JP6845169B2 - Image processing equipment, image processing methods and programs - Google Patents

Description

Embodiments of the present invention relate to image processing devices, image processing methods and programs.

As one of the maintenance and management of overhead lines (hereinafter referred to as overhead lines) stretched in the air such as power transmission lines, inspections are carried out using aerial images (videos) of the overhead lines.

In this inspection, the presence or absence of an abnormality in the overhead line is determined by visually checking the aerial image by the operator.

However, such inspection work requires a great deal of labor and time. Therefore, labor saving in the inspection work of overhead lines is desired.

On the other hand, in recent years, it has been studied to automatically determine the presence or absence of abnormalities in overhead lines by image processing on aerial images, but in order to perform efficient image processing, the aerial image It is necessary to detect the position of the overhead line (the area including the overhead line).

Japanese Patent No. 4255065

Therefore, an object to be solved by the present invention is to provide an image processing device, an image processing method, and a program capable of detecting the position of an overhead line from an image with high accuracy.

According to the embodiment, an image processing device for detecting the position of an overhead line in an image is provided. The image processing device includes acquisition means and detection means. The acquisition means acquires an image to be detected for detecting the position of the overhead line. The detection means detects a region on a straight line in which the direction of the edge is substantially homogeneous in the acquired image as the position of the overhead line. The detection means includes an extraction means, a voting means, a first calculation means, and a determination means. The extraction means extracts a feature amount composed of a plurality of elements at each of the plurality of feature amount extraction positions in the image. The voting means votes the extracted feature amount in the parameter space based on the feature amount extraction position where the feature amount is extracted. The first calculation means is an evaluation that reflects that the edge direction is a substantially uniform linear region for each parameter representing a straight line in the image based on the feature amount voted in the parameter space. Calculate the value. The determination means determines a parameter representing a straight line corresponding to the position of the overhead line based on the calculated evaluation value.

The figure for demonstrating the technique of detecting a linear region from an image. The figure which shows the relationship of the parameters θ and ρ which represent a straight line. The figure which shows an example of the overhead wire formed by twisting the strands. The block diagram which shows an example of the functional structure of the image processing apparatus which concerns on 1st Embodiment. The figure which shows an example of the hardware composition of an image processing apparatus. The flowchart which shows an example of the processing procedure of an image processing apparatus. The figure for demonstrating the feature amount extraction position. A diagram for conceptually explaining a structural tensor. The figure for conceptually explaining the parameter space. The figure for demonstrating the structure tensor which is converted from the sum of the features voted in the parameter space. The figure for demonstrating the process of calculating the width of an overhead line. The block diagram which shows an example of the functional structure of the image processing apparatus which concerns on 2nd Embodiment. The flowchart which shows an example of the processing procedure of an image processing apparatus. The figure for demonstrating a plurality of pattern recognition target areas set on the area containing a straight line represented by a parameter.

Hereinafter, each embodiment will be described with reference to the drawings.
(First Embodiment)
First, the first embodiment will be described. The image processing apparatus according to the present embodiment is used to obtain the position of an overhead line in an image when inspecting an overhead line such as a power transmission line using, for example, an aerial image (video).

Here, since the overhead line is generally stretched in the air by concrete columns or steel towers, it draws a catenary curve as a whole, but it is a part so that abnormalities can be confirmed during the above inspection. It is magnified and photographed (aerial photography). Therefore, the overhead line in the image can be regarded as a straight line. Therefore, the position of the overhead line can be obtained by detecting a linear region from the image.

The Hough transform is known as a method for detecting a linear region from an image. As shown in FIG. 1, assuming that a straight line passing through a point (x, y) in an image is represented by an angle θ and a distance ρ from an arbitrarily set coordinate origin, the formula of the straight line is ρ = xcosθ + ysinθ. It becomes.

There are innumerable straight lines passing through the point (x, y), but the relationship between the parameters θ and ρ representing the straight line passing through the point (x, y) is shown as shown in FIG. According to the relationship between the parameters θ and ρ, the value of ρ is uniquely determined with respect to θ.

In the Hough transform, parameters of various straight lines passing through each of all the points that are candidates for detection such as edge points in the image are calculated using the above equation and voted in the parameter space. By specifying the accumulation point from this voting result, the parameter of one straight line can be obtained.

However, the overhead wire is generally a stranded wire formed by twisting the strands as shown in FIG. Therefore, when detecting the position of the overhead line as described above, there is a possibility that stable detection cannot be performed by the Hough transform.

