Power distribution room video monitoring system based on wireless communication
Through the distribution room video surveillance system based on wireless communication, the gradient descent method optimizes video transmission and analysis, the problem of insufficient network broadband and insufficient recognition accuracy in the distribution room video surveillance system is solved, and efficient and stable Video surveillance and automatic recognition functions.
Patent Information
- Application Number
- CN202411328438.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-24
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing distribution room video surveillance system is insufficient in network broadband when transmitting large amounts of image data in real time, resulting in the impact of image transmission clarity and accuracy. The existing algorithms lack recognition accuracy and stability in complex scenarios, making it difficult to identify and locate People changes, production accidents, etc. cannot respond in a timely manner.
Design a distribution room video surveillance system based on wireless communication, including video capture module, wireless communication module, analysis module, training module and alarm display module, transmit video data through wireless network, and optimize machine learning process using gradient descent method Sensitivity coefficient, quantifies the sensitivity of the color values of the primary color channels of red, green and blue, extracts the pixel color change feature values, and realizes automatic analysis and recognition of the monitoring video.
Reduces data transmission delay and packet loss, meets the need to transmit large amounts of image data in real time, improves the recognition accuracy and stability of video surveillance, realizes flexible response to complex scenarios and automatic data recognition, and reduces the repetitive labor of human monitoring. .
Smart Images

Figure CN119094708A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of video monitoring, and in particular to a video monitoring system for a power distribution room based on wireless communication. Background Art
[0002] Video surveillance technology uses video equipment to shoot the monitored area in real time and transmit the captured images or video signals to the monitoring center. In the monitoring center, the received images or video signals are processed and finally stored through video recorders, hard disk recorders, network storage devices, etc. for later retrieval and playback. This process realizes real-time monitoring and recording of the monitored area. In recent years, with the rapid development of information technology, computer technology and the Internet, various social fields have accumulated huge data resources. With the widespread application of information systems, the algorithms related to how to make full use of large-scale data in surveillance videos and discover useful data have developed rapidly, which are derived from ancient data analysis and statistical techniques, combined with modern artificial intelligence, database and statistics related technologies.
[0003] Gradient descent method is an important branch of optimization theory. It was first used by the great mathematician Cauchy as early as 1847. It is the oldest analytical method, and most other analytical methods are its successors and form the basis of optimization methods. Gradient descent method is a common method for minimizing risk function and loss function, also known as the steepest descent method. Its main idea is to first initialize the coefficient vector, and then gradually iteratively update the coefficient vector value so that its loss function is iteratively updated in the direction of the fastest gradient descent, that is, along the negative direction of the gradient, until the local minimum is obtained.
[0004] The existing video surveillance technology for power distribution rooms has many shortcomings. First, the video surveillance system of the power distribution room needs to transmit a large amount of image data in real time, which places extremely high demands on network bandwidth. However, in actual applications, there is often a problem of insufficient network bandwidth resources, which affects the clarity and accuracy of image transmission.
[0005] Secondly, the data extraction method for video surveillance needs to be optimized. The existing algorithm may have problems with insufficient recognition accuracy and stability when facing complex scenes. It is difficult to meet higher accuracy and reliability requirements. Specifically, when faced with surveillance videos about personnel changes, production accidents, equipment damage or movement in the distribution room, the existing video surveillance system is difficult to identify and locate, and then respond in time to send alarms and notifications to management.
[0006] Therefore, it is necessary to design a distribution room video monitoring system based on wireless communication to solve the above technical problems. Summary of the invention
[0007] In order to overcome the shortcomings of the background technology, an embodiment of the present invention provides a distribution room video monitoring system based on wireless communication, which can effectively solve the problems involved in the above-mentioned background technology.
[0008] The purpose of the present invention can be achieved by the following technical solutions:
[0009] A distribution room video monitoring system based on wireless communication comprises a video capture module, a wireless communication module, an analysis module, a training module and an alarm display module.
[0010] The video capture module obtains the monitoring video of the distribution room through the cameras installed in various corners of the distribution room.
[0011] The wireless communication module transmits the surveillance videos captured by the cameras in each video capture module to the server where the analysis module and the training module are located through the wireless network card, and sends the data processed by the analysis module to the alarm display module in each administrator's personal computer through the wireless network.
