Smoke detection device

Abstract

To provide a smoke detection device that detects smoke generated in a monitoring target region, by using a feature amount focusing on a brightness fluctuation which is unique to smoke such that erroneous recognition of smoke can be suppressed.SOLUTION: A smoke detection device according to the present disclosure detects occurrence of smoke by performing image processing on a smoke candidate region in a monitoring target image taken by a monitoring camera. The smoke detection device includes a feature amount extraction unit that calculates a co-occurrence matrix about the monitoring target image taken by the monitoring camera, by using the luminance values of pixels in the smoke candidate region, and calculates a texture feature amount on the basis of the calculated co-occurrence matrix, and a smoke occurrence detection unit that determines whether or not smoke has occurred in the smoke candidate region in the monitoring target image on the basis of the texture feature amount calculated by the feature amount extraction unit.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a technique for detecting the occurrence of smoke by performing image processing on an image captured by a surveillance camera.

Background Art

[0002] There is a conventional device that can extract only the smoke region from the surveillance target region imaged by a surveillance camera by applying image processing technology (see, for example, Patent Document 1). Such a conventional device performs image processing on an image captured by a surveillance camera to extract feature amounts caused by smoke, and detects a state in which smoke has occurred as an abnormal state different from a steady state.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, the prior art has the following problems. In a conventional smoke detection method that applies image processing technology, on a multi-dimensional feature space based on a plurality of feature amounts, "smoke" and "non-smoke" are discriminated based on a discrimination boundary learned from teacher data, thereby improving the smoke detection accuracy.

[0005] However, as the types of feature amounts used in the conventional smoke detection method, those for evaluating the fluctuation of the luminance peculiar to smoke are insufficient. For example, there has been a problem that luminance changes due to lighting or sunlight, or luminance changes due to a person passing by are easily misrecognized as smoke. Therefore, it is desired to suppress the misrecognition of smoke by using a feature amount focusing on the fluctuation of the luminance peculiar to smoke.

[0006] The present disclosure has been made to solve the above-described problems, and an object thereof is to obtain a smoke detection device that can detect smoke generated in a monitoring target area using a feature amount focused on the fluctuation of the luminance peculiar to smoke and suppress misrecognition of smoke.

Means for Solving the Problems

[0007] The smoke detection device according to the present disclosure is a smoke detection device that detects the occurrence of smoke by performing image processing on a smoke candidate area in a monitoring target image captured by a monitoring camera, and for the monitoring target image captured by the monitoring camera, a co-occurrence matrix is calculated using the luminance values of each pixel in the smoke candidate area, and a feature amount extraction unit that calculates a texture feature amount based on the calculated co-occurrence matrix, and a smoke occurrence detection unit that determines whether or not smoke has occurred in the smoke candidate area in the monitoring target image based on the texture feature amount calculated by the feature amount extraction unit. A smoke detection device further includes an image memory that stores, as time-series monitoring target images, monitoring target images captured by a monitoring camera at predetermined sampling intervals. The feature extraction unit uses the luminance values of each pixel in the smoke candidate regions within two or more monitoring target images that are adjacent in time series and included in the time-series monitoring target images, and while considering the adjacency relationship between a target pixel and adjacent pixels with respect to the target pixel in one image, further considers the adjacency relationship between pixels at the same position as the target pixel in the frames before and after in time series, and calculates a co-occurrence matrix. is.

[0008] Further, the smoke detection device according to the present disclosure is a smoke detection device that detects the occurrence of smoke by performing image processing on a smoke candidate area in a monitoring target image captured by a monitoring camera, and at the time of monitoring, an image memory that stores the monitoring target image captured by the monitoring camera at predetermined sampling periods as a time-series monitoring target image, and for each pixel, an EMA image composed of pixels that is the result of performing exponential moving average using the time-series monitoring target image is sequentially calculated, and for each pixel, the luminance value of the monitoring target image captured when the EMA image is calculated or at the next sampling period is compared with the luminance value of the EMA image, and a preprocessing unit that generates a binary image based on the magnitude, and a smoke occurrence detection unit that determines whether or not smoke has occurred in the smoke candidate area in the monitoring target image based on the binary image generated in the smoke candidate area by the preprocessing unit.

