Intelligent self-cleaning security monitoring system

Abstract

The present application relates to a kind of intelligent self-cleaning security monitoring systems, including video acquisition module, video analysis server, cleaning execution module and database.Optical performance state tolerance zone is established by video analysis server, and the rate threshold of transmittance attenuation, edge sharpness loss and dispersion offset is set.Video acquisition module transmits real-time monitoring stream by RTSP protocol, video analysis server extracts transmittance, edge sharpness value and dispersion characteristic value, calculates its change rate and maps to three-dimensional characteristic space to generate real-time and predicted state point.Through calculating the instant offset and predicted offset of state point to tolerance zone boundary, combined with fourfold trigger condition, pollution state is intelligently judged.When any condition is met, video analysis server identifies pollution type through pollution classifier and generates cleaning instruction, drives cleaning module to execute targeted cleaning strategy, while automatically switching to backup camera to ensure monitoring continuity.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of video monitoring systems, and in particular to a self-cleaning system for improving the imaging quality of monitoring. BACKGROUND

[0002] In the long-term outdoor operation of a monitoring system, the monitoring lens is easily contaminated by dust, oil stains, water stains, etc., resulting in a serious decline in imaging quality. Traditional cleaning solutions mainly rely on regular manual inspection or simple timed triggering of the cleaning mechanism, which has significant defects:

[0003] 1. Existing technologies mostly use fixed thresholds to determine whether a single optical parameter exceeds the threshold, which cannot dynamically perceive the accumulation trend of pollution. For example, the light transmittance decreases slowly in the initial stage of dust adhesion, while sudden rain and fog are prone to trigger cleaning due to instantaneous parameter fluctuations. This passive detection results in a lag between cleaning action and actual pollution deterioration node, or produces invalid cleaning, which not only accelerates equipment wear and tear, but also causes video image quality to decline due to lens contamination.

[0004] 2. The cleaning method is seriously mismatched with the type of pollution. Most systems use a unified cleaning strategy, but the physical properties of dust, oil film, and water stains differ significantly. For example, oil film pollution needs to be dissolved with solvent, and if it is scraped forcibly, oil stains will be left behind; water stains will form stubborn water marks if not dried enough; dust will be hardened into dirt if cleaned with liquid. Existing technologies lack the ability to automatically identify the type of pollution, resulting in low cleaning efficiency and even secondary damage to the optical coating of the monitoring lens.

[0005] Therefore, there is an urgent need in the art to build a full-automatic closed-loop system of "detection-decision-execution-feedback" SUMMARY

[0006] To address the deficiencies of the prior art, the embodiments of the present application provide an intelligent self-cleaning security monitoring system, which comprises a video acquisition module, a video analysis server, a cleaning execution module, and a database, and there is a communication connection between each module;

[0007] The video analysis server establishes an optical performance state tolerance zone based on the pollution image samples obtained from the database, and extracts the light transmittance decay rate threshold, the edge sharpness loss rate threshold, and the dispersion shift rate threshold from the database;

[0008] The video acquisition module acquires a monitoring video stream and transmits the monitoring video stream to the video analysis server through the RTSP protocol;

[0009] The video analysis server extracts the light transmittance, edge sharpness value, and dispersion characteristic value of the monitoring image at the current time from the monitoring video stream, and calculates the light transmittance decay rate, edge sharpness loss rate, and dispersion shift rate based on the video data in the recent time window;

[0010] The video analysis server obtains a real-time instantaneous state point by mapping the transmittance, edge sharpness value and dispersion characteristic value of the current time to a transmittance-edge sharpness-dispersion three-dimensional characteristic space;

[0011] The video analysis server predicts the transmittance, edge sharpness value and dispersion characteristic value of each time node in the next time window, and generates a transmittance decay trajectory, an edge sharpness loss trajectory and a dispersion shift trajectory;

[0012] The video analysis server obtains a predicted instantaneous state point by mapping the transmittance, edge sharpness value and dispersion characteristic value of the end point of the next time window to a transmittance-edge sharpness-dispersion three-dimensional characteristic space;

[0013] The video analysis server calculates the distance from the real-time instantaneous state point to the boundary of the optical performance state tolerance region as an instantaneous offset, and calculates the distance from the predicted instantaneous state point to the boundary of the optical performance state tolerance region as a predicted offset;

[0014] When any of the following conditions is met, the video analysis server inputs the transmittance decay rate, edge sharpness loss rate, dispersion shift rate and transmittance, edge sharpness value and dispersion characteristic value of the current time into a pollution classifier to output a pollution type, generates a cleaning instruction according to the pollution type, and then sends the cleaning instruction to a cleaning execution module;

[0015] Condition 1, the instantaneous offset or the predicted offset is less than an instantaneous offset threshold; Condition 2, the transmittance decay rate, the edge sharpness loss rate and the dispersion shift rate are all greater than the corresponding rate threshold; Condition 3, the transmittance decay trajectory, the edge sharpness loss trajectory and the dispersion shift trajectory cross the optical performance state tolerance region; Condition 4, the ratio of the predicted offset to the instantaneous offset is greater than a preset proportion threshold, and the predicted offset is less than K times the instantaneous offset threshold;

[0016] The cleaning execution module executes a corresponding cleaning strategy according to the received cleaning instruction, and controls the video acquisition module to switch to a backup monitoring camera.

[0017] According to a preferred embodiment, the pollution type includes oil film pollution, dust pollution and water stain pollution.

[0018] According to a preferred embodiment, the cleaning execution module controls the monitoring camera to turn off infrared fill light through the ONVIF protocol before cleaning the monitoring camera.

[0019] According to a preferred embodiment, the video analysis server extracts the transmittance, edge sharpness value and dispersion characteristic value of the monitoring image of the current time from the monitoring video stream includes:

[0020] The monitoring image of the current time is obtained by decoding from the monitoring video stream;

[0021] The monitoring image is converted to HSV color space, and the pixel value of the brightness channel is extracted;

[0022] The arithmetic mean value of all pixel values of the brightness channel is calculated as the light transmittance;

[0023] The monitoring image is converted to a grayscale image, and the Sobel edge detection operator is applied to calculate the image gradient amplitude value; the pixels with a gradient amplitude value greater than the edge determination threshold value are counted as edge pixels; and the arithmetic mean value of the gradient amplitude values of all edge pixels is calculated as the edge sharpness value;

[0024] The monitoring image is converted to CIELAB color space, and the standard deviation of the pixel values of the red-green color axis and the yellow-blue color axis is calculated respectively; the standard deviation of the red-green color axis and the standard deviation of the yellow-blue color axis are added to obtain the color dispersion characteristic value representing the color dispersion state at the current time.

