CN105069461B - Based on image characteristic point collinearly with the insulator chain automatic positioning method of iso-distance constraint - Google Patents

Abstract

The invention discloses it is a kind of based on image characteristic point collinearly with the insulator chain automatic positioning method of iso-distance constraint, including the extraction of image preprocessing, curvature scale space angle point grid, conllinear equidistant points, hierarchical clustering and insulator chain positioning step.The present invention is simple, practical, utilize the conllinear and iso-distance constraint of insulator chain curvature scale space angle point, it can realize the automatic positioning of the insulator chain of arbitrary major axes orientation in complicated, noisy, low resolution Aerial Images, and it takes few, to partial occlusion and fall string with robustness, solves the problems, such as that current insulator chain positioning method accuracy is low, accidentally positioning and computation complexity are high.

Description

Insulator string automatic positioning method based on image feature point collineation and isometric constraint

Technical Field

The invention belongs to the field of maintenance of operation states of power transmission and transformation equipment, and particularly relates to an automatic insulator string positioning method based on collinear and equidistant constraint of image feature points.

Background

Insulators are indispensable elements in power transmission lines and have functions of mechanical support, electrical insulation and the like. Once the insulator is damaged, it loses its effectiveness, causing an irreparable and substantial loss. Therefore, timely inspection of the insulator is necessary. And the automatic positioning of the insulator string from the aerial image is an important premise for realizing the state detection and fault diagnosis of the insulator string. The existing automatic positioning method for aerial insulator strings can be roughly divided into 4 types: (1) dividing an original image into a plurality of regions based on a segmentation method, and marking the region of interest; (2) based on an edge detection method, finding out the outline of the target of interest to realize positioning; (3) analyzing the texture characteristics of the interested target by a texture-based method, and extracting the target position by taking the texture characteristics as a criterion; (4) and extracting the characteristics of the template and the test image for matching based on a matching method, and marking the matched region.

The segmentation-based method has low positioning accuracy, can not correctly process images with similar gray features, and is not suitable for complex, noisy and low-resolution aerial images. The edge-based method is sensitive to noise, false targets with similar edges cannot be correctly processed, and the running speed of the method can be greatly reduced by images with complex backgrounds and various edge changes. The texture-based method has high calculation complexity, cannot solve the problem of positioning the insulator string when the texture difference between the insulator string and the background is not large, and is difficult to distinguish certain pseudo targets which are similar to the texture features of the insulator string. The matching-based method has strong dependence on the template, and the large number of templates can greatly reduce the feature extraction and matching speed of the template and the test image.

In complex, noisy and low-resolution aerial images, the existing insulator string positioning method has the limitations of low precision, error positioning, high calculation complexity and the like, and the shape characteristics of the insulator string in the binary image are not considered. The insulator string has obvious shape difference with the targets such as a tower, a line and the like in the binary image.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: the aerial photography insulator string automatic positioning method based on image feature point collineation and isometric constraint is provided.

The technical scheme adopted by the invention is as follows:

an automatic positioning method for an aerial insulator string based on collinear and equidistant constraint of image feature points comprises the following steps:

step a: image preprocessing: preprocessing an image containing the aerial insulator string to obtain a binary image with a smooth edge after noise is filtered;

step b: curvature scale space corner extraction: extracting an edge image of the binary image, and extracting a contour curve from the edge image; calculating the curvature of each pixel point on the contour curve under the condition that the scale sigma is 3, and taking the point with the maximum local curvature as a candidate corner point; if the curvature value of the candidate corner point is greater than a preset curvature threshold value and is greater than 2 times of the local minimum value of the neighborhood, the candidate corner point is a correct corner point; and accurately positioning the corner points of the curvature scale space corresponding to the correct corner points in the original image.

Step c: collinear equidistant point extraction: and selecting any two curvature scale space angular points, searching a third point by utilizing collinear and equidistant constraints, and if the third point is the curvature scale space angular point, judging that the three points are collinear equidistant points.

Step d: hierarchical clustering: and performing hierarchical clustering on the directions of all the collinear equidistant points to ensure that the direction change of the collinear equidistant points in each class is smaller than a preset direction threshold value, and selecting the class with the largest quantity as a collinear equidistant point set of the insulator string.

