A substation multi-target recognition and tracking method based on image processing technology

Abstract

The present application relates to the field of image processing and power monitoring technology, and particularly relates to a substation multi-target identification and tracking method based on image processing technology, comprising the following steps: S1: real-time acquisition of substation monitoring video stream, output of filtered image; S2: extraction of moving target contour, generation of target contour set; S3: calculation of spatial gradient amplitude and electromagnetic interference quantization parameter, formation of feature vector; S4: input of feature vector into classifier, output of device target label or biological target label; S5: establishment of linear prediction model, output of predicted coordinate sequence; S6: when the predicted coordinate sequence intersects with the safety area of the device target, an intrusion alarm signal is generated. The present application, through the combination of target classification identification and trajectory prediction, realizes the accurate identification and early warning of biological targets in the substation under the electromagnetic interference environment, and improves the response capability and safety control level of the system to dynamic intrusion behavior.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing and power monitoring, and particularly relates to a substation multi-target identification and tracking method based on image processing technology. BACKGROUND

[0002] With the continuous improvement of the intelligence and automation level of substations, the safety protection and operation and maintenance monitoring of field devices are increasingly demanded; traditional video monitoring systems mainly rely on manual patrol and simple image change detection methods, which are difficult to accurately identify and track multiple types of targets in substations in a timely manner, especially in a strong electromagnetic interference environment, which is prone to false positives, false negatives and response lag and other problems.

[0003] In addition, in the substation scene, devices and personnel, biological targets interact frequently, and if the intrusion behavior cannot be timely determined, it is easy to cause device failure or safety accidents, which brings major hidden dangers to the stable operation of the power system; the existing technology in the aspects of multi-target identification and dynamic intrusion alarm generally has the problems of poor adaptability to complex electromagnetic interference, insufficient target classification accuracy, and lack of prediction and early warning mechanism, and it is difficult to meet the actual needs of the substation site for high reliability and intelligent prevention and control. Therefore, there is an urgent need for a substation multi-target identification and tracking method based on image processing technology to solve the above problems. SUMMARY

[0004] Based on the above purpose, the present application provides a substation multi-target identification and tracking method based on image processing technology.

[0005] A substation multi-target identification and tracking method based on image processing technology, comprising the following steps:

[0006] S1: real-time acquisition of substation monitoring video stream, and electromagnetic interference filtering of each frame of image, output of filtered image;

[0007] S2: adaptive threshold segmentation of the filtered image, extraction of the motion target contour, and generation of a target contour set;

[0008] S3: calculation of the spatial gradient amplitude and electromagnetic interference quantization parameters of each contour in the target contour set, formation of a feature vector;

[0009] S4: input of the feature vector into a classifier, and output of a device target label or a biological target label;

[0010] S5: calculation of the motion vector for the contour with the biological target label, construction of a linear prediction model based on the historical trajectory points, and output of a predicted coordinate sequence;

[0011] S6: generation of an intrusion alarm signal when the predicted coordinate sequence intersects with the safety area of the device target.

[0012] Optionally, the S1 specifically comprises:

[0013] S11: deploying a high-definition camera with network transmission function in the monitoring area of the substation, the camera being configured with a synchronous clock and a frame buffer structure, and outputting video image frames to the outside through a preset frame rate parameter, the video image frames being transmitted in real time to an image processing end through a wired Ethernet interface;

[0014] S12: performing electromagnetic interference filtering processing on the received video image frames, first performing two-dimensional Fourier transform on the image to obtain a frequency domain image, then suppressing the spectral components in the frequency domain image within a set interference frequency band range, and finally restoring the filtered image through inverse Fourier transform.

[0015] Optionally, the S2 specifically comprises:

[0016] S21: performing pixel-by-pixel difference operation on the filtered image obtained in S1 and its previous frame filtered image to obtain a difference image ;

[0017] S22: performing adaptive threshold segmentation on the difference image , calculating the best segmentation threshold of the whole image by using the Otsu method , and marking the positions with pixel values greater than as foreground regions to generate a binary segmentation image , which is expressed as: ;

[0018] S23: performing contour extraction processing on the binary segmentation image , detecting the edge pixels of the connected regions, extracting the outer boundaries of each foreground region, and constituting a target contour set.