[0025] According to a preferred embodiment, the video analysis server establishes the optical performance state tolerance zone based on the pollution image sample, which includes:

[0026] The video analysis server respectively collects the full cycle image sequence from the clean state to the optical failure of the dust pollution, oil film pollution and water stain pollution in the pollution image sample to obtain the dust image sequence, the oil film image sequence and the water stain image sequence;

[0027] The video analysis server respectively labels the light transmittance value, the edge sharpness value and the color dispersion characteristic value and the time stamp of each frame of pollution image in the dust image sequence, the oil film image sequence and the water stain image sequence frame by frame;

[0028] The video analysis server connects the light transmittance of the dust image sequence in time sequence to obtain the dust light transmittance curve; connects the light transmittance of the oil film image sequence in time sequence to obtain the oil film light transmittance curve; and connects the light transmittance of the water stain image sequence in time sequence to obtain the water stain light transmittance curve;

[0029] The video analysis server connects the edge sharpness value of the dust image sequence in time sequence to obtain the dust edge sharpness curve; connects the edge sharpness value of the oil film image sequence in time sequence to obtain the oil film edge sharpness curve; and connects the edge sharpness value of the water stain image sequence in time sequence to obtain the water stain edge sharpness curve;

[0030] The video analysis server connects the color dispersion characteristic value of the dust image sequence in time sequence to obtain the dust color dispersion characteristic value curve; connects the color dispersion characteristic value of the oil film image sequence in time sequence to obtain the oil film color dispersion characteristic value curve; and connects the color dispersion characteristic value of the water stain image sequence in time sequence to obtain the water stain color dispersion characteristic value curve.

[0031] According to a preferred embodiment, the video analysis server establishes the optical performance state tolerance region based on the pollution image sample includes:

[0032] Identify points in the dust transmittance curve where the absolute value of the first derivative is greater than a mutation threshold to obtain a number of dust transmittance key points; identify local minimum points of the first derivative of the dust edge sharpness curve to obtain a number of dust edge sharpness key points; identify points in the dust dispersion eigenvalue curve where the curvature value is greater than a curvature threshold to obtain a number of dust dispersion key points;

[0033] Identify points in the oil film transmittance curve where the absolute value of the second derivative is less than a convergence threshold to obtain a number of oil film transmittance key points; identify points in the oil film edge sharpness curve where the absolute value of the second derivative is less than a convergence threshold to obtain a number of oil film edge sharpness key points; identify points in the oil film dispersion eigenvalue curve where the absolute value of the second derivative is less than a convergence threshold to obtain a number of oil film dispersion key points;

[0034] Identify points in the water stain transmittance curve where the difference between the left limit and the right limit is greater than a step threshold to obtain a number of water stain transmittance key points; identify local minimum points of the second derivative of the water stain edge sharpness curve to obtain a number of water stain edge sharpness key points; identify points in the water stain dispersion eigenvalue curve where the derivative of the curvature changes sign to obtain a number of water stain dispersion key points;

[0035] Map all transmittance key points, edge sharpness key points, and dispersion key points of the three types of pollution, dust, oil film, and water stain, to the transmittance-edge sharpness-dispersion three-dimensional space according to the time stamp to form a three-dimensional key point set;

[0036] Generate a three-dimensional point cloud covering the full pollution state of dust, oil film, and water stain using an interpolation algorithm on the three-dimensional key point set, and construct a minimum convex polyhedron based on the three-dimensional point cloud, defining the interior region of the polyhedron as the optical performance state tolerance region;

[0037] Extract the boundary surface of the convex polyhedron, and perform implicit function fitting on the boundary surface using a radial basis function to obtain a decision function; the decision function satisfies that when the function value is less than or equal to zero, the point is located inside the optical performance state tolerance region; when the function value is greater than zero, the point is located outside the optical performance state tolerance region.

[0038] According to a preferred embodiment, the video analysis server detects whether the transmittance decay trajectory, the edge sharpness loss trajectory, and the dispersion shift trajectory cross the optical performance state tolerance region includes:

[0039] Align the transmittance decay trajectory, the edge sharpness loss trajectory, and the dispersion shift trajectory according to the time nodes, and generate a three-dimensional coordinate point containing the transmittance value, the edge sharpness value, and the dispersion eigenvalue at each time node;

[0040] Connect three-dimensional coordinate points in time sequence to form a spatial motion trajectory line, and discretely sample the spatial motion trajectory line to obtain a trajectory point set;

[0041] Map all trajectory points to a transmittance-edge sharpness-dispersion three-dimensional feature space to obtain a trajectory feature point set, and judge the spatial position relationship between the trajectory feature points and the optical performance state tolerance region point by point:

[0042] When there are two continuous trajectory feature points in the trajectory feature point set that satisfy the following conditions, it is determined that the tolerance region crossing occurs: the previous trajectory feature point is located inside the tolerance region and the latter trajectory feature point is located outside the tolerance region, or the previous trajectory feature point is located outside the tolerance region and the latter trajectory feature point is located inside the tolerance region.

[0043] The inside and outside positions of the trajectory feature points are determined by calculating the function value of the determination function; when the function value is greater than zero, it is determined that the trajectory feature point is located outside the optical performance state tolerance region; when the function value is less than or equal to zero, it is determined that the trajectory feature point is located inside the optical performance state tolerance region.

[0044] According to a preferred embodiment, the video analysis server calculates the transmittance decay rate, the edge sharpness loss rate and the dispersion shift rate based on the video data in the latest time window, which includes:

[0045] Obtain monitoring images at a plurality of continuous time points in the latest time window, extract the transmittance at each time point to form a transmittance sequence, extract the edge sharpness value at each time point to form an edge sharpness sequence, and extract the dispersion characteristic value at each time point to form a dispersion characteristic value sequence;

[0046] The following operations are performed on the transmittance sequence: calculate the difference value of the transmittance at adjacent time points in the transmittance sequence; divide each difference value by the corresponding time interval to obtain the instantaneous transmittance decay rate; and take the arithmetic mean of all instantaneous transmittance decay rates as the transmittance decay rate;

[0047] The following operations are performed on the edge sharpness sequence: calculate the difference value of the edge sharpness value at adjacent time points in the edge sharpness sequence; divide each difference value by the corresponding time interval to obtain the instantaneous edge sharpness loss rate; and take the arithmetic mean of all instantaneous edge sharpness loss rates as the edge sharpness loss rate;

[0048] The following operations are performed on the dispersion characteristic value sequence: calculate the difference value of the dispersion characteristic value at adjacent time points in the dispersion characteristic value sequence; divide each difference value by the corresponding time interval to obtain the instantaneous dispersion shift rate; and take the arithmetic mean of all instantaneous dispersion shift rates as the dispersion shift rate.

