JP7024669B2 - Overhead line wear inspection device - Google Patents

Description

The present invention relates to an overhead wire wear inspection device that detects a wear area of an overhead wire in consideration of a plurality of determination criteria.

Conventionally, two trolley wire measuring instruments including a floodlight unit that irradiates a trolley wire with a band-shaped slit light and a light receiving unit composed of a 3D camera are provided on the left and right sides of the vehicle, and the signal from the light receiving unit is image-processed. A trolley wire measuring device for measuring the shape of a worn portion of a trolley wire is known (see, for example, Patent Document 1 below).

In addition, a line sensor is installed vertically upward on the roof of the vehicle so that the scanning line of this line sensor crosses the trolley line, and the width of the worn part of the trolley line is obtained based on the image signal obtained from this line sensor. A wire wear measuring device is also known (see, for example, Patent Documents 2 to 4 below).

In Patent Document 2, another line sensor is installed diagonally upward on the roof of the vehicle so that the scanning line of this line sensor crosses the pantograph, and the pantograph is based on the image signal obtained from this line sensor. It is described that the width of the trolley wire wear portion is calculated from the width on the image of the trolley wire wear portion by measuring the height of the upper position and using this as the height from the line sensor to the trolley wire. ..

Further, in Patent Document 3, processing is performed so that the sliding surface width of the trolley wire can be detected robustly even when the sliding surface of the trolley wire is not uniformly bright and appears dark. Is described.

Further, Patent Document 4 describes that when a plurality of trolley wire candidates appear on an image, processing is performed so that a trolley wire worn portion can be reliably identified.

JP-A-2015-68675 Japanese Patent No. 4635557 Japanese Patent Application No. 2017-405146 Japanese Patent No. 6268382

C.M. Bishop, "Pattern Recognition and Machine Learning", pp.142-144, Springer Japan, 2007

Here, in the conventional trolley wire wear measuring method as described above, the same method is used for all line sensor images (for example, one method disclosed in any one of Patent Documents 2, 3 or 4). The width of the trolley wire worn part is calculated using.

However, since not all line sensor images are taken under the same conditions, it is optimal to apply the method of Patent Document 3, for example, to an image in which the trolley line appears locally dark. On the other hand, the optimum wear area detection method differs for each image, such that it is optimal to apply the method of Patent Document 2 to an image in which the trolley line is clearly shown as a whole, and all the images. There was a problem that it was difficult to detect the wear width by the optimum method.

Therefore, it is an object of the present invention to provide an overhead wire wear inspection device capable of detecting the wear width of the overhead wire by an optimum wear area detection method for each image.

The overhead wire wear inspection device according to the first invention for solving the above problems is
A database input unit that inputs data that specifies the wear area in each image for a large number of images acquired by photographing the wear area of the overhead wire in advance with a line sensor camera.
A plurality of judgment criteria input units for inputting a plurality of judgment criteria for extracting the wear region from the image, and a plurality of judgment criteria input units, and
Based on the data and the plurality of criteria, machine learning is used to determine the optimum mixing balance for determining the ratio of the plurality of criteria to be combined to best extract the wear region in each image. Multiple judgment criteria including the mixing balance determination unit
An image input unit for inputting an inspection image obtained by photographing the worn area of the overhead wire as an inspection target with the line sensor camera.
Multiple judgment criteria input unit for inputting the plurality of judgment criteria,
Multiple judgment criteria mixed balance input unit that inputs the optimum mixed balance,
A wear area detection method input unit for inputting a plurality of wear area detection methods for detecting the width of the wear area of the overhead wire from the inspection image, and
Based on the inspection image, the plurality of judgment criteria, the optimum mixing balance, and the wear region detection method, the wear region detection method most suitable for each of the plurality of inspection images is selected, and the wear region detection method is selected. It is characterized by including an overhead wire wear detection processing unit using a plurality of judgment criteria including an overhead wire wear area calculation unit in which the width of the wear area of the overhead wire obtained in the above method is used as a final inspection result.

Further, the overhead wire wear inspection device according to the second invention for solving the above-mentioned problems is
The overhead wire wear area calculation unit
The width of the wear area of the overhead wire is obtained from the inspection image by each wear area detection method, and the wear area of the overhead wire obtained by each wear area detection method is evaluated by using the judgment criteria and the optimum mixing balance, respectively. Wear width and evaluation value calculation unit to find
It is characterized by including an evaluation result comparison unit in which the wear region detection method having the highest evaluation value is selected and the width of the wear region of the overhead wire obtained by the selected wear region detection method is used as the final inspection result. And.