Effects of the Invention

[0009] According to the present disclosure, it is possible to obtain a smoke detection device that can detect smoke generated in a monitoring target area by using a feature amount focused on the fluctuation of the luminance peculiar to smoke, and suppress misrecognition of smoke.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Embodiments for Carrying Out the Invention

[0011] Hereinafter, a preferred embodiment of the smoke detection device of the present disclosure will be described with reference to the drawings. The present disclosure is characterized in that a feature amount focusing on the fluctuation of luminance peculiar to smoke is calculated using GLCM (Gray Level Co-occurrence Matrix), EMA (Exponential Moving Average), or a combination of GLCM and EMA.

[0012] Therefore, first, a smoke detection method based on a combination of GLCM and EMA will be described, and then a smoke detection method based on GLCM alone or EMA alone will be described.

[0013] Embodiment 1. FIG. 1 is a configuration diagram of the smoke detection device according to Embodiment 1 of the present disclosure. The smoke detection device according to Embodiment 1 includes an image memory 10 and a detection processing unit 20. The image memory 10 is configured as an image memory for a plurality of frames so that it can store surveillance target images captured by a surveillance camera 1 corresponding to an image pickup device at predetermined sampling periods for a past fixed period as time-series surveillance target images. An image for one frame is composed of a plurality of pixels.

[0014] Further, the detection processing unit 20 includes a preprocessing unit 21, a feature amount extraction unit 22, and a smoke generation detection unit 23, and detects the generation of smoke by performing image processing on a smoke candidate region in the surveillance target image captured by the surveillance camera 1. First, an outline of a series of processes of the detection processing unit 20 will be described.

[0015] The preprocessing unit 21 sequentially calculates an EMA image composed of pixels that are the results of performing exponential moving average using the time-series monitored images. Further, for each pixel, the preprocessing unit 21 compares the luminance value of the monitored image captured at the sampling period when the EMA image was calculated or the luminance value of the monitored image captured at the next sampling period with the luminance value of the EMA image, and generates a binary image based on the magnitude.

[0016] The feature extraction unit 22 calculates a co-occurrence matrix using the binary values of each pixel in the binary images sequentially generated within the smoke candidate region by the preprocessing unit 21. Further, the feature extraction unit 22 calculates a texture feature amount within the smoke candidate region by performing statistical processing for identifying the presence or absence of smoke on the calculated co-occurrence matrix.

[0017] The smoke generation detection unit 23 determines whether smoke has occurred in the smoke candidate region within the monitored image based on the texture feature amount calculated by the feature extraction unit 22.

[0018] Next, as a pre-step for obtaining a texture feature amount focusing on the fluctuation of luminance peculiar to smoke, the arithmetic processing for obtaining an EMA difference image and the arithmetic processing for obtaining a GLCM will be described in detail respectively.

[0019] <Regarding the EMA difference image> As preprocessing for calculating a feature amount with time-series information, the preprocessing unit 21 performs the calculation process of the EMA difference image. First, the method for creating the EMA image will be described, and then the method for creating the EMA difference image will be described. (1) Method for creating the EMA image The preprocessing unit 21 calculates an exponential moving average (hereinafter referred to as EMA) in the time-series direction for each luminance in each image of the time-series monitored images corresponding to a series of time-series images. In the present disclosure, an image with each value for which the EMA is calculated as the luminance value is defined as an EMA image.

[0020] The EMA image is an image that is continuously updated sequentially over time. If there is no change in the image to be monitored, there will be no change in the EMA image either. On the other hand, when a moving object moves across the field of view, the movement trajectory of the moving object remains as an afterimage on the EMA image. Also, when the illuminance of the entire screen changes gradually, the EMA image adapts sequentially and the luminance changes gradually.

[0021] The calculation formula of EMA is shown as the following formula (1) for each pixel. S t =α×Y t +(1-α)×S t-1 (1) However, S t : The exponential moving average at time t, which is the current sampling period S t-1 : The exponential moving average at time t-1, which is the previous sampling period Y t : The luminance value of the image to be monitored captured by the monitoring camera at time t α: The smoothing coefficient set as a value between 0 and 1

[0022] By appropriately setting the value of the smoothing coefficient α according to the environment of the object to be monitored, the speed of luminance change due to the influence of disturbing elements, etc., in the EMA image that is continuously updated sequentially over time, the degree of weighting of the value of EMA at the previous sampling time can be changed when obtaining the value of EMA at the current sampling time.