[0049] According to a preferred embodiment, the video analysis server predicts the transmittance, the edge sharpness value and the dispersion characteristic value at each time node in the next time window, and generates a transmittance decay trajectory, an edge sharpness loss trajectory and a dispersion shift trajectory, which includes:

[0050] The light transmittance decay rate, the edge sharpness loss rate and the dispersion shift rate and the light transmittance, the edge sharpness value and the dispersion characteristic value at the current moment are input into the LSTM prediction model trained based on the pollution image samples to output the light transmittance, the edge sharpness value and the dispersion characteristic value at each time node in the next time window, and the light transmittance, the edge sharpness value and the dispersion characteristic value at each time node are connected to generate the light transmittance decay trajectory, the edge sharpness loss trajectory and the dispersion shift trajectory, respectively.

[0051] According to a preferred embodiment, the full-cycle image sequences from the clean state to the optical failure of the dust pollution, the oil film pollution and the water stain pollution are collected in the pollution image samples respectively to obtain the dust image sequence, the oil film image sequence and the water stain image sequence;

[0052] For each image sequence, the light transmittance, the edge sharpness value, the dispersion characteristic value at each moment and the light transmittance decay rate, the edge sharpness loss rate and the dispersion shift rate closest to each moment are extracted;

[0053] The light transmittance, the edge sharpness value, the dispersion characteristic value at each moment and the light transmittance decay rate, the edge sharpness loss rate and the dispersion shift rate closest to each moment are combined into a feature vector, and the corresponding pollution type is labeled;

[0054] A neural network classification model is trained using the feature vector and the corresponding pollution type label to output the pollution type;

[0055] After the training is completed, the neural network classification model is deployed as a pollution classifier.

[0056] According to a preferred embodiment, the distance from the real-time instantaneous state point to the boundary of the optical performance state tolerance region comprises:

[0057] The function value of the decision function at the real-time instantaneous state point is calculated;

[0058] Starting from the real-time instantaneous state point, iterative step search is performed along the gradient direction of the decision function, and the function value of the decision function is recalculated after each step;

[0059] When the sign of the function value of the decision function changes for the first time during the search process, the current search point position is recorded;

[0060] The Euclidean distance between the real-time instantaneous state point and the sign change point is calculated as the distance to the boundary;

[0061] The gradient direction is determined by calculating the partial derivative of the decision function in the neighborhood of the state point through the central difference method.

[0062] According to a preferred embodiment, the calculating the distance from the predicted instantaneous state point to the boundary of the optical performance state tolerance region comprises:

[0063] calculating the function value of the decision function at the predicted instantaneous state point;

[0064] performing an iterative step search along the gradient direction of the decision function starting from the predicted instantaneous state point, and recalculating the function value of the decision function after each step;

[0065] recording the current search point position when the sign of the function value of the decision function changes for the first time during the search process;

[0066] calculating the Euclidean distance between the predicted instantaneous state point and the sign change point as the distance to the boundary;

[0067] wherein the gradient direction is determined by calculating the partial derivative of the decision function in the neighborhood of the state point by the central difference method.

[0068] The present application has the following beneficial effects:

[0069] 1. By constructing an optical performance tolerance region and real-time detecting the change trend of transmittance, edge sharpness and dispersion characteristics, the lens contamination state can be accurately perceived, and cleaning is actively triggered before the imaging quality significantly decreases, ensuring that the monitoring system continuously outputs high-quality video streams, avoiding image blurring, dispersion or detail loss caused by lens contamination. The image usability and accuracy of the monitoring system in complex environments are improved.

[0070] 2. The contamination classifier automatically identifies contamination types such as dust, oil film, water stains, and intelligently matches differentiated cleaning strategies, and the targeted cleaning solution significantly improves cleaning efficiency, avoids damage to the lens or mechanical structure by a single cleaning method, prolongs the service life of the equipment, reduces cleaning agent waste, and reduces operation and maintenance costs. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 A structural block diagram of an intelligent self-cleaning security monitoring system is provided for an exemplary embodiment. DETAILED DESCRIPTION

[0072] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, the same numbers refer to the same or similar elements throughout the drawings. The embodiments described in the following exemplary embodiments do not represent all the embodiments consistent with the present application. Rather, they are merely examples of devices and methods consistent with some aspects of the present application, as detailed in the appended claims.

[0073] The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in the present application and the following claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or," as used herein, refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0074] It is to be understood that, although the terms first, second, third, etc. can be used herein to describe various information, the information should not be limited to these terms. These terms are only used to distinguish one category of information from another category of information. For example, a first information can be termed a second information without departing from the scope of the present application. As used herein, the term "if' can be interpreted to mean "when" or "upon" or "in response to determining" depending on the context.

[0075] Referring to Figure 1 The intelligent self-cleaning security monitoring system comprises a video acquisition module, a video analysis server, a cleaning execution module and a database, and the modules are in communication connection;

[0076] The video analysis server establishes an optical performance state tolerance zone based on the pollution image samples obtained from the database, and extracts the light transmittance decay rate threshold, the edge sharpness loss rate threshold and the dispersion shift rate threshold from the database.

[0077] Optionally, the light transmittance decay rate threshold, the edge sharpness loss rate threshold and the dispersion shift rate threshold are set in advance according to actual conditions.

