A method for video monitoring in coal mines

Abstract

This application relates to the technical field of image denoising, and specifically relates to a method for video monitoring in coal mines. The method includes: collecting video images in coal mines using an explosion-proof camera; calculating the spatial proximity between each pixel point in the search window and the current pixel point; calculating the dust influence degree and light influence degree of the similarity window of each pixel point, and further analyzing the similarity of the directional dispersion degree of the similarity windows of any two pixel points; and calculating the weight value of each pixel point in the search window for the current pixel point in combination with the spatial proximity, and performing weighted summation on the pixel values of all pixel points in the search window to obtain the pixel value of the current pixel point after denoising, thereby constructing a denoised image. The purpose of this application is to eliminate the adverse effects of dust, light, and distance on similarity calculation without losing image details and edge information, reduce the noise of the monitoring video images in coal mines, and make the monitoring process more accurate.

Description

Technical Field

[0001] This application relates to the technical field of image denoising, and specifically relates to a method for video monitoring in coal mines. Background Art

[0002] Coal occupies an important position in the development of our country. Therefore, the safe exploitation of coal mines is of great significance to the social and economic development of our country. With the rapid development of the computer industry, great progress has been made in using intelligent devices to monitor the exploitation in coal mines and the safety of employees' lives.

[0003] However, due to insufficient light and a large amount of dust underground, a large amount of noise is contained in the collected video image information, which has an adverse impact on subsequent image processing tasks, and the application effects of using artificial intelligence for target recognition and detection are not very ideal. Summary of the Invention

[0004] In order to solve the above technical problems, this application provides a method for video monitoring in coal mines to solve the existing problems.

[0005] A method for video monitoring in coal mines of this application adopts the following technical solutions:

[0006] An embodiment of this application provides a method for video monitoring in coal mines, and this method includes the following steps:

[0007] S1, collecting video images in coal mines by using an explosion-proof camera;

[0008] S2, defining a similarity window and a search window for each pixel point in the video image of the coal mine; for each pixel point in the search window of the current pixel point, using the Gaussian function to analyze the distance between each pixel point and the current pixel point, and calculating the spatial proximity of each pixel point in the search window to the current pixel point;

[0009] S3, obtaining the dark channel value of each pixel point in the image and constructing a dark channel image; dividing the order of all dark channel values in the similarity window of each pixel point in the dark channel image, and analyzing different sizes of connected regions and their occurrence times at each divided order, as well as the distance between the centroid of each connected region and the center of the similarity window, and constructing the dust influence degree of the similarity window;

[0010] S4, obtaining the illumination component of each pixel point in the image and constructing an illumination image; calculating the illumination influence degree of the similarity window of each pixel point in the illumination image according to the same calculation method as the dust influence degree of the similarity window of each pixel point in the dark channel image;

[0011] S5. Based on the dust influence degree and the light influence degree, determine the value of the central pixel point corresponding to the similar window position in the comprehensive distribution image; calculate the directional dispersion degree of the similar window by using the dispersion situation of the gradient amplitudes of the values within the similar window of each pixel point in the comprehensive distribution image within different directional ranges; and analyze the similarity of the directional dispersion degrees of the similar windows of any two pixel points in the comprehensive distribution image.

[0012] S6. Combine the spatial proximity between each pixel point in the search window and the current pixel point, and the similarity of the directional dispersion degrees of the similar windows corresponding to these two pixel points in the comprehensive distribution image, and calculate the weight value of each pixel point in the search window for the current pixel point; based on the weight value, perform weighted summation on the pixel values of all pixel points in the search window to obtain the denoised pixel value of the current pixel point, thereby constructing a denoised image.

[0013] Preferably, in step S2, the similar window and the search window are each a square window centered on any pixel point, and the similar window is smaller than the search window, and the similar window slides within the search window.

[0014] Preferably, in step S3, the dark channel value of each pixel point is determined by the minimum value of the minimum values of each pixel point in the corresponding pixel point's similar window in each channel.

[0015] Preferably, in step S4, the illumination component of each pixel point in the illumination image is obtained by convolving the pixel value with a preset Gaussian kernel.

[0016] Preferably, in step S3, the construction method of the dust influence degree of the similar window is as follows:

[0017] Denote the dust influence degree of the similar window as , and the calculation formula is:

[0018]

[0019] In the formula, P is the number of dark channel value orders in the similar window, Q is the maximum value of the sizes of the connected regions with the same dark channel order in the similar window, M is the maximum number with the same connected region size in the similar window, p is the order p, q is the connected region size q, represents the occurrence times of the connected region with the dark channel order p and size q in the similar window, represents the distance between the centroid of the m-th connected region with the dark channel order p and size q in the similar window and the center of the similar window.

