by excluding particles located below the class centroid in the previous image, or by moving to the class centroid position and rearranging the class. By rearranging the class to be rearranged so as not to move it downward, it is possible to detect the movement trajectory of the class peculiar to smoke.

The processing frames 38-5 and 38-6 of FIG. 6 show the rearrangement of the class for the next difference image following the processing frame 38-4 of FIG. The processing frame 38-5 is in a state in which the particles are scattered so that the total number of particles is 40, centering on a predetermined position above the position (center of gravity) 46-1 of each particle at the time of the previous arrangement. As shown in the processing frame 38-6, the particles with low likelihood are eliminated, the particles indicated by the black circles with high likelihood are left, and the class centroid 46-2 of the particles with high likelihood is obtained to obtain class 44-2. Relocated.

In the rearrangement of the class, the distance L from the position (center of gravity) 46 of each particle placed last time to the scattering center 48 shown in FIG. 5 is changed according to the moving speed of smoke. The distance L can be made substantially constant by changing the number of frames for which the difference is taken in the difference image generation unit 24 shown in 2 according to the moving speed of smoke.

For example, assuming that the difference image generation unit 24 generates a difference image in units of 30 frames for smoke having a slow flow velocity and a difference image in units of 6 frames for smoke having a high flow velocity, the difference image is generated in units of 30 frames with a slow flow velocity. The amount of movement of the smoke region in a difference image having a high flow velocity, for example, in units of 6 frames, can be made substantially the same, and the processing of the class tracking unit 26 can be simplified.

(Initial placement of class after class relocation)
When all the classes have been rearranged in the image in which the classes are arranged in the previous image, the class tracking unit 26 randomly scatters 40 particles in the area other than the rearranged class area to make a new one. Place the class Perform the class initialization placement.

In the class rearrangement shown in the processing frames 38-5 and 38-6 of FIG. 6, the class placed in the immediately preceding image is one, and the class rearrangement is completed in one process. There is. When the class relocation is completed in this way, 40 particles 40 are randomly scattered in the area other than the class area 44-2 having the relocated class center of gravity 46-2 as shown in the processing frame 38-7. , The center of gravity is calculated by weighting the particles according to the likelihood of the pixel value in the same manner as the class initialization arrangement shown in FIG. 4, and if a new smoke region appears, a new class is found there. Can be placed.

(Detection of class movement trajectory)
The class tracking unit 26 performs a process of detecting the movement locus of a class that moves in time series by repeating class arrangement and class rearrangement in the difference image read in time series.

FIG. 7A shows processing frames 60-1 to 60-5 processed in time series at times t1 to t5, and classes C1 to C5 are sequentially arranged and increased by rearrangement.

As a result, the class tracking unit 26 can detect the movement locus of the class centers of gravity Gt1 to Gt5 in the class C1 shown in FIG. 7B, for example, for the class C1. Similarly, the movement locus of the class center of gravity is detected for the remaining classes C2 to C5.

(Erase of stagnant class)
As shown in the processing frame 60-5 of FIG. 7A, the class C1 moved due to the rise of smoke is not rearranged and stays when it reaches the upper end of the image.

When the class tracking unit 26 detects the stagnant class C1 as shown in the processing frame 60-5 of FIG. 8 when the class rearrangement processing is completed, the class C1 is deleted (as shown by the dotted line). Reset) and make it available for placement on new classes.

For reuse after deleting the accumulated classes, for example, assuming that the maximum number of classes is 5, as shown in FIG. 8, when the accumulation of class C1 is detected and deleted in the processing frame 60-5, the next sub As shown in the processing frame 60-6 for the image, the lower side of the smoke area of the class C2 to C5 existing at that time, for example, the lower side of the horizontal line 62 passing under the class C5 is set as the scattering area 64. , 40 particles 40 are randomly scattered to generate a new class, and the newly generated class is arranged as class C1 erased by retention.

Further, when class C1 stays and is erased, when other classes do not exist, a new class is arranged by using the lower side of the horizontal line passing under the erased class C1 as a scattering area 6. Perform initialization processing.

In the present embodiment, the maximum number of classes to be arranged by the class tracking unit 26 is set to, for example, 16 classes, but by erasing and rearranging such stagnant classes, the classes are arranged cyclically. Classes can be arranged without being restricted by the maximum number of classes.

(Effect of relocation based on class priority)
As shown in the processing frames 60-1 to 60-5 of FIG. 7, the class tracking unit 26 sets different priorities for each class when repeating the arrangement and rearrangement of the classes with respect to the smoke regions appearing one after another. From the likelihood of the remaining particles, excluding the particles that overlap in the smoke area of the upper class that has already been rearranged, among the multiple particles that were rearranged in order from the upper class and scattered for class rearrangement. Performs the process of relocating to find the center of gravity.

