JP7129053B2 - flame detector - Google Patents

Description

More particularly, the present invention relates to a flame detection device that uses image processing to detect flames generated by fire in a dark environment where it is difficult to detect the generation of smoke.

2. Description of the Related Art There is a conventional fire detection device that applies image processing technology and can extract only a flame region from a monitoring region captured by a monitoring camera, omitting an artificial light source (see, for example, Patent Document 1).

The conventional device according to Patent Document 1 includes a photographing means for photographing a monitored area and an image memory for storing images photographed by the photographing means, and processes the images stored in the image memory to detect a fire. To detect. More specifically, this conventional apparatus includes a fire candidate area extracting means for extracting a fire-like area from an image, and a black smoke detector for detecting whether or not black smoke is generated in the fire peripheral area around the fire-like area. A smoke detecting means and a fire discriminating means for discriminating an area likely to be a fire as a real fire area when the black smoke detecting means detects the black smoke are provided.

The black smoke detection means is composed of feature amount calculation means for calculating the area of the low luminance value area in the fire surrounding area. By providing such a configuration, the conventional apparatus extracts an area having brightness equal to or higher than a predetermined value as a fire-like area, and when the brightness of the fire-surrounding area around the fire-like area is equal to or less than a predetermined value, the fire is detected. It detects that black smoke is generated in the surrounding area.

Then, when black smoke is generated, the conventional apparatus finally determines that the fire-like area is the real fire area. Therefore, even if the monitoring area includes a mixture of light sources for illumination, light sources of vehicles, and the like, the conventional device can reliably distinguish between these light sources and flames. Furthermore, the conventional device makes it possible to detect the occurrence of fire more quickly by determining that black smoke is generated in the area around the fire when the brightness of the area around the fire tends to decrease.

JP-A-10-269471

However, the prior art has the following problems.
The prior art according to Patent Document 1 extracts an area having brightness equal to or higher than a predetermined value as a fire-like area, and specifies the periphery of the fire-like area as a fire surrounding area. However, in the two-stage process of identifying the fire-like area and then identifying the fire-surrounding area, there is a risk that the fire-like area will be identified in the first stage due to a false alarm source.

In particular, if dark environments such as tunnels or indoors after lights are turned off are assumed as monitoring areas, there is a risk that high-brightness areas due to false alarm sources other than flames may be identified as fire-like areas. Become.

Also, in a dark environment, it is difficult to detect the generation of smoke. Therefore, as a result of detecting black smoke, it is difficult to determine that a fire has finally occurred, unlike the conventional apparatus. Therefore, it is desired to perform flame detection with high accuracy without detecting a false alarm source even in such a dark environment.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and in particular, it enables high-precision flame detection using image processing in a dark environment where it is difficult to detect smoke. It is an object of the present invention to obtain a flame detection device that

A flame detection device according to the present invention is a flame detection device that detects the occurrence of a flame by subjecting an image captured by a camera to image processing. Identify the ambient light region by extracting pixels having preset ambient light luminance values to identify the ambient light region illuminated by the outbreak, and determine whether the flame is occurring in the identified ambient light region. In order to discriminate whether or not the flame is generated, the flame detecting unit calculates a feature value indicating behavior peculiar to the flame, and determines whether or not the flame is generated from the magnitude of the feature value calculated in the ambient light region .
Further, the flame detection device according to the present invention is a flame detection device for detecting the occurrence of a flame by performing image processing on an image picked up by a camera, wherein image data obtained by picking up a flame to be detected is A preprocessing unit that pre-sets ambient light luminance values for identifying the ambient light regions illuminated by the flame generation using a flame detection unit that identifies an ambient light region by extracting the flame, and determines whether or not a flame is generated by calculating a feature amount for identifying whether or not a flame is generated in the identified ambient light region; , wherein the preprocessing unit extracts pixels having luminance values within a predetermined range based on a flame extraction threshold in which a predetermined range is set as a luminance value suitable for identifying whether or not it is a flame. are collected to create a brightness histogram, a probability distribution for the brightness histogram is created, and brightness values having a probability of a certain value or more in the probability distribution are specified as ambient light brightness values.
Further, the flame detection device according to the present invention is a flame detection device that detects the occurrence of flames by performing image processing on an image captured by a camera, wherein the image captured by the camera includes: Specifying an ambient light region by extracting pixels having a preset ambient light luminance value, and calculating a feature amount for identifying whether or not a flame is generated in the specified ambient light region. The flame detection unit calculates the standard deviation of the horizontal coordinate of each pixel constituting the ambient light area as the spread feature value corresponding to the feature value, and determines the spread feature value. When the amount becomes equal to or greater than a preset horizontal width threshold value for judging flame, a first judging process for judging that flame has occurred is executed.