[0019] Optionally, the S23 specifically comprises:

[0020] S231: performing eight-neighbor connected region scanning on the binary segmentation image obtained in S22, sequentially traversing the foreground pixel points with a value of 1 through a breadth-first search algorithm, marking all the pixels connected to the corresponding pixel as the same connected region, and constructing a foreground region set;

[0021] S232: for each connected region, constructing its boundary contour, detecting its outer edge by using a boundary tracking algorithm, starting from the starting foreground pixel point in the clockwise direction, judging whether there is a pixel with a value of 1 and not visited in the eight-neighbor domain of the current pixel, if there is, adding the current contour path and continuing to search until returning to the starting point to form a complete contour path;

[0022] S233: take all the contour path sets as the target contour set, and perform a two-stage screening process on each contour;

[0023] First stage, calculate the number of contour points , if it satisfies , then the corresponding contour is eliminated;

[0024] Second stage, calculate the minimum circumscribed rectangle area of the contour , if it satisfies , then the corresponding contour is also eliminated; wherein, is the minimum threshold of the number of contour points; is the minimum threshold of the circumscribed rectangle area.

[0025] Optionally, the S3 specifically comprises:

[0026] S31: perform gray space gradient calculation on the image region corresponding to each contour in the target contour set, extract the image gradient in the horizontal direction and the vertical direction using the Sobel operator respectively, and calculate the spatial gradient amplitude of the corresponding pixel ;

[0027] S32: average the spatial gradient amplitudes of all pixels in the contour to obtain the average spatial gradient of the contour , which is taken as the structural feature index of the target contour;

[0028] S33: perform two-dimensional Fourier transform on the contour region image, and extract the frequency spectrum energy in the interference frequency band as the electromagnetic interference quantization parameter , which is defined as the sum of the power spectrum density of the band-stop frequency band in the frequency domain;

[0029] S34: combine the average spatial gradient of the contour and the electromagnetic interference quantization parameter to form a two-dimensional feature vector .

[0030] Optionally, the S4 specifically comprises:

[0031] S41: construct a binary classification model based on support vector machine, use the labeled feature vector samples in the historical images as the training set, including two types of biological targets and device targets, each sample is composed of a feature vector and its corresponding label;

[0032] S42: solve the optimal classification hyperplane based on the training set by the maximum margin criterion to obtain the model parameters , construct the discriminant function , the expression is , wherein, is the two-dimensional feature vector of the contour to be classified; is a normal vector of a hyperplane; is a bias term;

[0033] S43: the feature vector extracted in S3 is input into the trained classifier, if the following condition is met , the biological target label is output; if the following condition is met , the device target label is output, thereby completing the target type determination.

[0034] Optionally, the S5 specifically includes:

[0035] S51: for the contour determined as the biological target label in S4, first, the contour centroid coordinates in the current frame are extracted, and the corresponding coordinates are time-series paired with the centroid coordinates of the corresponding contour in the previous frame; based on the change of the centroid positions of the two consecutive frames, the motion vector of the current biological target is calculated;

[0036] S52: for each contour with the biological target label, the centroid coordinates of the recent several frames are recorded as a historical trajectory point sequence, and arranged in time sequence; based on the trajectory data, a linear prediction model of the time position relationship is established;

[0037] S53: the constructed linear prediction model is applied to the future several time points, the centroid coordinates of the target in the next multiple frames are predicted, and the prediction coordinates are output as a prediction coordinate sequence in time sequence.