[0078] Preferably, the video analysis server establishes the optical performance state tolerance zone based on the pollution image samples comprises:

[0079] The video analysis server respectively collects full-cycle image sequences of dust pollution, oil film pollution and water stain pollution from the cleaning state to the optical failure in the pollution image samples to obtain dust image sequences, oil film image sequences and water stain image sequences;

[0080] The video analysis server respectively labels the light transmittance value, the edge sharpness value and the dispersion characteristic value and the time stamp of each pollution image in the dust image sequences, the oil film image sequences and the water stain image sequences frame by frame;

[0081] The video analysis server connects the light transmittance of the dust image sequences in time sequence to obtain a dust light transmittance curve; connects the light transmittance of the oil film image sequences in time sequence to obtain an oil film light transmittance curve; and connects the light transmittance of the water stain image sequences in time sequence to obtain a water stain light transmittance curve;

[0082] The video analysis server connects the edge sharpness values of the dust image sequence in time sequence to obtain a dust edge sharpness curve; connects the edge sharpness values of the oil film image sequence in time sequence to obtain an oil film edge sharpness curve; and connects the edge sharpness values of the water stain image sequence in time sequence to obtain a water stain edge sharpness curve.

[0083] The video analysis server connects the dispersion feature values of the dust image sequence in time sequence to obtain a dust dispersion feature value curve; connects the dispersion feature values of the oil film image sequence in time sequence to obtain an oil film dispersion feature value curve; and connects the dispersion feature values of the water stain image sequence in time sequence to obtain a water stain dispersion feature value curve.

[0084] Preferably, the video analysis server establishes the optical performance state tolerance region based on the pollution image sample, which includes:

[0085] The points in the dust transmittance curve where the absolute value of the first derivative is greater than the mutation threshold are identified to obtain a plurality of dust transmittance key points; the local minimum points of the first derivative of the dust edge sharpness curve are identified to obtain a plurality of dust edge sharpness key points; and the points in the dust dispersion feature value curve where the curvature value is greater than the curvature threshold are identified to obtain a plurality of dust dispersion key points.

[0086] The points in the oil film transmittance curve where the absolute value of the second derivative is less than the convergence threshold are identified to obtain a plurality of oil film transmittance key points; the points in the oil film edge sharpness curve where the absolute value of the second derivative is less than the convergence threshold are identified to obtain a plurality of oil film edge sharpness key points; and the points in the oil film dispersion feature value curve where the absolute value of the second derivative is less than the convergence threshold are identified to obtain a plurality of oil film dispersion key points.

[0087] The points in the water stain transmittance curve where the difference between the left limit and the right limit is greater than the step threshold are identified to obtain a plurality of water stain transmittance key points; the local minimum points of the second derivative of the water stain edge sharpness curve are identified to obtain a plurality of water stain edge sharpness key points; and the points in the water stain dispersion feature value curve where the derivative of the curvature changes sign are identified to obtain a plurality of water stain dispersion key points.

[0088] All transmittance key points, edge sharpness key points and dispersion key points of the three types of pollution, i.e. dust, oil film and water stain, are uniformly mapped to a transmittance-edge sharpness-dispersion three-dimensional space according to the time stamp to form a three-dimensional key point set;

[0089] An interpolation algorithm is used to generate a three-dimensional point cloud covering the full pollution state of dust, oil film and water stain from the three-dimensional key point set, and a minimum convex polyhedron is constructed based on the three-dimensional point cloud, and the internal region of the polyhedron is defined as the optical performance state tolerance region.

[0090] The boundary surface of the convex polyhedron is extracted, and a radial basis function is used to perform implicit function fitting on the boundary surface to obtain a decision function; the decision function satisfies that when the function value is less than or equal to zero, the point is located inside the optical performance state tolerance region; and when the function value is greater than zero, the point is located outside the optical performance state tolerance region.

[0091] Optionally, the identification principle of the dust light transmittance key point is that dust discrete adhesion causes the light transmittance to decrease in steps. The first derivative reflects the change rate, and when dust suddenly accumulates, the absolute value of the first derivative suddenly increases. The mutation threshold is set to capture such steep drop points, which correspond to the explosive accumulation of dust in the actual scene.

[0092] Optionally, the identification principle of the dust edge sharpness key point is that dust scattering light causes the image edge blurring process to present an accelerated inflection point. The minimum point of the first derivative identifies the moment when the blurring rate is the fastest, which corresponds to the key node of the lens being quickly covered by dust.

[0093] Optionally, the identification principle of the dust dispersion key point is that dust causes the dispersion characteristic value to fluctuate intensively, and the curvature quantization curve is curved. The high-curvature point corresponds to the peak position of the abnormal mutation of dispersion (such as strong light diffraction caused by dust of a specific particle size), and the curvature threshold is used to filter noise interference.

[0094] Optionally, the identification principle of the oil film key point is that when the oil film uniformly covers the lens, the optical parameter change gradually flattens. The second derivative reflects the stability of the change rate, and when the absolute value tends to zero, it identifies that the oil film has formed a stable covering layer, and the convergence threshold determines the turning point from dynamic change to static balance.

[0095] Optionally, the identification principle of the water stain light transmittance key point is that the water stain forms a local lens effect, causing the light transmittance to change dramatically in a small area. The left and right limit difference quantifies the step jump amplitude (such as the refractive mutation of the water droplet at the edge), and the step threshold excludes natural gradual interference.

[0096] Optionally, the identification principle of the water stain edge sharpness key point is that the optical distortion of the water stain edge has an accelerated deterioration zone. The local minimum point of the second derivative corresponds to the peak value of the distortion rate, which identifies the moment when the water stain diffuses to the key imaging area.

[0097] Optionally, the identification principle of the water stain dispersion key point is that the water stain causes the dispersion curve to appear asymmetric oscillation. The sign change point of the curvature derivative (the curvature changes from increasing to decreasing or vice versa) locates the boundary inflection point of the abnormal dispersion area (such as the dispersion mutation zone of the water droplet edge).

[0098] Optionally, the mutation threshold, the convergence threshold, the step threshold, and the curvature threshold are obtained according to experiments.

[0099] Optionally, the optical performance state tolerance region is a dynamic optical parameter safety interval for ensuring imaging quality. Its core is to jointly construct a multi-dimensional safety space through the threshold boundaries of three types of parameters, namely, transmittance, edge sharpness, and dispersion characteristic value. When the real-time instantaneous state point is located within the region, the system determines that the imaging quality meets the requirements; if the state point approaches or breaks through the boundary due to pollution, a warning or cleaning mechanism is triggered. Essentially, it is a quantitative definition and dynamic guardian boundary of "acceptable imaging quality" by the system.

[0100] The video acquisition module acquires the monitoring video stream and transmits the monitoring video stream to the video analysis server through the RTSP protocol.