[0023] (2) Creation of the EMA difference image Next, the preprocessing unit 21 compares, for each pixel, the first luminance value, which is the luminance value of the image to be monitored captured at the sampling period or the next sampling period when calculating the EMA image, with the second luminance value, which is the luminance value of the EMA image. Then, the preprocessing unit 21 creates the EMA difference image according to the magnitude relationship between the first luminance value and the second luminance value.

[0024] As an example, for each pixel, when the first luminance value is higher than the second luminance value, the preprocessing unit 21 creates an EMA difference image by performing a binarization process in which it sets the pixel to white, and when the first luminance value is less than or equal to the second luminance value, it sets the pixel to black. Here, if each of the first luminance value and the second luminance value is set to 256 gradations from 0 to 255, white corresponds to 255 and black corresponds to 0.

[0025] Note that here, for the sake of simplicity of explanation, the EMA difference image is a binary image, but it is also possible to create the EMA difference image as a grayscale image with three or more values in which the difference between the first luminance value and the second luminance value is divided into multiple levels.

[0026] The features of the EMA difference image created in this way will be described. FIG. 2 is a diagram for explaining the differences in the EMA difference image according to the type of moving object in Embodiment 1 of the present disclosure. FIG. 2(a) shows an EMA difference image created for a scene with no movement in the monitoring target area, FIG. 2(b) shows an EMA difference image created when a smoke generation area is included in the monitoring target area, and FIG. 2(c) shows an EMA difference image created when a person, who is a moving object, is included in the monitoring target area.

[0027] The EMA difference image has a different appearance due to factors such as the way the subject moves in the monitoring target area and the periodicity of the surface texture change. When no change occurs within the screen, as shown in FIG. 2(a), random noise due to scanning lines or fine pixel value fluctuations due to AGC (Automatic Gain Control) appears over the entire screen on the EMA difference image.

[0028] Also, when the luminance changes within the monitoring target area due to a moving object such as a person, as shown in FIG. 2(c), a pattern corresponding to the brightness level of the object and an afterimage on the movement trajectory appear within the moving object area on the EMA difference image.

[0029] Also, when the brightness of the monitoring target area changes significantly due to changes in the illumination of the room or external light, on the EMA difference image, while the difference between the first luminance and the second luminance immediately after the change is large, objects within the screen appear as silhouettes according to the shading. After that, when the EMA image is applied to the screen after the illumination change, as the difference between the first luminance value and the second luminance value decreases, the EMA difference image returns to random noise.

[0030] On the other hand, when the smoke generation area, which is the original detection target, appears within the monitoring target area, as shown in Fig. 2(b), within the smoke generation area on the EMA difference image, a specific shading pattern caused by the flow and periodicity of the smoke appears.

[0031] <Regarding GLCM> In the EMA difference image calculated as described above, different textures appear depending on the presence or absence of moving objects within the screen or the difference in the moving directions of the moving objects. Therefore, by quantitatively identifying these differences using texture feature quantities, it becomes possible to determine whether there is an area where smoke has occurred. Thus, a method for creating a GLCM, which is used to extract texture feature quantities focusing on the flickering of the luminance peculiar to smoke, will be described.

[0032] Regarding GLCM, there are two cases to consider: obtaining the GLCM from a single image and obtaining the GLCM from a plurality of images arranged in time series. In the present disclosure, the former will be referred to as a two-dimensional GLCM, and the latter will be referred to as a three-dimensional GLCM. The specific method for creating the GLCM performed by the feature quantity extraction unit 22 and the calculation of texture feature quantities suitable for smoke detection based on the GLCM will be described below separately for two dimensions and three dimensions.

[0033] (1) Creation of two-dimensional GLCM FIG. 3 is an image diagram for explaining a method of creating a GLCM from a single image input by the feature extraction unit 22 in Embodiment 1 of the present disclosure. FIG. 3(a) shows a binary image of 5×5 pixels as an example of the input image. Further, FIG. 3(b) shows a matrix A corresponding to the GLCM created by the feature extraction unit 22 based on the input image.

[0034] The GLCM corresponds to a matrix A that quantitatively represents the degree of variation in luminance based on the positional relationship between pixels. When creating a two-dimensional GLCM, the luminance value of the target pixel is taken for the row, and the luminance value around the target pixel is taken for the column. That is, when each element of the matrix A shown in FIG. 3(b) is denoted as A i、j when i represents the density value of the target pixel and j represents the density value of the neighboring pixel.