[0012] As a preferred embodiment of the invention, the analysis module analyzes the surveillance video at a preset frame interval k, specifically: obtain the resolution of the surveillance video: image width W×image height H, and perform coordinate positioning on all pixels in the surveillance video. The coordinate positioning is (w, h), where 0≤w≤W; 0≤h≤H. At a preset frame interval k, the color of each pixel point (w, h) in the surveillance video is decomposed into three channels of R, G and B according to the three color channels of red, green and blue, and the three-channel color values of each pixel point are obtained, including the red channel color value R(k, w, h), the green channel color value G(k, w, h) and the blue channel color value B(k, w, h).
[0013] Furthermore, the surveillance video is obtained at a preset frame interval k by the formula Calculate the sum of the color values αR(k), αG(k), and αB(k) corresponding to the three channels of R, G, and B in the surveillance video of the kth frame.
[0014] Further, at every preset frame interval k, the current time T and the local weather type are obtained through the Internet. The local weather types include sunny, cloudy, light rain, moderate rain, heavy rain, sleet, hail, haze, etc. At every preset frame interval k, the corresponding preset weather characteristic value U(k) is matched according to the local weather type.
[0015] As a preferred embodiment of the invention, three sensitivity coefficients are set, namely λ1, λ2 and λ3.
[0016] Set the sensitivity coefficient output model: in is a set of weight coefficients, where b1, b2, and b3 are bias coefficients. The specific values of all weight coefficients and bias coefficients are obtained from the machine learning results of the training module.
[0017] Based on the values of the three sensitivity coefficients λ1, λ2 and λ3, every preset frame interval k, the formula
[0018] Calculate the pixel color change feature value corresponding to each point (w, h) in the Kth frame Where w and h are the coordinate locations of the pixel points, where 0≤w≤W; 0≤h≤H.
[0019] As a preferred embodiment of the present invention, an initial abnormal pixel feature number Count=0 is preset.
[0020] By formula Calculate the number of abnormal pixel features Count(k) corresponding to each frame number k, where is the preset pixel color change feature value threshold, where P h w is the pixel status code of the pixel point (w, h). In a frame of surveillance video, the calculation formula is traversed over all the pixel points (w, h), and the abnormal pixel feature number Count (k) of each frame is calculated by continuous addition.
[0021] Furthermore, a statistical cycle T is preset, and every T×K frames, the number of abnormal pixel features generated successively is counted: Count(k)=Count(0), Count(1×K), Count(2×K), ..., Count(T×K). By formula Calculate the average abnormal pixel feature number U, where i = 1, 2, 3, ..., T.
[0022] When the average abnormal pixel feature number U is greater than the preset threshold Umax, it is considered that there is an obvious moving object in the surveillance video within the previous T×K frames, and the surveillance video within the previous T×K frames is marked as an abnormal video clip.
[0023] Furthermore, the abnormal video clips are highlighted, specifically:
[0024] In each abnormal video clip obtained, every preset frame number k, locate the coordinates (w, h) of all pixels whose pixel status code value is 1, and record them as abnormal coordinates. Perform numerical judgment, locate the point with the smallest width coordinate w and the largest height coordinate h among all abnormal coordinates, and record it as corner point 1; locate the point with the largest width coordinate w and the largest height coordinate h among all abnormal coordinates, and record it as corner point 2; draw a rectangle with corner point 1 and corner point 2 as the two positioning points of the diagonal line of the rectangle.
[0025] The abnormal video clips that have been highlighted are sent to the training module and the alarm display module.
[0026] As a preferred embodiment of the invention, the training module trains the output model of the sensitivity coefficients λ1, λ2 and λ3, including the weight coefficients, through a machine learning process based on the gradient descent method. And the bias coefficients b1, b2 and b3, specifically including the following stages:
[0027] Phase 1, training material preparation: Obtain a set of surveillance videos recorded under different shooting conditions, numbered as video j=1,2,3,...,m; where m is the total number of surveillance video segments. Locate the video frame kj containing information about personnel, equipment or abnormal situations in each surveillance video j=1,2,3,...,m, where kj=k1,k2,k3,...,km. Subsequently, manually mark the moving personnel, equipment or abnormal situations in the kj=k1,k2,k3,...,km frames in each surveillance video j=1,2,3,...,m, and draw a selection rectangle to divide and limit the area where the moving personnel, equipment or abnormal situations are located. The shooting conditions include: the sum of the color values αR(k), αG(k) and αB(k) corresponding to the three channels R, G and B in the surveillance video, and the time T and weather characteristic value U(k) corresponding to the shooting of the surveillance video. Get the pixel coordinates of the four corner points of the selection rectangle (w1, h1), (w2, h2), (w3, h3) and (w4, h4). Among the four corner points, locate the point with the smallest width coordinate w and the largest height coordinate h, and record it as reference corner point 1; locate the point with the largest width coordinate w and the largest height coordinate h, and record it as reference corner point 2. Get the coordinates of reference corner point 1 and reference corner point 2 (wj01, hj01) and (wj02, hj02), where j = 1, 2, 3, ..., m.
