Disconnecting switch closing identification device based on space angle conversion

Abstract

The invention discloses an isolation switch closing identification device based on space angle conversion, and the device comprises a matrix generation module which is used for obtaining a homography matrix which is used for determining the mapping relation of two planes, and after the installation position of the device is determined, calling a primary matrix generation module to generate the homography matrix; the knife arm extraction module is used for extracting knife arm linear parameter characteristics of the disconnecting switch in the image data; the perspective correction module is used for realizing coordinate mapping of a camera plane-a disconnecting switch motion plane through the homography matrix generated in the matrix generation module; the linear parameter characteristics of the knife arm of the disconnecting switch are converted into corrected linear parameter characteristics through mapping; and the angle calculation module is used for calculating an actual angle value of the knife arm of the disconnecting switch according to the rectified linear parameter characteristics and outputting a closing identification result of the disconnecting switch. The method is suitable for detecting the opening and closing state of the disconnecting switch, can improve the measurement precision in a non-orthogonal shooting scene, and meets the industrial detection requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image processing, and in particular to an isolating switch closing recognition device based on spatial angle conversion. Background Art

[0002] In industrial inspection and automation control, measuring the spatial angle of a target object through images is one of the key technologies. The traditional methods have the following problems:

[0003] 1. Strict installation position: In traditional methods, to ensure the accuracy of the device, the installation position requirements are extremely strict. However, in high-risk industrial environments such as high-voltage isolating switches, the placement position of the camera device is greatly restricted by safety, and in most cases, the strict installation position conditions cannot be met.

[0004] 2. Perspective distortion effect: When the camera optical axis is not perpendicular to the target plane, the geometric distortion in the image will distort the real angle. The existing methods rely on manual calibration or fixed viewing angles, with poor flexibility.

[0005] 3. Low degree of automation: It is necessary to manually select feature points or rely on specific calibration tools, and it is difficult to adapt to dynamic scenarios.

[0006] Therefore, there is an urgent need for a precise recognition and measurement device with high precision, anti-interference, and capable of automatically correcting perspective distortion. Summary of the Invention

[0007] The purpose of the present invention is to overcome the deficiencies of the prior art, and provide an isolating switch closing recognition device based on spatial angle conversion. By integrating edge detection, Hough transform, and homography matrix technology, problems such as inaccurate edge extraction and difficult perspective distortion correction in complex scenarios are solved, and high-precision and fully automatic angle measurement are achieved, especially suitable for angle measurement and state judgment of components such as high-voltage isolating switches and robotic arms in industrial scenarios.

[0008] To achieve the above purpose, the technical solution provided by the present invention is: An isolating switch closing recognition device based on spatial angle conversion, comprising:

[0009] A matrix generation module, configured to obtain a homography matrix, which is used to establish the mapping relationship between two planes. After determining the installation position of the device, the matrix generation module is called once to generate a homography matrix;

[0010] A knife arm extraction module, configured to extract the straight line parameter features of the isolating switch knife arm from the image data;

[0011] The perspective correction module is used to implement the coordinate mapping between the camera plane and the disconnector motion plane through the homography matrix generated by the matrix generation module, where the camera plane refers to the image data plane captured by the camera, and the disconnector motion plane refers to the actual motion plane of the disconnector; through the above mapping, the linear parameter feature of the disconnector arm is converted into the corrected linear parameter feature.

[0012] The angle calculation module is used to calculate the actual angle value of the disconnector arm according to the corrected linear parameter feature and output the disconnector closing recognition result.

[0013] Furthermore, the matrix generation module includes a point pair generation unit and a matrix calculation unit, where:

[0014] The point pair generation unit is used to generate the point pair data between the camera plane and the disconnector motion plane;

[0015] The matrix calculation unit is used to generate a high-precision homography matrix for the point pair data.