[0035] The feature extraction unit 22 calculates the matrix A by obtaining the cumulative addition result for each element of A i、j Specifically, for A 0、0 A 0、1 A 1、0 A 1、1 each element is incremented when the following conditions are satisfied. A 0、0 : It is incremented by 1 when the luminance value of the target pixel is 0 and the luminance value of the neighboring pixel is 0. A 0、1 : It is incremented by 1 when the luminance value of the target pixel is 0 and the luminance value of the neighboring pixel is 1. A 1、0 : It is incremented by 1 when the luminance value of the target pixel is 1 and the luminance value of the neighboring pixel is 0. A 1、1 : It is incremented by 1 when the luminance value of the target pixel is 1 and the luminance value of the neighboring pixel is 1.

[0036] For example, the case where the pixel enclosed by the double line in the binary input image shown in FIG. 3(a) is taken as the target pixel, and the pixel on the right and the pixel on the left are taken as the neighboring pixels respectively will be described. In this case, the density value of the target pixel is "0", the density value of the neighboring pixel on the right is "1", and the density value of the neighboring pixel on the lower side is "0". Therefore, among the four elements of the matrix A, A0、0 and A 0、1 1 will be added to it.

[0037] For all the pixels within the block, by sequentially repeating the same process while moving the pixel of interest, the feature extraction unit 22 can obtain the cumulative addition result for each element. Further, the feature extraction unit 22 can obtain the normalized matrix A by dividing the cumulative addition result for each element in matrix A by the sum of the cumulative addition results of all elements.

[0038] By performing normalization, each element A i、j will represent the probability that the luminance values of the pixels are adjacent as (0, 0), (0, 1), (1, 0), (1, 1).

[0039] In the specific example described with reference to FIG. 3, since the luminance values of the EMA difference image are binary values of 0 and 1, the matrix A corresponding to the GLCM is a 2×2 matrix. However, this is just an example. For example, when the EMA difference image has four values of 0, 1, 2, and 3, the matrix A corresponding to the GLCM will be a 4×4 matrix.

[0040] Also, in the specific example described with reference to FIG. 3, the two pixels on the right and below are used as the adjacent image, but it is also possible to use pixels other than the right and below as the adjacent pixels. For example, it is also possible to use the two pixels on the left and above as the adjacent image. It is also possible to use diagonal pixels as the adjacent pixels.

[0041] The feature extraction unit 22 can calculate the matrix A corresponding to the two-dimensional GLCM by executing the above-described arithmetic processing on the input image.

[0042] (2) Creation of three-dimensional GLCM The above-mentioned two-dimensional GLCM will be calculated using a single differential image created at regular intervals as the input image according to the period for calculating the feature amount. In this case, there is a problem that the feature amount is not stable depending on the bias of the random noise. Alternatively, there is also a problem that a sufficient number of feature amounts cannot be obtained because feature amount extraction is not performed in most frames where the differential image is not created.

[0043] Therefore, in order to improve the smoke detection accuracy, it is desirable to create a GLCM for extracting statistically stable feature amounts. Therefore, in order to solve these problems, instead of the two-dimensional GLCM that quantitatively expresses the degree of variation in luminance based on the positional relationship between pixels in a single input image, it is conceivable to create a three-dimensional GLCM considering the adjacent relationship in the front and back directions in time series.

[0044] As an example, it is conceivable to accumulate 32 frames of the EMA differential image and perform three-dimensional GLCM creation and texture analysis for each cubic voxel composed of 32 pixels × 32 pixels × 32 frames.

[0045] FIG. 4 is an image diagram for explaining a method of creating a GLCM from a plurality of sequentially arranged image inputs by the feature amount extraction unit 22 in Embodiment 1 of the present disclosure. FIG. 4(a) shows a cubic voxel as a time-series binary image in which 32 frames of the EMA differential image are accumulated.

[0046] Further, FIG. 4(b) illustrates a case where, in addition to the input image at the sampling time t, the input image at the previous sampling time t-1 in time series and the input image at the next sampling time t+1 in time series are considered, and a three-dimensional GLCM is obtained using a total of three input images.

[0047] In creating the two-dimensional GLCM described with reference to FIG. 3 above, the adjacency relationship between a target pixel and two adjacent pixels to the right and below in a single image was considered. In contrast, as shown in FIG. 4, in creating the three-dimensional GLCM, the adjacency relationship is further considered for pixels at the same position as the target pixel in the frames before and after in time series, and the degree of variation in luminance based on the positional relationship between pixels is quantitatively evaluated.