[0028] Stage 2: Define the machine learning process: Initialize coefficients and preset weight coefficients The initial values of the bias coefficients b1, b2 and b3 are all 1. The kj=k1, k2, k3, ..., km frames of the surveillance video j=1, 2, 3, ..., m in the training material preparation stage are sent to the analysis module for calculation, and a set of corner point coordinates (wj1, hj1) and (wj2, hj2) are obtained by calculation, where j=1, 2, 3, ..., m.
[0029] Setting the loss function
[0030]
[0031] Stage 3: Execute machine learning training:
[0032] The weight coefficient and bias coefficient are optimized by the gradient descent method, with the minimum loss function as the optimization guide, specifically:
[0033] Set the objective function: MINJ(M).
[0034] Setting update rules
[0035] The weight coefficient corresponding to the minimum value of the objective function is obtained through iterative training. And bias coefficients b1, b2, b3. And all weight coefficients and bias coefficients are sent to the analysis module as calculation parameters in the sensitive coefficient output model. γ+1 ) and M( γ ) are the values of the parameters in the γ+1th and γth iterations respectively; β is the preset learning rate; ▽J(M( γ )) is the loss function J(M) when M=M( γ ) when the gradient function is Indicates that when M=M( γ ), the loss function J(M) is about M( γ ) is the partial derivative of .
[0036] The alarm display module generates an alarm signal when the analysis module identifies an abnormal video, displays the alarm signal through the display screen of the administrator's personal computer and plays the abnormal video clip.
[0037] Compared with the prior art, the present invention has the following beneficial effects:
[0038] 1. The present invention establishes an information transmission network in the camera, server and administrator's personal computer through a wireless communication module, which reduces data transmission delay and packet loss, meets the demand of the distribution room video monitoring system for real-time transmission of a large amount of image data, reduces the demand for network bandwidth of the entire system, alleviates the problem of insufficient network bandwidth resources, and avoids the clarity and accuracy of video and image transmission being affected;
[0039] 2. The present invention quantifies the system's sensitivity to the color values of the three primary color channels of red, green, and blue by setting the sensitivity coefficients λ1, λ2, and λ3, ensuring that the surveillance video can accurately capture details under any lighting conditions. The data extraction method for video surveillance is optimized, and the recognition accuracy and stability of automatic analysis of video surveillance are improved. By setting the sensitivity coefficient output model, factors such as time, ambient light, and weather are comprehensively considered, achieving a more flexible adjustment strategy that can dynamically respond to environmental changes. The gradient descent method is set through the training module to optimize the various coefficients in the sensitivity coefficient output model, ensuring that the entire system still performs well when facing complex scenes.
[0040] 3. The present invention extracts and calculates the pixel color change feature value and the abnormal pixel feature number through the training module, and extracts the color value of each pixel in the monitoring video through the three channels of R, G and B, so as to obtain quantitative data including the movement characteristics of people and objects, and highlight the abnormal pixels in the identified abnormal video; it realizes automatic data recognition of monitoring videos, improves the interpretability of data, and replaces the tedious and repetitive labor in manual monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0042] Figure 1 is a system block diagram of the present invention;
[0043] Figure 2 It is a working schematic diagram of the wireless communication module of the present invention. DETAILED DESCRIPTION
[0044] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0045] See also Figure 1 As shown, a distribution room video monitoring system based on wireless communication includes a video capture module, a wireless communication module, an analysis module, a training module and an alarm display module.
[0046] The video capture module obtains the monitoring video of the distribution room through the cameras installed in various corners of the distribution room.
[0047] See also Figure 2As shown, the wireless communication module transmits the surveillance videos captured by the cameras in each video capture module to the server where the analysis module and the training module are located through the wireless network card, and sends the data processed by the analysis module to the alarm display module in each administrator's personal computer through the wireless network.