[0101] The video analysis server extracts the transmittance, edge sharpness value, and dispersion characteristic value of the monitoring image at the current time from the monitoring video stream, and calculates the transmittance decay rate, edge sharpness loss rate, and dispersion shift rate based on the video data in the recent time window.

[0102] The video analysis server extracts the transmittance, edge sharpness value, and dispersion characteristic value of the monitoring image at the current time from the monitoring video stream, and calculates the transmittance decay rate, edge sharpness loss rate, and dispersion shift rate based on the video data in the recent time window.

[0103] Decode the monitoring image at the current time from the monitoring video stream;

[0104] Convert the monitoring image to the HSV color space and extract the pixel values of the brightness channel;

[0105] Calculate the arithmetic mean of all pixel values of the brightness channel as the transmittance;

[0106] Convert the monitoring image to a grayscale image and apply the Sobel edge detection operator to calculate the image gradient amplitude; count the pixels with gradient amplitude greater than the edge determination threshold as edge pixels; calculate the arithmetic mean of the gradient amplitude of all edge pixels as the edge sharpness value;

[0107] Convert the monitoring image to the CIELAB color space and calculate the standard deviation of the pixel values of the red-green hue axis and the yellow-blue hue axis, respectively; add the standard deviation of the red-green hue axis and the yellow-blue hue axis to obtain the dispersion characteristic value representing the dispersion state at the current time.

[0108] The video analysis server calculates the transmittance decay rate, edge sharpness loss rate, and dispersion shift rate based on the video data in the recent time window, which includes:

[0109] Obtain the monitoring images at a plurality of consecutive times within the recent time window, extract the transmittance at each time to form a transmittance sequence, extract the edge sharpness value at each time to form an edge sharpness sequence, and extract the dispersion characteristic value at each time to form a dispersion characteristic value sequence;

[0110] The following operations are performed on the transmittance sequence: the difference between the transmittance at adjacent time points in the transmittance sequence is calculated; each difference is divided by the corresponding time interval to obtain an instantaneous transmittance decay rate; and the arithmetic mean of all the instantaneous transmittance decay rates is taken as the transmittance decay rate.

[0111] The following operations are performed on the edge sharpness sequence: the difference between the edge sharpness values at adjacent time points in the edge sharpness sequence is calculated; each difference is divided by the corresponding time interval to obtain an instantaneous edge sharpness loss rate; and the arithmetic mean of all the instantaneous edge sharpness loss rates is taken as the edge sharpness loss rate.

[0112] The following operations are performed on the dispersion eigenvalue sequence: the difference between the dispersion eigenvalues at adjacent time points in the dispersion eigenvalue sequence is calculated; each difference is divided by the corresponding time interval to obtain an instantaneous dispersion shift rate; and the arithmetic mean of all the instantaneous dispersion shift rates is taken as the dispersion shift rate.

[0113] Optionally, the time window refers to the length of continuous data sampling for calculating the dynamic parameter, and its essence is the "observation time scale" of the system analyzing the optical degradation trend.

[0114] The video analysis server obtains a real-time instantaneous state point by mapping the transmittance, edge sharpness value, and dispersion eigenvalue at the current time to a transmittance-edge sharpness-dispersion three-dimensional feature space.

[0115] Optionally, the real-time instantaneous state point represents the optical performance of the monitoring camera at a certain precise time.

[0116] The video analysis server predicts the transmittance, edge sharpness value, and dispersion eigenvalue at each time node in the next time window, and generates a transmittance decay trajectory, an edge sharpness loss trajectory, and a dispersion shift trajectory.

[0117] Optionally, the video analysis server predicts the transmittance, edge sharpness value, and dispersion eigenvalue at each time node in the next time window, and generates a transmittance decay trajectory, an edge sharpness loss trajectory, and a dispersion shift trajectory, including:

[0118] The video analysis server inputs the transmittance decay rate, the edge sharpness loss rate, and the dispersion shift rate, and the transmittance, the edge sharpness value, and the dispersion eigenvalue at the current time into an LSTM prediction model trained based on contaminated image samples to output the transmittance, the edge sharpness value, and the dispersion eigenvalue at each time node in the next time window, and connects the transmittance, the edge sharpness value, and the dispersion eigenvalue at each time node to generate the transmittance decay trajectory, the edge sharpness loss trajectory, and the dispersion shift trajectory, respectively.

[0119] The video analysis server obtains a predicted instantaneous state point by mapping the transmittance, edge sharpness value, and dispersion eigenvalue at the end of the next time window to a transmittance-edge sharpness-dispersion three-dimensional feature space.

[0120] Optionally, the predicted instantaneous state point represents the optical performance of the monitoring camera at a certain precise time.

[0121] The video analysis server calculates the distance from the real-time instantaneous state point to the boundary of the optical performance state tolerance region as the real-time offset, and calculates the distance from the predicted instantaneous state point to the boundary of the optical performance state tolerance region as the predicted offset.

[0122] Preferably, the calculation of the distance from the real-time instantaneous state point to the boundary of the optical performance state tolerance region comprises:

[0123] calculating the function value of the decision function at the real-time instantaneous state point;

[0124] starting from the real-time instantaneous state point, performing iterative step search along the gradient direction of the decision function, and recalculating the function value of the decision function after each step;

[0125] when the sign of the function value of the decision function changes for the first time during the search process, recording the position of the current search point;

[0126] calculating the Euclidean distance between the real-time instantaneous state point and the sign change point as the distance to the boundary;

[0127] wherein the gradient direction is determined by calculating the partial derivative of the decision function in the neighborhood of the state point by the central difference method.

[0128] Preferably, the calculation of the distance from the predicted instantaneous state point to the boundary of the optical performance state tolerance region comprises:

[0129] calculating the function value of the decision function at the predicted instantaneous state point;

[0130] starting from the predicted instantaneous state point, performing iterative step search along the gradient direction of the decision function, and recalculating the function value of the decision function after each step;

[0131] when the sign of the function value of the decision function changes for the first time during the search process, recording the position of the current search point;

[0132] calculating the Euclidean distance between the predicted instantaneous state point and the sign change point as the distance to the boundary;

[0133] wherein the gradient direction is determined by calculating the partial derivative of the decision function in the neighborhood of the state point by the central difference method.