[0016] Furthermore, the point pair generation unit includes the following steps:

[0017] 1) Given the camera installation height h, the elevation angle θ of a specific point relative to the camera, and the distance d between the camera and the disconnector, the height h of the specific point from the camera is determined through geometric calculation 1 = tanθ·d, so as to obtain the relative position of this point in the disconnector motion plane; based on the geometric relationship, the height L of the specific point from the horizontal plane is determined as L = tanθ·d + h, so as to obtain the specific point t = (x t = 0, y t = L), where its x t and y t are the X and Y axis coordinate values of the specific point in the disconnector motion plane;

[0018] 2) The spatial relationship of the disconnector is a relative coordinate set Q = {q s |q s ∈R 2 , s = 1, 2, 3,..., N}, where R 2 represents the two-dimensional real number set, the set value q s represents the relative coordinate of the s-th point with respect to the specific point, N represents the number of points, and the set Q is determined by the model of the disconnector; the coordinate set K = {k s = q s + t|q s ∈Q, s = 1, 2, 3,..., N} in the disconnector motion plane coordinate system is calculated from the specific point coordinates determined in step 1) and the set Q, and the set value k s represents the s-th coordinate in the disconnector motion plane coordinate system;

[0019] 3) Read the coordinate set file in the camera plane coordinate system and transform it to generate the coordinate set P in the camera plane coordinate system;

[0020] 4) Convert the sets P and K into a JSON file and output the JSON file.

[0021] Further, the matrix calculation unit includes the following steps:

[0022] 1) Normalize the point pair data to improve numerical stability;

[0023] 1a) First, perform a translation operation to move the point pair data to the origin;

[0024] 1b) Then, perform a scaling operation on the point pair data processed in step 1a) so that the root mean square distance of all points to the origin is a fixed value;

[0025] 2) Input the point pair data processed in step 1) into the RANSAC iterative method;

[0026] 2a) Randomly sample 4 groups of point pair data g1, g2, g3, g4 from the point pair data, where the point pair data is g i = ((x″ i , y″ i ), (X″ i , Y″ i )); (x″ i , y″ i ) and (X″ i , Y″ i ) are the i-th X and Y axis coordinate values in the set P and the i-th X and Y axis coordinate values in the set K after translation and scaling coordinate processing in step 1) respectively;

[0027] 2b) Construct a system of linear equations:

[0028]

[0029] Each point pair data provides two equations, and a total of 8 equations are generated from four pairs of point data, forming a homogeneous equation system AH K = 0; where the matrix A is the coefficient matrix of the above system of linear equations, with a size of 8×9, and each row corresponds to an equation; H K is the matrix solution vector of the equation, H K = [h 11 , h 12 , h 13 , h 21 , h 22 , h 23 , h 31 , h 32 , h33 T is the expansion vector of H K , is the value of the K th row and th column of the H matrix .

[0030] 2c) Reproject the set P with H K , that is, perform a projection transformation on all points in the set P;

[0031] 2d) Count the number of inliers. An inlier is a data point that satisfies the minimum error between the model prediction value and the actual observation value under the given model homography matrix; The projection error measure ∈ < 3px, where px is the pixel value, and a point with a reprojection point error less than the projection error measure is determined to be an inlier;

[0032] 2e) Update the number of inliers and the homography matrix. Initially, the number of inliers is 0. When the result of inlier calculation in step 2d) is greater than the current number of inliers, update the current number of inliers and update H best = H K , where H best is the optimal homography matrix, that is, the matrix with the largest number of inliers in step 2d); Otherwise, neither the number of inliers nor H best is updated;

[0033] 2f) Execute steps 2a), 2b), 2c), 2d), and 2e) in a loop D times, where D is the loop count parameter. After the loop ends, output H best ;

[0034] 3) In each iteration of step 2), a homography matrix is calculated once, and the optimal homography matrix in all iterations is output.

[0035] Furthermore, the tool arm extraction module includes a Gaussian filter unit, a Sobel operator unit, a double-threshold detection unit, and a Hough transform unit, where:

[0036] The Gaussian filter unit uses a 3×3 or 5×5 convolution kernel to suppress noise in the image data, generates the convolution kernel through a two-dimensional Gaussian function, and the weights are determined by the distance of the pixel from the center point:

[0037]

[0038] I Gau = I RAW * G(u,v)

[0039] In the formula, G(u,v) is the Gaussian kernel function, σ is the standard deviation, which controls the smoothness, u and v are the independent variables of the X-axis and Y-axis of the Gaussian kernel function, and I RAW ​is the original image input, I Gau is the Gaussian filtered image, and * is the image convolution operation;

[0040] The Sobel operator unit calculates the gradient magnitude and direction in the horizontal and vertical directions of the image data after noise suppression processing, including the following steps:

[0041] 1) Obtain the kernel matrix:

[0042]

[0043]

[0044] In the formula, S x is the horizontal direction kernel, and S y is the vertical direction kernel;

[0045] 2) Gradient calculation:

[0046] G x = I Gau * S x , G y = I Gau * S y

[0047] In the formula, G x , G y are the horizontal and vertical direction gradients respectively;

[0048] 3) Obtain the gradient magnitude and gradient direction:

[0049]

[0050] In the formula, Grdlen is the gradient magnitude, and Grddir is the gradient direction;

[0051] The double-threshold detection unit sets high and low thresholds to perform edge screening on the gradient magnitude, retains the coordinate data with high gradient magnitude, and generates an edge data set Q EDGE , where Q EDGE is the set of all coordinate values that meet the edge screening conditions;

[0052] The Hough transform unit discretizes the edge data set Q EDGE in the parameter space, and screens out the straight line parameter features of the disconnector arm, including the following steps:

[0053] 1) Map all edge data points (`x,`y) in the set Q EDGE to the point coordinates (ρ,θ′) in the polar coordinate parameter space:

[0054] ρ = `x cosθ′ + `y sinθ′

[0055] In the formula, `x` and `y` are the X and Y axis coordinate values in the edge data set Q EDGE ρ is the radial distance, representing the distance from the point to the origin; θ′ is the angular coordinate, representing the angle between the point and the positive X axis;

[0056] 2) Each edge data point traverses all possible θ′, calculates the corresponding ρ, and counts in the accumulator;

[0057] 3) Select the (ρ, θ′) with the highest count in the accumulator as the feature of the arm straight line parameter.

[0058] Furthermore, the perspective correction module includes a straight line parameter input unit, a coordinate transformation unit, and a straight line parameter output unit, where:

[0059] The straight line parameter input unit is used to generate points on the camera plane through the given feature (ρ, θ′) of the arm straight line parameter, where ρ is the radial distance, representing the distance from the point to the origin, and θ′ is the angular coordinate, representing the angle between the point and the positive X axis:

[0060] ρ = x cosθ′ + y sinθ′

[0061] In the formula, x and y are the X and Y axis coordinate values in the camera plane system;

[0062] The coordinate transformation unit is used to perform a homography matrix projection transformation to map the points on the camera plane to the isolating switch motion plane:

[0063]

[0064] In the formula, H is the homography matrix, W′ is the homogeneous component, and are the X and Y axis coordinate values of the un-scaled isolating switch motion plane system, and are the X and Y axis coordinate values of the isolating switch motion plane system;

[0065] The straight line parameter output unit is used to convert the points on the isolating switch motion plane into the corrected straight line parameter features and output them.

[0066] Furthermore, the angle calculation module includes a vector generation unit and an angle analysis unit, where:

[0067] The vector generation unit generates a spatial vector according to the corrected straight line parameter feature;

[0068] The angle analysis unit performs angle calculation formula calculation according to the spatial vector and outputs the isolating switch closing recognition result:

[0069]

[0070] In the formula, is the true angle of the spatial vector, and are the spatial vectors generated by the vector generation unit. When the angle value is less than or equal to 3 degrees, the judgment result is closing; otherwise, the judgment result is non-closing.

[0071] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0072] 1. High precision: By combining a Gaussian filter, a Sobel operator, a double-threshold detection unit, and Hough transform to extract the straight-line parameters of the disconnector arm, it has anti-noise interference and accurate edge positioning, and the homography matrix calculation is accurate;

[0073] 2. Adaptive perspective correction: Dynamically correct the deformation through the homography matrix without the need to fix the camera perspective;

[0074] 3. Fully automated process: From edge extraction to angle calculation, no manual intervention is required, which is suitable for industrial on-line detection.

[0075] In the present invention, the homography matrix only needs to be calculated once by the matrix generation module after determining the installation position of the device, and there is no need to calculate it again later. After the relative positions of the camera and the disconnector are determined, the mapping relationship between the two planes has been fixed, and accurate recognition and measurement with a high degree of automation can be achieved. In short, the present invention realizes high-precision edge detection and anti-noise ability through the arm extraction module, combines the dynamic homography matrix calculation to correct the perspective deformation, solves the problems of strong perspective dependence and much manual intervention in the traditional method, and finally realizes high-precision and high-efficiency industrial on-line detection with a fully automated process, significantly improving the positioning accuracy and system adaptability in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is the architecture diagram of the system of the present invention.