[0038] Optionally, the S52 specifically includes:

[0039] S521: for each contour with the biological target label, the centroid coordinates in the consecutive frame images are extracted, and a historical trajectory point sequence is constructed , wherein represents the frame number, represents the centroid horizontal and vertical coordinates in the corresponding frame;

[0040] S522: two independent linear fitting models are constructed respectively with time as the independent variable, and horizontal coordinate and vertical coordinate as the dependent variables, the slope and intercept of the linear function are calculated by the least square method, and the linear relationship between time and spatial position is obtained;

[0041] The horizontal prediction model is: ;

[0042] The vertical prediction model is: ; wherein, represents the predicted centroid coordinates at the future time ;​ is a motion speed estimation value in the transverse direction and the longitudinal direction; is an initial position offset.

[0043] Optionally, S6 specifically comprises:

[0044] S61: based on the preset device target region in the monitoring picture, two-dimensional boundary coordinates of each designated device are extracted respectively, and a corresponding safety region boundary box is constructed;

[0045] S62: the biological target prediction coordinate sequence obtained in S5 is judged frame by frame whether it falls into any device safety region, if any coordinate point in the prediction coordinate sequence is located in the boundary box corresponding to a device safety region, it is regarded that the biological target invades the device region, and it is determined as an intersection state;

[0046] S63: when at least one prediction coordinate point and any device safety region have a spatial intersection relationship, an invasion alarm signal is immediately sent to the system.

[0047] Optionally, the alarm signal comprises the number of the biological target, the prediction path, and the identification information of the potential invasion device region.

[0048] The beneficial effects of the present application are:

[0049] The present application, by performing frequency domain filtering processing on the monitoring image in a strong electromagnetic interference environment, combining adaptive threshold segmentation and contour extraction, a stable target contour set is constructed, and a feature vector is formed by a spatial gradient feature and a frequency spectrum energy feature, and a support vector machine model is used to realize high-precision classification and identification of device targets and biological targets, solving the problem of poor target detection accuracy of the existing method in a complex interference background.

[0050] The present application, by combining the history trajectory point to calculate the motion vector of the biological target, constructing a linear prediction model to output the coordinate sequence at the future time, and judging the spatial intersection with the safety boundary of the device, triggering the invasion alarm signal in advance, forming an intelligent protection process of identification, prediction and alarm linkage, effectively improving the response timeliness and identification reliability of the dynamic invasion target of the substation. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0052] Fig. 1 It is a schematic diagram of the multi-target recognition and tracking method of the embodiment of the present application.

[0053] Fig. 2 The output prediction coordinate sequence flowchart of the embodiment of the present application. DETAILED DESCRIPTION

[0054] The present application will be described in detail below with reference to the drawings and specific embodiments. It should be noted that, in order to make the embodiments more detailed, the following embodiments are the best, preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement some known technologies; and the drawings are only used to more specifically describe the embodiments, and are not intended to specifically limit the present application.

[0055] It should be noted that in the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiments can include a specific feature, structure or property, but not necessarily every embodiment includes the specific feature, structure or property. In addition, when a specific feature, structure or property is described in combination with an embodiment, it should be within the knowledge of those skilled in the related art to implement such a feature, structure or property in combination with other embodiments (whether or not explicitly described).

[0056] Generally, the terms can be understood at least in part from the context of their use. For example, depending on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular or can be used to describe combinations of features, structures, or characteristics, in the plural, that are collectively described in the singular. In addition, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but instead can allow for existence of other factors that are not necessarily explicitly described.

[0057] As shown in Figs. 1-2 A substation multi-target recognition and tracking method based on image processing technology, comprising the following steps:

[0058] S1: Real-time acquisition of substation monitoring video stream, and electromagnetic interference filtering of each frame of image, output of filtered image;

[0059] S1 specifically includes:

[0060] S11: Deploying a high-definition camera with network transmission function in the substation monitoring area, the camera being configured with a synchronous clock and a frame buffer structure, and outputting video image frames outside through a preset frame rate parameter, the video image frames being transmitted in real time to an image processing end through a wired Ethernet interface;

[0061] S12: Perform electromagnetic interference filtering processing on the received video image frame, first perform two-dimensional Fourier transform on the image to obtain a frequency domain graph, then suppress the frequency spectrum components in the frequency band range within the set interference frequency band in the frequency domain graph, and finally restore it to a filtered image through inverse Fourier transform; Through the above scheme, high-quality monitoring video can be stably collected in the strong electromagnetic environment of the substation, and the influence of the specific electromagnetic interference frequency band is effectively suppressed through frequency domain band-stop filtering, ensuring the accuracy and stability of subsequent image segmentation and target recognition processing.