According to the overhead wire wear inspection device according to the present invention, the wear width can be detected by the optimum wear area detection method for each image.

It is explanatory drawing which shows the apparatus outline of the overhead wire wear inspection apparatus which concerns on embodiment of this invention. FIG. 3 is a block diagram showing a configuration of a plurality of determination criteria optimum mixing ratio determination processing unit included in the processing apparatus shown in FIG. 1. FIG. 3 is a block diagram showing a configuration of an overhead wire wear detection processing unit using a plurality of judgment criteria included in the processing apparatus shown in FIG. 1. It is explanatory drawing which shows an example of the image data with a wear area label. It is explanatory drawing which shows an example of the relationship between the judgment criterion a and the matching rate t. It is explanatory drawing which introduces the polynomial φ (a) and shows an example of the relationship between the judgment criterion a and the matching rate t. It is explanatory drawing which shows the example which the function expression power by a model parameter is insufficient. It is explanatory drawing which shows the example which the function expression power by a model parameter is appropriate. It is explanatory drawing which shows the example that the function expression power by a model parameter is overfitting. It is explanatory drawing which shows the example when there is an undetected area with respect to a true value. It is explanatory drawing which shows the example of the case where the detection area matches the true value. It is explanatory drawing which shows the example when there is an over-detection area with respect to a true value.

Hereinafter, the overhead wire wear inspection apparatus according to the present invention will be described with reference to the drawings.

The details of the overhead wire wear inspection apparatus according to the embodiment of the present invention will be described with reference to FIGS. 1 to 7C.

As shown in FIG. 1, in this embodiment, the overhead wire wear inspection device includes a first line sensor camera 11 and a second line sensor camera installed on the roof of a train vehicle (hereinafter, simply referred to as a vehicle) 10. A 12-series, a lighting device 13, and a processing device 20 installed inside the vehicle 10 are provided.

The first line sensor camera 11 is installed vertically upward on the roof of the vehicle 10 so that the direction 18 of the scanning line is the same as the direction of the sleepers. As a result, the scanning line of the first line sensor camera 11 crosses the trolley line 16.

Further, the second line sensor camera 12 is installed on the roof of the vehicle 10 diagonally upward toward the pantograph 14, and the scanning line direction 19 is installed in the same direction as the vehicle vertical direction. .. As a result, the scanning line of the second line sensor camera 12 crosses the pantograph 14. In this embodiment, the height of the trolley wire 16 (the upper position 15 of the pantograph 14) used when the width of the wear region on the image of the trolley wire 16 (hereinafter referred to as the wear width) is obtained by the wear region detection method described later. A second line sensor camera 12 is used as a means for taking a picture for detecting the height).

The image data acquired by the first and second line sensor cameras 11 and 12 is input to the processing device 20.

The illuminating device 13 illuminates the trolley line 16 in the area imaged by the first line sensor camera 11. Note that 17 shown in the figure is a structure that supports the trolley wire.

The processing device 20 is, for example, a device such as a computer, and the device configuration includes an arithmetic unit, a storage device, an input / output device, and the like, and the functional configuration is a plurality of judgment criteria optimum mixing ratio determination processing unit shown in FIG. 21 and the overhead wire wear detection processing unit 22 using a plurality of determination criteria shown in FIG. 3 are included in the configuration.

Hereinafter, the details of the processing in the processing apparatus 20 will be described with reference to FIGS. 2 to 7C.

As shown in FIG. 2, the plurality of determination criteria optimum mixing rate determination processing unit 21 includes a past database input unit 211, a plurality of determination criteria input units 212, a plurality of determination criteria mixing balance determination units 213, and a storage unit 214.

The past database input unit 211 manually captures, for example, a large number of images (for example, thousands of images acquired in the past) acquired by photographing the trolley line 16 with the first line sensor camera 11 in advance. The data in which the trolley wire wear region (for example, the white region A _W shown in FIG. 4) is designated (hereinafter, image data with a wear region label) is stored in the storage unit 214.

The multiple judgment criteria input unit 212 is asked to "whether overhead wires are detected for the number of overhead wires given as equipment information", "whether the estimated width of the trolley wire wear portion has an appropriate thickness", and "wear area". The wear area of the trolley wire 16 from the image, such as "Is the brightness value low on both sides of the image?", "Is the wear area continuous?", "Is the width of the wear area significantly different before and after?" A plurality of determination criteria for extraction are stored in the storage unit 214 as determination criteria data.

