Method for monitoring and early warning of hazardous waste treatment based on video analysis

Abstract

The application belongs to the field of video recognition, and particularly relates to a dangerous waste treatment monitoring and early warning method based on video analysis, which comprises the following steps: performing mixed Gaussian background modeling on a video to obtain a predicted gray value of a target position in a current frame; obtaining scene local stability of the target position in the current frame; determining dynamic abnormality of the target position in the current frame, and marking a candidate abnormal area in the current frame according to the dynamic abnormality; obtaining a motion vector of each pixel point in the current frame, and determining motion consistency of the candidate abnormal area; determining abnormal confidence of the candidate abnormal area according to the motion consistency and average dynamic abnormality of the pixel points in the candidate abnormal area; and performing dangerous waste risk assessment according to the abnormal confidence. The application solves the technical problem that video analysis of dangerous waste treatment is easily disturbed by dynamic interference generated in normal production, and improves the accuracy and reliability of dangerous waste treatment monitoring and early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of video recognition. More specifically, the present application relates to a hazardous waste treatment monitoring and early warning method based on video analysis. BACKGROUND

[0002] The hazardous waste treatment link is related to environmental and personnel safety, so a video monitoring system is usually deployed at the treatment site to find leaks, burns and other abnormal situations in real time. In the prior art, video analysis methods based on image features are mainly used, such as identifying abnormalities through color models, texture features or morphological means, in order to provide monitoring and early warning at the first time. These methods have certain effect in general industrial scenes, but are limited when facing the complex working conditions of hazardous waste treatment.

[0003] Existing video analysis methods often rely on fixed thresholds or preset rules to judge the light, color or texture features of the image. This approach is easily disturbed by normal production activities in the treatment site, such as dynamic changes in water vapor, dust or light reflection, which can be misjudged as risk events. At the same time, when the characteristics of abnormal events are weak or atypical, these methods may not be able to accurately identify them, resulting in missed detection. At the same time, when the real leakage or burning event is in the early stage, its visual features are weak or atypical, and these methods may not be able to accurately identify them due to the signal not reaching the preset intensity, resulting in missed detection.

[0004] The root cause of this problem is that traditional algorithms lack the ability to adapt to dynamic environments. The hazardous waste treatment site is not a stable background environment, but a constantly changing dynamic scene. Algorithms based on fixed parameters and prior rules cannot distinguish between normal unstable phenomena and real abnormalities. As a result, false positives are frequent when the environmental noise is high, and real risks are easily missed when the signal is not obvious, making it difficult to meet safety needs. SUMMARY

[0005] To solve the technical problem that the video analysis of hazardous waste treatment is easily disturbed by dynamic interference from normal production, the present application provides a hazardous waste treatment monitoring and early warning method based on video analysis, comprising:

[0006] A Gaussian mixture (GJ) background model is performed on videos of key monitoring points at hazardous waste treatment facilities to obtain a GJ model for each location in the video. Taking any pixel location in the video as the target location, the mean parameter of the sub-Gaussian model that best matches the grayscale value of the target location in the previous frame is obtained and used as the predicted grayscale value of the target location in the current frame. The grayscale values ​​of the target location in the current frame and several previous frames are used to construct a grayscale sequence for the target location. Based on the mean and standard deviation of the grayscale sequence, and the mean and standard deviation of the differences between two adjacent grayscale values ​​in the grayscale sequence, the scene of the target location in the current frame is determined. Local stability; based on the difference between the predicted and actual grayscale values ​​of the target location in the current frame, and the local stability of the scene, the dynamic anomaly degree of the target location in the current frame is determined, and candidate anomaly regions are marked in the current frame according to the magnitude of the dynamic anomaly degree; the motion vector of each pixel in the current frame is obtained using optical flow, and the motion consistency of the candidate anomaly regions is determined based on the proportion of pixels in the candidate anomaly regions whose vertical components of the motion vectors are greater than 0; the anomaly confidence of the candidate anomaly regions is determined based on the motion consistency and the average dynamic anomaly degree of the pixels in the candidate anomaly regions; and a hazardous waste risk assessment is performed based on the magnitude of the anomaly confidence.