For example, if the older the class is, the higher the priority is set. Taking the relocation from the processing frame 60-4 in FIG. 7 to the processing frame 60-5 as an example, the higher priority is given in the order of classes C1 to C4. The degree is set.

At this time, the class tracking unit 26 first rearranges the class C1 having the highest priority, but at this time, the scattering range of the particles is not limited by the other classes C2 to C4 having the lower priority. Next, the class placement unit 26 rearranges the next high-priority class C2. At this time, the rearranged high-priority class C1 smoke region is masked, the particles scattered for the class C2 rearrangement are removed from the class C1 smoke region, and the class C1 and C2 smoke regions are removed. Do not overlap.

As a result, the classes can be rearranged without overlapping in response to the continuously rising smoke region, and the movement trajectory of the classes can be accurately detected.

[Fire Judgment Department]
The fire determination unit 28 determines a fire by detecting a characteristic predetermined change due to smoke based on the movement locus of the center of gravity position of each class detected by the class tracking unit 26. That is, the fire determination unit 28 determines that if the movement locus of the class rises with a lateral fluctuation within a predetermined width, for example, with a lateral fluctuation width of a predetermined value or more, it is determined as smoke, and the movement locus becomes less than the predetermined width and is linear. If it rises to, it is judged that it is not smoke.

Such a linear rise of the class movement locus appears in the class movement locus generated from the difference image of the movement of a person or the like. In the high-luminance portion of the difference image generated by the movement of a person or the like, the color density distribution is solid and even, so that the movement trajectory of the class is linear.

FIG. 9 is an explanatory diagram showing a linear class movement locus, and when a person moves from the front side of the surveillance camera 10 to the front side in the optical axis direction, it is linear as shown in the processing frame 60-11. It becomes the movement locus of the center of gravity Gt1 to Gt5 of the class that rises to, and when a person moves diagonally forward, the movement locus of the center of gravity Gt1 to Gt5 of the class that moves linearly diagonally upward like the processing frame 60-12. Therefore, when a person crosses the front, the movement locus of the center of gravity Gt1 to Gt5 of the class that moves linearly in the lateral direction is obtained as in the processing frame 60-13. It should be noted that the processing frames 60-13 are unlikely to appear because the particles are scattered around the upper side of the center of gravity and the classes are rearranged.

In the determination of the linearity of the movement locus by the fire determination unit 28, in addition to the lateral shake width, for example, the inclination (vector) between each class in the movement locus of the class is detected, and the difference between the maximum value and the minimum value of the inclination is detected. May be determined to be smoke if is greater than or equal to or exceeds a predetermined threshold value, and may be determined not to be smoke if is less than or equal to the threshold value. Here, the difference between the maximum value and the minimum value of the inclination indicates the amount of lateral shake (degree of lateral shake) of the class movement locus, and if there is no or small amount of lateral shake, it can be determined that the movement is linear.

The determination of the linearity of the movement locus is not limited to the above method, and an appropriate method such as a method of determining from the amount of deviation of each class with respect to the straight line connecting the start point and the end point of the class movement locus can be applied.

Further, the fire judgment unit 28 sets the accumulation condition for the fire judgment based on the movement locus of the class. For example, when the smoke judgment of the class movement locus is consecutive a plurality of times, or the smoke judgment of the class movement locus is a plurality of frames. When it is continuous, it is judged as a fire and a fire alarm is output and a fire detection signal is transmitted to the outside.

[Processing operation of smoke detector]
10 is a flowchart showing an outline of smoke detection processing by the smoke detection device of FIG. 2, FIG. 11 is a flowchart showing details of class tracking by the particle filter of step S2 of FIG. 10, and FIG. 12 is step S3 of FIG. It is a flowchart which showed the detail of the fire judgment based on the class tracking locus of.

(Overview of smoke detection processing operation)
As shown in FIG. 10, the control unit 20 of the smoke detection device 10 shown in FIG. 2 operates the difference image generation unit 24 in step S1, and the predetermined number of pixels photographed by the surveillance camera 10 from the image input unit 22. The maximum value and the minimum value of each pixel are detected, the maximum value image and the minimum value image between a plurality of frames are created, and the difference image is generated from the maximum value image and the minimum value image.