[0062] The steps of performing electromagnetic interference filtering processing are as follows:

[0063] Two-dimensional Fourier transform: Perform two-dimensional Fourier transform on the received video image frame to obtain a frequency domain image, and the calculation formula is: , wherein, is the gray value of the original image in the spatial domain; is the width and height of the image; is the complex spectrum of the image in the frequency domain; is the frequency coordinate; is the exponential term, which is used to map the image from the spatial domain to the frequency domain, and the parameters are defined as follows, is the imaginary unit, which represents ; is the constant, which is the constant; is the frequency domain coordinate, which represents the frequency components in the horizontal and vertical directions; is the spatial domain coordinate, which represents the position of the pixel point in the image; is the total number of pixels in the horizontal direction (width) of the image; is the total number of pixels in the vertical direction (height) of the image; is the normalized mapping relationship between the spatial position and the frequency in the horizontal and vertical directions, respectively;

[0064] Band-stop filter design: For the specific electromagnetic interference frequency band of the substation , a band-stop filter function is constructed, which is defined as: , wherein, and are the inner and outer diameter thresholds of the interference frequency band, respectively, which are set according to the field interference spectrum; represents the suppression of the spectrum components in the band-stop region, and the original spectrum is retained in the remaining region;

[0065] Spectrum filtering processing: Perform point-by-point multiplication operation on the frequency domain image and the band-stop filter to obtain the filtered frequency domain image ;

[0066] Two-dimensional inverse Fourier transform: Perform two-dimensional inverse Fourier transform on the filtered frequency domain image to restore the spatial domain image , which is the filtered image.

[0067] S2: performing adaptive threshold segmentation on the filtered image to extract the contour of the moving target and generate a target contour set;

[0068] S2 specifically includes:

[0069] S21: performing pixel-by-pixel difference operation on the filtered image obtained in S1 and the previous frame filtered image to obtain a difference image , and the calculation formula is: , wherein, is the gray value of the current frame image; is the gray value of the previous frame image;

[0070] S22: performing adaptive threshold segmentation on the difference image , calculating the best segmentation threshold of the whole image by using the Otsu method , and marking the positions with pixel values greater than as foreground regions to generate a binary segmentation image , which is expressed as: ; wherein, is the global threshold value adaptively calculated by the maximum inter-class variance criterion;

[0071] S23: performing contour extraction processing on the binary segmentation image , detecting the edge pixels of the connected regions, extracting the outer boundary of each foreground region, and constituting a target contour set; through the above steps, the accurate target region extraction operation can be performed on the continuous monitoring image frames without manual intervention, wherein the foreground extraction method based on frame difference adapts to the dynamic scene, and the global optimal segmentation threshold of the Otsu method ensures the stability of the foreground region, and finally the contour set provides a structured input basis for the subsequent feature extraction and classification recognition steps.

[0072] The steps of calculating the best segmentation threshold of the whole image by using the Otsu method are as follows:

[0073] Statistically counting all pixel values of the difference image to construct its gray level histogram; assuming that the image gray level is (taking the value as 256), the gray value is , and the number of pixels corresponding to each gray level is , then the total number of pixels is: ; and the corresponding normalized probability distribution is: , wherein represents the probability of the gray level , satisfying ;

[0074] The gray value According to the threshold value Is divided into foreground And background Two categories, for each possible , the following statistics are calculated:

[0075] The total probability of foreground pixels: ;

[0076] The total probability of background pixels: ;

[0077] The foreground mean: ;

[0078] The background mean: ;

[0079] The overall mean of the image: ;

[0080] For each , the inter-class variance , the expression is: ;

[0081] Traverse all , find the threshold value , the expression is:

[0082] , wherein , is the best global threshold value of image segmentation; by using Otsu method to calculate the adaptive threshold value, the optimal segmentation point can be automatically selected based on the image gray scale distribution, so as to dynamically adjust the threshold value under different light conditions and electromagnetic interference environment, realize the robust foreground extraction of moving target, and enhance the accuracy and consistency of subsequent contour detection.