[0020] Preferably, in step S5, the product result of the dust influence degree of the similarity window of each pixel point in the dark channel image and the illumination influence degree of the similarity window of the corresponding pixel point in the illumination image is used as the value of the corresponding pixel point in the comprehensive distribution image.

[0021] Preferably, in step S5, the calculation method of the direction dispersion degree of the similarity window is as follows:

[0022] Obtain the gradient direction and gradient amplitude of each pixel point in the comprehensive distribution map;

[0023] Divide the gradient direction into several gradient direction ranges; accumulate and statistically analyze the gradient amplitudes belonging to the same gradient direction range to obtain a gradient direction histogram;

[0024] Take the sum of squares of the accumulated results of the gradient amplitudes under all gradient direction ranges in the gradient direction histogram as the direction dispersion degree of the similarity window.

[0025] Preferably, in step S5, the similarity of the direction dispersion degrees of the similarity windows of any two pixel points in the comprehensive distribution image is determined by the result of negatively correlating and mapping the difference between the direction dispersion degrees of the similarity windows of these two pixel points.

[0026] Preferably, in step S6, the calculation method of the weight of each pixel point in the search window for the current pixel point is as follows:

[0027] Add the spatial proximity of each pixel point in the search window to the current pixel point and the similarity of the direction dispersion degrees of the similarity windows of these two corresponding pixel points in the comprehensive distribution image to obtain the weight of each pixel point in the search window for the current pixel point.

[0028] Preferably, the method further includes performing target recognition and tracking processing on the denoised image.

[0029] This application has at least the following beneficial effects:

[0030] This application utilizes the redundant information in the video images in the coal mine underground. According to the characteristics of more dust and weak illumination in the coal mine underground, taking the search window in the image as a unit, analyzes the distribution characteristics of dust and illumination in all internal similarity windows, considers the influence magnitudes of dust and illumination on different pixel points, and combines the spatial proximity, dark channel value proximity, and illumination situation proximity of the image to estimate each pixel value in the image. Without losing the image details and edge information, it eliminates the adverse effects of dust, illumination, and distance on the similarity calculation, reduces the noise of the monitoring video images in the coal mine underground, and makes the monitoring process more accurate. Description of the Drawings

[0031] To more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0032] Figure 1 It is a flowchart of a coal mine underground video monitoring method provided by the present application;

[0033] Figure 2 It is a flowchart for constructing an index of the weights of each pixel point in the search window provided by the present application for the current pixel point. Specific Embodiments

[0034] To further elaborate on the technical means and effects adopted by the present application to achieve the intended invention purpose, the following, in combination with the drawings and preferred embodiments, details the specific embodiments, structures, features, and effects of a coal mine underground video monitoring method proposed according to the present application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs.

[0036] The following specifically describes the specific solution of a coal mine underground video monitoring method provided by the present application with reference to the drawings.

[0037] A coal mine underground video monitoring method provided by an embodiment of the present application.

[0038] Specifically, a coal mine underground video monitoring method is provided as follows. Please refer to Figure 1 and the method includes the following steps:

[0039] S1, Use an explosion-proof camera to collect coal mine underground video images.

[0040] The present application needs to denoise the coal mine underground video images and needs to first use an explosion-proof camera to collect the images.

[0041] Cameras are needed for daily inspections, documentary filming, geological surveys, etc. in underground coal mines. However, in underground coal mines, there are flammable and explosive gases and substances such as gas and coal dust. When underground equipment is operating normally or malfunctioning, it will generate electric sparks, arcs, hot surfaces, and hot particles, etc. They all have thermal energy and may become ignition sources and heat sources for igniting mine gas and coal dust. Ordinary cameras have not been improved with explosion-proof technology and may generate sparks with combustible gases underground, thus leading to the risk of explosion. Explosion-proof cameras can be used in dangerous places in coal mines and are mainly used for investigating and collecting evidence of coal mine disaster accidents and daily geological recording, recording production safety conditions, operating status of mechanical and electrical equipment, roof support conditions, and geological features underground.

[0042] Accordingly, this application uses an explosion-proof camera to collect video images underground in coal mines. Among them, the explosion-proof camera needs to meet the explosion-proof standards underground in coal mines and is used to safely collect video images in an environment containing gas or coal dust. The specific model of the explosion-proof camera is set by the implementer himself.