[0007] This invention improves the accuracy and reliability of video monitoring and early warning in complex and dynamic environments such as hazardous waste treatment. First, it calculates a dynamic anomaly score that can suppress normal production disturbances such as water vapor by using a Gaussian mixture model and scene local stability analysis. Second, it uses motion consistency to assess whether the motion pattern of candidate anomaly areas conforms to the physical law of smoke rising upwards. By combining the anomaly score with the smoke rising model, this invention can effectively distinguish between real risk events and physically meaningless artifacts, overcoming the limitations of traditional fixed threshold algorithms. It can improve the accuracy and reliability of monitoring real hazardous events while reducing the false alarm rate.

[0008] Preferably, the step of obtaining the mean parameter of the sub-Gaussian model in the Gaussian mixture model of the target location that best matches the grayscale value of the target location in the previous frame, and using it as the predicted grayscale value of the target location in the current frame, includes: if the grayscale value of the target location in the previous frame has the smallest difference from the mean of a certain sub-Gaussian model, and is between the mean-centered value of the sub-Gaussian model with the smallest mean difference... A match is considered successful if the target position is within a confidence interval of one-times the standard deviation, and the mean of the matched sub-Gaussian model is taken as the predicted gray value of the target position in the current frame. If the gray value of the target position in the previous frame does not match any of the sub-Gaussian models, the predicted gray value of the nearest temporally matched pixel at the same position is selected as the predicted gray value of the target position in the current frame.

[0009] The application matches the gray value of the previous frame with multiple sub-models in the Gaussian mixture model, and selects the mean value of the most matched sub-model as the prediction value of the current frame, the mean value represents the most stable and core gray value of the background state, using the mean value as the prediction value can effectively filter out random noise that may exist in the previous frame image, thereby providing a more stable and more accurate background reference than directly using the previous frame gray value; when all sub-Gaussian models do not match, the prediction value of the last successful matching at this position is selected as the current prediction value, ensuring that even in the case that the pixel point does not belong to any known background model, an effective prediction value can still be provided, avoiding the problem of subsequent calculation interruption caused by matching failure, and ensuring the continuity of monitoring and analysis.

[0010] Preferably, the expression of the scene local stability is: ; wherein, is the frame number of the current frame, is the target position The scene local stability in the current frame, is a preset reference fluctuation benchmark, and respectively represent the mean value and the standard deviation of the gray value of the target position in a window with the current frame as the end frame, and are the mean value and the standard deviation of the absolute value of the gray value change between adjacent frames of the target position in a short-time window with the current frame as the end frame, is a hyperparameter to prevent the denominator from being 0.

[0011] The application evaluates the scene local stability by calculating the mean value and the standard deviation of the gray value of the target position in the time window, as well as the mean value and the standard deviation of the absolute value of the gray value change between adjacent frames, in a high brightness area, a larger gray fluctuation may be considered as normal noise, while in a darker area, the same fluctuation may be a significant event, the application can also distinguish between smooth and uniform changes and sudden and irregular disturbances, so that when calculating the dynamic anomaly degree, it can effectively distinguish between gray value instability caused by normal production activities and instability caused by real anomalies, thereby reducing false positives in complex dynamic environments and improving the accuracy of monitoring and early warning.

[0012] Preferably, the expression of the dynamic anomaly degree is: ; wherein, is the frame number of the current frame, is the target position The dynamic anomaly degree in the current frame, is the gray value of the target position in the current frame; is the prediction gray value of the target position in the current frame, is the target position is the scene local stability of the current frame, is the parameter for adjusting the influence of the scene local stability, is the absolute value symbol.

