Anomaly detection system and method for overhead line connectors

Abstract

This invention provides a system and method for detecting abnormalities in overhead line connectors that analyze images of overhead line cables, measure the three-dimensional shape of the overhead line connectors, and enable prediction of deterioration of the overhead line connectors. [Solution] This system detects abnormalities in overhead wire connectors from images captured by a first camera and a second camera, respectively, which are installed on the roof of a railway vehicle in a direction perpendicular to the direction of travel and positioned to look up at the overhead wires from both the left and right sides. The system determines the movement of the first and second cameras in each of the two frames captured by the first and second cameras at predetermined intervals from the moving railway vehicle, and determines the three-dimensional coordinates along the overhead wire connector from the parallax between the images. The abnormality detection method includes an image input step of obtaining each of the two frames, a movement amount calculation step of determining the movement of the first and second cameras from the parallax between the images, and a three-dimensional coordinate determination step of determining the three-dimensional coordinates along the overhead wire connector from the parallax between the images.

Description

Technical Field

[0001] The present invention relates to a system and method for analyzing an image of an electric train line to detect an abnormality of an electric train line connector, and more particularly to an abnormality detection system and detection method for an electric train line connector that enables prediction of deterioration of the electric train line connector.

Background Art

[0002] Among the electric train lines that supply power to electric railways, those that are erected overhead on the line are called overhead electric train lines (hereinafter referred to as overhead lines), and generally, are composed of a plurality of wire strands such as trolley wires and suspension wires for suspending them. One of the components that electrically connect these wire strands is an electric train line connector (hereinafter simply referred to as a "connector").

[0003] Here, the trolley wire and the suspension wire are pushed up by the pantograph of the passing train and vibrated. Therefore, in the connector, both flexibility that can withstand deformation and current-carrying performance are required, and a copper wire or the like is used. On the other hand, over a long period of time, even a copper wire or the like may break due to metal fatigue caused by natural deterioration such as oxidation and repeated vibration.

[0004] For example, Patent Document 1 discloses a method for performing fatigue characteristic evaluation based on a transfer function obtained by a random wave excitation test as a quantitative evaluation of the fatigue durability of a connector. A connector is attached to a wire strand and fitting vibration tester having ears on a vibration table, a pseudo-random wave is input, and strain is measured with a strain gauge attached to the connector or the like.

[0005] On the other hand, as a method for efficiently detecting an abnormality in the attachment state of accessories of an electric train line, it has also been proposed to perform abnormality detection by imaging the electric train line while running a vehicle with a camera attached to the roof of the vehicle and performing image analysis.

[0006] For example, Non-Patent Document 1 describes a method for detecting abnormalities in the installation status of connectors from images of overhead power lines by combining a laser sensor and image analysis technology. Automatically extracting wires like overhead power lines from images is not easy because various structures are also captured in the background. Therefore, the approximate location of the wires is measured using a laser sensor, and based on that result, the analysis range is narrowed to find the wires, and then the overhead power line accessories such as connectors are measured. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2012-1100 [Non-patent literature]

[0008] [Non-Patent Document 1] Matsumura, Shu; Nezu, Kazuyoshi; "Reducing Labor in Overhead Line Inspection Using Laser Sensors and Image Analysis Technology"; RRR Journal, Vol. 78, No. 8, pp. 4-7, 2021 / 8 [Overview of the project] [Problems that the invention aims to solve]

[0009] To predict the deterioration of a connector based on its installation condition, it was considered necessary to measure the curvature of the installed connector. However, since the connector has a linear shape (curved shape) and extends in three-dimensional space, measuring its curvature requires three-dimensional shape measurement. Measuring this using a laser sensor as described above from a moving vehicle is difficult. Furthermore, the three-dimensional shape of the connector cannot be directly measured from planar images captured by an on-board camera.