[0077] Figure 2 is the schematic diagram of the relationship of the perspective correction module.

[0078] Figure 3 is the schematic diagram of the relationship of the point pair generation unit.

[0079] Figure 4 is the flowchart of the matrix calculation unit. DETAILED DESCRIPTION OF THE INVENTION

[0080] The present invention will be further described in detail below in conjunction with the embodiments and the drawings, but the embodiments of the present invention are not limited thereto.

[0081] AsFigure 1 As shown in the figure, this embodiment discloses a disconnector closing recognition device based on spatial angle conversion, including:

[0082] A matrix generation module, configured to obtain a homography matrix, which is used to establish the mapping relationship between two planes. After determining the installation position of the device, the matrix generation module is called once to generate the homography matrix;

[0083] An arm extraction module, configured to extract the straight line parameter features of the disconnector arm from the image data;

[0084] A perspective correction module, configured to implement the coordinate mapping between the camera plane and the disconnector movement plane through the homography matrix generated in the matrix generation module, where the camera plane refers to the image data plane captured by the camera, and the disconnector movement plane refers to the actual movement plane of the disconnector; through the above mapping, the straight line parameter features of the disconnector arm are converted into the corrected straight line parameter features;

[0085] An angle calculation module, configured to calculate the actual angle value of the disconnector arm according to the corrected straight line features and output the disconnector closing recognition result.

[0086] The data flow direction is clearly marked by a solid arrow, and the matrix parameter flow direction is clearly marked by a dashed arrow. After the device is installed, the homography matrix parameters only need to be calculated once, and no further calculation is required. In the preparation stage of the device of the present invention, the point pair generation unit is first called to generate point pair data, and then the point pair data is input into the matrix calculation unit to obtain the homography matrix, and the homography matrix is output to the perspective correction module for storage; after the homography matrix is calculated and stored, the device of the present invention enters the ready stage. The arm extraction module receives the disconnector image data and extracts the straight line parameter features of the disconnector arm through the Gaussian filter unit, Sobel operator unit, double threshold detection unit, and Hough transform unit; the perspective correction module receives the straight line parameter features of the disconnector arm, and through the homography matrix projection transformation, maps the points on the camera plane to the disconnector movement plane and outputs the corrected straight line parameter features; the angle calculation module receives the corrected straight line parameter features, executes the angle calculation formula, and outputs the disconnector closing recognition result.

[0087] Specifically, the matrix generation module includes a point pair generation unit and a matrix calculation unit, where:

[0088] The point pair generation unit is configured to generate point pair data between the camera plane and the disconnector movement plane, including the following steps:

[0089] 1) As Figure 3 shown, given the camera installation height h, the elevation angle θ of a specific point relative to the camera, and the distance d between the camera and the disconnector, the height h of the specific point from the camera is determined through geometric calculation1 = tanθ·d, thereby obtaining the relative position of this point in the isolation switch movement plane; based on the geometric relationship, determine the height L of the specific point from the horizontal plane as L = tanθ·d + h, thereby obtaining the specific point t = (x t = 0, y t = L), where x t and y t are the X and Y axis coordinate values of the specific point in the isolation switch movement plane;

[0090] 2) The spatial relationship of the isolation switch is a relative coordinate set Q = {q s |q s ∈R 2 , s = 1, 2, 3,..., N}, where R 2 represents the two-dimensional real number set, the set value q s represents the relative coordinate of the s-th point with respect to the specific point, N represents the number of points, and the set Q is determined by the model of the isolation switch; calculate the corresponding coordinate set K = {k s = q s + t|q s ∈Q, s = 1, 2, 3,..., N} in the isolation switch movement plane coordinate system based on the specific point coordinates determined in step 1) and the set Q, and the set value k s represents the s-th coordinate in the isolation switch movement plane coordinate system;

[0091] 3) Read the coordinate set file in the camera plane coordinate system and convert it to generate the coordinate set P in the camera plane coordinate system;

[0092] 4) Convert the set P and the set K into a JSON file and output this JSON file.