[0013] The application can keep the system highly sensitive to weak real abnormal signals by calculating the dynamic anomaly degree, and the weight of unstable regions with normal water vapor, dust and other dynamic interference can be reduced, thereby reducing false positives of normal fluctuations, and distinguishing real risks from normal environmental interference.

[0014] Preferably, the candidate abnormal region is marked in the current frame according to the size of the dynamic anomaly degree, comprising: in response to the dynamic anomaly degree of the target position in the current frame being greater than a preset abnormal threshold, marking the target position in the current frame; using a connected domain algorithm to aggregate spatially adjacent marked points into a candidate abnormal region.

[0015] Preferably, the motion consistency of the candidate abnormal region is determined, comprising: using an optical flow method on the current frame and the previous continuous video frame to obtain the motion vector of each pixel point; calculating the ratio of the number of motion vectors with positive vertical components in the candidate abnormal region to the total number of motion vectors in the candidate abnormal region as the motion consistency of the candidate abnormal region.

[0016] Preferably, the expression of the abnormal confidence is: ; wherein, is the frame number of the current frame, represents the candidate abnormal region in the current frame, represents the candidate abnormal region pixel number, represents the pixel point in the candidate abnormal region, is the dynamic anomaly degree of the pixel point in the current frame, is the motion consistency of the candidate abnormal region in the current frame. The abnormal confidence of the application is obtained by multiplying the average dynamic anomaly degree and the motion consistency of the candidate abnormal region. Only when a candidate abnormal region meets both conditions of being abnormal and conforming to the real smoke model, the abnormal confidence will be high, ensuring that the early warning system is only for real and physically meaningful risk events, and improving the accuracy and reliability of monitoring.

[0017]

[0018] ​Preferably, the risk assessment of the hazardous waste according to the abnormal confidence size comprises: presetting a plurality of risk threshold values; collecting a large amount of abnormal confidence under normal working conditions, and performing threshold value classification on the data to obtain 95% quantile, 99% quantile and 99.9% quantile; in response to the abnormal confidence of the candidate abnormal region being less than or equal to the 95% quantile, the hazardous waste treatment is risk-free; in response to the abnormal confidence of the candidate abnormal region being greater than the 95% quantile and less than the 99% quantile, the hazardous waste treatment is low-risk; in response to the abnormal confidence of the candidate abnormal region being greater than or equal to the 99% quantile and less than the 99.9% quantile, the hazardous waste treatment is medium-risk; and in response to the abnormal confidence of the candidate abnormal region being greater than or equal to the 99.9% quantile, the hazardous waste treatment is high-risk.

[0019] Preferably, the video of the key monitoring point of the hazardous waste treatment facility is subjected to mixed Gaussian background modeling, and the number of sub-Gaussian models is 5.

[0020] Preferably, the gray value of the target position in the current frame and a plurality of previous frames is used to form a gray sequence of the target position, and the gray value of the target position in the current frame and 50 previous frames is used to form the gray sequence of the target position.

[0021] The present application has the following advantages:

[0022] The present application can make the sensitivity of the anomaly detection adapt to the changes of environmental factors such as light and water vapor in real time, solve the problem of high false alarm rate of the traditional fixed threshold algorithm in a complex dynamic environment, and has strong environmental adaptability and robustness.

[0023] The present application can distinguish between real physical events and non-physical artifacts such as changes in light, accurately distinguish between real risks and normal environmental interference, and reduce the false alarm rate caused by random noise and non-critical events and the missed detection rate of features that are not obvious.

[0024] The present application can effectively distinguish between real physical events and non-physical artifacts such as changes in light, accurately distinguish between real risks and normal environmental interference, and reduce the false alarm rate caused by random noise and non-critical events and the missed detection rate of features that are not obvious. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart schematically showing a hazardous waste treatment monitoring and early warning method based on video analysis in the present application;

[0026] Figure 2 is a flowchart schematically showing S4 in the present application. DETAILED DESCRIPTION

[0027] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of the present application.