[0048] The analysis module analyzes the surveillance video at every preset frame interval k, specifically: obtain the resolution of the surveillance video: image width W×image height H, and locate the coordinates of all pixels in the surveillance video. The coordinate location is (w, h), where 0≤w≤W; 0≤h≤H. At every preset frame interval k, the color of each pixel (w, h) in the surveillance video is decomposed into three channels of R, G and B according to the three color channels of red, green and blue, and the three-channel color values of each pixel are obtained, including the red channel color value R(k, w, h), the green channel color value G(k, w, h) and the blue channel color value B(k, w, h).
[0049] Furthermore, the surveillance video is obtained at a preset frame interval k by the formula Calculate the sum of the color values αR(k), αG(k), and αB(k) corresponding to the three channels of R, G, and B in the surveillance video of the kth frame.
[0050] Further, at every preset frame interval k, the current time T and the local weather type are obtained through the Internet. The local weather types include sunny, cloudy, light rain, moderate rain, heavy rain, sleet, hail, haze, etc. At every preset frame interval k, the corresponding preset weather characteristic value U(k) is matched according to the local weather type.
[0051] Three sensitivity coefficients are set, namely λ1, λ2 and λ3.
[0052] It should be noted that the settings of the sensitivity coefficients λ1, λ2, and λ3 are intended to quantify the sensitivity to the color values of the three primary color channels of red, green, and blue. Given that the proportion of the three-channel color values of people and objects in surveillance videos will fluctuate with natural lighting conditions. The settings of the sensitivity coefficients λ1, λ2, and λ3 are intended to dynamically reallocate the consideration weights of changes in the color values of the three channels of red, green, and blue as environmental conditions change, ensuring that surveillance videos can capture accurate details under any lighting conditions.
[0053] It should be further explained that, in view of the significant time-varying nature of sunlight in the distribution room, it is obviously unreasonable to set the three sensitivity coefficients λ1, λ2 and λ3 as fixed constants. Similarly, it is unreliable to determine the values of these coefficients by constructing fixed functions that change the sensitivity coefficients with time t, especially considering the complexity of environmental conditions and seasonal changes. Specifically, if three fixed function relationships such as λ1=f1(t), λ2=f2(t) and λ3=f3(t) are used to predict these coefficients, there will be significant limitations: although the fixed functions can theoretically provide specific values of the three sensitivity coefficients for a specific time t, in actual operation, due to the time difference between sunset and sunrise and the significant fluctuation of light intensity with the seasons, this prediction based on a fixed time point will lose its accuracy and reliability. Therefore, in order to ensure the accuracy and adaptability of the sensitivity coefficients, it is necessary to explore more flexible calculation or adjustment strategies that can dynamically respond to environmental changes.
[0054] Set the sensitivity coefficient output model: in is a set of weight coefficients, where b1, b2, and b3 are bias coefficients. The specific values of all weight coefficients and bias coefficients are obtained from the machine learning results of the training module.
[0055] Based on the values of the three sensitivity coefficients λ1, λ2 and λ3, every preset frame interval k, the formula
[0056] Calculate the pixel color change feature value corresponding to each point (w, h) in the Kth frame Where w and h are the coordinate locations of the pixel points, where 0≤w≤W; 0≤h≤H.
[0057] It should be noted that this formula aims to extract features of the color value of each pixel in the surveillance video through the three channels of R, G and B to obtain quantitative data containing the movement characteristics of people and objects.
[0058] An initial abnormal pixel feature number Count=0 is preset.
[0059] By formula Calculate the number of abnormal pixel features Count(k) corresponding to each frame number k, where is the preset pixel color change feature value threshold, where P h w is the pixel status code of the pixel point (w, h). In a frame of surveillance video, the calculation formula is traversed over all the pixel points (w, h), and the abnormal pixel feature number Count (k) of each frame is calculated by continuous addition.
[0060] It should be noted that the abnormal pixel feature number is intended to count the number of pixels in each frame of surveillance video that have changed significantly compared to the previous preset frame interval.
[0061] Furthermore, a statistical cycle T is preset, and every T×K frames, the number of abnormal pixel features generated successively is counted: Count(k)=Count(0), Count(1×K), Count(2×K), ..., Count(T×K). By formula Calculate the average abnormal pixel feature number U, where i = 1, 2, 3, ..., T.