The multi-judgment standard mixed balance determination unit 213 uses the wear area labeled image data and the judgment standard data input from the storage unit 214 to determine the wear area in each image by combining a plurality of judgment criteria at what specific weight. The optimum mixing balance for determining whether the calculation can be performed most correctly is obtained by machine learning, and is stored in the storage unit 214 as the optimum mixing balance data.

The processing in the plurality of judgment criteria mixed balance determination unit 213 will be described in detail below.
First, when trying to introduce multiple criteria, the following two problems should be solved.
<Problem 1>
How to determine the balance of weights given to each judgment criterion (optimal mixing balance) when the weights given to each of multiple judgment criteria are added together as one evaluation axis.
<Problem 2>
How to define the matching rate t between the evaluation value obtained using the above-mentioned evaluation axis and the true value (image data with wear area label created from the past database).

Regarding <Problem 1>, for example, the linear combination function in which the weights and biases w ₀ , w ₁ , w ₂ , and w ₃ when there are three judgment criteria a, b, and c are adjusted by the following equation (1). ), Linear regression (parameter w is estimated by maximum likelihood estimation or MAP estimation so that the match rate t (details will be described later) matches the output of this function and the image data with wear area label created from the past database. The simplest way is to ask.
f (a, b, c) = w ₀ + w ₁ a + w ₂ b + w ₃ c ... (1)

However, the above equation (1) may not approach the true value. As an example, it is not necessary to consider the judgment criteria b and c because almost the same values are output for any image, and consider the case where the judgment criteria a and the matching rate t have a relationship shown in FIG. 5A.

As shown in FIG. 6A, it is self-evident that it is impossible to express this relationship by a linear regression equation such as the following equation (2).
f (a) = w ₀ + w ₁ a ... (2)

Therefore, in such a case, as shown in FIG. 5B, one solution is to introduce a basis function into the linear regression model. Therefore, for example, consider introducing the polynomial shown in the following equation (3).
φ _i (a) = a ⁱ ... (3)

At this time, the function f can be extended to the following equation (4).

If M is a model parameter of the basis function to be used (here, what degree polynomial is used), the number of required parameters w increases, but the degree of freedom of expression of the function f is greatly improved. However, as the number of model parameters increases, there is a possibility of a state called overfitting, which corresponds to actual data but cannot correspond to unknown data (see FIG. 6C).

Further, in the above description, for the sake of simplicity, an example in which only the judgment criterion a is involved in the function f is shown, but since many judgment criteria are actually involved, FIGS. 6A to 6C (FIG. 6B shows the function expressive power by model parameters). It is difficult to judge whether or not the expressive power of the function is appropriate by illustrating it as shown in (Example) in which the data is appropriately expressed. Therefore, it is necessary to judge whether the expressive power of the function is appropriate by some means.

Here, machine learning is known as a general means of learning from labeled past data and extracting its features. Among the methods of machine learning that can avoid overfitting and also enable non-linear feature extraction, deep learning, kernel regression method, and L1 regularization or lasso regularization, which suppress overfitting and express with a small amount of features, are not described above. There is a method (sparse modeling) described in Patent Document 1.

Table 1 shows the advantages and disadvantages of learning from labeled past data and extracting its characteristics using deep learning, kernel regression, and sparse modeling.

As shown in Table 1, it can be said that sparse modeling is excellent in clarifying which judgment criteria are effective and how effective, but when it is judged that the judgment criteria may be a black box (which judgment criteria are effective). If it is determined that it is not necessary to clarify), machine learning methods such as deep learning or kernel regression can be applied.

Regarding <Problem 2>, the difference between the image group of the past database given the true value (for example, the pixel coordinates of the wear area of the overhead wire) and the result when the analysis is performed with the predetermined analysis parameters is obtained. Ask. The "predetermined analysis parameter" referred to here is a parameter used for calculating the wear width for the image group of the past database. Specifically, for example, a Gaussian filter having a filter size of V is performed on each image group in the past database, binarization processing is performed at the threshold value W, X times closing processing is performed, and Y times. If the area is Z or less, the opening process is performed, and if the area is Z or less, it is removed as noise, and the combination of processes such as calculating the length (wear width) from the left end to the right end of each label and the parameters of each process are set to "predetermined". It is called "analysis parameter". It should be noted that known processes shall be applied to processes such as Gaussian filter, binarization process, closing process, opening process, labeling, etc., and detailed description thereof will be omitted.