Subsequently, the control unit 20 proceeds to step S2, operates the class tracking unit 26, scatters a plurality of particles in the image according to a predetermined rule for each difference image generated by the difference image generation unit 24, and each particle. By weighting with the likelihood of indicating smoke-likeness, the process of finding the center of gravity of the smoke region and arranging the classes is repeated, and the class is tracked to detect the movement locus indicating the movement of the position of the center of gravity of the class.

Subsequently, the control unit 20 proceeds to step S3, operates the fire determination unit 28, and if the movement trajectory of the class center of gravity detected by the class tracking unit 26 continues for a predetermined time or longer, causes a fire as smoke due to a fire. to decide.

Subsequently, the control unit 20 proceeds to step S4, determines whether or not the fire determination in step S3 satisfies a predetermined accumulation condition, and proceeds to step S5 when, for example, the accumulation condition such that the fire determination is continuous a predetermined number of times is satisfied. Proceed, the alarm display unit 30 and the acoustic alarm unit 32 are operated to output a fire alarm, and a fire detection signal is transmitted to the fire alarm system 18 shown in FIG. 1 to output a fire alarm.

(Class tracking process)
The class tracking unit 26, which operates according to the instruction of the control unit 20 in step S2 of FIG. 10, performs the class tracking process shown in FIG. The class tracking unit 26 reads the difference image generated by the difference image generation unit 24 in step S11, and if the smoke region first appears in the difference image, the class has not been arranged in the processing up to that point. , Step S12 determines that there is no existing class, and proceeds to step S13 to arrange the class as the initialization process.

In the class arrangement as the initialization process by the class tracking unit 26, for example, 40 particles are randomly scattered in the difference image in step S13 as shown in FIG. 4, and the density value of each particle is set in step S14. Set the weight as the likelihood, erase the particles with low likelihood, leave the particles with high likelihood, calculate the center of gravity of the weight of the particles left in step S15, and place a new class with the obtained centroid. Then, the center of gravity (coordinate position) of the class arranged in step S16 is stored, and the process returns to the main routine of FIG.

On the other hand, when the class tracking unit 26 determines the existence of an existing class in step S12, the class tracking unit 26 proceeds to step S17 and performs a class rearrangement process. The class rearrangement process by the class tracking unit 26 takes out the center of gravity of the existing class in step S17. Here, when a plurality of classes have already been arranged, the priorities are set for the classes in the order of oldest, so that the class tracking unit 26 has the center of gravity of the class with the highest priority (the class with the oldest appearance). Take out.

Subsequently, the class tracking unit 26 proceeds to step S18, determines whether or not the class from which the center of gravity is taken out is a non-smoke class, and if it is a non-smoke class, proceeds to step S19, deletes the class, and creates a new class. Make it available for placement. For example, as shown in the processing frame 60-5 of FIG. 7, the class that reaches the upper end of the screen and stays is erased as a non-smoke class. Also, if the width of the particle distribution used to calculate the center of gravity of the class is wider than the predetermined range (when the standard deviation exceeds the predetermined value), it is deleted as a non-smoke class.

When the class tracking unit 26 determines that the class is smoke-like in step S18, the process proceeds to step S20, and as shown in FIG. 5, in addition to the particles left in the previous process, the class tracking unit 26 scatters the particles at a predetermined distance L above the class center of gravity 76. 48 is set, and the number of particles erased in the previous process is scattered by generating coordinates with a normal distribution random number, weights are set with the density value of each particle as the likelihood in step S21, and in step S22. Eliminate particles with low likelihood, leave particles with high likelihood, calculate the center of gravity of the weight of the remaining particles, rearrange the class, and proceed to step S23 to relocate the center of gravity (coordinate position) of the rearranged class. Remember.

Subsequently, the class tracking unit 26 proceeds to step S24 to determine whether or not the rearrangement of all classes has been completed, and if not, returns to step S17 and takes out the center of gravity of the next high-priority class. , Steps S20 to S23 relocate the class.

When the class tracking unit 26 determines in step S24 that the relocation of all the classes has been completed, the process proceeds to step S26, and when it determines that the class has been deleted in the class relocation processes of steps S17 to S25, the class tracking unit 26 proceeds to step S27. As shown in the processing frame 60-6 of FIG. 8, 40 particles are randomly scattered under the existing class, and the density value of the particles is weighted as the likelihood in steps S14 to S16 to determine the likelihood. The initialization process for arranging a new class is performed by erasing low particles and finding the position of the center of gravity while leaving particles with high likelihood.

When the particles are scattered in step S27, if there is no other class other than the erased class, 40 particles are randomly scattered under the erased class, and the particle densities are scattered in steps S14 to S16. A new class is placed by weighting the value as the likelihood, erasing the particles with low likelihood, leaving the particles with high likelihood, and finding the center of gravity.