[0083] S23 specifically comprises:

[0084] S231: Perform eight-neighbor connected region scanning on the binary segmentation image obtained in S22 , traverse the foreground pixel points with value 1 in turn through the breadth-first search algorithm, mark all the pixels connected with the corresponding pixel as the same connected region, and construct a foreground region set;

[0085] S232: For each connected region, construct its boundary contour, detect its outer edge by using the boundary tracking algorithm, and judge whether there is a pixel with value 1 and not visited in the eight-neighborhood of the current pixel in the clockwise direction from the starting foreground pixel point, if there is, add the current contour path and continue searching until returning to the starting point, forming a complete contour path;

[0086] S233: Use all contour paths as the target contour set, and perform two-level filtering on each contour;

[0087] Level 1: Calculate the number of contour points If satisfied If so, the corresponding outline will be removed;

[0088] The second level calculates the area of ​​the minimum bounding rectangle of the contour. If satisfied Similarly, the corresponding contours are also removed; among them, This is the minimum threshold for the number of contour points; The minimum threshold (in pixels) for the area of ​​the circumscribed rectangle is used. The above steps, by combining connected component scanning and boundary tracking, can accurately separate independent moving targets in binary images and extract their complete outer boundary contours. This avoids contour information loss caused by edge breaks or polygon missingness, laying a precise contour foundation for subsequent target feature extraction and classification recognition, and improving the overall recognition accuracy of the system.

[0089] S3: Calculate the spatial gradient magnitude and electromagnetic interference quantization parameters of each contour in the target contour set to form a feature vector;

[0090] S3 specifically includes:

[0091] S31: Perform grayscale spatial gradient calculation on the image region corresponding to each contour in the target contour set. Use the Sobel operator to extract the image gradients in the horizontal and vertical directions respectively, and calculate the spatial gradient magnitude of the corresponding pixel. Its expression is: ,in, For pixels Spatial gradient magnitude; The horizontal gradient value calculated for the Sobel operator; The vertical gradient value calculated for the Sobel operator; Indicates the pixel coordinates in the corresponding contour;

[0092] S32: Spatial gradient magnitude for all pixels within the contour By averaging, the average spatial gradient of the contour is obtained. As a structural feature index of the target contour, its expression is:

[0093] ,in, The average spatial gradient of the contour; This represents the total number of pixels in the outline.

[0094] S33: Perform a two-dimensional Fourier transform on the contour region image and extract the spectral energy within the interference frequency band as the electromagnetic interference quantization parameter. It is defined as the sum of the power spectral densities in the band-stop frequency band in the frequency domain; the calculation formula is: ,in, Electromagnetic interference quantization parameters for the profile; The result is a two-dimensional Fourier transform of the contour region image; These are the horizontal and vertical frequency coordinates in the frequency domain; These are the lower and upper limits of the preset electromagnetic interference frequency band range; The magnitude of the frequency amplitude is used for frequency band selection;

[0095] S34: Calculate the average spatial gradient corresponding to the contour. Electromagnetic interference quantization parameters Combining to form a two-dimensional feature vector The above steps jointly extract spatial structure features and frequency domain interference features in each contour region to construct a discriminative two-dimensional feature vector, which can effectively improve the classification accuracy and environmental adaptability of subsequent equipment targets and biological targets in the complex background of substations.