As an example, let the white region _{A W} in FIG. 4 be the true value of the overhead wire wear region, and the regions _DA , DB, and DC shown in FIGS. 7A to _7C , respectively, as the overhead wire wear region as a result obtained by analysis. Since there is no difference between the regions D and _B shown in FIG. 7B and the white regions A _W shown in FIG. 4, the matching rate t is 100%. On the other hand, it can be seen that the matching rate t does not reach 100% in the regions _DA and DC shown in FIGS. 7A and _7C because there are undetected regions and there are over-detected regions.

It is necessary to give a standard for how to evaluate each of the undetected and overdetected, but in this embodiment, both overdetected and undetected are treated equally. That is, for example, when the area of the white area A _W in FIG. 4 is 100, the area 90 of the white area A _W is detected (the area 10 of the white area A _W is not detected). In the case of), the matching rate t is 90%, the matching rate t is 90% when an area of 10 is over-detected in addition to all the white areas A _W , and an area of 90 is detected in the white area A _W. When an area with an area of 10 is over-detected in other areas, the matching rate t is -10% for undetected and -10% for over-detected, for a total of 80%.

In this way, the plurality of judgment criteria mixed balance determination unit 213 determines the optimum mixing balance of the plurality of judgment criteria by applying the above-mentioned solutions to the above-mentioned <problem 1> and <problem 2>. .. Here, in Table 2, the optimum mixed balance (weight) when the average value of each judgment criterion is normalized so as to have an average of 0 and a variance of 1, with the number of wears, the derivative of the overhead wire displacement, and the derivative of the wear width as the judgment criteria. An example of the coefficient) is shown.

Such an optimum mixing balance is used as a weighting coefficient for each judgment standard, and the final evaluation value in one evaluation axis described above (hereinafter, simply final) is obtained by multiplying the evaluation values of each judgment standard and adding all of them. Evaluation value).

Further, as shown in FIG. 3, the overhead wire wear detection processing unit 22 using multiple judgment criteria includes an image input unit 221, a plurality of judgment reference input units 222, a plurality of judgment standard mixed balance input units 223, and a wear area detection method input unit 224. It includes an overhead wire wear area calculation unit 225 and a storage unit 226.

The image input unit 221 inputs an image (inspection image) obtained by photographing the worn area of the trolley wire 16 as an inspection target with the first line sensor camera 11 and arranging the image signals in chronological order, and image data. It is stored in the storage unit 226 as.

The multiple judgment standard input unit 222 is asked to "whether overhead wires are detected for the number of overhead wires given as equipment information", "whether the estimated width of the trolley wire wear portion has an appropriate thickness", and "wear area". The storage unit uses multiple judgment criteria such as "whether the brightness values are low on both sides", "whether the wear area is continuous", and "whether the width of the wear area has changed significantly before and after". Save to 226.

The plurality of judgment criteria mixed balance input unit 223 inputs the optimum mixed balance data obtained by the plurality of judgment criteria mixed balance determination unit 213 of the plurality of judgment criteria optimal mixing ratio determination processing unit 21 and stores it in the storage unit 226.

The wear area detection method input unit 224 inputs a plurality of wear area detection methods for detecting a wear area from image data, as shown in Patent Documents 2 to 4, and stores them as wear area detection method data. Save to 226.

The overhead wire wear area calculation unit 225 selects the most suitable wear area detection method for each image data based on the image data, the judgment standard data, the optimum mixed balance data, and the wear area detection method data, and the selected wear area detection method. The wear width of the trolley wire 16 obtained by the above method is used as the final trolley wire wear width (inspection result), and includes a wear width and evaluation value calculation unit 225a and an evaluation result comparison unit 225b. ..

The wear width and evaluation value calculation unit 225a calculates the wear width of the trolley wire 16 by performing wear width detection processing by each wear area detection method for each of the image data stored in the storage unit 226, and each of them. For the trolley wire wear area obtained by the wear area detection method, the final evaluation value is obtained by using the judgment standard data and the optimum mixed balance data.

That is, in the trolley wire wear region obtained by each wear region detection method, for example, when the overhead wire is detected for the number of overhead wires given as equipment information, the estimated width of the trolley wire wear portion has an appropriate thickness. If the wear area has low brightness values on both sides, the wear area is continuous, and the width of the wear area does not change significantly before and after, the evaluation value of each criterion is high. If not, the evaluation value of each criterion will be low.

For the trolley wire wear area obtained by each wear area detection method, the evaluation value of each judgment criterion obtained as described above is multiplied by the weight based on the optimum mixing balance, and all of them are added to each final result. Find the evaluation value.