[0093] Specifically, the matrix calculation unit is used to generate a high-precision homography matrix for the point pair data, as Figure 4 shown, including the following steps:

[0094] 1) Normalize the point pair data to improve numerical stability;

[0095] 1a) First perform a translation operation to translate the point pair data to the origin;

[0096] 1b) Then perform a scaling operation on the point pair data processed in step 1a) so that the root mean square distance of all points to the origin is a fixed value;

[0097] 2) Input the point pair data processed in step 1) into the RANSAC iterative method;

[0098] 2a) Randomly sample 4 groups of point pair data g1, g2, g3, g4 from the point pair data, where the point pair data is g i= ((x i ″, y i ″), (X i ", Y i ″)); (x i ″, y i ″) and (X i ″, Y i ″) are the i-th X and Y axis coordinate values in set P and the i-th X and Y axis coordinate values in K after translation and scaling coordinate processing in step 1), respectively;

[0099] 2b) Construct a system of linear equations:

[0100]

[0101] Each pair of points provides two equations, and four pairs of point data generate a total of 8 equations, forming a homogeneous system of equations AH K = 0; where the matrix A is the coefficient matrix of the above system of linear equations, with a size of 8×9, and each row corresponds to an equation; H K is the matrix solution vector of the equation, H K = [h 11 , h 12 , h 13 , h 21 , h 22 , h 23 , h 31 , h 32 , h 33 T is the expansion vector of H K , and is the value of the K -th row and -th column of the H

[0102] 2c) Reproject set P with H K , that is, perform a projection transformation on all points in set P;

[0103] 2d) Count the number of inliers. An inlier is a data point that satisfies the minimum error between the model prediction value and the actual observation value under the given model homography matrix; The projection error measure ∈ < 3px, where px is the pixel value, and a point with a reprojection error less than the projection error measure is determined to be an inlier;

[0104] 2e) Update the number of inliers and the homography matrix. Initially, the number of inliers is 0. When the inlier calculation result in step 2d) is greater than the current number of inliers, update the current number of inliers and update H best = H K , where H best ​is the optimal homography matrix, that is, the matrix with the largest number of inliers in step 2d); otherwise, neither the number of inliers nor H best is updated;

[0105] 2f) Execute steps 2a), 2b), 2c), 2d), and 2e) in a loop D times, where D is the loop count parameter. After the loop ends, output H best ;

[0106] 3) In each iteration of step 2), a homography matrix is calculated once, and the optimal homography matrix among all iterations is output.

[0107] Specifically, the knife arm extraction module is used to extract the straight-line features of the disconnector knife arm in the image data, and includes a Gaussian filter unit, a Sobel operator unit, a double-threshold detection unit, and a Hough transform unit, where:

[0108] The Gaussian filter unit uses a 3×3 or 5×5 convolution kernel to suppress noise in the image data, generates a convolution kernel through a two-dimensional Gaussian function, and the weights are determined by the distance of the pixel from the center point:

[0109]

[0110] I Gau = I RAW * G(u,v)

[0111] In the formula, G(u,v) is the Gaussian kernel function, σ is the standard deviation, which controls the smoothing degree, u and v are the independent variables of the Gaussian kernel function on the X-axis and Y-axis, and I RAW is the original image input, and I Gau is the image processed by Gaussian filtering, and * is the image convolution operation;

[0112] The Sobel operator unit calculates the gradient amplitude and direction of the image data after noise suppression processing in the horizontal and vertical directions, including the following steps:

[0113] 1) Obtain the kernel matrix:

[0114]

[0115]

[0116] In the formula, S x is the horizontal direction kernel, and S y is the vertical direction kernel;

[0117] 2) Gradient calculation:

[0118] G x = I Gau * S x , G y= I Gau *S y

[0119] Wherein, G x and G y are the horizontal and vertical direction gradients respectively;

[0120] 3) Obtain the gradient magnitude and gradient direction:

[0121]

[0122]

[0123] Wherein, Grdlen is the gradient magnitude and Grddir is the gradient direction;

[0124] The double-threshold detection unit sets high and low thresholds to perform edge screening on the gradient magnitude, retains the coordinate data with high gradient magnitude, and generates an edge data set Q EDGE , where Q EDGE is the set of all coordinate values that meet the edge screening conditions;