[0096] S4: Input the feature vector into the classifier and output the device target label or biological target label;

[0097] S4 specifically includes:

[0098] S41: Construct a binary classification model based on support vector machines, using feature vector samples labeled in historical images as the training set, including two classes: biological targets and device targets. Each sample consists of a feature vector. and its corresponding tags Composition; its expression is:

[0099] ,in, For the training sample set; For the first Two-dimensional feature vectors of training samples; For the corresponding target category label, +1 represents biological targets and -1 represents equipment targets;

[0100] S42: Solve for the optimal classification hyperplane based on the training set using the maximum margin criterion to obtain the model parameters. Construct the discriminant function Its expression is ,in, The two-dimensional feature vector of the contour to be classified; Let be the normal vector parameter of the hyperplane; For bias terms;

[0101] S43: Extract the feature vector from S3 Input the trained classifier, if it satisfies If the condition is met, then output the biological target label; if the condition is met, then output the biological target label. Then, the device target label is output, thereby completing the target type determination. The above steps use the support vector machine model to construct the decision boundary, and input the two-dimensional feature vector containing structural information and electromagnetic interference characteristics into the trained classifier for discrimination. It can stably distinguish between equipment targets and intrusive biological targets in substations under different interference backgrounds, providing an accurate classification basis for subsequent tracking and alarm processing.

[0102] S5: For contours with biological target labels, calculate motion vectors, construct a linear prediction model based on historical trajectory points, and output the predicted coordinate sequence;

[0103] S5 specifically includes:

[0104] S51: For the contour identified as a biological target in S4, firstly, extract its centroid coordinates in the current frame, and then temporally pair these coordinates with the centroid coordinates of the corresponding contour in the previous frame; based on the change in centroid position between two consecutive frames, calculate the motion vector of the current biological target, reflecting its direction and velocity; the calculation formula is as follows: ,in, For biological targets Motion vector in the current frame; For the current frame Biological Targets The coordinates of the centroid; Previous frame Biological Targets The coordinates of the centroid;

[0105] S52: For each contour with a biological target label, record the centroid coordinates of its most recent several frames as a historical trajectory point sequence and arrange them in chronological order; based on the trajectory data, establish a linear prediction model of temporal position relationship, using the historical centroid position as input sample to fit the spatial position change trend of the target at subsequent times.

[0106] S53: Apply the constructed linear prediction model to several future moments to predict the centroid coordinates of the target in the next few frames, and output these predicted coordinates in chronological order as a predicted coordinate sequence for subsequent determination of whether to enter the device's safe zone.

[0107] S52 specifically includes:

[0108] S521: For each contour labeled with a biological target, extract its continuous... Centroid coordinates in frame images, construct historical trajectory point sequence , wherein represents frame number, represents the centroid horizontal and vertical coordinates in the corresponding frame;

[0109] S522: respectively taking time as independent variable, and horizontal coordinate and vertical coordinate as dependent variable, construct two independent linear fitting models, and calculate the slope and intercept of the linear function by using least square method to obtain the linear relationship between time and spatial position;

[0110] The horizontal prediction model is: ;

[0111] The vertical prediction model is: ; wherein, represents the predicted centroid coordinates at future time ; is the horizontal and vertical motion speed estimate (i.e. slope); is the initial position offset (i.e. intercept);

[0112] The above parameters are calculated by least square method, and the calculation expression of horizontal slope and intercept is as follows: , and the calculation method of vertical is the same as the above form, only is replaced by ; the above steps can efficiently establish the prediction model of target centroid position change with time by applying least square method for linear fitting on historical centroid trajectory points, realize the time sequence inference of future position of biological target, and improve the forward-looking judgment ability of the system on dynamic intrusion behavior.

[0113] S6: when the predicted coordinate sequence intersects with the safety area of the device target, generate an intrusion alarm signal;

[0114] S6 specifically includes:

[0115] S61: based on the preset device target area in the monitoring picture, extract the two-dimensional boundary coordinates of each specified device, and construct the corresponding safety area boundary box, the boundary box is set with a fixed buffer distance based on the geometric outer edge of the device, forming a rectangular or polygonal area with spatial constraint, which is used to judge whether there is intrusion behavior;

[0116] S62: judging whether the biological target prediction coordinate sequence obtained in S5 falls into any device safety area frame by frame, if any coordinate point in the prediction coordinate sequence is located in the boundary box corresponding to a device safety area, it is considered that the biological target invades the device area, and the intersection state is determined;

[0117] S63: when at least one prediction coordinate point and any device safety area have a spatial intersection relationship, an invasion alarm signal is immediately sent to the system; the above-mentioned step can judge the spatial intersection between the prediction coordinate sequence and the device safety boundary, can predict the potential invasion risk of the biological target before it really approaches the device, can generate an alarm signal in time, can realize the early intervention and response control, and can improve the identification and disposal ability of the substation security system to the dynamic invasion target.