Step e: positioning an insulator string: and marking the collinear equidistant point set of the insulator string by using the minimum external rectangle so as to realize the automatic positioning of the aerial photography insulator string.

The specific steps in the step a are as follows:

step a-1: carrying out binarization processing on the image containing the aerial insulator string to obtain a binary image of the insulator string;

step a-2: performing morphological corrosion and expansion on the insulator string binary image to obtain a filtered insulator string binary image;

step a-3: and filtering small regions with areas smaller than a preset area threshold value in the filtered insulator string binary image to obtain a preprocessing result image.

The specific steps in the step b are as follows:

step b-1: extracting the canny edge in the preprocessing result image to generate an edge image;

step b-2: extracting a contour curve from the edge image, the contour curve being represented as a functional form Γ (μ, σ) at a scale σ, parameterized by an arc length μ:

Γ(μ,σ)＝(x(μ,σ),y(μ,σ)) (1)

wherein g (mu, sigma) is a Gaussian function with the scale sigma, x (mu), y (mu) are coordinate expressions with the arc length mu as a parameter,is a convolution operation;

step b-3: calculating the curvature of each pixel point on the contour curve under the condition that the scale sigma is 3, and finding out a point with the maximum local curvature as a candidate angular point;

wherein,

representing the first and second derivatives of g (mu, sigma), respectively,is a convolution operation.

Step b-4: if the curvature value of the candidate corner point is greater than a preset curvature threshold value and is greater than 2 times of the local minimum value of the neighborhood, the candidate corner point is a correct corner point;

step b-5: and accurately positioning the corner points of the curvature scale space corresponding to the correct corner points in the original image.

The specific steps in the step c are as follows:

step c-1: establishing a two-dimensional array A (N,2), wherein N is the number of angular points, and array elements of the two-dimensional array A are coordinates of angular points in a curvature scale space in an original image;

step c-2: assigning each array element in the array A to a point (x) in sequence_p,y_p)，(x_p,y_p) E.g. A, and for each point (x)_p,y_p) Repeating the steps c-3 to c-4;

step c-3: for each one different from (x)_p,y_p) Point (x) of_q,y_q) E.g. A, calculating (x)_p,y_p) And (x)_q,y_q) A distance d between_pqAnd a direction o_pq；

Step c-4: sequentially differ from (x)_p,y_p) And (x)_q,y_q) Is calculated for (x, y) and (x, y) at a point (x, y) ∈ A_p,y_p) A distance d between_pAnd a direction o_pIf d is satisfied_pAnd d_pqThe relative difference being less than a certain minimum value epsilon₁，o_pAnd o_pqAbsolute difference less than a certain minimumε₂：

Then, the three point X is determined to be { (X)_p,y_p),(x_q,y_q) And (x, y) is a collinear equidistant point, the step c-3 is turned to, and otherwise, the step c-4 is repeatedly executed.

The specific steps in the step d are as follows:

step d-1: for each group of collinear equidistant points X_j＝{(x_j1,y_j1),(x_j2,y_j2),(x_j3,y_j3) J is more than or equal to 1 and less than or equal to M, and the direction o is calculated_jM is the number of collinear equidistant point groups;

step d-2: each group of collinear equidistant points X_jSet as a cluster;

step d-3: calculating the root mean square between any two clusters to obtain a distance matrix O of the direction ═ O_ij},1≤i,j≤M；

o_ij＝o_i-o_j(8)

Step d-4: will o_ijCombining the two clusters corresponding to the minimum value into a new cluster;

step d-5: and d-3-d-4 are repeated, and when the difference between the maximum value and the minimum value of the directions of the collinear equidistant points in the clusters is larger than a preset direction threshold value, cluster merging is ended.

Adopt the produced beneficial effect of above-mentioned technical scheme to lie in:

1. according to the method, collineation and equidistant constraint of the angular points of the curvature scale space of the insulator string are utilized, so that automatic accurate positioning of the insulator string in any main shaft direction in a complex aerial image is realized, and the problems of low precision, error positioning and high calculation complexity of the existing insulator string positioning method are solved;

2. the method has the advantages of less time consumption, robustness for partial shielding and string dropping, improved positioning precision and improved automation performance.