[0043] S2, define a similarity window and a search window for each pixel point in the video image of the underground coal mine; for each pixel point in the search window of the current pixel point, use the Gaussian function to analyze the distance between each pixel point and the current pixel point, and calculate the spatial proximity of each pixel point in the search window to the current pixel point.

[0044] Considering the self-similar nature of the image, the redundant information in the image can be used to denoise the image, and the detailed features of the image can be retained to the greatest extent while denoising.

[0045] First, define a similarity window and a search window for each pixel point in the image, which are used to analyze the distribution of the similarity between pixel points in a local range, so as to facilitate subsequent accurate denoising processing of the pixel value of the central pixel point.

[0046] It should be noted that the similarity window and the search window are each a square window centered on any pixel point, and the similarity window is smaller than the search window, and the similarity window slides within the search window.

[0047] Among them, the size of the similarity window is , in this embodiment, f takes the value of 4, that is, the size of the similarity window is 9×9; the size of the search window When, in this embodiment, s takes the value of 6, that is, the size of the search window is 13×13.

[0048] It is worth noting that when the pixel points in the similarity window or the search window exceed the image range, zero-padding operations are performed on them.

[0049] Let the x - coordinate of the current pixel be (k, l), and the y - coordinate of a pixel in the search window of the current pixel x be (i, j). The similarity between the two points depends on spatial proximity, dark - channel value proximity, and illumination - condition proximity.

[0050] When a pixel is far from another pixel, the influence on the other pixel is smaller. Spatial proximity describes the difference in spatial positions between two pixels. The distance between pixels can be analyzed using a Gaussian function. The spatial proximity between the current pixel x and the pixel y in its search window is represented as follows:

[0051]

[0052] In the formula, e is the natural constant, h is the smoothing parameter used to control the attenuation degree of the Gaussian function. The larger h is, the flatter the Gaussian function changes, and the higher the influence of pixel y on the denoising level of the current pixel x, but it will also cause image blurring; the smaller h is, the lower the influence of pixel y on the denoising level of the current pixel x, and the more edge - detail components are retained, but there will be excessive noise remaining. In this embodiment, h is taken as 1.

[0053] S3. Obtain the dark - channel value of each pixel in the image and construct a dark - channel image; divide the orders of all dark - channel values in the similarity window of each pixel in the dark - channel image, and analyze the connected regions of different sizes and their occurrence times at each divided order, as well as the distance between the centroid of each connected region and the center of the similarity window, and construct the dust influence degree of the similarity window.

[0054] In real life, the images observed or collected are often not the true colors of objects after being affected by light intensity, dust, etc. When performing image processing, it is necessary to consider and eliminate the influence of these external factors, and then calculate the similarity between different pixels.

[0055] Considering that the observed or collected images are affected by dust and light, the similarity between two pixels can be judged by the dark - channel value of the image and the similarity of light intensity.

[0056] Color images all contain three channels, namely the R, G, and B channels. In a clear and fog - free picture, for any local area except the sky area, at least one channel value of the pixels is very low because dust, shadows, etc. affect the object. This channel value is the dark - channel value. The dark - channel value of a pixel is equal to the minimum of the minimum values of each pixel in its local window in each channel, which is used to reflect the degree of influence of the pixel by dust.

[0057] Accordingly, based on the dark channel prior theory, for any pixel point in the image, the minimum value of the minimum values of each pixel point in the similarity window of the any pixel point in each channel is obtained as the dark channel value of the any pixel point.

[0058] Calculate the dark channel value of each pixel point in the image according to the above formula to obtain a dark channel image.

[0059] Dust often appears in patches and has certain regional characteristics. For the similarity window of each pixel point in the dark channel image, construct its dark channel region size matrix and dark channel distance size matrix.

[0060] The size of the dark channel region size matrix is P×Q. P is the number of orders of the dark channel values in the similarity window. In this embodiment, the value is 32, that is, the pixel values 0 - 255 are evenly mapped to the dark channel orders 1 - 32 every 8 pixels. For example: map 0 - 7 to order 1, 8 - 15 to order 2, 248 - 255 to order 32. Q is the maximum value of the size of the connected region with the same dark channel order in the similarity window, that is, the size of the region where the connected same order is located. The elements in the dark channel region size matrix represent the occurrence times of the connected region with the dark channel order p and size q in the similarity window.