[0118] The alarm signal includes the number of the biological target, the prediction path and the identification information of the potential invasion device area.

[0119] The present application covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, the specific details are described in the following preferred embodiments of the present application, and the present application can also be completely understood without the description of these details for those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, the well-known methods, processes, flows, elements and circuits are not described in detail.

[0120] The above-mentioned is only the preferred embodiment of the present application, and it should be pointed out that for those skilled in the art, some improvements and refinements can be made without departing from the principle of the present application, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for multi-target recognition and tracking in a substation based on image processing technology, characterized in that, Includes the following steps: S1: Real-time acquisition of substation monitoring video stream, electromagnetic interference filtering of each frame, and output of filtered image; S2: Perform adaptive threshold segmentation on the filtered image to extract the contours of moving targets and generate a set of target contours; S3: Calculate the spatial gradient magnitude and electromagnetic interference quantization parameters of each contour in the target contour set to form a feature vector; S4: Input the feature vector into the classifier, and output the device target label or biological target label; specifically including: S41: Construct a binary classification model based on support vector machine, using feature vector samples labeled in historical images as the training set, including two categories: biological targets and device targets. Each sample consists of a feature vector and its corresponding label. S42: Solve for the optimal classification hyperplane based on the training set using the maximum margin criterion to obtain the model parameters. Construct the discriminant function Its expression is ,in, The two-dimensional feature vector of the contour to be classified; Let be the normal vector parameter of the hyperplane; For bias terms; S43: Extract the feature vector from S3 Input the trained classifier, if it satisfies If the condition is met, then output the biological target label; if the condition is met, then output the biological target label. If so, the device target label is output, thus completing the target type determination; S5: For contours with biological target labels, calculate the motion vector, construct a linear prediction model based on historical trajectory points, and output a predicted coordinate sequence; specifically including: S51: For the contour identified as a biological target in S4, first extract its centroid coordinates in the current frame, and then perform temporal pairing between the corresponding coordinates and the centroid coordinates of the corresponding contour in the previous frame; based on the change in centroid position between two consecutive frames, calculate the motion vector of the current biological target. S52: For each contour with a biological target label, record the centroid coordinates of its most recent few frames as a historical trajectory point sequence and arrange them in chronological order; based on the trajectory data, establish a linear prediction model of the temporal and positional relationship. S53: Apply the constructed linear prediction model to several future moments to predict the centroid coordinates of the target in the next few frames, and output these predicted coordinates as a predicted coordinate sequence in chronological order; Specifically, S52 includes: S521: For each contour labeled with a biological target, extract its continuous... Centroid coordinates in frame images are used to construct a sequence of historical trajectory points. ,in Indicates the frame number. Indicates the horizontal and vertical coordinates of the centroid in the corresponding frame; S522: By time As the independent variable, with the horizontal axis as the coordinate. with vertical coordinate With the dependent variable as the variable, two independent linear fitting models are constructed. The slope and intercept of the linear function are calculated using the least squares method to obtain the linear relationship between time and spatial location. The lateral prediction model is: ; The longitudinal prediction model is as follows: ;in, Indicates a future moment The predicted centroid coordinates; These are estimates of the lateral and longitudinal motion velocities; This is the initial position offset; S6: When the predicted coordinate sequence intersects with the safe area of ​​the device target, an intrusion alarm signal is generated.