3. The method is simple and feasible, obtains higher positioning precision, needs shorter time and does not need manual participation.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a schematic diagram of collinear and equidistant constraint of corner points in curvature scale space of an insulator string image according to the present invention;

FIG. 3 is an aerial insulator string image according to embodiment 1 of the present invention;

FIG. 4 shows the curvature scale space corner extraction result in embodiment 1 of the present invention;

FIG. 5 shows the collinear equidistant points extracted in example 1 of the present invention;

FIG. 6 shows the hierarchical clustering results of embodiment 1 of the present invention;

FIG. 7 is a minimum circumscribed rectangle frame in embodiment 1 of the present invention;

fig. 8 shows the positioning result of embodiment 1 of the present invention.

Detailed Description

The invention is further illustrated by the following figures and examples.

Example 1:

as shown in fig. 1, an automatic positioning method for an aerial insulator string based on collineation and equidistant constraint of image feature points comprises the following steps:

step a: image preprocessing: preprocessing an image containing the aerial insulator string to obtain a binary image with a smooth edge after noise is filtered;

step b: curvature scale space corner extraction: extracting an edge image of the binary image, and extracting a contour curve from the edge image; calculating the curvature of each pixel point on the contour curve under the condition that the scale sigma is 3, and taking the point with the maximum local curvature as a candidate corner point; if the curvature value of the candidate corner point is greater than a preset curvature threshold value and is greater than 2 times of the local minimum value of the neighborhood, the candidate corner point is a correct corner point; and accurately positioning the corner points of the curvature scale space corresponding to the correct corner points in the original image.

Step c: collinear equidistant point extraction: and selecting any two curvature scale space angular points, searching a third point by utilizing collinear and equidistant constraints, and if the third point is the curvature scale space angular point, judging that the three points are collinear equidistant points. Taking 3 umbrella disks as an example, as shown in fig. 2, l is the main axis direction of the insulator string, a, B, C are the corner points of the curvature scale space of the insulator string, d_AB，d_BCRespectively, the distance between AB and BC. A, B, C are approximately located on a straight line l₁Above, and l₁Parallel to l; d_AB，d_BCApproximately equal. Therefore, A, B, C are determined to be collinear equidistant points.

Step d: hierarchical clustering: and performing hierarchical clustering on the directions of all the collinear equidistant points to ensure that the direction change of the collinear equidistant points in each class is smaller than a preset direction threshold value, and selecting the class with the largest quantity as a collinear equidistant point set of the insulator string.

Step e: positioning an insulator string: and marking the collinear equidistant point set of the insulator string by using the minimum external rectangle so as to realize the automatic positioning of the aerial photography insulator string.

The specific steps in the step a are as follows:

step a-1: carrying out binarization processing on the image containing the aerial insulator string to obtain a binary image of the insulator string;

step a-2: performing morphological corrosion and expansion on the insulator string binary image to obtain a filtered insulator string binary image;

step a-3: and filtering small regions with areas smaller than a preset area threshold value in the filtered insulator string binary image to obtain a preprocessing result image.

The specific steps in the step b are as follows:

step b-1: extracting the canny edge in the preprocessing result image to generate an edge image;

step b-2: extracting a contour curve from the edge image, the contour curve being represented as a functional form Γ (μ, σ) at a scale σ, parameterized by an arc length μ:

Γ(μ,σ)＝(x(μ,σ),y(μ,σ)) (1)

wherein g (mu, sigma) is a Gaussian function with the scale sigma, x (mu), y (mu) are coordinate expressions with the arc length mu as a parameter,is a convolution operation;

step b-3: calculating the curvature of each pixel point on the contour curve under the condition that the scale sigma is 3, and finding out a point with the maximum local curvature as a candidate angular point;

wherein,

representing the first and second derivatives of g (mu, sigma), respectively,is a convolution operation.

Step b-4: if the curvature value of the candidate corner point is greater than a preset curvature threshold value and is greater than 2 times of the local minimum value of the neighborhood, the candidate corner point is a correct corner point;

step b-5: and accurately positioning the corner points of the curvature scale space corresponding to the correct corner points in the original image.