Specifically, the Hough transform is a method of detecting a single straight line from the entire image, but when the image (aerial image) contains a miscellaneous background, the influence of many edge points on the background. Incorrect parameters (ie, straight lines) may be detected in. On the other hand, a measure may be taken to limit the range of parameters to vote (for example, θ) by finding the direction of the edge at the edge point, but in the case of a stranded wire such as an overhead wire, the stranded wire is twisted. This measure does not work effectively because the combined strands form an edge diagonal to the direction of the overhead line.

Therefore, in the image processing apparatus according to the present embodiment, by accumulating and analyzing the directions of edges scattered in the image in a parameter space representing a straight line, a linear region in which the direction of the edges is substantially uniform in the image is formed. Shall be detected as the position of the overhead line.

Hereinafter, the image processing apparatus according to the present embodiment will be described. FIG. 4 is a block diagram showing an example of the functional configuration of the image processing apparatus according to the present embodiment.

As shown in FIG. 4, the image processing device 10 includes an image acquisition unit 11, a feature extraction unit 12, a voting unit 13, an evaluation value calculation unit 14, and a parameter determination unit 15.

The image acquisition unit 11 acquires an image (hereinafter referred to as a target image) to be detected for detecting the position of the overhead line. The target image is an image of an overhead line taken aerial, and may be an image taken from an aircraft such as a helicopter, or an image taken using a drone or the like.

The feature extraction unit 12, the voting unit 13, the evaluation value calculation unit 14, and the parameter determination unit 15 described below detect a linear region having substantially uniform edge directions as the position of an overhead line in the above image. Functions as a department.

The feature extraction unit 12 extracts a feature amount composed of a plurality of elements at each of the plurality of feature amount extraction positions in the target image.

The voting unit 13 votes the feature amount extracted by the feature extraction unit 12 in the parameter space based on the feature amount extraction position from which the feature amount is extracted.

The evaluation value calculation unit 14 calculates an evaluation value for each parameter representing a straight line in the target image based on the voting result (feature amount voted in the parameter space) by the voting unit 13. In this embodiment, the parameter includes the above-mentioned set of θ and ρ.

The parameter determination unit 15 determines a parameter representing a straight line corresponding to the position of the overhead line based on the evaluation value calculated by the evaluation value calculation unit 14.

FIG. 5 shows an example of the hardware configuration of the above-mentioned image processing device. As shown in FIG. 5, the image processing device 10 includes a CPU 101, a non-volatile memory 102, a main memory 103, a communication device 104, and the like.

The CPU 101 is a hardware processor that controls the operation of various components in the image processing device 10. The CPU 101 executes various programs loaded from the non-volatile memory 102, which is a storage device, into the main memory 103. The program executed by the CPU 101 includes an operating system (OS), a program for detecting the position of an overhead line from the above-mentioned image (hereinafter referred to as an image processing program), and the like. The CPU 101 also executes, for example, a basic input / output system (BIOS), which is a program for hardware control.

A part or all of the image acquisition unit 11, the feature extraction unit 12, the voting unit 13, the evaluation value calculation unit 14, and the parameter determination unit 15 described above are stored in the CPU 101 (that is, the computer of the image processing device 10) as an image processing program. That is, it shall be realized by software. The image processing program may be stored and distributed in a computer-readable storage medium, or may be downloaded to the image processing device 10 via a network. A part or all of each of these parts 11 to 15 may be realized by hardware such as an IC (Integrated Circuit), or may be realized as a combination configuration of the software and hardware.

The communication device 104 is a device configured to perform, for example, wired or wireless communication with an external device.

Although only the CPU 101, the non-volatile memory 102, the main memory 103, and the communication device 104 are shown in FIG. 5, the image processing device 10 is, for example, an HDD (Hard Disk Drive) and an SSD (Solid State Drive). Such other storage devices may be provided, or an input device, an output device, and the like may be provided.

Next, an example of the processing procedure of the image processing apparatus 10 according to the present embodiment will be described with reference to the flowchart of FIG. The process shown in FIG. 6 is executed, for example, when inspecting an overhead line, but it may be executed in advance before the inspection.

First, the image acquisition unit 11 acquires the target image (the target image for detecting the position of the overhead line) (step S1). The target image is an image in which an overhead line is aerial photographed as described above, but it may be acquired (received) from an external device via a network or the like, or a storage unit (storage device) in the image processing device 10. ) May be obtained.

When the target image acquired in step S1 is a multi-channel image such as a color image, the image acquisition unit 11 shall convert the target image into an image represented by a single color (1 channel image). The image represented by a single color includes a monochrome image or the like having one pixel value for one pixel.