[0028] The specific implementation of the present application will be described in detail below with reference to the drawings.

[0029] The embodiments of the present application disclose a hazardous waste treatment monitoring and early warning method based on video analysis, referring to Figure 1 , comprising steps S1 to S5:

[0030] S1, Gaussian mixture background modeling is performed on the video of the key monitoring point of the hazardous waste treatment facility, and the mean value parameter of the sub-Gaussian model that best matches the gray value of the target position in the previous frame is obtained as the predicted gray value of the target position in the current frame.

[0031] Specifically, a visible light camera is deployed at the key monitoring point of the hazardous waste treatment facility, and the key monitoring point includes the feeding port, the outer wall of the incinerator, the periphery of the chemical reaction kettle, and the ground of the waste storage area.

[0032] The visible light camera continuously acquires video stream data, and decodes the video stream into a sequence of image frames indexed by frame number.

[0033] Each frame of image is preprocessed, including image denoising and contrast enhancement, to obtain clear image data for subsequent analysis. In this embodiment, the image denoising adopts a Gaussian denoising algorithm, and the contrast enhancement adopts a histogram equalization algorithm. Implementers can also use other denoising and contrast enhancement algorithms, such as bilateral filtering algorithm and gamma transformation algorithm.

[0034] The video is subjected to Gaussian mixture background modeling, and the number of sub-Gaussian models is set to 5 to obtain the Gaussian mixture model of each pixel position in the video. Taking any pixel position as a target position, if the gray value of the target position in the previous frame is the smallest difference with the mean value of a certain sub-Gaussian model, and is within the confidence interval of twice the standard deviation centered on the mean value of the sub-Gaussian model with the smallest mean value difference, it is considered as a successful match. The mean value of the matched sub-Gaussian model is regarded as the predicted gray value of the target position in the current frame. If the gray value of the target position in the previous frame does not match all the sub-Gaussian models, the predicted gray value of the successfully matched pixel point of the same position in the nearest time sequence is selected as the predicted gray value of the target position in the current frame. In this embodiment, five sub-Gaussian models and twice the standard deviation of the confidence interval are used for illustration. Implementers can also adjust the number of sub-Gaussian models and the multiple of standard deviation of the confidence interval according to actual conditions.

[0034] The video is subjected to Gaussian mixture background modeling, and the number of sub-Gaussian models is set to 5 to obtain the Gaussian mixture model of each pixel position in the video. Taking any pixel position as a target position, if the gray value of the target position in the previous frame is the smallest difference with the mean value of a certain sub-Gaussian model, and is within the confidence interval of twice the standard deviation centered on the mean value of the sub-Gaussian model with the smallest mean value difference, it is considered as a successful match. The mean value of the matched sub-Gaussian model is regarded as the predicted gray value of the target position in the current frame. If the gray value of the target position in the previous frame does not match all the sub-Gaussian models, the predicted gray value of the successfully matched pixel point of the same position in the nearest time sequence is selected as the predicted gray value of the target position in the current frame. In this embodiment, five sub-Gaussian models and twice the standard deviation of the confidence interval are used for illustration. Implementers can also adjust the number of sub-Gaussian models and the multiple of standard deviation of the confidence interval according to actual conditions.

[0035] S2, the gray value of the target position in the current frame and several previous frames constitutes a gray sequence of the target position, and the scene local stability of the target position in the current frame is determined according to the mean and standard deviation of the gray sequence, the mean and standard deviation of the difference between two adjacent gray values in the gray sequence.

[0036] It should be noted that when the hazardous waste treatment does not appear to be abnormal, the gray value of the target position in the corresponding time window is stably distributed, and when the hazardous waste treatment may appear to be abnormal, the gray value of the target position in the corresponding time window will fluctuate, so the stability of the gray value of the target position in the corresponding time window is measured.