[0125] The Hough transform unit discretizes the parameter space of the edge data set Q EDGE , and screens out the straight-line parameter features of the disconnector arm, including the following steps:

[0126] 1) Map all edge data points (`x, `y) in the set Q EDGE to the point coordinates (ρ, θ′) in the polar coordinate parameter space:

[0127] ρ = `x cosθ′ + `y sinθ′

[0128] Wherein, `x, and `y are the X and Y axis coordinate values in the edge data set Q EDGE , ρ is the radial distance, representing the distance from the point to the origin; θ′ is the angular coordinate, representing the angle between the point and the positive axis;

[0129] 2) Each edge data point traverses all possible θ′, calculates the corresponding ρ, and counts in the accumulator;

[0130] 3) Select the group of (ρ, θ′) with the highest count in the accumulator as the straight-line parameter feature of the arm.

[0131] Specifically, the perspective correction module is used to implement the coordinate mapping between the camera plane and the disconnector moving plane through the homography matrix generated by the matrix generation module, where the camera plane refers to the image data plane captured by the camera, and the disconnector moving plane refers to the actual moving plane of the disconnector; through this mapping, the linear feature of the disconnector arm is transformed into the corrected linear feature, including a linear parameter input unit, a coordinate transformation unit, and a linear parameter output unit, where:

[0132] The linear parameter input unit is used to generate points on the camera plane through a given set of linear parameter features (ρ, θ′) of the disconnector arm, where ρ is the radial distance, representing the distance from the point to the origin, and θ′ is the angular coordinate, representing the angle between the point and the positive X-axis:

[0133] ρ = x cosθ′ + y sinθ′

[0134] In the formula, x and y are the coordinate values of the X and Y axes in the camera plane system;

[0135] The coordinate transformation unit is used to perform the homography matrix projection transformation to map the points on the camera plane to the disconnector moving plane;

[0136] As Figure 2 shown, this figure is divided into two planes, the disconnector moving plane and the camera plane. Specifically, the disconnector moving plane is the spatial plane where the disconnector actually moves and is perpendicular to the horizontal plane. The camera plane refers to the camera plane formed after the disconnector passes through the principle of pinhole imaging, which is equivalent to the image data. There will be a projection center between the two planes. The included angle of the disconnector arm in the camera plane is α2, and the included angle of the disconnector arm in the disconnector moving plane is α1. Generally, the included angles of the disconnector arms in the two planes are not equal. Specifically, the angles will also be distorted after perspective distortion, while the angle in the disconnector moving plane is the real angle. The homography matrix obtained through the matrix calculation unit can perform a double mapping of any two points between the two planes, transform the points on the camera plane to the disconnector moving plane, and the angle in the disconnector moving plane is the real angle, thereby realizing the correction of the perspective distortion caused by different perspectives of the camera. The specific formula is as follows:

[0137]

[0138] In the formula, H is the homography matrix, and W′ is the homogeneous component, and are the unscaled coordinate values of the X and Y axes in the disconnector moving plane system, and are the coordinate values of the X and Y axes in the disconnector moving plane system;

[0139] The straight line parameter output unit is used to convert the points in the moving plane of the disconnector into the corrected straight line parameter features and output them.

[0140] Specifically, the angle calculation module is used to calculate the actual angle value of the disconnector arm according to the corrected straight line feature data and output the disconnector closing recognition result, including a vector generation unit and an angle analysis unit, where:

[0141] The vector generation unit generates a space vector according to the corrected straight line parameter features;

[0142] The angle analysis unit performs angle calculation according to the space vector and outputs the disconnector closing recognition result:

[0143]

[0144] In the formula, is the true angle of the space vector, and are the space vectors generated by the vector generation unit. When the angle value is less than or equal to 3 degrees, the judgment result is closing, otherwise the judgment result is not closing.

[0145] The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. An isolating switch closing recognition device based on spatial angle conversion, characterized in that, Including: A matrix generation module, configured to obtain a homography matrix, which is used to establish a mapping relationship between two planes. After determining the installation position of the device, the matrix generation module is called once to generate a homography matrix; An arm extraction module, configured to extract the straight-line parameter features of the disconnector arm from the image data; A perspective correction module, configured to implement coordinate mapping between the camera plane and the disconnector motion plane through the homography matrix generated by the matrix generation module, where the camera plane refers to the image data plane captured by the camera, and the disconnector motion plane refers to the actual motion plane of the disconnector; through the above mapping, the straight-line parameter features of the disconnector arm are converted into corrected straight-line parameter features; An angle calculation module, configured to calculate the actual angle value of the disconnector arm according to the corrected straight-line parameter features, and output a disconnector closing recognition result.