[0062] When the average abnormal pixel feature number U is greater than the preset threshold Umax, it is considered that there is an obvious moving object in the surveillance video within the previous T×K frames, and the surveillance video within the previous T×K frames is marked as an abnormal video clip.
[0063] Furthermore, the abnormal video clips are highlighted, specifically:
[0064] In each abnormal video clip obtained, every preset frame number k, locate the coordinates (w, h) of all pixels whose pixel status code value is 1, and record them as abnormal coordinates. Perform numerical judgment, locate the point with the smallest width coordinate w and the largest height coordinate h among all abnormal coordinates, and record it as corner point 1; locate the point with the largest width coordinate w and the largest height coordinate h among all abnormal coordinates, and record it as corner point 2; draw a rectangle with corner point 1 and corner point 2 as the two positioning points of the diagonal line of the rectangle.
[0065] The abnormal video clips that have been highlighted are sent to the training module and the alarm display module.
[0066] The training module trains the output model of the sensitivity coefficients λ1, λ2 and λ3, including the weight coefficients, through a machine learning process based on the gradient descent method. And the bias coefficients b1, b2 and b3, specifically including the following stages:
[0067] Phase 1, training material preparation: Obtain a set of surveillance videos recorded under different shooting conditions, numbered as video j=1,2,3,...,m; where m is the total number of surveillance video segments. Locate the video frame kj containing information about personnel, equipment or abnormal situations in each surveillance video j=1,2,3,...,m, where kj=k1,k2,k3,...,km. Subsequently, manually mark the moving personnel, equipment or abnormal situations in the kj=k1,k2,k3,...,km frames in each surveillance video j=1,2,3,...,m, and draw a selection rectangle to divide and limit the area where the moving personnel, equipment or abnormal situations are located. The shooting conditions include: the sum of the color values αR(k), αG(k) and αB(k) corresponding to the three channels R, G and B in the surveillance video, and the time T and weather characteristic value U(k) corresponding to the shooting of the surveillance video. Get the pixel coordinates of the four corner points of the selection rectangle (w1, h1), (w2, h2), (w3, h3) and (w4, h4). Among the four corner points, locate the point with the smallest width coordinate w and the largest height coordinate h, and record it as reference corner point 1; locate the point with the largest width coordinate w and the largest height coordinate h, and record it as reference corner point 2. Get the coordinates of reference corner point 1 and reference corner point 2 (wj01, hj01) and (wj02, hj02), where j = 1, 2, 3, ..., m.
[0068] It should be noted that the process aims to establish the following relationship: Relationship 1, surveillance video + shooting conditions = coordinates of corner point 1 and corner point 2. This relationship is a connection obtained through artificial definition and will serve as a standard reference for machine learning.
[0069] Stage 2: Define the machine learning process: Initialize coefficients and preset weight coefficients The initial values of the bias coefficients b1, b2 and b3 are all 1. The kj=k1, k2, k3, ..., km frames of the surveillance video j=1, 2, 3, ..., m in the training material preparation stage are sent to the analysis module for calculation, and a set of corner point coordinates (wj1, hj1) and (wj2, hj2) are obtained by calculation, where j=1, 2, 3, ..., m.
[0070] Setting the loss function
[0071]
[0072] It should be noted that the process aims to establish the following relationship: Relationship 2, surveillance video + shooting conditions = coordinates of corner point 1 and corner point 2. This relationship is obtained through the operation of the analysis module. Then the position difference between the reference corner point obtained by relationship 1 and the corner point obtained by relationship 2 is quantified by the loss function. This loss function not only serves as a quantitative standard for measuring the effect of machine learning, but also directly guides the learning process: when the value of the loss function reaches the lowest point, it means that the machine learning has reached the optimal state, which is considered a learning success.
[0073] Stage 3: Execute machine learning training:
[0074] The weight coefficient and bias coefficient are optimized by the gradient descent method, with the minimum loss function as the optimization guide, specifically:
[0075] Set the objective function: MINJ(M).
[0076] Setting update rules
[0077] The weight coefficient corresponding to the minimum value of the objective function is obtained through iterative training. And bias coefficients b1, b2, b3. And all weight coefficients and bias coefficients are sent to the analysis module as calculation parameters in the sensitive coefficient output model. γ+1 ) and M( γ ) are the values of the parameters in the γ+1th and γth iterations respectively; β is the preset learning rate; ▽J(M( γ )) is the loss function J(M) when M=M( γ ) when the gradient function is Indicates that when M=M( γ ), the loss function J(M) is about M( γ ) is the partial derivative of .