In the following description, the image that has been converted by the image acquisition unit 11 is also referred to as a target image for convenience.

Next, the feature extraction unit 12 extracts a feature amount calculated based on the brightness gradient vector of the image around the feature amount extraction position at each of a large number of predetermined feature amount extraction positions arranged on the target image. (Step S2).

Hereinafter, the process of step S2 will be specifically described. First, the feature amount extraction position will be described with reference to FIG. 7. In the present embodiment, the feature amount extraction positions 201 are arranged at substantially equal intervals on the target image 200 as shown in FIG. 7. The feature amount extraction positions 201 may be arranged so as to have a one-to-one correspondence with the pixels constituting the target image 200, or may be arranged at intervals larger than the intervals of the pixels.

Further, for the feature amount extraction position 201, a feature amount extraction area 202, which is a certain area including the feature amount extraction position 201, is set. As shown in FIG. 7, the feature amount extraction area 202 may be set so as to overlap with other adjacent feature amount extraction areas.

Although the feature amount extraction region 202 is shown to have a circular shape centered on the feature amount extraction position 201, it may have another shape or the like.

The feature extraction unit 12 obtains a structural tensor described below for each feature extraction position set as described above.

Here, horizontal differentiation at a point of the target image (x, y), respectively vertical differentiation _{l x (x, y),} l y (x, y) to. The horizontal derivative and the vertical derivative (elements of the luminance gradient vector) can be obtained by a filter calculation using a Sobel filter or the like. Moreover, the set of pixels included in the feature amount extraction area set for the i-th feature quantity extraction position of the arranged feature quantity extraction position on the target image and P _i. It is assumed that the number of feature amount extraction positions arranged on the target image is L, and i = {1, 2, ..., L}.

In this case, the structure tensor A _i in the i-th feature quantity extraction position can be determined by the following equation (1). Note that w (w, y) in the equation (1) is a predetermined weight in (x, y).

上記した式（１）によって求められる構造テンソルＡ_ｉは、対象画像中のｉ番目の特徴量抽出位置について設定されている特徴量抽出領域における輝度勾配ベクトルの統計量であり、２×２の対称行列となる。

Structure tensor A _i obtained by the equation (1) described above is a statistic of the intensity gradient vectors in the feature amount extraction area set for the i-th feature quantity extraction position in the target image, the 2 × 2 symmetric It becomes a matrix.

Specifically, the brightness gradient vector is a vector representing a characteristic of a change in luminance as shown in the left side of FIG. 8 (e.g., high brightness direction), the structure tensor A _i is the as shown in the right side of FIG. 8 Represents the distribution of the luminance gradient vector.

Here, assuming that the distribution of the brightness gradient vector represented by the structural tensor _Ai is represented by an elliptical shape, the direction of the major axis of the elliptical shape is the first eigenvector and the length (major axis) of the major axis. The square corresponds to the first eigenvalue, the direction of the elliptical minor axis (direction orthogonal to the first eigenvector) corresponds to the second eigenvector, and the square of the length (minor axis) of the minor axis corresponds to the second eigenvalue.

In this case, the first eigenvalue of the structure tensor A _i represents the variance of the direction of the intensity gradient vectors of the first eigenvector of the structure tensor A _i. The second eigenvalue of the structure tensor A _i represents the variance of the direction of the intensity gradient vectors of the second eigenvector of the structure tensor A _i. Thus, for example, if the first eigenvalue of the structure tensor A _i is significantly larger with respect to the second eigenvalue, intensity gradient vectors of the feature extraction region determined the structure tensor A _i is be biased in a particular direction I understand.

Feature extraction unit 12, a linear combination of the elements of the structure tensor A _i described above, is extracted as a feature quantity of the i-th feature quantity extraction position. Here, since the structural tensor _Ai is a 2 × 2 symmetric matrix as described above, only the three elements of the upper triangular component (1 row 1 column component, 1 row 2 column component, 2 row 2 column component) are used. It can be expressed by. When the _{a 12,} 2 × 2 matrix components of one row and one column component _{a 11,} 1 row 2 column component of the upper triangular component is _{a 22,} feature extraction unit 12, the _{a 11, a 12} and a _A vector consisting of 22 three elements is extracted as a feature quantity.

That is, the feature amount extracted by the feature extraction unit 12 in the present embodiment (feature amount extracted at each of the plurality of feature amount extraction positions) is an element of the gradient vector (brightness gradient vector) at the feature amount extraction position. It includes a feature quantity consisting of a plurality of elements represented by the sum of quadratic equations.