[0134] When any one of the following conditions is met, the video analysis server inputs the transmittance decay rate, the edge sharpness loss rate, the dispersion shift rate, and the transmittance, the edge sharpness value, and the dispersion characteristic value at the current time into the pollution classifier to output the pollution type, and generates a cleaning instruction according to the pollution type, and then sends the cleaning instruction to the cleaning execution module;

[0135] Condition 1, the instantaneous offset or the predicted offset is less than the instantaneous offset threshold; Condition 2, the transmittance decay rate, the edge sharpness loss rate and the chromatic dispersion offset rate are all greater than the corresponding rate threshold; Condition 3, the transmittance decay trajectory, the edge sharpness loss trajectory and the chromatic dispersion offset trajectory cross the optical performance state tolerance region; Condition 4, the ratio of the predicted offset to the instantaneous offset is greater than the preset proportion threshold, and the predicted offset is less than K times the instantaneous offset threshold.

[0136] Optionally, the transmittance decay rate, the edge sharpness loss rate and the chromatic dispersion offset rate are all greater than the corresponding rate threshold, that is, the transmittance decay rate is greater than the transmittance decay rate threshold, the edge sharpness loss rate is greater than the edge sharpness loss rate threshold, and the chromatic dispersion offset rate is greater than the chromatic dispersion offset rate threshold at the same time.

[0137] Optionally, K can be 2.

[0138] Optionally, the instantaneous offset threshold and the preset proportion threshold are obtained according to experiments.

[0139] Optionally, when the instantaneous offset is less than the instantaneous offset threshold, it indicates that the current lens has been contaminated and needs to be cleaned immediately to prevent imaging failure.

[0140] Optionally, when the predicted offset is less than the instantaneous offset threshold, it indicates that the lens will be contaminated in the future time window, and needs to be cleaned immediately to prevent imaging failure.

[0141] Optionally, the transmittance decay rate, the edge sharpness loss rate and the chromatic dispersion offset rate are all greater than the corresponding rate threshold, indicating that the contamination has entered an exponential deterioration stage and needs to be cleaned immediately.

[0142] Optionally, the transmittance decay trajectory, the edge sharpness loss trajectory and the chromatic dispersion offset trajectory cross the optical performance state tolerance region indicate that the contamination continues to break through the safety zone and needs to be cleaned to restore the cleaning state.

[0143] Preferably, the video analysis server detecting whether the transmittance decay trajectory, the edge sharpness loss trajectory and the chromatic dispersion offset trajectory cross the optical performance state tolerance region comprises:

[0144] Aligning the transmittance decay trajectory, the edge sharpness loss trajectory and the chromatic dispersion offset trajectory according to time nodes, and generating a three-dimensional coordinate point containing a transmittance value, an edge sharpness value and a chromatic dispersion characteristic value at each time node;

[0145] Connecting the three-dimensional coordinate points in time sequence to form a space motion trajectory line, and discretely sampling the space motion trajectory line to obtain a trajectory point set;

[0146] Mapping all the trajectory points to the transmittance-edge sharpness-dispersion three-dimensional feature space obtains a trajectory feature point set, and the spatial positional relationship between the trajectory feature points and the optical performance state tolerance zone is judged point by point:

[0147] When there are two continuous trajectory feature points in the trajectory feature point set satisfying the following conditions, it is determined that the tolerance zone crossing occurs: the previous trajectory feature point is located inside the tolerance zone and the latter trajectory feature point is located outside the tolerance zone, or the previous trajectory feature point is located outside the tolerance zone and the latter trajectory feature point is located inside the tolerance zone.

[0148] The inside and outside positions of the trajectory feature points are determined by calculating the function value of the determination function; when the function value is greater than zero, it is determined that the trajectory feature point is located outside the optical performance state tolerance zone; when the function value is less than or equal to zero, it is determined that the trajectory feature point is located inside the optical performance state tolerance zone.

[0149] Preferably, full-cycle image sequences from a clean state to an optical failure of dust pollution, oil film pollution and water stain pollution are collected in the contaminated image samples respectively to obtain dust image sequences, oil film image sequences and water stain image sequences;

[0150] For each image sequence, the transmittance, edge sharpness value, dispersion characteristic value and distance from the nearest time window of each time point transmittance decay rate, edge sharpness loss rate and dispersion shift rate are extracted;

[0151] The transmittance, edge sharpness value, dispersion characteristic value and distance from the nearest time window of each time point transmittance decay rate, edge sharpness loss rate and dispersion shift rate are combined into a feature vector, and the corresponding pollution type is labeled;

[0152] A neural network classification model is trained using the feature vector and the corresponding pollution type label to output the pollution type;

[0153] After training, the neural network classification model is deployed as a pollution classifier.

[0154] Optionally, the pollution type includes oil film pollution, dust pollution and water stain pollution.

[0155] The cleaning execution module executes the corresponding cleaning strategy according to the received cleaning instruction, and controls the video acquisition module to switch to the standby monitoring camera.

[0156] Optionally, the cleaning strategy includes: dust pollution is cleaned by activating ion wind; oil film pollution is cleaned by triggering organic solvent spraying and silica gel scraping; water stain pollution is cleaned by starting hot air drying and ultrasonic vibration.

[0157] Before cleaning: turn off the infrared fill light of the monitoring camera, switch to short exposure mode to prevent overexposure of the cleaning liquid reflection; during cleaning: enable the standby camera relay monitoring; after cleaning: re-detect the light transmittance, edge sharpness value and dispersion characteristic value to detect whether the cleaning is successful, and if not, start secondary cleaning or report failure.

[0158] The cleaning execution module controls the monitoring camera to turn off the infrared fill light through the ONVIF protocol before cleaning the monitoring camera.

[0159] The application can accurately perceive the lens pollution state by constructing an optical performance tolerance zone and detecting the change trend of light transmittance, edge sharpness and dispersion characteristics in real time, and actively triggers cleaning before the imaging quality significantly decreases, ensures the continuous output of high-quality video stream of the monitoring system, and avoids image blur, dispersion or detail loss caused by lens pollution. The image usability and accuracy of the monitoring system in complex environments are improved.

[0160] The pollution classifier automatically identifies pollution types such as dust, oil film and water stains, and intelligently matches differentiated cleaning strategies, so that the targeted cleaning solution significantly improves the cleaning efficiency, avoids damage to the lens or mechanical structure caused by single cleaning method, prolongs the service life of the equipment, and reduces cleaning agent waste and operation and maintenance costs.