2. The disconnector closing recognition device based on spatial angle conversion according to claim 1, characterized in that The matrix generation module includes a point pair generation unit and a matrix calculation unit, where: The point pair generation unit is configured to generate point pair data between the camera plane and the disconnector motion plane; The matrix calculation unit is configured to generate a high-precision homography matrix for the point pair data.

3. The disconnector closing identification device based on spatial angle conversion according to claim 2, wherein The point pair generation unit includes the following steps: 1) Given the installed height h of the camera, the elevation angle θ of a specific point relative to the camera, and the distance d between the camera and the disconnector, the height h of the specific point from the camera is determined through geometric calculation 1 = tanθ · d, thereby obtaining the relative position of this point in the movement plane of the disconnector; based on the geometric relationship, the height L of the specific point from the horizontal plane is determined as L = tanθ · d + h, thereby obtaining the specific point t = (x t = 0, y t = L), where its x t and y t are the X and Y axis coordinate values of the specific point in the movement plane of the disconnector; 2) The spatial relationship of the disconnector is a set of relative coordinates \(Q = \{q\) s |q s \(\in R\) 2 , \(s = 1, 2, 3, \ldots, N\}\), \(R\) 2 represents the two-dimensional real number set, and the set value \(q\) s represents the relative coordinate of the \(s\)-th for a specific point position. \(N\) represents the number of points. The set \(Q\) is determined by the type of the disconnector; the coordinate of the specific point position determined by step 1) and the set \(Q\) are used to calculate the coordinate set \(K=\{k\) s \(=q\) s \(+t|q\) s \(\in Q, s = 1, 2, 3, \ldots, N\}\), and the set value \(k\) s represents the \(s\)-th coordinate of the disconnector motion plane coordinate system; 3) Read the coordinate set file in the camera plane coordinate system, and convert it to generate the coordinate set P in the camera plane coordinate system; 4) Convert the set P and the set K into a JSON file, and output the JSON file.

4. The disconnector closing identification device based on spatial angle conversion according to claim 2, wherein The matrix calculation unit includes the following steps: 1) Normalize the point pair data to improve numerical stability; 1a) First perform a translation process to translate the point pair data to the origin; 1b) Then perform a scaling process on the point pair data processed in step 1a) so that the root mean square distance of all points to the origin is a fixed value; 2) Input the point pair data processed in step 1) into the RANSAC iterative method; 2a) Randomly sample 4 groups of point pairs g1, g2, g3, g4 from the point pair data, where the point pair data is g i = ((x i ″, y i ″), (X i ″, Y i ″)); (x i ″, y i ″) and (X i ″, Y i ″) are respectively the i-th X and Y axis coordinate values in the set P and the i-th X and Y axis coordinate values in K after the translation and scaling coordinate processing in step 1); 2b) Construct a linear equation system: Each pair of points provides two equations. Four pairs of point data generate a total of 8 equations, forming a homogeneous system of equations AH K = 0; where the matrix A is the coefficient matrix of the above linear system of equations, with a size of 8×9, and each row corresponds to an equation; H K is the matrix solution vector of the equation, H K = [h 11 , h 12 , h 13 , h 21 , h 22 , h 23 , h 31 , h 32 , h 33 T is the expansion vector of H K , and is the value of the K matrix at the th row and ​ 2c) With H K Reproject the set P, that is, perform a projection transformation on all points in the set P; 2d) Count the number of inliers. An inlier is a data point that satisfies the minimum error between the model prediction value and the actual observation value under the given model homography matrix; use the projection error measure ∈ < 3px, where px is the pixel value, and a point with a reprojection point error less than the projection error measure is determined as an inlier; 2e) Update the number of inliers and the homography matrix. Initially, the number of inliers is 0. When the result of inlier calculation in step 2d) is greater than the current number of inliers, update the current number of inliers and update H best = H K , where H best is the optimal homography matrix, that is, the matrix with the largest number of inliers in step 2d); otherwise, neither the number of inliers nor H best is updated; 2f) Perform steps 2a), 2b), 2c), 2d), and 2e) in a loop D times, where D is a loop count parameter, and output H after the loop ends best ; 3) Calculate a homography matrix for each iteration in step 2), and output the best homography matrix among all the iteration times.