The feature amount extracted by the feature extraction unit 12 may be any one capable of calculating an evaluation value described later, for example, a vector consisting of two elements _{a 11} −a ₂₂ and a _{12 and the like.} It may be present or it may be something else.

By performing such processing for each feature amount extraction position (feature amount extraction area), the feature extraction unit 12 can extract the feature amount for each of the feature amount extraction positions arranged on the target image.

The feature extraction unit 12 does not have to extract the feature amount for all the feature amount extraction positions arranged on the target image. Specifically, if the position (existence range) of the overhead line in the target image can be predicted by the position of the overhead line detected from the frame (image) acquired before the target image, the prediction is made. The feature amount may be extracted only for the feature amount extraction position existing in the specified range. In addition, when there is an area in which specific information (for example, time) is displayed in the target image, the feature amount is not extracted for the feature amount extraction position existing in the area (that is, the feature amount). It may be excluded from the extraction target).

Returning to FIG. 6 again, the voting unit 13 votes the feature amount extracted in step S12 in the parameter space based on each feature amount extraction position from which the feature amount is extracted (step S3).

Here, with reference to FIG. 9, the parameter space (voting space) in which the feature amount is voted in step S3 will be conceptually described.

Assuming that the straight line in the target image is represented by two parameters of the angle θ and the distance ρ as described above, in the present embodiment, the range of θ is divided into M sections, and each of the divided sections is divided in this way. Let the representative value of the interval be θ _j (j = {1, 2, ..., M}). Similarly, the range of ρ is divided into N intervals, and the representative value of each interval divided in this way is ρ _k (k = {1, 2, ..., N}). The representative value of each section is, for example, the median value of the value corresponding to the section.

_{In the present embodiment, a storage area v jk} for voting the feature quantity is prepared _{for all the parameters (θ j} , ρ _k ) of the parameter space 300 quantized in this way. In FIG. 9, for example, the storage area 301 _{shows the storage area v 11 prepared for the parameters (θ 1} , ρ ₁ ), and for example, the storage area 302 shows the storage area v prepared for the _{parameters (θ M} , ρ _N ). _{Shows MN} .

投票部１３は、上記した処理をｉ＝｛１，２，…，Ｌ｝とｊ＝｛１，２，…，Ｍ｝の組み合わせ（つまり、特徴量抽出位置とθの組み合わせ）毎に実行する。

The voting unit 13 executes the above processing for each combination of i = {1, 2, ..., L} and j = {1, 2, ..., M} (that is, a combination of the feature amount extraction position and θ). ..

For example, when the feature amount extraction position is a point (x ₁ , y ₁ ), the voting unit 13 uses the feature amount extraction position (x ₁ , y ₁ ) and the value of ρ corresponding to _{θ 1} based on the equation (2). _{Is calculated, and ρ k ′} closest to the calculated value of ρ is specified. The voting unit 13 has the feature amount f ₁ extracted for the feature amount extraction position (x ₁ , y ₁ ) in the storage area v 1 _{k'prepared for the parameters (θ 1} , ρ _k' ) of the parameter space 300 described above. To vote.

_{Next, the voting unit 13 calculates the value of ρ corresponding to θ 2} based on the feature amount extraction position (x ₁ , y ₁ ) and the equation (2), and ρ closest to the calculated value of ρ. Identify _k'. The voting unit 13 votes the feature amount f ₁ extracted for the feature amount extraction position (x ₁ , y ₁ ) in the storage area v 1 _{k'prepared for the parameters (θ 2} , ρ _{k') of the parameter space 300.} To do.

Note that the _k'[rho identified from the value of [rho corresponding to ₂ [rho _k'and theta identified from the value of [rho corresponding to theta _1, may be the same or may be different.

When the same process is _{executed up to θ M} , the voting process for the feature extraction position (x ₁ , y _{1) is completed.} By executing such a voting process for all the feature amount extraction positions (that is, the first to Lth feature amount extraction positions), the voting unit 13 was extracted for each feature amount extraction position in step S2. Features can be voted on for each parameter in the parameter space 300.

According to the voting result by the voting unit 13, for example, the parameter space 300 includes a parameter in which the feature amount of a plurality of feature amount extraction positions is voted (added) and a parameter in which the feature amount is not voted. To do.

Although the processing is described here assuming that the processing is executed for all combinations of the feature amount extraction position and θ, the position of the overhead line in the target image is predicted by, for example, the position of the overhead line detected from the previous frame. If possible, the features may be voted on only for the parameters corresponding to the predicted range. According to this, it is possible to reduce the processing amount in the voting unit 13.