[0037] Specifically, the scene local stability of the target position of the current frame image of the video is constructed as follows:

[0038] ;

[0039] Wherein, is the frame number of the current frame, is the target position in the scene local stability of the current frame, is a preset reference fluctuation benchmark, an empirical value , to avoid division by zero error in the extreme case of complete stillness, and as a normalization benchmark, the comprehensive fluctuation in the brackets is mapped to the final stability score, and respectively represent the mean and standard deviation of the gray value of the target position in a 50-frame window with the current frame as the end frame, and is the mean and standard deviation of the absolute value of the gray value change between adjacent frames of the target position in a short-time window with the current frame as the end frame, is a hyperparameter to prevent the denominator from being 0, an empirical value In the embodiment , , the length of the short-time window is 50 frames, and in other embodiments, the reference fluctuation benchmark, the size of the hyperparameter, and the length of the short-time window can be adjusted according to actual conditions.

[0040] Wherein, represents the relative change of the gray value of the target position. For example, for a bright area with an average gray value of 200, a fluctuation of 10 gray values may only be normal noise; but for a dark area with an average gray value of 20, the same fluctuation of 10 gray values may be a significant event, so that the stability evaluation has adaptability to different gray areas;

[0041] The stationarity of the change is used to measure the change. If the scene changes uniformly and slowly, the inter-frame gray value change of the target position in each frame will be close to each other, and the standard deviation of the inter-frame gray value change will be small, and the value of the item will also be small; conversely, if the scene changes unevenly and rapidly, the value of the inter-frame gray value change will be large and small alternately, the standard deviation of the inter-frame gray value change will be large, and the value of the item will be large, The stationarity of the scene local region of the target position in the current frame is in the range of (0, 1].

[0042] The closer the value is to 1, the more stable the target position is in the time window; The closer the value is to 0, the less stable the target position is in the time window.

[0043] S3, according to the difference between the predicted gray value and the actual gray value of the target position in the current frame, and the stationarity of the scene local region, the dynamic abnormality degree of the target position in the current frame is determined, which is used to distinguish between stable and normal instability and abnormal instability, for example, the water vapor generated during the treatment of hazardous waste is normal instability, and the water vapor mixed with color abnormal smoke is abnormal instability.

[0044] Specifically, the dynamic abnormality degree satisfies the expression:

[0045] ;

[0046] wherein, is the frame sequence number of the current frame, is the dynamic abnormality degree of the target position in the current frame, is the gray value of the target position in the current frame; is the predicted gray value of the target position in the current frame, is the stationarity of the scene local region of the target position in the current frame, is a parameter for adjusting the influence of the stationarity of the scene local region, and the empirical value is 1, is the absolute value symbol. In the embodiment , in other embodiments, the implementer can also adjust according to the actual situation.

[0047] When the local region of the scene is highly stable, tends to 1, the operator also tends to , and is low, and the output dynamic abnormality degree is close to 0; when the local region of the scene is unstable tends to 0, tends to 0, when the local area of the scene is less stable, the instability is divided into normal instability and abnormal instability, for example, the water vapor generated during the treatment of hazardous waste is normal instability, and the water vapor mixed with smoke with abnormal color is abnormal instability. The normal instability has been classified as normal through the learning of the Gaussian mixture model, The difference will be very small. For abnormal instability, the new change does not conform to the normal water vapor model identified by the GMM, The difference will be larger.

[0048] Further, the dynamic anomaly degree output in the normal instability state is less than the dynamic anomaly degree output in the abnormal instability state, because even if they both belong to the instability situation, The values are similar, but The values are quite different, so the dynamic anomaly degree output in the abnormal instability state is greater than the dynamic anomaly degree output in the normal instability state .

[0049] It needs to be further explained that when a weak real anomaly appears on an originally stable background, The change is small, but due to the appearance of the anomaly, its value will be slightly lower than that when it is absolutely stable. At this time, the enhancement operator will provide an amplification factor greater than 1 to amplify the originally weak signal.