[0061] The size of the dark channel distance size matrix is P×Q×M. M is the maximum number of connected regions with the same size in the similarity window. Among them, the element represents the distance between the centroid of the m-th connected region with the dark channel order p and size q in the similarity window and the center of the similarity window.

[0062] Furthermore, by analyzing the connected regions of different sizes and their occurrence times at each divided order, and the distance between the centroid of each connected region and the center of the similarity window, construct the dust influence degree of the similarity window , and the calculation formula is:

[0063]

[0064] In the formula, P is the number of orders of the dark channel values in the similarity window, Q is the maximum value of the size of the connected region with the same dark channel order in the similarity window, M is the maximum number of connected regions with the same size in the similarity window, p is the order p, q is the connected region size q, represents the occurrence times of the connected region with the dark channel order p and size q in the similarity window, represents the distance between the centroid of the m-th connected region with the dark channel order p and size q in the similarity window and the center of the similarity window.

[0065] It should be understood that the larger the value of p, the larger the order of the dark channel, indicating a greater degree of influence by dust; the larger the value of q, the larger the connected region of the dark channel, the larger it is, the more times the connected region appears, indicating a wider range affected by dust; the larger it is, the closer the connected region is to the central pixel point, indicating a greater influence of the central pixel point by the dust region.

[0066] For the connected regions in the dark channel image with a larger order of the dark channel, a larger area of the connected region, and a closer distance to the center of the similarity window, they are more affected by dust and should be given a greater weight.

[0067] S4. Obtain the illumination component of each pixel point in the image and construct an illumination image; according to the same calculation method for the dust influence degree of the similarity window of each pixel point in the dark channel image, calculate the illumination influence degree of the similarity window of each pixel point in the illumination image.

[0068] In real life, the observed or captured image is composed of an illumination image and a reflection image. The color of an object is determined by the object's ability to reflect light, rather than by the absolute value of the illumination intensity. To obtain the true color of the object, it is necessary to eliminate the influence of different illuminations on the captured image information.

[0069] Based on the Retinex theory, the illumination component of the ambient light can be approximately represented by the convolution of the original image and a Gaussian kernel, so that the illumination image can be obtained. The value of each pixel point in the illumination image is obtained by convolving the pixel value of the corresponding pixel point with the Gaussian kernel, and it is denoted as the illumination component of the corresponding pixel point. In this embodiment, the size of the Gaussian kernel is set to 5×5. Among them, the process of using the Gaussian kernel to convolve the pixel value to obtain the illumination component in the Retinex theory is a well-known technology and will not be elaborated here.

[0070] Similarly, according to the illumination image, according to the dust influence degree of the similarity window of each pixel point in the dark channel image with the same calculation method, calculate the illumination influence degree of the similarity window of each pixel point in the illumination image .

[0071] S5. Based on the dust influence degree and the illumination influence degree, determine the value of the central pixel point at the corresponding similarity window position in the comprehensive distribution image; use the dispersion of the gradient amplitude of the values within the similarity window of each pixel point in the comprehensive distribution image in different direction ranges to calculate the direction dispersion degree of the similarity window; and analyze the similarity of the direction dispersion degrees of the similarity windows of any two pixel points in the comprehensive distribution image.

[0072] By synthesizing the effects of dust and light, the comprehensive distribution influence degree of a certain pixel can be obtained. , the comprehensive distribution influence degree is the product result of the dust influence degree and the light influence degree within the similarity window corresponding to its pixel.

[0073] Calculate the comprehensive distribution influence degree for each pixel in the image, and then, according to the corresponding positions of the pixels in the image, form a comprehensive distribution image with the comprehensive distribution influence degrees of all pixels.

[0074] Since the directions and intensities of the dust and light effects on each pixel are different, that is, the gradient directions and amplitudes are different, the HOG operator can be used to statistically analyze the local gradient amplitudes and directions of the comprehensive distribution image to form a histogram based on gradient characteristics.

[0075] To obtain the gradient histogram, in this embodiment, the sobel operator is first used to obtain the gradient direction and gradient amplitude of each pixel in the comprehensive distribution map. Among them, the sobel operator is a well-known technology and will not be elaborated here. In other embodiments of the present application, the Prewitt operator, Canny edge detection, etc. can also be used to obtain the gradient direction and gradient amplitude.