_{Further, ρ k ′} closest to the value of ρ corresponding to _{θ j} calculated based on the equation (2) is specified, and the storage area v _{jk ′ prepared for the parameters (θ j} , ρ _{k ′} ) is characteristic. Although the amount is explained as being voted _{, a plurality of ρ k'that} are close to the value of ρ corresponding to _{θ j} are specified, and the feature amount is apportioned based on the difference between the value of ρ and the value of ρ _k'. However, a plurality of parameters (θ _j , ρ _{k ′} ) may be voted for respectively. That is, for example _{, assuming that a plurality of ρ k ′} close to the value of ρ corresponding to _{θ j} are represented by ρ _{k ′} 1 and ρ _{k ′} 2, for one θ _j , (θ _j , ρ _{k ′} 1) and ( The feature quantity may be voted for two sets of parameters of _{θ j} and ρ _{k ′ 2).}

Returning to FIG. 6 again, the evaluation value calculation unit 14 calculates the evaluation value for each parameter based on the feature amount voted for each parameter in step S3 (step S4).

Hereinafter, the process of step S4 will be specifically described. Here, it is assumed that the evaluation value calculation unit 14 calculates the evaluation value of the parameter (θ _j , ρ _k ). In this case, the evaluation value calculation unit 14 calculates the evaluation value with respect to the sum of the feature quantities voted for the _{parameters (θ j} , ρ _{k) in the parameter space.}

Here, in the case of using a vector of three elements of _{a 11, a 12} and _{a 22} is a triangular component on the structure tensor _{A i} described above as the characteristic amount, the parameter (θ _j, ρ _k) to It is assumed that the sum of the features voted in the above is a vector consisting of three elements _{, for example, b 11} , b ₁₂ and b _22. In this case, the parameter (θ _j, ρ _k) the sum of the voting feature amount with respect to the first row of the first column component _{b 11,} one row two rows component _{b 12,} 2 rows and one column component _{b 12,} It can be converted into a _{structure tensor A jk} which is a 2 × 2 symmetric matrix having a 2-row 2-column component as b _22.

The above-mentioned structural tensor A _i represented the distribution of the brightness gradient vector in the i-th feature amount extraction region in the target image, but the structural tensor A _jk has parameters (θ _j , ρ _k ) as shown in FIG. It can be regarded as a structural tensor calculated in the region 400 along the straight line represented by, and can be said to represent the distribution of the brightness gradient vector in the region 400.

That is, if the first eigenvalue of such a structural tensor A _jk is significantly larger than the second eigenvalue, an aligned edge is formed on the region 400 along the straight line represented by _{(θ j} , ρ _k). You can see that they are lined up.

Therefore, in the present embodiment, the evaluation value calculation unit 14 calculates _{the difference s jk} between the first eigenvalue and the second eigenvalue of the _{structural tensor A jk} as the evaluation value.

Calculating the triangular component on the structure tensor _{A jk} and _{b 11, b 12, b 22} as described above, the difference _{s jk} of the first eigenvalue and a second eigenvalue of the structure tensor _{A jk} is the following formula (3) Will be done.

評価値算出部１４は、上記したパラメータ空間のパラメータの全てについて上記した式（３）に基づいて評価値（第１固有値及び第２固有値の差分）を算出する。なお、評価値の算出処理は、パラメータ空間の全てのパラメータのうち、ステップＳ３において特徴量が投票されたパラメータについてのみ実行されるようにしてもよい。

The evaluation value calculation unit 14 calculates the evaluation value (difference between the first eigenvalue and the second eigenvalue) based on the above equation (3) for all the parameters in the above parameter space. The evaluation value calculation process may be executed only for the parameter whose feature amount is voted in step S3 among all the parameters in the parameter space.

In this case, the evaluation value can be calculated if there are two values, the difference between the diagonal components of the structural tensor and the off-diagonal component. Therefore, the feature amount extracted in step S2 is the structural tensor as described above. It may be a two-dimensional vector whose elements are the difference between the diagonal components (that is, a ₁₁ −a ₂₂ ) and the off-diagonal component (that is, a _12). According to such a configuration, the amount of memory and the amount of processing (calculation amount) can be reduced.

Further, here, the difference s _jk shown in the equation (3) is calculated as the evaluation value, but if the evaluation value is the _{square of the difference s jk} , the square root calculation becomes unnecessary, so further calculation is performed. It is possible to reduce the amount.