The specific steps in the step c are as follows:

step c-1: establishing a two-dimensional array A (N,2), wherein N is the number of angular points, and array elements of the two-dimensional array A are coordinates of angular points in a curvature scale space in an original image;

step c-2: assigning each array element in the array A to a point (x) in sequence_p,y_p)，(x_p,y_p) E.g. A, and for each point (x)_p,y_p) Repeating the steps c-3 to c-4;

step c-3: for each one different from (x)_p,y_p) Point (x) of_q,y_q) E.g. A, calculating (x)_p,y_p) And (x)_q,y_q) A distance d between_pqAnd a direction o_pq；

Step c-4: sequentially differ from (x)_p,y_p) And (x)_q,y_q) Is calculated for (x, y) and (x, y) at a point (x, y) ∈ A_p,y_p) A distance d between_pAnd a direction o_pIf d is satisfied_pAnd d_pqThe relative difference being less than a certain minimum value epsilon₁，o_pAnd o_pqAbsolute difference less than a certain minimum value epsilon₂：

Then, the three point X is determined to be { (X)_p,y_p),(x_q,y_q) And (x, y) is a collinear equidistant point, the step c-3 is turned to, and otherwise, the step c-4 is repeatedly executed.

The specific steps in the step d are as follows:

step d-1: for each group of collinear equidistant points X_j＝{(x_j1,y_j1),(x_j2,y_j2),(x_j3,y_j3) J is more than or equal to 1 and less than or equal to M, and the direction o is calculated_jM is the number of collinear equidistant point groups;

step d-2: each group of collinear equidistant points X_jSet as a cluster;

step d-3: calculating the root mean square between any two clusters to obtain a distance matrix O of the direction ═ O_ij},1≤i,j≤M；

o_ij＝o_i-o_j(8)

Step d-4: will o_ijCombining the two clusters corresponding to the minimum value into a new cluster;

step d-5: and d-3-d-4 are repeated, and when the difference between the maximum value and the minimum value of the directions of the collinear equidistant points in the clusters is larger than a preset direction threshold value, cluster merging is ended.

In this embodiment, the aerial original image of the single string insulator string is shown in fig. 3(a), and the aerial original image of the double string insulator string is shown in fig. 3 (b). After the two images are respectively preprocessed, the curvature scale space corner points of the two images are extracted as shown in fig. 4(a) and 4(b), so that the corner points contain rich local features and shape information, and the shape features can be preliminarily described. And removing corner points which do not meet the constraint by using the unique collinear and equidistant constraint conditions of the insulator string, and extracting collinear equidistant points which meet the constraint, as shown in fig. 5(a) and 5 (b). The maximum class is obtained by hierarchical clustering, and the direction of the collinear equidistant points of the maximum class is approximately consistent with the direction of the principal axis, as shown in fig. 6(a) and 6 (b). Finally, the collinear equidistant points of the insulator string are marked with a minimum circumscribed rectangle, as shown in fig. 7(a) and 7 (b); the result of displaying the rectangular positioning frame in the original image and automatically positioning the insulator string is shown in fig. 8(a) and 8 (b).

The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (5)

1. An automatic positioning method for an aerial insulator string based on collinear and equidistant constraint of image feature points comprises the following steps:

step a: image preprocessing: preprocessing an image containing the aerial insulator string to obtain a binary image with a smooth edge after noise filtering, wherein the binary image is used as a preprocessing result image;

step b: curvature scale space corner extraction: extracting an edge image of the binary image, and extracting a contour curve from the edge image; calculating the curvature of each pixel point on the contour curve under the condition that the scale sigma is 3, and taking the point with the maximum local curvature as a candidate corner point; if the curvature value of the candidate corner point is greater than a preset curvature threshold value and is greater than 2 times of the local minimum value of the neighborhood, the candidate corner point is a correct corner point; accurately positioning the curvature scale space corner corresponding to the correct corner in the original image;

step c: collinear equidistant point extraction: selecting any two curvature scale space angular points, searching a third point by utilizing collinear and equidistant constraints, and if the third point is the curvature scale space angular point, judging that the three points are collinear and equidistant points;

step d: hierarchical clustering: performing hierarchical clustering on the directions of all the collinear equidistant points to ensure that the direction change of the collinear equidistant points in each class is smaller than a preset direction threshold value, and selecting the class with the largest quantity as a collinear equidistant point set of the insulator string;

step e: positioning an insulator string: and marking the collinear equidistant point set of the insulator string by using the minimum external rectangle so as to realize the automatic positioning of the aerial photography insulator string.