2. The method for multi-target recognition and tracking in a substation based on image processing technology according to claim 1, characterized in that, S1 specifically includes: S11: Deploy high-definition cameras with network transmission capabilities in the substation monitoring area. The cameras are equipped with a synchronous clock and frame buffer structure, and output video image frames to the outside through preset frame rate parameters. The video image frames are transmitted to the image processing terminal in real time through a wired Ethernet interface. S12: Perform electromagnetic interference filtering on the received video image frame. First, perform a two-dimensional Fourier transform on the image to obtain a frequency domain image. Then, suppress the spectral components in the frequency domain image whose frequency band is within the set interference frequency band. Finally, restore the filtered image through an inverse Fourier transform.

3. The method for multi-target recognition and tracking in a substation based on image processing technology according to claim 1, characterized in that, S2 specifically includes: S21: Perform a pixel-by-pixel difference operation between the filtered image obtained in S1 and the filtered image of the previous frame to obtain the difference image. ; S22: For the difference image Adaptive thresholding is performed, and the optimal segmentation threshold for the entire image is calculated using Otsu's method. and pixel values ​​greater than The location is marked as the foreground region, and a binary segmentation image is generated. Its expression is: ; S23: Segmenting the binary image Perform contour extraction processing, detect the edge pixels of connected regions, extract the outer boundary of each foreground region, and form a target contour set.

4. The method for multi-target recognition and tracking in a substation based on image processing technology according to claim 3, characterized in that, S23 specifically includes: S231: The binary segmented image obtained in S22 Perform an eight-neighbor connected region scan, and use a breadth-first search algorithm to traverse the foreground pixels with a value of 1 in turn, mark all pixels connected to the corresponding pixel as the same connected region, and construct a set of foreground regions; S232: For each connected region, construct its boundary contour, use the boundary tracking algorithm to detect its outer edge, start from the starting foreground pixel in a clockwise direction, determine whether there is a pixel with a value of 1 that has not been visited in the eight neighborhoods of the current pixel, if so, add it to the current contour path and continue searching until returning to the starting point to form a complete contour path. S233: Use all contour paths as the target contour set, and perform two-level filtering on each contour; Level 1: Calculate the number of contour points If satisfied If so, the corresponding outline will be removed; The second level calculates the area of ​​the minimum bounding rectangle of the contour. If satisfied Similarly, the corresponding contours are also removed; among them, This is the minimum threshold for the number of contour points; This is the minimum threshold for the area of ​​the circumscribed rectangle.

5. The method for multi-target recognition and tracking in a substation based on image processing technology according to claim 1, characterized in that, S3 specifically includes: S31: Perform grayscale spatial gradient calculation on the image region corresponding to each contour in the target contour set. Use the Sobel operator to extract the image gradients in the horizontal and vertical directions respectively, and calculate the spatial gradient magnitude of the corresponding pixel. ; S32: Spatial gradient magnitude for all pixels within the contour By averaging, the average spatial gradient of the contour is obtained. , as a structural feature indicator of the target contour; S33: Perform a two-dimensional Fourier transform on the contour region image and extract the spectral energy within the interference frequency band as the electromagnetic interference quantization parameter, which is defined as the sum of the power spectral density of the band-stop frequency band in the frequency domain. S34: Calculate the average spatial gradient corresponding to the contour. Electromagnetic interference quantization parameters Combining to form a two-dimensional feature vector .

6. The method for multi-target recognition and tracking in a substation based on image processing technology according to claim 1, characterized in that, S6 specifically includes: S61: Based on the preset target area of ​​the device in the monitoring screen, extract the two-dimensional boundary coordinates of each specified device and construct the corresponding safe area boundary box; S62: For the biological target prediction coordinate sequence obtained in S5, determine frame by frame whether it falls into any device security area. If any coordinate point in the prediction coordinate sequence is located within the bounding box corresponding to a certain device security area, it is considered that the biological target has invaded the device area and is judged as an intersection state. S63: When at least one predicted coordinate point spatially intersects with the security zone of any device, an intrusion alarm signal is immediately sent to the system.

7. A method for multi-target recognition and tracking in a substation based on image processing technology according to claim 6, characterized in that, The alarm signal includes the biological target's ID, predicted path, and identification information of the potential intrusion device area.

Images