Returning to FIG. 6 again, the parameter determination unit 15 determines a parameter representing a straight line corresponding to the position of the overhead line in the target image based on the evaluation value calculated for each parameter in step S4 (step S5). ..

Here, it is assumed that among the evaluation values calculated for each parameter in step S4, for example, the evaluation value _{sop calculated for the parameter (θ o} , ρ _p ) is the maximum value, and the parameter (θ _o , ρ _p). ) Is determined as a parameter representing a straight line corresponding to the position of the overhead line. Note that θ _o is one of the above-mentioned θ _{1 to θ M} , and ρ _p is one of the above-mentioned ρ _{1 to ρ N.}

The evaluation value calculated for each parameter is regarded as a function s (θ, ρ), and s (θ) is based on the evaluation value of the parameter located around the _{parameter (θ o} , ρ _{p) in the parameter space. The vicinity of o} , ρ _p ) may be approximated by parabolic fitting, and θ, ρ, which can obtain the maximum value of the approximated function, may be determined as a parameter representing a straight line corresponding to the position of the overhead line.

The parameter determined in step S5 may be output to an external device that executes a process of determining the presence or absence of an abnormality in the overhead line via the network, or is output to a storage device such as the non-volatile memory 102. , May be managed in the image processing apparatus 10.

As described above, in the present embodiment, the target image (target image) for detecting the position of the overhead line is acquired, and the position of the overhead line is defined as a linear region in which the edge direction is substantially homogeneous in the target image. Detect as. That is, in the present embodiment, as shown in FIG. 3 described above, the Hough transform is extended by using the structural tensor, paying attention to the point that the direction of the edge becomes uniform in the direction along the overhead line (longitudinal direction). As a result, a linear region in which the directions of the edges are aligned is detected as the position where the overhead line exists.

In the present embodiment, such a configuration makes it possible to detect the position of the overhead line from the image with high accuracy without human intervention.

In the present embodiment, a feature amount composed of a plurality of elements is extracted at each of a plurality of feature amount extraction positions in the target image, and the extracted feature amount is used as the feature amount extraction position from which the feature amount is extracted. The parameter space is voted based on the above, and the evaluation value is calculated for each parameter representing a straight line in the target image based on the feature amount voted in the parameter space. In the present embodiment, the parameter representing the straight line corresponding to the position of the overhead line is determined based on the evaluation value calculated in this way. The feature amount extracted at each of the plurality of feature amount extraction positions includes a feature amount composed of a plurality of elements represented by the sum of the quadratic equations of the elements of the gradient vector at the feature amount extraction position. Further, in the present embodiment, the structural tensor is calculated based on the sum of the feature quantities voted for the parameters in the parameter space, and the difference between the first eigenvalue and the second eigenvalue of the calculated structural tensor is used as the parameter. Calculated as an evaluation value. In the present embodiment, by executing such a process, it is possible to realize the detection of a linear region having substantially uniform edge directions in the above-mentioned target image.

Here, when the target image is a multi-channel image such as a color image, it is difficult to define the luminance gradient vector in the present embodiment. Therefore, when the target image is a color image or the like, it is easy to define light and darkness such as an image in which the target image is represented by a single color (a monochrome image having one pixel value for one pixel). Convert to. According to such a configuration, the luminance gradient vector can be appropriately defined in the target image, and the detection accuracy of the position of the overhead line can be improved.

Further, in the present embodiment, the angle θ and the distance ρ are determined (output) as parameters representing the straight line corresponding to the position of the overhead line, but the width of the overhead line in the target image is further increased as the parameters. You may ask for it.

Here, the evaluation value s _op of the evaluation value s _jk are calculated for each parameter as described above is assumed to be the maximum value, the straight line parameters (θ _o, ρ _p) corresponds to the position of the overhead wire It shall be determined as a parameter to be represented. In this case, in the target image, the overhead line exists in the direction of the _{angle θ o} from the coordinate origin and at the position of the _{distance ρ p.}

In this case, assuming that the above-mentioned evaluation value s _jk is regarded as a pixel value with respect to the position (j, k), the _{evaluation value obtained when the angle θ o} is fixed and the distance ρ is changed is shown in FIG. It changes as shown. In FIG. 11, the portion having a low evaluation value corresponds to the region other than the overhead line (background), and the portion having a high evaluation value corresponds to the region where the overhead line exists. According to this, the point that first falls below the threshold value in the left-right direction when viewed from the position where the evaluation value is maximum is regarded as the boundary between the overhead line and the background, and the overhead line is specified by specifying the distance ρ corresponding to each point. It becomes possible to calculate the width of.