[0050] S4, marking the current frame candidate anomaly region according to the dynamic anomaly degree, obtaining the motion vector of each pixel point in the current frame by using the optical flow method, determining the motion consistency of the candidate anomaly region according to the vertical component of the motion vector in the candidate anomaly region, and determining the anomaly confidence of the candidate anomaly region according to the motion consistency and the average dynamic anomaly degree of the pixel points in the candidate anomaly region.

[0051] It needs to be explained that the dynamic anomaly degree mainly focuses on visual anomalies, and cannot distinguish non-physical artifacts, such as noise appearing on the camera sensor for a moment, and small blocks caused by encoding errors, which can cause the local dynamic anomaly degree value to be very high, but the noise and small blocks are not the dangerous waste treatment anomalies that need to be concerned. Moreover, the dynamic anomaly degree lacks real changes in context, for example, the momentary switching of a lamp in the distance is a real light and shadow change, but it does not constitute an event that needs to be warned in the scene of hazardous waste. During the treatment of hazardous waste, smoke leakage is an event that needs to be warned by the present application, so the present application performs real-time evaluation on potential abnormal signals according to the smoke rising model, thereby reducing false positives caused by random noise and non-key events and other unreal anomalies.

[0052] The flowchart of step S4 refers to Figure 2 , including steps S401 to S404, specifically:

[0053] S401, screening out a candidate abnormal region from a current frame of the video by using a dynamic abnormality degree.

[0054] Specifically, mark the pixel points with a dynamic abnormality degree greater than a preset abnormal threshold value, aggregate adjacent marked points into an independent candidate abnormal region using a connected domain algorithm, and the experience value of the abnormal threshold value is 30. The implementer can set the abnormal threshold value according to the actual situation.

[0055] S402, obtaining a motion vector of each pixel point of a current frame image of the video by using an optical flow method.

[0056] Specifically, input consecutive video frames, for example, a current frame and a previous frame, calculate the motion direction and speed of each pixel point in the image between the consecutive frames using the optical flow method, and form a two-dimensional motion vector. In this embodiment, the Gunnar Farnebäck optical flow method is adopted to realize it. In other embodiments, the implementer can select an algorithm for obtaining a motion vector of each pixel point of a current frame image of the video according to the actual situation, for example, the Horn-Schunck optical flow method.

[0057] It should be noted that the optical flow method can generate a dense motion vector field for all pixel points in the entire image including the candidate abnormal region. A candidate abnormal region outputs a motion field, which contains the motion vector information of all points in the candidate abnormal region.

[0058] S403, calculating a ratio of the number of motion vectors with a positive vertical component in a candidate abnormal region to the total number of motion vectors in the candidate abnormal region as the motion consistency of the candidate abnormal region.

[0059] Smoke will form a volumetric and continuous region, and in an indoor environment without strong wind, the overall motion trend of the smoke is upward due to the influence of buoyancy; while non-physical artifacts such as random noise points may appear at a certain position in a frame and disappear in the next frame, and the motion is random, disordered and discontinuous. Therefore, the motion consistency of the candidate abnormal region is obtained according to the smoke characteristics.

[0060] Specifically, the total number of motion vectors contained in a candidate abnormal region is denoted as , the total number of motion vectors with a positive vertical component is denoted as , and the motion consistency of the candidate abnormal region is calculated as In the current frame:

[0061] ;

[0062] wherein, is the frame number of the current frame, is the candidate abnormal region the total number of motion vectors with positive vertical component within the candidate abnormal region, is the total number of motion vectors contained in the candidate abnormal region, the greater the ratio of the number of motion vectors with positive vertical component within the candidate abnormal region to the total number of motion vectors in the candidate abnormal region, the more likely the candidate abnormal region is a true smoke uplift model, and vice versa.

[0063] S404, obtaining the abnormal confidence of the candidate abnormal region by using the motion consistency of the candidate abnormal region and the average abnormality of the candidate abnormal region.