[0048] The feature extraction unit 22 creates a matrix A using four adjacent pixels by considering, for the target pixel in the input image at time t, two adjacent pixels on the right and below (in the x and y directions) in the input image at time t, as well as two adjacent pixels at the same position as the target pixel in the input image at time t in the input images at times t - 1 and t + 1.

[0049] In this way, the feature extraction unit 22 can create a three-dimensional GLCM using four adjacent pixels based on three input images arranged in time series. As a result, compared to the case of creating a two-dimensional GLCM, it is possible to create a GLCM for calculating more stable feature amounts by considering the monitored target images in more sampling periods.

[0050] (3) Extraction process of texture feature amounts based on GLCM Next, after the matrix A is created as a two-dimensional GLCM or a three-dimensional GLCM, a specific statistical processing example for extracting texture feature amounts used for smoke detection determination from the A matrix by the feature extraction unit 22 will be described.

[0051] In the present disclosure, as a specific example, the case of extracting three texture feature amounts, namely entropy, non-uniformity, and energy, by performing statistical processing on the GLCM as texture feature amounts will be described below. Note that other texture feature amounts can also be used when extracting texture feature amounts.

[0052] (3-1) Regarding entropy When the GLCM is set as P(A), the entropy can be calculated by the following formula (2).

[0053]

Equation

[0054] Taking the difference image shown in Figure 2 above as an example, the entropy becomes a higher value as the arrangement of white and black in the difference image is more random, like the "static scene" in Figure 2(a). On the other hand, when the arrangement of white and black is regular, or when white or black is uniform, such as the afterimage part behind the person in the "moving object (person)" in Figure 2(c), the entropy becomes a lower value.

[0055] Figure 5 is an image visualizing the high and low levels of entropy for each of the smoke scene and the scene where people move in Embodiment 1 of the present disclosure. Specifically, Figure 5(a) is an image visualizing the high and low levels of entropy in the smoke scene, and Figure 5(b) is an image visualizing the high and low levels of entropy in the scene where people move.

[0056] Furthermore, Figure 6 is a diagram showing the probability density distribution of entropy in the smoke region and the non-smoke region in Embodiment 1 of the present disclosure. Specifically, Figure 6(a) is the probability density distribution of entropy obtained for the 2D GLCM, and Figure 6(b) is the probability density distribution of entropy obtained for the 3D GLCM.

[0057] Also, in each of Figure 6(a) and Figure 6(b), the probability density distribution of entropy within the smoke region is shown as "Distribution 1", and the probability density distribution of entropy within the non-smoke region is shown as "Distribution 2".

[0058] As shown in FIG. 6, distribution 1, which is the probability density distribution of entropy in the smoke region, tends to have a lower entropy. For this reason, in FIGS. 6(a) and 6(b), it can be seen that distribution 1 has a lower peak and the peak is shifted to the left compared to distribution 2. Also, from the comparison between FIGS. 6(a) and 6(b), it can be understood that the difference between distribution 1 and distribution 2 is more clearly shown in the texture feature amount calculated based on the three-dimensional GLCM.

[0059] Therefore, the smoke generation detection unit 23 can quantitatively determine whether smoke has occurred in the monitoring target region from the probability density distribution of the texture feature amount regarding entropy calculated by the feature amount extraction unit 22.

[0060] (3-2) Regarding non-uniformity When the GLCM is set as P(A), the non-uniformity can be calculated by the following formula (3).

[0061] [Number]

[0062] When the probability that white or black pixels are continuous in the difference image is low, the non-uniformity becomes a high value. Therefore, the value of the non-uniformity tends to be low inside the smoke and high in the region where random noise is generated.

[0063] FIG. 7 is an image visualizing the high and low levels of non-uniformity obtained for each of the smoke scene and the scene where people move in Embodiment 1 of the present disclosure. Specifically, FIG. 7(a) is an image visualizing the high and low levels of non-uniformity in the smoke scene, and FIG. 7(b) is an image visualizing the high and low levels of non-uniformity in the scene where people move.

[0064] Furthermore, FIG. 8 is a diagram showing the probability density distribution of non-uniformity in the smoke region and the non-smoke region in Embodiment 1 of the present disclosure. Specifically, FIG. 8(a) is the probability density distribution of non-uniformity obtained for the two-dimensional GLCM, and FIG. 8(b) is the probability density distribution of non-uniformity obtained for the three-dimensional GLCM.