The width of the overhead line calculated in this way can be used in the above-mentioned process of determining the presence or absence of an abnormality in the overhead line (hereinafter referred to as an abnormality determination process), and contributes to improving the accuracy of the abnormality determination. be able to.

The abnormality discrimination process may be executed by an external device, or may be executed by a discrimination unit (not shown) included in the image processing device 10. The abnormality discrimination process may be executed using, for example, a learning model constructed by learning an image of an overhead line having an abnormality and an image of an overhead line without an abnormality, or may be executed by another method. I do not care. According to such an abnormality determination process, it is possible to significantly reduce the labor and time required for the operator to visually check the aerial image (video), for example, and to save the labor for the inspection work of the overhead line. ..

In the present embodiment, the image processing device 10 has been described as one device, but for example, the image processing device 10 is realized by a plurality of devices, and each of the parts 11 to 15 shown in FIG. 4 is distributed to the plurality of devices. It doesn't matter if it is done.

(Second embodiment)
Next, the second embodiment will be described. In this embodiment, detailed description of the same parts as those of the first embodiment described above will be omitted, and different parts will be mainly described.

FIG. 12 is a block diagram showing an example of the functional configuration of the image processing apparatus according to the present embodiment. In FIG. 12, the same reference numerals are given to the same parts as those in FIG.

As shown in FIG. 12, the image processing apparatus 10 according to the present embodiment includes the image acquisition unit 11, the feature extraction unit 12, the voting unit 13, the evaluation value calculation unit 14, and the parameter determination described in the first embodiment described above. In addition to the unit 15, the determination unit 16 is included.

The determination unit 16 determines whether or not an overhead line exists on a straight line represented by the parameter determined by the parameter determination unit 15 (that is, a region in the target image corresponding to the parameter).

The image processing device 10 according to the present embodiment is different from the above-described first embodiment in that the image processing device 10 includes such a determination unit 16.

Next, an example of the processing procedure of the image processing apparatus 10 according to the present embodiment will be described with reference to the flowchart of FIG.

First, the processes of steps S11 to S15 corresponding to the processes of steps S1 to S5 shown in FIG. 6 are executed.

When the process of step S15 is executed, the determination unit 16 determines whether or not an overhead line exists in the region including the straight line represented by the parameter determined in step S15 (hereinafter referred to as the target region). (Step S16).

Here, the process of step S16 will be described with reference to FIG. FIG. 14 shows a target region 501 in the target image 500.

In step S16, the determination unit 16 sets a plurality of pattern recognition target areas on the target area 501. In FIG. 14, areas 502 to 506 are set as a plurality of pattern authentication target areas.

The determination unit 16 determines whether or not the images in the plurality of pattern recognition target areas 502 to 506 set on the target area 501 are overhead lines (the images include the overhead lines). For this determination process, for example, a general two-class classifier using a neural network or the like can be used. The classifier shall learn a large number of overhead line images and non-overhead line images in advance. According to such a discriminator, it is possible to determine whether or not the image in each pattern recognition target area 502 to 506 is an overhead line. The classifier may be a discriminator by another method such as a support vector machine (SVM).

When the determination unit 16 determines that the image in the pattern recognition target area equal to or larger than the predetermined ratio of the pattern recognition target areas 502 to 506 described above is an overhead line, the determination unit 16 establishes an overhead line in the target area 501. Determined to exist. Specifically, assuming that the predetermined ratio is, for example, 0.8, an overhead line exists in the target area 501 when the image is determined to be an overhead line in four or more pattern recognition target areas. Then it is judged.

Returning to FIG. 13 again, when it is determined in step S16 that an overhead line exists in the target area (YES in step S16), the parameter determined in step S15 is output as a parameter representing a straight line corresponding to the position of the overhead line. (Step S17).

On the other hand, when it is determined that the overhead line does not exist in the target area (NO in step S16), the process of step S17 is not executed, and the parameter determined in step S15 is not output.

In the present embodiment, it corresponds to the position of the overhead line by the configuration of determining whether or not the overhead line exists in the target area (on the straight line represented by the parameter determined by the parameter determination unit 15) in the target image. It is possible to output a parameter representing a straight line with a high probability of being performed, and it is possible to improve the detection accuracy of the position of the overhead line.