[0064] Specifically, the expression of the abnormal confidence is:

[0065] ;

[0066] wherein, is the frame number of the current frame, denotes the abnormal confidence of the candidate abnormal region in the current frame, denotes the number of pixel points of the candidate abnormal region , denotes the number of pixel points of the candidate abnormal region , denotes the pixel point within the candidate abnormal region , is the dynamic abnormality of the pixel point in the current frame, is the motion consistency of the candidate abnormal region in the current frame.

[0067] It should be noted that, the average value of the dynamic abnormality of all pixel points within the candidate abnormal region is calculated, the product of the motion consistency of the candidate abnormal region and the average value of the dynamic abnormality of all pixel points within the candidate abnormal region is calculated. Only when the candidate abnormal region conforms to the smoke uplift model and the abnormality of the candidate abnormal region is high, a higher abnormal confidence is output, indicating that the abnormality of the candidate abnormal region is a true abnormality. Otherwise, the abnormal region does not conform to the true physical model or the abnormality of the candidate abnormal region is low, and a lower abnormal confidence is output, indicating that the abnormality of the candidate abnormal region is not a true abnormality. The true abnormality refers to the abnormality such as smoke, and the untrue abnormality refers to the abnormality such as random noise.

[0068] It needs to be pointed out that although the water vapor during the treatment of hazardous waste may also conform to the smoke floating model in the image of the current frame of the video, the water vapor has been excluded when the candidate abnormal area is screened out from the current frame of the video by using the dynamic anomaly degree in S401, and the candidate abnormal area will not appear water vapor.

[0069] S5, risk assessment of hazardous waste treatment according to the size of the abnormal confidence.

[0070] Specifically, a plurality of risk threshold values are preset: a large number of abnormal confidence under normal working conditions are collected, and the data are classified threshold values: 95% quantile, 99% quantile and 99.9% quantile.

[0071] In response to the abnormal confidence of the candidate abnormal area being less than or equal to 95% quantile, the hazardous waste treatment is risk-free; in response to the abnormal confidence of the candidate abnormal area being greater than 95% quantile and less than 99% quantile, the hazardous waste treatment is low-risk; in response to the abnormal confidence of the candidate abnormal area being greater than or equal to 99% quantile and less than 99.9% quantile, the hazardous waste treatment is medium-risk; and in response to the abnormal confidence of the candidate abnormal area being greater than or equal to 99.9% quantile, the hazardous waste treatment is high-risk.

[0072] Thus, the risk assessment of hazardous waste treatment is realized.