According to the present invention, the ambient light region illuminated by the flame is specified without directly detecting the flame, and based on the feature amount calculated from the ambient light region, false detection is suppressed, and the original It has a configuration for detecting a flame, which is a detection target. As a result, it is possible to obtain a flame detection device that can detect flames with high precision using image processing, especially in a dark environment where it is difficult to detect smoke.

1 is a configuration diagram of a flame detection device according to Embodiment 1 of the present invention; FIG. FIG. 4 is a diagram showing a luminance histogram and a luminance probability distribution calculated by the ambient light region extracting section in Embodiment 1 of the present invention; FIG. 4 is a diagram comparing an image of light emitted by a flame under a low-illuminance environment and an image of light emitted by a light under a low-illuminance environment in Embodiment 1 of the present invention. In the first embodiment of the present invention, an image in which a peripheral light region is specified for an image of light emitted by a flame under a low-illuminance environment, and a peripheral light region for an image of light emitted by a light under a low-illuminance environment are identified. It is the figure which contrasted with the specified image. In Embodiment 1 of the present invention, an image obtained by calculating the time-series difference for an image of light emitted by a flame under a low-illuminance environment, and an image taken when the shutter is switched from the closed state to the open state under the low-illuminance environment. FIG. 10 is a diagram comparing an image obtained by calculating a time-series difference with respect to an image that has been obtained.

BEST MODE FOR CARRYING OUT THE INVENTION Preferred embodiments of the flame detection device of the present invention will be described below with reference to the drawings. A technical feature of the present invention is to specify an ambient light region illuminated by a flame without directly detecting the flame, and to perform flame detection with high accuracy based on the feature amount calculated from the ambient light region. It is something to do.

Embodiment 1.
FIG. 1 is a configuration diagram of a flame detection device according to Embodiment 1 of the present invention. The flame detection device according to Embodiment 1 includes an image memory 10 , a preprocessing section 20 and a flame detection section 30 . The image memory 10 is configured as an image memory for a plurality of frames so that images captured by the camera 1 can be stored as time-series data for a predetermined period in the past.

The preprocessing section 20 is configured including a peripheral light region extraction section 21 . Based on the image of the flame stored in the image memory 10 or the imaging data of the flame collected in advance, the preprocessing unit 20 defines the surrounding bright area illuminated by the generation of the flame as the ambient light area. It has the function of identifying

The flame detection unit 30 includes a spread feature amount calculation unit 31 , a fluctuation feature amount calculation unit 32 , and a flame determination unit 33 . The flame detection unit 30 uses the ambient light region specified by the preprocessing unit 20 to calculate the spread feature amount and the fluctuation feature amount for detecting the presence or absence of flame generation, and detects flame generation based on the calculation results. It has a function to judge whether or not

Here, the general steps of the detection principle of the present invention will be described.
First, the preprocessing unit 20 uses imaging data obtained by imaging a flame to be detected, and identifies an ambient light luminance value for identifying an ambient light region illuminated by the generation of the flame by prior learning (step 1). In this case, it is most desirable to use data captured by the camera 1 when a fire breaks out in an actual monitoring environment. However, if the ambient light luminance value can be specified, it is also possible to use imaging data obtained under another environment.