[0161] Computer program instructions for carrying out operations of the present application can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, and procedural programming languages such as "C" language or similar programming languages. Computer readable program instructions can be executed entirely on a user computer, partially on a user computer, as a separate software package, partially on a user computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, connected through the Internet using an Internet service provider). In some embodiments, by utilizing the state information of computer readable program instructions to individualize customize electronic circuits, such as programmable logic circuits, field programmable gate arrays (FPGA) or programmable logic arrays (PLA), the electronic circuits can execute computer readable program instructions to implement various aspects of the present application.

[0162] The non-transitory computer readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the processor executes the method.

[0163] Those skilled in the art can understand that all or part of the steps of the above method can be instructed by a program to complete the relevant hardware (for example, a processor, an FPGA, an ASIC, etc.), and the program can be stored in a readable storage medium, such as a read-only memory, a magnetic disk or an optical disk, etc. All or part of the steps of the above embodiment can also be implemented by using one or more integrated circuits. Accordingly, each module in the above embodiment can be implemented in the form of hardware, for example, by an integrated circuit to implement its corresponding function, or can be implemented in the form of a software function module, for example, by a processor executing a program / instruction stored in a memory to implement its corresponding function. The embodiments of the present application are not limited to any specific form of combination of hardware and software.

[0164] In addition, each functional unit in each embodiment herein can be integrated in one processing unit, or each unit can exist physically independently, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0165] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical solutions herein, essentially or in the form of a contribution to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment herein. The foregoing storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, etc.

[0166] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of protection of the present application.

Claims

1. An intelligent self-cleaning security monitoring system, characterized in that, The system comprises a video acquisition module, a video analysis server, a cleaning execution module and a database, and the modules are communicatively connected; The video analysis server establishes an optical performance state tolerance zone based on pollution image samples obtained from the database, and extracts light transmittance decay rate threshold, edge sharpness loss rate threshold and chromatic dispersion shift rate threshold from the database; The video acquisition module acquires a monitoring video stream and transmits the monitoring video stream to the video analysis server through an RTSP protocol; The video analysis server extracts the light transmittance, edge sharpness value and chromatic dispersion characteristic value of a monitoring image at the current time from the monitoring video stream, and calculates the light transmittance decay rate, edge sharpness loss rate and chromatic dispersion shift rate based on video data in a recent time window; The video analysis server obtains a real-time instantaneous state point by mapping the light transmittance, edge sharpness value and chromatic dispersion characteristic value at the current time to a light transmittance-edge sharpness-chromatic dispersion three-dimensional feature space; The video analysis server predicts the light transmittance, edge sharpness value and chromatic dispersion characteristic value at each time node in the next time window, and generates a light transmittance decay trajectory, an edge sharpness loss trajectory and a chromatic dispersion shift trajectory; The video analysis server obtains a predicted instantaneous state point by mapping the light transmittance, edge sharpness value and chromatic dispersion characteristic value at the end of the next time window to the light transmittance-edge sharpness-chromatic dispersion three-dimensional feature space; The video analysis server calculates the distance from the real-time instantaneous state point to the boundary of the optical performance state tolerance zone as an instantaneous offset, and calculates the distance from the predicted instantaneous state point to the boundary of the optical performance state tolerance zone as a predicted offset; When any one of the following conditions is met, the video analysis server inputs the light transmittance decay rate, edge sharpness loss rate, chromatic dispersion shift rate and light transmittance, edge sharpness value and chromatic dispersion characteristic value at the current time into a pollution classifier to output a pollution type, generates a cleaning instruction according to the pollution type, and then sends the cleaning instruction to the cleaning execution module; Condition 1: the instantaneous offset or the predicted offset is less than an instantaneous offset threshold; Condition 2: the light transmittance decay rate, edge sharpness loss rate and chromatic dispersion shift rate are all greater than the corresponding rate threshold; Condition 3: the light transmittance decay trajectory, edge sharpness loss trajectory and chromatic dispersion shift trajectory cross the optical performance state tolerance zone; Condition 4: the ratio of the predicted offset to the instantaneous offset is greater than a preset proportion threshold, and the predicted offset is less than K times the instantaneous offset threshold; The cleaning execution module executes a corresponding cleaning strategy according to the received cleaning instruction, and controls the video acquisition module to switch to a backup monitoring camera. The pollution type includes oil film pollution, dust pollution and water stain pollution.

2. The system of claim 1, wherein, The cleaning execution module controls the monitoring camera to turn off infrared fill light through an ONVIF protocol before cleaning the monitoring camera.

3. The system of claim 2, wherein, The video analysis server extracts the light transmittance, edge sharpness value and chromatic dispersion characteristic value of a monitoring image at the current time from the monitoring video stream, which includes:

4. The system of claim 3, wherein, Decoding the monitoring image at the current time from the monitoring video stream; Converting the monitoring image to an HSV color space and extracting pixel values of a brightness channel; Calculating the arithmetic mean of all pixel values of the brightness channel as the light transmittance; ​ The monitoring image is converted into a gray image, and a Sobel edge detection operator is applied to calculate the gradient amplitude of the image; the pixels with gradient amplitude greater than the edge determination threshold are counted as edge pixels; the arithmetic mean of the gradient amplitude of all edge pixels is calculated as the edge sharpness value; The monitoring image is converted to CIELAB color space, and the standard deviation of the pixel values of the red-green color axis and the yellow-blue color axis is calculated respectively; the standard deviation of the red-green color axis and the standard deviation of the yellow-blue color axis are added to obtain the color dispersion characteristic value representing the color dispersion state at the current time.

5. The system of claim 4, wherein, The video analysis server establishes an optical performance state tolerance region based on the pollution image sample, which includes: The video analysis server collects dust pollution, oil film pollution and water stain pollution from the clean state to the full cycle image sequence of optical failure in the pollution image sample to obtain the dust image sequence, the oil film image sequence and the water stain image sequence; The video analysis server respectively labels the transmittance value, edge sharpness value and color dispersion characteristic value and time stamp of each frame of pollution image in the dust image sequence, oil film image sequence and water stain image sequence; The video analysis server connects the transmittance of the dust image sequence in time sequence to obtain the dust transmittance curve; connects the transmittance of the oil film image sequence in time sequence to obtain the oil film transmittance curve; connects the transmittance of the water stain image sequence in time sequence to obtain the water stain transmittance curve; The video analysis server connects the edge sharpness value of the dust image sequence in time sequence to obtain the dust edge sharpness curve; connects the edge sharpness value of the oil film image sequence in time sequence to obtain the oil film edge sharpness curve; connects the edge sharpness value of the water stain image sequence in time sequence to obtain the water stain edge sharpness curve; The video analysis server connects the color dispersion characteristic value of the dust image sequence in time sequence to obtain the dust color dispersion characteristic value curve; connects the color dispersion characteristic value of the oil film image sequence in time sequence to obtain the oil film color dispersion characteristic value curve; connects the color dispersion characteristic value of the water stain image sequence in time sequence to obtain the water stain color dispersion characteristic value curve.