[0065] Also, in each of FIGS. 8(a) and 8(b), the probability density distribution of non-uniformity within the smoke region is shown as "Distribution 1", and the probability density distribution of non-uniformity within the non-smoke region is shown as "Distribution 2".

[0066] As shown in FIG. 8, Distribution 1, which is the probability density distribution of non-uniformity within the smoke region, tends to have a smaller non-uniformity. Therefore, in each of FIGS. 8(a) and 8(b), it can be seen that the peak of Distribution 1 is shifted to the left compared to Distribution 2. Also, from the comparison between FIGS. 8(a) and 8(b), it can be understood that the difference between Distribution 1 and Distribution 2 is more clearly shown in the texture feature amount calculated based on the three-dimensional GLCM.

[0067] Therefore, the smoke generation detection unit 23 can quantitatively determine whether smoke has occurred in the monitoring target region from the probability density distribution of the texture feature amount regarding non-uniformity calculated by the feature amount extraction unit 22.

[0068] (3-2) Regarding Energy When the GLCM is set as P(A), the energy can be calculated by the following formula (4).

[0069]

Equation

[0070] When the difference image is uniform in white or black, or when a specific pattern is repeated, the energy becomes a high value. Therefore, the value of the energy tends to be high inside the smoke and low in the region where random noise is generated.

[0071] FIG. 9 is an image visualizing the levels of energy obtained for each of a smoke scene and a scene where people move in Embodiment 1 of the present disclosure. Specifically, FIG. 9(a) is an image visualizing the level of energy in the smoke scene, and FIG. 9(b) is an image visualizing the level of energy in the scene where people move.

[0072] Furthermore, FIG. 10 is a diagram showing the probability density distributions of energy in the smoke region and the non-smoke region in Embodiment 1 of the present disclosure. Specifically, FIG. 10(a) is the probability density distribution of energy obtained for the two-dimensional GLCM, and FIG. 10(b) is the probability density distribution of energy obtained for the three-dimensional GLCM.

[0073] Also, in each of FIGS. 10(a) and 10(b), the probability density distribution of non-uniformity within the smoke region is shown as "Distribution 1", and the probability density distribution of non-uniformity within the non-smoke region is shown as "Distribution 2".

[0074] As shown in FIG. 10, Distribution 1, which is the probability density distribution of energy within the smoke region, tends to have a larger energy. Therefore, in each of FIGS. 10(a) and 10(b), it can be seen that the peak of Distribution 1 is shifted to the right compared to Distribution 2. Also, from the comparison between FIGS. 10(a) and 10(b), it can be understood that the difference between Distribution 1 and Distribution 2 is more clearly shown in the texture feature amount calculated based on the three-dimensional GLCM.

[0075] Therefore, the smoke generation detection unit 23 can quantitatively determine whether smoke has occurred in the monitoring target region from the probability density distribution of the texture feature amount related to the energy calculated by the feature amount extraction unit 22.

[0076] In addition, in the first embodiment, in order to calculate the feature amount focusing on the fluctuation of the luminance peculiar to smoke, the GLCM is created using the EMA difference image as the input image, and statistical processing is performed to quantitatively determine whether smoke has occurred in the monitoring target area. However, the smoke detection device according to the present disclosure is not limited to such a configuration.

[0077] Therefore, a supplementary explanation will be given below for each of the configuration that does not use the GLCM using the EMA difference image and the configuration that uses the GLCM without using the EMA difference image.

[0078] <Configuration that does not use the GLCM using the EMA difference image> Such a configuration can be realized by configuring the detection processing unit 20 as the following preprocessing unit 21 and smoke generation detection unit 23.

[0079] In the image memory 10, during monitoring, the monitoring target images captured by the monitoring camera 1 at predetermined sampling periods are stored as time-series monitoring target images.

[0080] The preprocessing unit 21 sequentially calculates an EMA image composed of pixels that are the results of performing exponential moving average using the time-series monitoring target images for each pixel. Further, the preprocessing unit 21 compares the luminance value of the monitoring target image captured at the time when the EMA image is calculated or at the next sampling period with the luminance value of the EMA image for each pixel, and generates a binary image based on the magnitude.

[0081] Then, the smoke generation detection unit 23 determines whether smoke has occurred in the smoke candidate area in the monitoring target image based on the binary image generated in the smoke candidate area by the preprocessing unit 21.