[0078] The alarm display module generates an alarm signal when the analysis module identifies an abnormal video, displays the alarm signal through the display screen of the administrator's personal computer and plays the abnormal video clip.
[0079] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.
[0080] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and the claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and the claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations;
[0081] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A distribution room video monitoring system based on wireless communication, comprising an analysis module and a training module, characterized in that: The analysis module analyzes the monitoring video at a preset frame interval k, and locates the coordinates of all pixels in the monitoring video; at a preset frame interval, the color value of each pixel in the monitoring video is decomposed into three channels of R, G and B according to the three color channels of red, green and blue, to obtain the three-channel color value of each pixel, including the red channel color value R(k, w, h), the green channel color value G(k, w, h) and the blue channel color value B(k, w, h); the shooting condition parameters are obtained by data capture and calculation: the sum of the color values corresponding to the three channels, the current time and weather characteristic values, and the shooting condition parameters are input into the sensitivity coefficient output model to determine the sensitivity coefficient; the pixel color change characteristic value, the abnormal pixel characteristic number and the average abnormal pixel characteristic number are obtained by calculation in combination with the value of the sensitivity coefficient; the abnormal video clip is marked by judging the value of the average abnormal pixel characteristic number, and the abnormal video clip is highlighted; The training module trains the output model of the sensitivity coefficients λ1, λ2 and λ3, including the weight coefficients, through a machine learning process based on the gradient descent method. and bias coefficients b1, b2 and b3; the work of the training module is divided into three stages, including stage one, preparation of training materials; stage two, definition of machine learning process; And stage three, perform machine learning training.
2. A distribution room video monitoring system based on wireless communication according to claim 1, characterized in that: It also includes a video capture module, a wireless communication module and an alarm display module; The video capture module obtains the monitoring video of the power distribution room through the cameras installed in various corners of the power distribution room; The wireless communication module transmits the surveillance videos captured by the cameras in each video capture module to the server where the analysis module and the training module are located through the wireless network card, and sends the data processed by the analysis module to the alarm display module in each administrator's personal computer through the wireless network; The alarm display module generates an alarm signal when the analysis module identifies an abnormal video, displays the alarm signal through the display screen of the administrator's personal computer and plays the abnormal video clip.
3. A distribution room video monitoring system based on wireless communication according to claim 1, characterized in that: The specific process of obtaining shooting condition parameters through data capture and calculation is as follows: Obtain the resolution of the surveillance video: image width W×image height H, and perform coordinate positioning for all pixels in the surveillance video; the coordinate positioning is (w, h), where 0≤w≤W; 0≤h≤H; every preset frame interval k, decompose the color of each pixel (w, h) in the surveillance video into three channels of R, G and B according to the three color channels of red, green and blue, and obtain the three-channel color value of each pixel, including the red channel color value R(k, w, h), the green channel color value G(k, w, h) and the blue channel color value B(k, w, h); Get the surveillance video every preset frame interval k through the formula Calculate the sum of the color values αR(k), αG(k) and αB(k) corresponding to the three channels of R, G and B in the surveillance video of the kth frame; At every preset frame interval k, the current time T and local weather type are obtained through the Internet; the local weather types include sunny, cloudy, light rain, moderate rain, heavy rain, sleet, hail, haze, etc.; at every preset frame interval k, the corresponding preset weather characteristic value U(k) is matched according to the local weather type.
4. A distribution room video monitoring system based on wireless communication according to claim 1, characterized in that: The specific process of inputting the shooting condition parameters into the sensitivity coefficient output model to determine the sensitivity coefficient is as follows: Set three sensitivity coefficients, namely λ1, λ2 and λ3; Set the sensitivity coefficient output model: in is a set of weight coefficients, where b1, b2, and b3 are bias coefficients.