As for each wear region detection method for determining the wear width of the trolley wire 16, for example, the known methods described in the above-mentioned Patent Documents 2 to 4 and the like are used, and detailed description thereof is omitted here.

The evaluation result comparison unit 225b selects a wear region detection method for obtaining the inspection result having the highest final evaluation value for each image data from a plurality of wear region detection methods, and uses the selected wear region detection method. The obtained wear width of the trolley wire 16 is stored in the storage unit 226 as the final wear width (trolley wire wear width data) in each image data.

In this embodiment, the image data with the wear area label is manually given by the user who can judge the wear area, but some other means (for example, other measuring instrument such as laser data) is used. Image data with a wear area label may be created.

Further, in the present embodiment, the matching rate t is defined by treating over-detection and undetected equally, but the present invention can be evaluated by weighting one or both of over-detection and undetected as necessary. Various changes can be made without departing from the spirit.

Further, in this embodiment, an example in which the processing device 20 is installed inside the vehicle 10 is shown, but the processing device 20 may be installed outside the vehicle 10.
Further, in this embodiment, an example in which the height of the trolley wire 16 is detected by the second line sensor camera 12 is shown, but the height of the trolley wire 16 may be obtained by another known method.

According to the overhead wire wear inspection device according to the present embodiment configured in this way, by providing the multiple judgment criteria optimum mixing ratio determination processing unit 21, one can be combined with a plurality of judgment criteria at any specific weight. The optimum mixed balance for determining whether the optimum evaluation axis can be obtained can be automatically extracted by machine learning using the past database, and the most suitable wear area detection for each image data can be detected. Since it is possible to select the method in consideration of a plurality of judgment criteria and adopt the wear width obtained by using this wear area detection method as the final overhead wire wear inspection result, the wear inspection of the trolley wire 16 is performed. It is possible to carry out with higher accuracy than in the past.

10 Inspection vehicle 11 First line sensor camera 12 Second line sensor camera 13 Lighting 14 Pantograph 15 Pantograph top position 16 Trolley line 17 Structure 18 Direction of scanning line of first line sensor camera 19 Second line sensor Camera scanning line direction 20 Processing device 21 Multiple judgment criteria optimum mixing rate determination processing unit 22 Multiple judgment criteria utilization Overhead line wear detection processing unit 211 Past database input unit 212 Multiple judgment criteria input unit 213 Multiple judgment criteria mixing balance determination unit 214 Storage 221 Image input unit 222 Multiple judgment standard input unit 223 Multiple judgment standard mixed balance input unit 224 Wear area detection method input unit 225 Overhead wire wear area calculation unit 225a Wear width and evaluation value calculation unit 225b Evaluation result comparison unit 226 Storage unit

Claims (2)

A database input unit that inputs data that specifies the wear area in each image for a large number of images acquired by photographing the wear area of the overhead wire in advance with a line sensor camera.
A plurality of judgment criteria input units for inputting a plurality of judgment criteria for extracting the wear region from the image, and a plurality of judgment criteria input units, and
Based on the data and the plurality of criteria, machine learning is used to determine the optimum mixing balance for determining the ratio of the plurality of criteria to be combined to best extract the wear region in each image. Multiple judgment criteria including the mixing balance determination unit
An image input unit for inputting an inspection image obtained by photographing the worn area of the overhead wire as an inspection target with the line sensor camera.
Multiple judgment criteria input unit for inputting the plurality of judgment criteria,
Multiple judgment criteria mixed balance input unit that inputs the optimum mixed balance,
A wear area detection method input unit for inputting a plurality of wear area detection methods for detecting the width of the wear area of the overhead wire from the inspection image, and
Based on the inspection image, the plurality of judgment criteria, the optimum mixing balance, and the wear region detection method, the most suitable wear region detection method for each of the plurality of inspection images is selected, and the wear region detection method is selected. An overhead wire wear inspection apparatus comprising: an overhead wire wear detection processing unit using a plurality of judgment criteria including an overhead wire wear area calculation unit whose final inspection result is the width of the wear area of the overhead wire obtained in the above-mentioned method. The overhead wire wear area calculation unit
The width of the wear area of the overhead wire is obtained from the inspection image by each wear area detection method, and the wear area of the overhead wire obtained by each wear area detection method is evaluated by using the judgment criteria and the optimum mixing balance, respectively. Wear width and evaluation value calculation unit to find
It is characterized by including an evaluation result comparison unit in which the wear region detection method having the highest evaluation value is selected and the width of the wear region of the overhead wire obtained by the selected wear region detection method is used as the final inspection result. The overhead wire wear inspection device according to claim 1.

Images