Claims

1. A hazardous waste treatment monitoring and early warning method based on video analysis, characterized in that, The application relates to a method for monitoring a hazardous waste treatment facility. The method comprises the following steps: Gaussian mixture background modeling is performed on a video of a key monitoring point of the hazardous waste treatment facility to obtain a Gaussian mixture model of each position in the video; any pixel position in the video is taken as a target position, and a mean value parameter of a sub-Gaussian model in the Gaussian mixture model of the target position which is most matched with a gray value of the target position in a previous frame is taken as a predicted gray value of the target position in a current frame; The gray value of the target position in the current frame and several previous frames is constituted as a gray sequence of the target position, and the local stability of the target position in the current frame is determined according to the mean value and the standard deviation of the gray sequence, the mean value and the standard deviation of the difference between two adjacent gray values in the gray sequence, and the local stability of the scene in the current frame, comprising: ; wherein, is the frame number of the current frame, is the target position in the current frame, is the local stability of the scene in the current frame, is a preset reference fluctuation benchmark, and respectively represent the mean value and the standard deviation of the gray value of the target position in a window with the current frame as the end frame, and is an over-parameter to prevent the denominator from being 0. According to the difference between the predicted gray value and the actual gray value of the target position in the current frame and the local stability of the scene, a dynamic abnormality degree of the target position in the current frame is determined, comprising: ; wherein, is the gray value of the target position in the current frame, is the gray value of the target position in the current frame; is the predicted gray value of the target position in the current frame, is a parameter for adjusting the influence of the local stability of the scene, is an absolute value symbol, and the candidate abnormal area in the current frame is marked according to the size of the dynamic abnormality degree. The method comprises the following steps: a motion vector of each pixel point in the current frame is obtained by using an optical flow method; motion consistency of the candidate abnormal region is determined according to a proportion of pixel points with a vertical component greater than 0 in the motion vector of the candidate abnormal region; abnormal confidence of the candidate abnormal region is determined according to the motion consistency and an average dynamic abnormality degree of the pixel points in the candidate abnormal region; and hazardous waste risk assessment is performed according to the abnormal confidence. 2.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, The method comprises the following steps: a mean value parameter of a sub-Gaussian model in the Gaussian mixture model of the target position which is most matched with a gray value of the target position in a previous frame is taken as a predicted gray value of the target position in a current frame. If the target position is in the confidence interval of the mean value of the sub-Gaussian model with the minimum difference between the gray value of the previous frame and the mean value of the sub-Gaussian model, i.e. the matching is successful, the mean value of the matched sub-Gaussian model is regarded as the predicted gray value of the target position in the current frame. If the target position is not matched with all sub-Gaussian models, the predicted gray value of the target position in the current frame is selected as the predicted gray value of the successfully matched pixel point in the same position in the nearest time sequence. If the target position is in the confidence interval of the mean value of the sub-Gaussian model with the minimum difference between the gray value of the previous frame and the mean value of the sub-Gaussian model, i.e. the matching is successful, the mean value of the matched sub-Gaussian model is regarded as the predicted gray value of the target position in the current frame. If the target position is not matched with all sub-Gaussian models, the predicted gray value of the target position in the current frame is selected as the predicted gray value of the successfully matched pixel point in the same position in the nearest time sequence. 3.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, The method comprises the following steps: the target position is marked in the current frame when the dynamic abnormality degree of the target position in the current frame is greater than a preset abnormal threshold; and connected domain algorithm is used to aggregate spatially adjacent marked points into a candidate abnormal region. The method comprises the following steps: a motion vector of each pixel point is obtained by using an optical flow method on the current frame and a previous continuous video frame; and a ratio of a number of motion vectors with a positive vertical component to a total number of motion vectors in the candidate abnormal region is taken as the motion consistency of the candidate abnormal region. 4.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, The expression of the abnormal confidence is as follows: The method comprises the following steps: a plurality of hierarchical risk thresholds are preset; a large number of abnormal confidences under normal working conditions are collected, and hierarchical thresholds are set for the data, that is, 95% quantile, 99% quantile and 99.9% quantile; the hazardous waste treatment is risk-free when the abnormal confidence of the candidate abnormal region is less than or equal to the 95% quantile; the hazardous waste treatment is low-risk when the abnormal confidence of the candidate abnormal region is greater than the 95% quantile and less than the 99% quantile; the hazardous waste treatment is medium-risk when the abnormal confidence of the candidate abnormal region is greater than or equal to the 99% quantile and less than the 99.9% quantile; and the hazardous waste treatment is high-risk when the abnormal confidence of the candidate abnormal region is greater than or equal to the 99.9% quantile. 5.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, The method comprises the following steps: the number of sub-Gaussian models is set to be 5. ； wherein, is a frame number of the current frame, denotes the candidate abnormal region in the current frame, denotes the candidate abnormal region , denotes the pixel points in the candidate abnormal region , is the pixel point in the current frame, is a dynamic abnormality degree of the candidate abnormal region in the current frame. 6.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, The method comprises the following steps: the gray values of the target position in the current frame and in previous frames are taken as a gray sequence of the target position. The method comprises the following steps: the gray values of the target position in the current frame and in previous 50 frames are taken as a gray sequence of the target position. ​ 7.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, ​ ​ 8.The video analysis based hazardous waste treatment monitoring and early warning method according to claim 1, wherein, ​ ​

Images