When the class tracking unit 26 determines that there is no class deleted in step S26, the process proceeds to step S28, and as shown in the processing frame 38-7 of FIG. 6, the class tracking unit 26 randomly fills the free area of the relocated existing class. Forty particles are scattered in the space, and the initialization process for arranging a new class is performed in steps S14 to S16. The particles in step S28 may be scattered by randomly scattering 40 particles while masking the smoke region of the existing class.

(Fire judgment processing)
The fire determination unit 28, which operates according to the instruction of the control unit 20 in step S3 of FIG. 10, performs the fire determination process shown in FIG.

As shown in FIG. 12, the fire determination unit 28 reads the time-series data of the center of gravity position showing the tracking locus of the class center of gravity shown in the processing frame 60 of FIG. 7B in step S31, and class in step S32. The movement direction of the center of gravity position between them is detected, and it is determined in step S33 whether or not the movement is linear.

The fire determination unit 28 detects the inclination of the straight line connecting the center of gravity between each class as the detection of the moving direction of the center of gravity position in step S32, and obtains the difference between the maximum value and the minimum value of the detected inclination as the amount of lateral movement. , In step S33, the amount of lateral shake is compared with a predetermined threshold value, and if it is equal to or more than the threshold value, the process proceeds to step S34 to determine that it is a smoke candidate. Exclude.

Subsequently, the fire determination unit 36 proceeds to step S36, and if it determines that the number of smoke candidates determined in step S34 is equal to or greater than a predetermined threshold value, it proceeds to step S37, determines that it is smoke, and performs fire countermeasure processing. For example, a fire alarm is output and a fire detection signal is transmitted.

[Modification of the present invention]
(Class placement)
In the above embodiment, when the accumulated class is deleted and a new class is placed, the lower side of the area where the other class exists, or the lower side of the deleted class when the other class does not exist. Classes are arranged by randomly scattering particles on the side, but the present invention is not limited to this, and for example, a predetermined number of particles may be uniformly and randomly scattered over the entire screen.

(Fire detection)
Further, in the above embodiment, as a characteristic time-series change due to smoke, a class is arranged and tracked in the characteristic area by a particle filter, but the present invention is not limited to this, and other appropriate filter processing may be applied. A fire may be determined by tracking a time-series change that indicates a characteristic movement due to smoke.

(Image processing during normal operation and when smoke is generated)
In addition, the number of images to be processed is normally reduced (for example, one image for every five images taken), and when it is judged that there is a high possibility of smoke, the number of images to be processed is increased (images of all images taken). You may try to process it. Here, the number of images specifically means the number of frames.

In addition, when it is judged that there is a high possibility of smoke, it is judged that there is a high possibility of smoke, including images that were not processed (for example, the remaining 4 images at the time of 5 imagings), and from the time point. All image processing may be performed again from a short time ago.

Further, as another image processing, a process of simply changing the imaging interval, for example, may increase the imaging interval in normal times and shorten the imaging interval when it is determined that there is a high possibility of smoke.

(Other)
Further, the present invention is not limited to the above-described embodiment, includes appropriate modifications that do not impair its purpose and advantages, and is not further limited by the numerical values shown in the above-described embodiment.

10: Surveillance camera 12: Smoke detection device 14: Surveillance area 15: Fire source 16: Smoke 16a: Smoke image 16b: Smoke difference image 18: Fire alarm system 20: Control unit 22: Image input unit 24: Difference image generation unit 26: Class tracking unit 28: Fire judgment unit 30: Alarm display unit 32: Acoustic alarm unit 34: Frame 36: Difference frame 40: Particle 42: Smoke estimation particles 44-1, 44-2: Class 46-1, 46- 2: Center of gravity 48: Scattering center 50, 58: Scattering area 54: Non-scattering area

Claims (14)