6. The system of claim 5, wherein, The video analysis server establishes an optical performance state tolerance region based on the pollution image sample, which includes: Identify the points in the dust transmittance curve whose first derivative absolute value is greater than the mutation threshold to obtain a plurality of dust transmittance key points; identify the local minimum points of the first derivative of the dust edge sharpness curve to obtain a plurality of dust edge sharpness key points; identify the points on the dust color dispersion characteristic value curve whose curvature value is greater than the curvature threshold to obtain a plurality of dust color dispersion key points; Identify the points in the oil film transmittance curve whose second derivative absolute value is less than the convergence threshold to obtain a plurality of oil film transmittance key points; identify the points in the oil film edge sharpness curve whose second derivative absolute value is less than the convergence threshold to obtain a plurality of oil film edge sharpness key points; identify the points in the oil film color dispersion characteristic value curve whose second derivative absolute value is less than the convergence threshold to obtain a plurality of oil film color dispersion key points; The left limit and the right limit difference in the water penetration light transmittance curve are identified to obtain a plurality of water penetration light transmittance key points; the local minimum points of the second derivative in the water penetration edge sharpness curve are identified to obtain a plurality of water penetration edge sharpness key points; the points where the curvature derivative changes sign in the water penetration color dispersion characteristic value curve are identified to obtain a plurality of water penetration color dispersion key points; All light transmittance key points, edge sharpness key points and color dispersion key points of the three types of pollution of dust, oil film and water penetration are uniformly mapped to a light transmittance-edge sharpness-color dispersion three-dimensional space according to time stamps to form a three-dimensional key point set; An interpolation algorithm is used on the three-dimensional key point set to generate a three-dimensional point cloud covering the full pollution state of dust, oil film and water penetration, and a minimum convex polyhedron is constructed based on the three-dimensional point cloud, and the internal region of the polyhedron is defined as an optical performance state tolerance region; The boundary surface of the convex polyhedron is extracted, and a radial basis function is used to perform implicit function fitting on the boundary surface to obtain a decision function; the decision function satisfies that when the function value is less than or equal to zero, the point is located inside the optical performance state tolerance region; when the function value is greater than zero, the point is located outside the optical performance state tolerance region.

7. The system of claim 6, wherein, The video analysis server detects whether the light transmittance attenuation trajectory, the edge sharpness loss trajectory and the color dispersion offset trajectory cross the optical performance state tolerance region, including: Aligning the light transmittance attenuation trajectory, the edge sharpness loss trajectory and the color dispersion offset trajectory according to time nodes, and generating a three-dimensional coordinate point containing the light transmittance value, the edge sharpness value and the color dispersion characteristic value at each time node; Connecting the three-dimensional coordinate points in time sequence to form a space motion trajectory line, and discretely sampling the space motion trajectory line to obtain a trajectory point set; Mapping all trajectory points to a light transmittance-edge sharpness-color dispersion three-dimensional feature space to obtain a trajectory feature point set, and judging the spatial position relationship between the trajectory feature points and the optical performance state tolerance region point by point: When there are two continuous trajectory feature points in the trajectory feature point set that satisfy the following conditions, it is determined that the tolerance region crossing occurs: the previous trajectory feature point is located inside the tolerance region and the next trajectory feature point is located outside the tolerance region, or the previous trajectory feature point is located outside the tolerance region and the next trajectory feature point is located inside the tolerance region; The inside and outside positions of the trajectory feature points are determined by calculating the function value of the decision function; when the function value is greater than zero, it is determined that the trajectory feature point is located outside the optical performance state tolerance region; when the function value is less than or equal to zero, it is determined that the trajectory feature point is located inside the optical performance state tolerance region.

8. The system of claim 7, wherein, The video analysis server calculates the light transmittance attenuation rate, the edge sharpness loss rate and the color dispersion offset rate based on the video data in the latest time window, including: Obtaining a plurality of monitoring images at consecutive time points in the latest time window, extracting the light transmittance at each time point to form a light transmittance sequence, extracting the edge sharpness value at each time point to form an edge sharpness sequence, and extracting the color dispersion characteristic value at each time point to form a color dispersion characteristic value sequence; The following operations are performed on the light transmittance sequence: calculating the difference between the light transmittance at adjacent time points; dividing each difference by the corresponding time interval to obtain the instantaneous light transmittance attenuation rate; and taking the arithmetic average of all instantaneous light transmittance attenuation rates as the light transmittance attenuation rate; The edge sharpness sequence is operated as follows: the difference between the edge sharpness values at adjacent time points in the edge sharpness sequence is calculated; each difference is divided by the corresponding time interval to obtain an instantaneous edge sharpness loss rate; and the arithmetic mean of all the instantaneous edge sharpness loss rates is taken as the edge sharpness loss rate; The dispersion characteristic value sequence is operated as follows: the difference between the dispersion characteristic values at adjacent time points in the dispersion characteristic value sequence is calculated; each difference is divided by the corresponding time interval to obtain an instantaneous dispersion shift rate; and the arithmetic mean of all the instantaneous dispersion shift rates is taken as the dispersion shift rate.

9. The system of claim 8, wherein, The video analysis server predicts the light transmittance, edge sharpness value and dispersion characteristic value at each time node in the next time window, and generates the light transmittance decay trajectory, edge sharpness loss trajectory and dispersion shift trajectory, comprising: The light transmittance decay rate, edge sharpness loss rate and dispersion shift rate and the light transmittance, edge sharpness value and dispersion characteristic value at the current time are input into the LSTM prediction model trained based on the pollution image sample to output the light transmittance, edge sharpness value and dispersion characteristic value at each time node in the next time window, and the light transmittance, edge sharpness value and dispersion characteristic value at each time node are connected to generate the light transmittance decay trajectory, edge sharpness loss trajectory and dispersion shift trajectory, respectively.

Images