5. The disconnector closing identification device based on spatial angle conversion according to claim 1, characterized in that The arm extraction module includes a Gaussian filter unit, a Sobel operator unit, a double-threshold detection unit, and a Hough transform unit, where: The Gaussian filter unit uses a 3×3 or 5×5 convolution kernel to suppress noise in the image data, generates a convolution kernel through a two-dimensional Gaussian function, and the weights are determined by the distance of the pixel from the center point: I Gau = I RAW * G(u, v) where \(G(u, v)\) is the Gaussian kernel function, \(\sigma\) is the standard deviation that controls the degree of smoothing, \(u\) and \(v\) are the independent variables of the Gaussian kernel function on the \(X\)-axis and \(Y\)-axis, and \(I\) RAW is the input of the original image, and \(I\) Gau is the image processed by Gaussian filtering, and \(*\) represents the image convolution operation; The Sobel operator unit calculates the gradient amplitude and direction in the horizontal and vertical directions of the image data after the noise suppression process, including the following steps: 1) Obtain the kernel matrix: Wherein, S x is the horizontal direction kernel, and S y is the vertical direction kernel; 2) Calculate the gradient: G x = I Gau * S x , G y = I Gau * S y where G x and G y are the horizontal and vertical directional gradients, respectively; 3) Obtain the gradient amplitude and gradient direction: where Grdlen is the gradient amplitude and Grddir is the gradient direction; The double-threshold detection unit sets high and low thresholds to perform edge screening on the gradient magnitude, retains the coordinate data with high gradient magnitude, and generates an edge data set Q EDGE , where Q EDGE is a set of all coordinate values that meet the edge screening conditions; The Hough transform unit discretizes the edge data set Q in the parameter space EDGE , and filters out the linear parameter features of the disconnector arm, including the following steps: 1) Map all the edge data points (`x`, `y`) in the set Q EDGE to the point coordinates (ρ, θ′) in the polar coordinate parameter space: ρ = `xcosθ′ + `ysinθ′ Wherein, `x, and` y are the X and Y axis coordinate values in the edge data set Q EDGE ; ρ is the radial distance, representing the distance from the point to the origin; θ′ is the angular coordinate, representing the angle between the point and the positive X axis 2) Traverse all possible θ′ for each edge data point, calculate the corresponding ρ, and count in the accumulator; 3) Select the (ρ, θ′) with the highest count in the accumulator as the linear parameter feature of the tool arm.

6. The disconnector closing recognition device based on spatial angle conversion according to claim 1, characterized in that The perspective correction module includes a linear parameter input unit, a coordinate transformation unit, and a linear parameter output unit, where: The linear parameter input unit is used to generate points on the camera plane through a given set of linear parameter features (ρ, θ′) of the tool arm, where ρ is the radial distance representing the distance from the point to the origin, and θ′ is the angular coordinate representing the angle between the point and the positive X-axis: ρ = xcosθ′ + ysinθ′ In the formula, x and y are the coordinate values of the X and Y axes in the camera plane system; The coordinate transformation unit is used to perform a homography matrix projection transformation to map the points on the camera plane to the isolating switch motion plane: Wherein, H is a homography matrix, and W′ is a homogeneous component, and are the coordinate values of the X and Y axes of the un-scaled isolating switch motion plane system, and are the coordinate values of the X and Y axes of the isolating switch motion plane system; The linear parameter output unit is used to convert the points on the isolating switch motion plane into corrected linear parameter features and output them.

7. The disconnector closing identification device based on spatial angle conversion according to claim 1, wherein The angle calculation module includes a vector generation unit and an angle analysis unit, where: The vector generation unit generates a spatial vector according to the corrected linear parameter features; The angle analysis unit performs angle calculation according to the spatial vector and outputs the isolating switch closing recognition result: In the formula, is the true angle of the spatial vector, and are the spatial vectors generated by the vector generation unit. When has an angle value less than or equal to 3 degrees, the judgment result is closing; otherwise, the judgment result is not closing.

Images