An imaging means that sequentially captures images in the monitoring area,
For each image generated by the imaging means, a class tracking means that repeats the process of arranging the class by obtaining the center of gravity of the smoke region in the image, and tracks the class to detect the movement locus, and a class tracking means.
A fire determination means for determining a fire by detecting a characteristic predetermined change due to smoke from the movement locus of the class detected by the class tracking means.
Is provided,
Further, the fire determination means is characterized in that when the movement locus of the class is rising with lateral fluctuation, it is determined as smoke, and when it is rising linearly, it is determined that it is not smoke. Smoke detector.
In the smoke detection device according to claim 1, the fire determination means detects the inclination between each class in the movement locus of the class, and the difference between the maximum value and the minimum value of the inclination is equal to or more than a predetermined threshold value. A smoke detection device, characterized in that if it exceeds the threshold value, it is determined to be smoke, and if it is less than or equal to the threshold value, it is determined not to be smoke.
In the smoke detection device according to claim 1,
The class tracking means scatters a predetermined number of particles in an image according to a predetermined rule for each image captured by the imaging means, weights each particle by a likelihood indicating smokeiness, and is less than a predetermined value. Alternatively, the process of erasing the particles having a likelihood of less than or equal to the predetermined value and arranging the classes while leaving the particles having a likelihood of more than or equal to the predetermined value is repeated, and the class is tracked to detect the movement locus. ,
Furthermore, the class tracking means
In the case of an image in which the class is not arranged, the predetermined number of particles are randomly scattered in the image and a new class is arranged.
In the case of an image in which the class is arranged, a predetermined number of particles are scattered according to a probability density distribution around a predetermined position above the particles left in the previous class arrangement, and the class is rearranged to relocate all the classes. A smoke detection device, characterized in that a predetermined number of particles are randomly scattered in an area other than the above-mentioned class to arrange a new class after rearranging the particles.
In the smoke detection device according to claim 3,
Further, a difference image generation means for generating a difference image is provided for each of a plurality of images captured by the imaging means.
The class tracking means traces the class by arranging and rearranging the class based on the particles for each difference image generated by the difference image generating means, and detects a movement locus of smoke. Detection device.
In the smoke detection device according to claim 3, the class tracking means can arrange up to a predetermined maximum number of new classes, and when the retention of the class is detected, the retained class is detected. A smoke detector characterized by erasing and using for new class placement.
In the smoke detection device according to claim 5, when the class tracking means erases the stagnant class, it is below the area where the other class exists, or when the other class does not exist. A smoke detector characterized by randomly sprinkling multiple particles and placing a new class underneath the erased class.
The smoke detection device according to claim 3, wherein the class tracking means weights the pixel values of the particles as the likelihood.
Images in the monitoring area are sequentially captured by the imaging means, and
The class tracking means repeats the process of finding the center of gravity of the smoke region in the image and arranging the class for each image generated by the imaging means, and tracks the class to detect the movement locus.
The fire determination means detects a characteristic predetermined change due to smoke from the movement locus of the class detected by the class tracking means and determines that it is smoke.
Further, the fire determination means is characterized in that when the movement locus of the class is rising with lateral fluctuation, it is determined as smoke, and when it is rising linearly, it is determined that it is not smoke. Smoke detection method.
In the smoke detection method according to claim 8, the fire determination means detects the inclination between each class in the movement locus of the class, and the difference between the maximum value and the minimum value of the inclination is equal to or more than a predetermined threshold value. A method for detecting smoke, characterized in that if it exceeds the threshold value, it is determined to be smoke, and if it is less than or equal to the threshold value, it is determined that it is not smoke.
In the smoke detection method according to claim 8,
The class tracking means scatters a predetermined number of particles in the image according to a predetermined rule for each image captured by the imaging means, weights each particle by a likelihood indicating smokeiness, and is less than a predetermined value. Alternatively, the process of erasing the particles having a likelihood of less than or equal to the predetermined value and arranging the classes while leaving the particles having a likelihood of more than or equal to the predetermined value is repeated, and the class is tracked to detect the movement locus. ,
Furthermore, the class tracking means
In the case of an image in which the class is not arranged, the predetermined number of particles are randomly scattered in the image and a new class is arranged.
In the case of an image in which the class is arranged, a predetermined number of particles are scattered according to a probability density distribution around a predetermined position above the particles left in the previous class arrangement, and the class is rearranged to relocate all the classes. A method for detecting smoke, which comprises arranging a new class by randomly scattering a predetermined number of particles in an area other than the class after rearranging the particles.
In the smoke detection method according to claim 10,
Further, a difference image generation means for generating a difference image is provided for each of a plurality of images captured by the imaging means.
The class tracking means traces the class by arranging and rearranging the class based on the particles for each difference image generated by the difference image generating means, and detects a movement locus of smoke. Detection method.
In the smoke detection method according to claim 10, the class tracking means can arrange up to a predetermined maximum number of new classes, and when the retention of the class is detected, the retained class is detected. A smoke detection method characterized by erasing and rearranging as a new class.
In the smoke detection method according to claim 12, when the class tracking means erases the stagnant class and rearranges it as a new class, a predetermined number of the class tracking means is located below the area where the other class exists. A method for detecting smoke, which comprises randomly scattering particles of smoke and rearranging the erased class.
The smoke detection method according to claim 10, wherein the class tracking means weights the particles with the likelihood of the particles as a pixel value.

Images