2. The automatic positioning method for the aerial insulator string based on the collinear and equidistant constraint of the image feature points according to claim 1, which is characterized in that: the specific steps in the step a are as follows:

step a-1: carrying out binarization processing on the image containing the aerial insulator string to obtain a binary image of the insulator string;

step a-2: performing morphological corrosion and expansion on the insulator string binary image to obtain a filtered insulator string binary image;

step a-3: and filtering small regions with areas smaller than a preset area threshold value in the filtered insulator string binary image to obtain a preprocessing result image.

3. The automatic positioning method for the aerial insulator string based on the collinear and equidistant constraint of the image feature points as claimed in claim 1, which is characterized in that: the specific steps in the step b are as follows:

step b-1: extracting the canny edge in the preprocessing result image to generate an edge image;

step b-2: extracting a contour curve from the edge image, the contour curve being represented as a functional form Γ (μ, σ) at a scale σ, parameterized by an arc length μ:

Γ(μ,σ)＝(x(μ,σ),y(μ,σ)) (1)

wherein g (mu, sigma) is a Gaussian function with the scale sigma, x (mu), y (mu) are coordinate expressions with the arc length mu as a parameter,is a convolution operation;

step b-3: calculating the curvature of each pixel point on the contour curve under the condition that the scale sigma is 3, and finding out a point with the maximum local curvature as a candidate angular point;

wherein,

representing the first and second derivatives of g (mu, sigma), respectively,is a convolution operation;

step b-4: if the curvature value of the candidate corner point is greater than a preset curvature threshold value and is greater than 2 times of the local minimum value of the neighborhood, the candidate corner point is a correct corner point;

step b-5: and accurately positioning the corner points of the curvature scale space corresponding to the correct corner points in the original image.

4. The automatic positioning method for the aerial insulator string based on the collinear and equidistant constraint of the image feature points as claimed in claim 1, which is characterized in that: the specific steps in the step c are as follows:

step c-1: establishing a two-dimensional array A (N,2), wherein N is the number of angular points, and array elements of the two-dimensional array A are coordinates of angular points in a curvature scale space in an original image;

step c-2: assigning each array element in the array A to a point (x) in sequence_p,y_p)，(x_p,y_p) E.g. A, and for each point (x)_p,y_p) Repeating the steps c-3 to c-4;

step c-3: for each one different from (x)_p,y_p) Point (x) of_q,y_q) E.g. A, calculating (x)_p,y_p) And (x)_q,y_q) A distance d between_pqAnd a direction o_pq；

Step c-4: sequentially differ from (x)_p,y_p) And (x)_q,y_q) Is calculated for (x, y) and (x, y) at a point (x, y) ∈ A_p,y_p) A distance d between_pAnd a direction o_pIf d is satisfied_pAnd d_pqThe relative difference being less than a certain minimum value epsilon₁，o_pAnd o_pqAbsolute difference less than a certain minimum value epsilon₂：

Then, the three point X is determined to be { (X)_p,y_p),(x_q,y_q) And (x, y) is a collinear equidistant point, the step c-3 is turned to, and otherwise, the step c-4 is repeatedly executed.

5. The automatic positioning method for the aerial insulator string based on the collinear and equidistant constraint of the image feature points as claimed in claim 1, which is characterized in that: the specific steps in the step d are as follows:

step d-1: for each group of collinear equidistant points X_j＝{(x_j1,y_j1),(x_j2,y_j2),(x_j3,y_j3) J is more than or equal to 1 and less than or equal to M, and the direction o is calculated_jM is the number of collinear equidistant point groups;

step d-2: each group of collinear equidistant points X_jSet as a cluster;

step d-3: calculating the root mean square between any two clusters to obtain a distance matrix O of the direction ═ O_ij},1≤i,j≤M；

o_ij＝o_i-o_j(8)

Step d-4: will o_ijCombining the two clusters corresponding to the minimum value into a new cluster;

step d-5: and d-3-d-4 are repeated, and when the difference between the maximum value and the minimum value of the directions of the collinear equidistant points in the clusters is larger than a preset direction threshold value, cluster merging is ended.