[0082] That is, the detection processing unit 20 performs such arithmetic processing, and without obtaining the GLCM for the EMA difference image as shown in the previous Figure 2, by calculating the feature amount focusing on the luminance fluctuation peculiar to smoke from the EMA difference image, the state shown in Figure 2(b) can be specified, and the misrecognition of smoke can be suppressed.

[0083] <Configuration using GLCM without using EMA difference image> Such a configuration can be realized by configuring the detection processing unit 20 as the following feature amount extraction unit 22 and smoke generation detection unit 23.

[0084] The feature amount extraction unit 22 creates a co-occurrence matrix using the luminance values of each pixel in the smoke candidate region for the monitoring target image captured by the monitoring camera 1. Further, the feature amount extraction unit 22 calculates a texture feature amount based on the created co-occurrence matrix.

[0085] Specifically, instead of the EMA difference image, the feature amount extraction unit 22 uses the monitoring target image captured by the monitoring camera 1, converts an image with, for example, 256 gradations of luminance as the monitoring target image into a grayscale image with 2 or more and 256 gradations or less, and then creates a co-occurrence matrix.

[0086] Then, the smoke generation detection unit 23 determines whether smoke has occurred in the smoke candidate region in the monitoring target image based on the texture feature amount calculated by the feature amount extraction unit.

[0087] That is, the detection processing unit 20 performs such arithmetic processing, and by calculating the feature amount focusing on the luminance fluctuation peculiar to smoke even by creating the GLCM without using the EMA difference image, it can be determined whether smoke has occurred, and the misrecognition of smoke can be suppressed.

Explanation of symbols

[0088] 1 Monitoring camera, 10 Image memory, 20 Detection processing unit, 21 Preprocessing unit, 22 Feature amount extraction unit, 23 Smoke generation detection unit.

Claims

1. A smoke detection device that detects the occurrence of smoke by performing image processing on a smoke candidate region in a monitoring target image captured by a monitoring camera, a feature amount extraction unit that calculates a co-occurrence matrix using the luminance values of each pixel in the smoke candidate region for the monitoring target image captured by the monitoring camera, and calculates a texture feature amount based on the calculated co-occurrence matrix; a smoke occurrence detection unit that determines whether smoke has occurred in the smoke candidate region in the monitoring target image based on the texture feature amount calculated by the feature amount extraction unit A smoke detection device comprising: further comprising an image memory that stores the monitoring target image captured by the monitoring camera as a time-series monitoring target image at a predetermined sampling period; The feature amount extraction unit uses the luminance values of each pixel in the smoke candidate region in two or more monitoring target images that are adjacent in time series included in the time-series monitoring target image, and considers the adjacency relationship between a pixel of interest in one image and an adjacent pixel to the pixel of interest, and further considers the adjacency relationship between the pixel of interest and the pixel at the same position in the frames before and after in time series, and calculates the co-occurrence matrix Smoke detection device.

2. A smoke detection device that detects the occurrence of smoke by performing image processing on a smoke candidate region in a monitoring target image captured by a monitoring camera, an image memory that stores the monitoring target image captured by the monitoring camera as a time-series monitoring target image at a predetermined sampling period during monitoring; a preprocessing unit that sequentially calculates an EMA image composed of pixels that are the result of performing exponential moving average using the time-series monitoring target image for each pixel, and for each pixel, compares the luminance value of the monitoring target image captured when the EMA image is calculated or in the next sampling period with the luminance value of the EMA image, and generates a binary image based on the magnitude; a smoke occurrence detection unit that determines whether smoke has occurred in the smoke candidate region in the monitoring target image based on the binary image generated in the smoke candidate region by the preprocessing unit A smoke detection device comprising:

3. further comprising a feature amount extraction unit that calculates a co-occurrence matrix using the binary values of each pixel in the binary image sequentially generated in the smoke candidate region by the preprocessing unit, and calculates a texture feature amount based on the calculated co-occurrence matrix Instead of using the binary image, the smoke generation detection unit determines whether smoke has occurred in the smoke candidate region in the monitoring target image based on the texture feature amount calculated by the feature amount extraction unit. The smoke detection device according to claim 2.

4. The preprocessing unit stores the binary image sequentially generated for each sampling period as a time-series binary image in the image memory. The feature amount extraction unit calculates the co-occurrence matrix using the binary values of each pixel in the smoke candidate region of two or more binary images that are adjacent in time series in the time-series binary image. The smoke detection device according to claim 3.

Images