[0076] To facilitate the statistical analysis of the gradient directions and gradient amplitudes of the comprehensive distribution of pixels within the similarity window, the 0-180 degree range of the gradient direction is equally divided into N equal parts, that is, divided into N gradient direction ranges. In this embodiment, the value of N is 9, which are 0, 20, 40,..., 160, corresponding to [0, 20) degrees, [20, 40) degrees,..., [160 - 180) degrees respectively. Then, the gradient amplitudes belonging to the same gradient direction range are accumulated and statistically analyzed to obtain the gradient direction histogram, that is, a vector composed of 9 values , n takes values from 1 to 9, representing 9 gradient direction ranges, represents the sum of the gradient amplitudes of the nth angular range and normalizes it to , and the direction dispersion degree of this similarity window can be expressed as LR:

[0077]

[0078] The magnitude of the LR value represents the comprehensive distribution, that is, the dispersion degree of the dust and light distribution directions. When there are more distribution directions, that is, when the dust and light distributions are more dispersed, LR is larger; when the distribution direction is more single, that is, when the dust and light distributions are more concentrated, LR is smaller.

[0079] The result of performing a negative correlation mapping on the difference in the direction dispersion of the similarity windows of any two pixel points in the comprehensive distribution image is taken as the similarity F of the similarity windows of any two pixel points in the comprehensive distribution image. It should be understood that the larger the result of the negative correlation mapping, the more unified the dust and light distribution directions in the similarity windows of the two pixel points, indicating that the areas around the two pixel points have more similar features and the similarity of the similarity windows of the two pixel points is higher.

[0080] Optionally, the negative correlation mapping can be implemented by methods such as negative linear mapping, negative exponential mapping, or setting adjustment parameters. Specifically, in this embodiment, the opposite of the absolute value of the difference in the direction dispersion of the similarity windows of two pixel points is used as the exponent of an exponential function with the natural constant e as the base, and the calculation result of the exponential function is used as the similarity of the similarity windows of the two pixel points.

[0081] S6. Combining the spatial proximity of each pixel point in the search window to the current pixel point and the similarity of the direction dispersion of the similarity windows corresponding to these two pixel points in the comprehensive distribution image, calculate the weight of each pixel point in the search window for the current pixel point; based on the weights, perform a weighted sum of the pixel values of all pixel points in the search window to obtain the pixel value of the current pixel point after denoising, thereby constructing the denoised image.

[0082] In this embodiment, taking the pixel point y in the search window of the current pixel point x in the image as an example, the spatial proximity between the current pixel point x and the pixel point y in its search window and the similarity of the similarity windows corresponding to these two pixel points in the comprehensive distribution image are used to calculate the similarity between the pixel point y in the search window and the current pixel point x; that is, through the distance information between the two pixel points and the similar features of the distributions around the two pixel points, the similarity between the pixel point y in the search window and the current pixel point x is analyzed.

[0083] Further, after calculating the similarity between each pixel point in the search window and the current pixel point, normalization processing can be performed first to be used for weighted summation of the pixel values of each pixel point. Accordingly, the weight of the pixel point y in the search window for the current pixel point x is calculated ; ; where is the sum of the similarities between all pixel points in the search window and the current pixel point x. The larger the weight, the greater the influence of the pixel point y on denoising the current pixel point x.

[0084] Thus, the weights of each pixel point in the search window for the current pixel point can be calculated.

[0085] In this application, the flowchart for constructing the index of the weights of each pixel point in the search window for the current pixel point is as shown in the appendix Figure 2 as follows.

[0086] Then, based on the weights of each pixel point in the search window for the current pixel point, the pixel values of all pixel points in the search window are weighted and summed to obtain the pixel value of the current pixel point after denoising. Calculate the pixel values of all pixel points in the image after denoising, thereby constructing the denoised image.

[0087] The pixel value of each pixel point in the image is determined by the weighted sum of all pixel values in its search window, and its weight is not unique, which is determined by the similarity between two pixel points. The pixel values around the current pixel point are used to determine the final output value of this pixel point, thereby completing the denoising process of the image.

[0088] After removing noises such as dust and light from the image, the denoised image is further subjected to target recognition and tracking processing. By multi-object tracking (MOT), specific targets (such as miners, equipment) can be recognized and continuously tracked, and abnormal behaviors or potential dangerous areas can be accurately detected, thereby improving the safety and standardization of production. The specific multi-object tracking (MOT) can be implemented by combining Kalman filtering and Hungarian algorithms (such as SORT, DeepSORT algorithms, etc.), and the above algorithms are all well-known technologies and will not be elaborated here. Specifically, it is set by the implementer himself.