5. A distribution room video monitoring system based on wireless communication according to claim 4, characterized in that: The specific process of obtaining the pixel color change feature value, abnormal pixel feature number and average abnormal pixel feature number through calculation based on the value of the sensitivity coefficient is as follows: Based on the values of the three sensitivity coefficients λ1, λ2 and λ3, every preset frame interval k, the formula Calculate the pixel color change feature value corresponding to each point (w, h) in the Kth frame Where w and h are the coordinates of the pixel, where 0≤w≤W; 0≤h≤H; Preset an initial abnormal pixel feature number Count = 0; through the formula Calculate the number of abnormal pixel features Count(k) corresponding to each frame number k, where is the preset pixel color change feature value threshold, where P h w is the pixel status code of the pixel point (w, h); In a frame of surveillance video, the calculation formula is traversed over all pixel points (w, h), and the number of abnormal pixel features Count (k) of each frame is calculated by continuous addition; A statistical cycle T is preset, and every T×K frames, the number of abnormal pixel features generated successively is counted: Count(k)=Count(0), Count(1×K), Count(2×K), ..., Count(T×K); by the formula Calculate the average abnormal pixel feature number U, where i = 1, 2, 3, ..., T.
6. A distribution room video monitoring system based on wireless communication according to claim 5, characterized in that: The specific process of marking abnormal video clips by judging the value of the average abnormal pixel feature number is as follows: When the average abnormal pixel feature number U is greater than the preset threshold Umax, it is considered that there is an obvious moving object in the surveillance video within the previous T×K frames, and the surveillance video within the previous T×K frames is marked as an abnormal video clip.
7. A distribution room video monitoring system based on wireless communication according to claim 1, characterized in that: The specific process of highlighting abnormal video clips is as follows: In each abnormal video clip obtained, every preset frame number k, locate the coordinates (w, h) of all pixel points whose pixel status code value is 1, and record them as abnormal coordinates; perform numerical judgment to locate the point with the smallest width coordinate w and the largest height coordinate h among all abnormal coordinates, and record it as corner point one; locate the point with the largest width coordinate w and the largest height coordinate h among all abnormal coordinates, and record it as corner point two; draw a rectangle with corner point one and corner point two as the two positioning points of the diagonal of the rectangle.
8. A distribution room video monitoring system based on wireless communication according to claim 1 or 4, characterized in that: The specific process of training the output model of the sensitivity coefficients λ1, λ2 and λ3 through the machine learning process based on the gradient descent method is: Training material preparation: Obtain a set of surveillance videos recorded under different shooting conditions, numbered as video j = 1, 2, 3, ..., m; Where m is the total number of surveillance video segments; the first video frame kj containing information about personnel, equipment or abnormal situations in each surveillance video segment j=1,2,3,...,m is located, where kj=k1,k2,k3,...,km; then, the moving personnel, equipment or abnormal situations are manually marked in the kj=k1,k2,k3,...,km frames in each surveillance video segment j=1,2,3,...,m, and a selection rectangle is drawn to divide and limit the area where the moving personnel, equipment or abnormal situations are located; the shooting condition parameters include: the corresponding R, G and B channels in the surveillance video The sum of color values αR(k), αG(k) and αB(k), as well as the time T and weather characteristic value U(k) corresponding to the shooting of the surveillance video; obtaining the pixel coordinates (w1, h1), (w2, h2), (w3, h3) and (w4, h4) of the four corner points of the frame selection rectangle; among the four corner points, locating the point with the smallest width coordinate w and the largest height coordinate h, recorded as the reference corner point one; locating the point with the largest width coordinate w and the largest height coordinate h, recorded as the reference corner point two; obtaining the coordinates (wj01, hj01) and (wj02, hj02) of the reference corner point one and the reference corner point two, where j = 1, 2, 3, ..., m; Define the machine learning process: initialize coefficients and preset weight coefficients The initial values of the bias coefficients b1, b2 and b3 are all 1. The kj=k1, k2, k3, ..., km frames of the surveillance video j=1, 2, 3, ..., m in the training material preparation stage are sent to the analysis module for calculation, and a set of corner point coordinates (wj1, hj1) and (wj2, hj2) are obtained by calculation, where j=1, 2, 3, ..., m; Setting the loss function Perform machine learning training: Set the objective function: MINJ(M); Setting update rules The weight coefficient corresponding to the minimum value of the objective function is obtained through iterative training. and bias coefficients b1, b2, b3; and all weight coefficients and bias coefficients are sent to the analysis module as calculation parameters in the sensitive coefficient output model; where M( γ+1 ) and M( γ ) are the values of the parameters in the γ+1th and γth iterations respectively; β is the preset learning rate; is the loss function J(M) when M=M( γ ) when the gradient function is Indicates that when M=M( γ ), the loss function J(M) is about M( γ ) is the partial derivative of .