Next, when actually monitoring, the preprocessing unit 20 extracts a pixel group having the ambient light luminance value specified by prior learning from the image acquired from the camera 1 in a dark environment. to identify the ambient light region (step 2).

Next, the spread feature amount calculation unit 31 calculates the spread feature amount based on the specified ambient light region, and the flame determination unit 33 determines the flame-likeness based on the spread feature amount to determine fire. (step 3).

Furthermore, the fluctuation feature amount calculation unit 32 calculates the fluctuation feature amount based on the specified ambient light region, and the flame determination unit 33 determines the flame-likeness based on the fluctuation feature amount to determine fire. (Step 4).

Accordingly, a specific flame detection method of the present invention will be described along these steps.

<Step 1: Specifying Ambient Light Luminance Value by Pre-learning>
The ambient light region extracting unit 21 in the preprocessing unit 20 obtains the ambient light luminance value that should be possessed by the ambient light pixels forming the ambient light region by prior learning. Specifically, the ambient light region extracting unit 21 first collects pixels that satisfy the luminance value condition of the following formula (1) from video data obtained by photographing flames.
T/3<luminance value<T (1)

Here, T is a flame extraction threshold, which can be obtained, for example, from the dynamic range of an image in which flames are generated, using the following equation (2).
(max−min)×0.8+min (2)
max: maximum brightness value min: minimum brightness value

The above equation (1) is an example, and T/3 on the left side indicating the lower limit is appropriately set as a value smaller than T according to the characteristics of the environment to be monitored, the characteristics of the flame, etc. be able to. Also, T on the right-hand side indicating the upper limit can be appropriately set as a value other than T according to the characteristics of the environment to be monitored, the characteristics of the flame, and the like.

However, as a result of verification using a sample image of a flame, it was confirmed that an appropriate ambient light region can be extracted according to the luminance value condition of the above formula (1). One of the features of the present invention is that the ambient light region is specified by setting, as the luminance value condition, a region including a range lower than the threshold value T for directly detecting the flame itself.

Then, the ambient light region extracting unit 21 creates a luminance histogram by counting the number of pixels for each luminance with respect to each of the extracted pixels. Furthermore, the ambient light region extraction unit 21 creates a luminance probability distribution from the luminance histogram. FIG. 2 is a diagram showing a luminance histogram and a luminance probability distribution calculated by the ambient light region extraction section 21 in Embodiment 1 of the present invention.

Note that the ambient light region extraction unit 21 can apply kernel density estimation, as an example, when creating the luminance probability distribution from the luminance histogram.

Next, the ambient light region extracting unit 21 uses the created luminance probability distribution as learning data, and identifies luminance values having a probability of equal to or greater than a certain value as ambient light luminance values. An appropriate value is set for this constant value according to the characteristics of the environment to be monitored, the characteristics of the flame, and the like. Through such a series of pre-learning, the pre-processing unit 20 can specify in advance the ambient light luminance values of the ambient light pixels forming the ambient light region.

<Step 2: Processing for Extracting Ambient Light Area During Monitoring>
When performing actual monitoring, the ambient light region extracting unit 21 extracts, as ambient light pixels, pixels having previously specified ambient light brightness values from one image captured by the monitoring camera 1. . As a result, the ambient light region extraction unit 21 can easily identify the ambient light region in the image acquired during monitoring.

In the preprocessing unit 20, only the process of specifying the ambient light luminance value by pre-learning in step 1 is performed. Therefore, it is also possible to employ a configuration that is implemented on the flame detection section 30 side.

<Step 3: Flame Detection Processing Based on Spread Feature Amount>
When detecting flames in a dark environment, that is, in a low illumination environment, lighting fixtures such as headlights are a possible source of false alarms. FIG. 3 shows an image of light emitted by a flame under a low-illuminance environment (FIG. 3A) and an image of light emitted by a light under a low-illuminance environment (FIG. 3A) in Embodiment 1 of the present invention. 3(b)).

As is clear from the comparison between Figs. 3(a) and 3(b), the light emitted by the flame, which is the detection target, has no directivity, and the light emitted by the lighting equipment such as the light, which is the source of the false alarm, has no directivity. , it can be seen that there is directivity.