In the present embodiment, it is possible to determine whether or not an overhead line exists in each of the plurality of images constituting the aerial image by using the determination result by the determination unit 16. According to this, even if the section (tracking section) in which each overhead line is photographed is unknown in a long-time aerial image in which a large number of overhead lines are tracked in sequence, the overhead line exists. By extracting only the frame (image) to be processed and executing the abnormality determination process, it is possible to realize a significant labor saving in the inspection work of the overhead line.

According to at least one embodiment described above, it is possible to provide an image processing device, an image processing method, and a program capable of detecting the position of an overhead line from an image with high accuracy.

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, as well as in the scope of the invention described in the claims and the equivalent scope thereof.

10 ... Image processing device, 11 ... Image acquisition unit, 12 ... Feature extraction unit, 13 ... Voting unit, 14 ... Evaluation value calculation unit, 15 ... Parameter determination unit, 16 ... Judgment unit, 101 ... CPU, 102 ... Non-volatile memory , 103 ... Main memory, 104 ... Communication device.

Claims (9)

In an image processing device that detects the position of an overhead line in an image
An acquisition means for acquiring an image to be detected for the position of the overhead line, and
It is provided with a detecting means for detecting a linear region having substantially uniform edge directions in the acquired image as an overhead line position.
The detection means
An extraction means for extracting a feature amount composed of a plurality of elements at each of a plurality of feature amount extraction positions in the image, and an extraction means.
A voting means for voting the extracted feature amount in the parameter space based on the feature amount extraction position where the feature amount was extracted, and
The first calculation for calculating an evaluation value reflecting that the edge direction is a substantially homogeneous linear region for each parameter representing a straight line in the image based on the feature amount voted in the parameter space. Means and
As a determination means for determining a parameter representing a straight line corresponding to the position of the overhead line based on the calculated evaluation value.
including
Images processing device. Feature amount extracted in each of the plurality of feature quantity extraction position, according to claim 1, further comprising a feature amount including a plurality of elements is expressed by the sum of the quadratic components of the gradient vector in the feature quantity extraction position Image processing equipment. The first calculation means calculates a structural tensor based on the sum of the feature quantities voted for the parameter in the parameter space, and calculates the difference between the first eigenvalue and the second eigenvalue of the calculated structural tensor of the parameter. The image processing apparatus according to claim 1 or 2, which is calculated as an evaluation value. The image processing apparatus according to any one of claims 1 to 3 , wherein the acquisition means includes a conversion means for converting the acquired image into an image represented by a single color. The image processing apparatus according to any one of claims 1 to 3 , wherein the detection means further includes a second calculation means for calculating the width of an overhead line in the image based on the determined parameter. The image processing apparatus according to any one of claims 1 to 3 , further comprising a determination means for determining whether or not an overhead line exists on a straight line represented by the determined parameter in the image. The image processing according to any one of claims 1 to 6 , further comprising a discriminating means for determining the presence or absence of an abnormality in the overhead line by image processing on the acquired image based on the position of the detected overhead line. apparatus. An image processing method executed by an image processing device that detects the position of an overhead line in an image.
A step of acquiring an image to be detected for the position of the overhead line, and
It includes a step of detecting a linear region having substantially uniform edge directions in the acquired image as an overhead line position .
The step to detect is
A step of extracting a feature amount composed of a plurality of elements at each of a plurality of feature amount extraction positions in the image, and a step of extracting the feature amount.
A step of voting the extracted feature amount in the parameter space based on the feature amount extraction position where the feature amount was extracted, and
Based on the feature amount voted in the parameter space, a step of calculating an evaluation value reflecting that the edge direction is a substantially homogeneous linear region for each parameter representing a straight line in the image, and a step of calculating the evaluation value.
With the step of determining the parameter representing the straight line corresponding to the position of the overhead line based on the calculated evaluation value.
including
Images processing method. A program executed by the computer of an image processing device that detects the position of an overhead line in an image.
On the computer
A step of acquiring an image to be detected for the position of the overhead line, and
The step of detecting a linear region having substantially uniform edge directions in the acquired image as the position of an overhead line is executed .
The step to detect is
A step of extracting a feature amount composed of a plurality of elements at each of a plurality of feature amount extraction positions in the image, and a step of extracting the feature amount.
A step of voting the extracted feature amount in the parameter space based on the feature amount extraction position where the feature amount was extracted, and
Based on the feature amount voted in the parameter space, a step of calculating an evaluation value reflecting that the edge direction is a substantially homogeneous linear region for each parameter representing a straight line in the image, and a step of calculating the evaluation value.
With the step of determining the parameter representing the straight line corresponding to the position of the overhead line based on the calculated evaluation value.
including
Program.

Images