[0089] Each embodiment in this application is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments.

[0090] It should be noted that unless otherwise specified and limited, terms such as "including", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a circuit structure, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the article or device including the element. In addition, the term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0091] Those skilled in the art will readily think of other implementation schemes of this application after considering the specification and practicing the invention herein. This application is intended to cover any variations, uses or adaptations of this application, and these variations, uses or adaptations follow the general principles of this application and include common general knowledge or conventional technical means in the technical field not invented by this application.

[0092] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A video monitoring method for underground coal mines, characterized in that: The method comprises the following steps: A similarity window and a search window are defined for each pixel point in the video image of the underground coal mine; for each pixel point in the search window of the current pixel point, the distance between each pixel point and the current pixel point is analyzed by using a Gaussian function, and the spatial proximity between each pixel point in the search window and the current pixel point is calculated; Obtain the dark channel value of each pixel in the image and construct a dark channel image; divide all dark channel values ​​in the similarity window of each pixel in the dark channel image into orders, and analyze the connected areas of different sizes and their occurrence times under each order of division, as well as the distance between the centroid of each connected area and the center of the similarity window, to construct the dust influence degree of the similarity window; Obtain the illumination component of each pixel in the image and construct an illumination image; calculate the illumination influence of the similar window of each pixel in the illumination image in the same way as the dust influence of the similar window of each pixel in the dark channel image; Based on the dust influence and the illumination influence, the value of the central pixel point corresponding to the similar window position in the comprehensive distribution image is determined; the directional dispersion of the similar window is calculated by using the dispersion of the gradient amplitude of the value in the similar window of each pixel point in the comprehensive distribution image in different direction ranges; and the similarity of the directional dispersion of the similar windows of any two pixel points in the comprehensive distribution image is analyzed; The weight of each pixel in the search window to the current pixel is calculated by combining the spatial proximity of each pixel in the search window to the current pixel and the similarity of the directional dispersion of the similar windows of the two corresponding pixels in the comprehensive distribution image; based on the weight, the pixel values ​​of all pixels in the search window are weighted summed to obtain the denoised pixel value of the current pixel, thereby constructing a denoised image.

2. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S2, the similarity window and the search window are respectively a square window centered on any pixel point, and the similarity window is smaller than the search window, and the similarity window slides in the search window.

3. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S3, the dark channel value of each pixel is determined by the minimum value of the minimum values ​​of each pixel in each channel in the similarity window corresponding to the pixel.

4. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S4, the illumination component of each pixel in the illumination image is obtained by convolving the pixel value with a preset Gaussian kernel.

5. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S3, the dust influence degree of the similar window is constructed by: The dust influence degree of the similar window is recorded as , the calculation formula is: Where P is the number of dark channel value orders in the similar window, Q is the maximum size of the connected area with the same dark channel order in the similar window, M is the maximum number of connected areas with the same size in the similar window, p is the order p, q is the size of the connected area q, It represents the number of occurrences of connected regions with dark channel order p and size q in the similarity window. It represents the distance between the centroid of the mth connected region with dark channel order p and size q in the similarity window and the center of the similarity window.

6. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S5, the product of the dust influence degree of the similar window of each pixel point in the dark channel image and the illumination influence degree of the similar window of the corresponding pixel point in the illumination image is used as the value of the corresponding pixel point in the comprehensive distribution image.

7. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S5, the calculation method of the directional dispersion of the similarity window is: Obtain the gradient direction and gradient amplitude of each pixel in the comprehensive distribution map; The gradient direction is divided into several gradient direction ranges; the gradient amplitudes belonging to the same gradient direction range are accumulated and counted to obtain a gradient direction histogram; The square sum of the accumulated gradient amplitudes in all gradient direction ranges in the gradient direction histogram is taken as the directional dispersion of the similarity window.

8. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S5, the similarity of the directional dispersion of similar windows of any two pixels in the comprehensive distribution image is determined by the result of negative correlation mapping of the difference in the directional dispersion of the similar windows of the two pixels.

9. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: In step S6, the weight of each pixel in the search window to the current pixel is calculated as follows: The spatial proximity of each pixel in the search window to the current pixel is added to the similarity of the directional dispersion of the similar windows of the two corresponding pixels in the comprehensive distribution image to obtain the weight of each pixel in the search window to the current pixel.

10. A method for underground video monitoring in a coal mine according to claim 1, characterized in that: The method further includes performing target recognition and tracking processing on the denoised image.

Images