Depending on the presence or absence of directivity, there is a difference in how ambient light spreads. Therefore, in the present invention, the feature amount indicating the spread of the ambient light is utilized to perform the flame detection determination. FIG. 4 shows an image (FIG. 4(a)) in which a peripheral light region is specified with respect to an image obtained by capturing light emitted by a flame under a low-illuminance environment, and a light under the low-illuminance environment, according to the first embodiment of the present invention. It is a figure which contrasted the image (FIG.4(b)) which specified the peripheral light area|region regarding the image which imaged the emitted light.

As is clear from the comparison of FIGS. 4(a) and 4(b), comparing the respective ambient light regions, flames tend to spread laterally, but false alarm sources such as lights tend to spread. tend to spread vertically. Therefore, the spread feature amount calculation unit 31 calculates, as the spread feature amount, the standard deviation, which is an index of the lateral variation, with respect to the horizontal axis coordinates of the ambient light pixels forming the ambient light region.

In the first embodiment, the case where the standard deviation is used as the spread feature amount is exemplified. be.

Then, when the spread feature amount is equal to or greater than a preset width threshold, the spread feature amount calculation unit 31 determines that it is a flame. In this way, the spread feature amount calculation unit 31 can distinguish between the states of FIGS. It is possible to realize a flame detection device capable of suppressing flames and detecting a flame to be detected with high accuracy.

It should be noted that the flame detection processing based on the spread feature amount in step 3 is based on the processing for one image. However, it is also possible to finally determine the presence or absence of flames by taking into consideration the results of spread feature values obtained individually for each of the time-series images.

<Step 4: Flame Detection Processing Based on Fluctuation Feature Amount>
False alarm sources in a dark environment include not only lighting fixtures such as light lamps, but also, for example, the case where a shutter provided in a monitored space is changed from a closed state to an open state. Switching the shutter from the closed state to the open state may cause a sudden change in brightness in most of the surveillance area imaged by the camera 1 .

When such a phenomenon occurs, the possibility of erroneous determination increases with the determination of only the spread feature amount of the ambient light described in step 3. Therefore, the fluctuation feature amount calculator 32 calculates the fluctuation feature amount by paying attention to the time-series change of the luminance value in order to capture the fluctuation peculiar to the flame.

Specifically, the fluctuation feature amount calculation unit 32 calculates the time-series difference within a certain period of time for ambient light pixels in the ambient light region, and counts the number of times the difference value is equal to or greater than a preset fluctuation determination value. count each. Here, the time-series difference is a difference value for each pixel between two images that are time-sequentially adjacent among a plurality of images captured in time-series by the camera 1 .

Further, the fluctuation feature amount calculation unit 32 obtains the number of pixels indicating the number of times the count value is equal to or greater than the preset count value for flame determination, and furthermore, the calculated number of pixels is the total number of pixels constituting the ambient light region. A ratio to the number is calculated as a fluctuation feature amount.

FIG. 5 shows an image ( FIG. 5A ) obtained by calculating the time-series difference with respect to the image of the light emitted by the flame under the low-illuminance environment in the first embodiment of the present invention, and the It is a figure which contrasted with the image (54(b)) which calculated the time-series difference regarding the image which imaged when it switched from the closed state to the open state.

In addition, in FIG. 5, the background difference is shown together with the time-series difference. Here, the background difference is the difference value for each pixel between the background image captured by the camera 1 when no flame is generated and each of the multiple images captured in time series by the camera 1. That is.

In the case of flames, many time-series differences are extracted near the edge of the ambient light region. As a result, the ratio of the number of pixels indicating the number of times equal to or greater than the count value for flame determination to the total number of pixels forming the ambient light region, that is, the value of the fluctuation feature amount increases.

On the other hand, when the shutter is opened and closed, the time-series difference is extracted less, and the background difference area and the time-series difference area overlap. As a result, the ratio of the number of pixels indicating the number of times equal to or greater than the count value for flame determination to the total number of pixels forming the ambient light region, that is, the value of the fluctuation feature amount is extremely small.

If there is a revolving lamp, a flashlight, or the like that serves as a false alarm source, the area from which only the time-series difference is extracted becomes large because the difference between light and dark is large. As a result, the ratio of the number of pixels indicating the number of times equal to or greater than the count value for flame determination to the total number of pixels forming the ambient light region, that is, the value of the fluctuation feature amount becomes extremely large.

Therefore, when the fluctuation feature amount falls within a desired range, the fluctuation feature amount calculator 32 determines that a flame has occurred, thereby eliminating the false alarm source due to opening and closing of the shutter, the revolving light, the flash light, and the like. It is possible to realize a flame detection device that can prevent the flame from being determined and can detect the flame to be detected with higher accuracy.

As described above, according to Embodiment 1, flame detection is performed by focusing on the ambient light region illuminated by the flame without directly detecting the flame. Specifically, a configuration is provided in which the ambient light luminance value for specifying the ambient light region can be calculated by pre-learning. Furthermore, during actual monitoring, the configuration is such that the ambient light region is specified by extracting the pixels having the ambient light luminance value, and the feature amount in the ambient light region is calculated to detect the flame. .

As the feature quantity, a spread feature quantity that indicates behavior peculiar to flame is used. Further, in order to remove the cause of false detection when using the spread feature, the fluctuation feature is also used as a further feature. As a result, it is possible to realize a flame detection device that can detect flames with high precision using image processing in a dark environment where it is difficult to detect smoke.

Although the presence or absence of flame generation is determined using the ambient light region, which is a feature of the present invention, flame determination may be performed by combining, for example, an algorithm capable of determining flame.

1 camera (surveillance camera), 10 image memory, 20 preprocessing unit, 21 ambient light region extraction unit, 30 flame detection unit, 31 feature amount calculation unit, 32 feature amount calculation unit, 33 flame determination unit.

Claims (3)

A flame detection device that detects the occurrence of a flame by performing image processing on an image captured by a camera,
identifying the ambient light region by extracting pixels having an ambient light luminance value preset to identify the ambient light region illuminated by the flame generation from the image captured by the camera; In the specified ambient light region, a feature quantity indicating a behavior peculiar to flame is calculated in order to identify whether or not a flame is generated, and flame generation is performed from the magnitude of the feature quantity calculated in the ambient light region. A flame detection device comprising a flame detection part for determining the presence or absence of a flame. A flame detection device that detects the occurrence of a flame by performing image processing on an image captured by a camera,
a preprocessing unit that sets in advance an ambient light luminance value for specifying an ambient light region illuminated by the generation of the flame, using imaging data obtained by imaging the flame to be detected;
Identifying the ambient light region by extracting pixels having the ambient light luminance value from the image captured by the camera, and identifying whether or not a flame is generated in the identified ambient light region. A flame detection unit that determines whether or not a flame is generated by calculating a feature amount for
with
The preprocessing unit collects pixels having luminance values within a predetermined range based on a flame extraction threshold set with a predetermined range of luminance values suitable for identifying whether or not there is a flame. creating a brightness histogram by doing so, creating a probability distribution for the brightness histogram, and specifying a brightness value having a probability of a certain value or more in the probability distribution as the ambient light brightness value.
Flame detector. A flame detection device that detects the occurrence of a flame by performing image processing on an image captured by a camera,
A peripheral light region is specified by extracting pixels having a preset ambient light luminance value from the image captured by the camera, and whether or not a flame is generated in the specified peripheral light region. Equipped with a flame detection unit that determines the presence or absence of a flame by calculating a feature value for identifying whether
The flame detection unit calculates a standard deviation of abscissa coordinates of each pixel constituting the ambient light area as a spread feature amount corresponding to the feature amount, and calculates a width for flame determination with which the spread feature amount is set in advance. Execute a first judgment process for judging that a flame has occurred when the threshold value is exceeded.
Flame detector.

Images