[0010] The present invention has been made in view of the above circumstances, and its purpose is to provide an abnormality detection system and detection method for overhead line connectors that analyze images taken of overhead line wires, measure the three-dimensional shape of the overhead line connectors, and enable the prediction of deterioration of the overhead line connectors. [Means for solving the problem]

[0011] The abnormality detection system according to the present invention is a system for detecting abnormalities in an overhead line connector from images captured by a first camera and a second camera, respectively, which are mounted on the roof of a railway vehicle in a direction perpendicular to the direction of travel and are positioned to look up at the overhead line from both the left and right sides. The system is characterized by determining the amount of movement of the first camera and the second camera in two frames of images captured by the first camera and the second camera at predetermined intervals from the moving railway vehicle, and determining the three-dimensional coordinates along the overhead line connector from the parallax between the images.

[0012] With these features, it is possible to analyze images of overhead lines and measure the three-dimensional shape of the overhead line connectors, thereby providing a prediction of the deterioration of the overhead line connectors.

[0013] In the invention described above, a reference point captured in common to all of the images may be determined, and the amount of movement of the first camera and the second camera may be determined from the parallax between the images. Alternatively, the reference point may correspond to a metal fitting fixed to the overhead wire, and the reference point may be determined for each of the images based on the known shape of the metal fitting. With such features, the three-dimensional shape of the overhead wire connector can be easily measured by analyzing images of the overhead wire, and the deterioration of the overhead wire connector can be predicted.

[0014] The present invention relates to an abnormality detection method for detecting an abnormality in an overhead line connector from images captured by a first camera and a second camera, respectively, which are mounted on the roof of a railway vehicle in a direction perpendicular to the direction of travel and positioned to look up at the overhead line from both the left and right sides. The method comprises: an image input step of obtaining two frames each of images captured by the first camera and the second camera at predetermined intervals from the moving railway vehicle; a movement amount calculation step of determining a reference point captured in common to all of the images and determining the amount of movement of the first camera and the second camera from the parallax between the images; and a three-dimensional coordinate determination step of determining three-dimensional coordinates along the overhead line connector from the parallax between the images.

[0015] With these features, it is possible to analyze images of overhead lines and measure the three-dimensional shape of the overhead line connectors, thereby providing a prediction of the deterioration of the overhead line connectors.

[0016] In the invention described above, the step of calculating the amount of movement may be characterized by using a metal fitting fixed to the overhead wire as the reference point, and determining the reference point in each of the images based on the known shape of the metal fitting. With this feature, the three-dimensional shape of the overhead wire connector can be easily measured by analyzing the image of the overhead wire, and the deterioration of the overhead wire connector can be predicted.

[0017] In the invention described above, the three-dimensional coordinate determination step may be characterized by including a shape extraction step of extracting the shape of the wires of the overhead line connector in each of the images, and a step of determining the three-dimensional coordinates for each corresponding point along the shape of the wires between the images, assuming that changes in the height position of the overhead line connector can be ignored. Alternatively, the shape extraction step may be characterized by including a step of extracting the region of the overhead line connector by semantic segmentation and performing a thinning process to extract the shape of the wires of the overhead line connector. According to this feature, the three-dimensional shape of the overhead line connector can be measured more easily by analyzing images of overhead lines, and the deterioration of the overhead line connector can be predicted.

[0018] In the above-described invention, the three-dimensional coordinate determination step may include a curvature calculation step of calculating the curvature of the tram line connector by applying an approximation formula using the pixel position in the vertical direction of the image as an intermediate variable for each of the three-dimensional directions of the three-dimensional coordinates. According to such a feature, it is possible to analyze an image of the tram line and measure the three-dimensional shape and its curvature of the tram line connector, and thus it is possible to predict the deterioration of the tram line connector.

Brief Description of the Drawings

[0019] [Figure 1] It is a block diagram of an abnormality detection system according to the present invention. [Figure 2] It is a flowchart showing an abnormality detection method according to the present invention. [Figure 3] It is a principle diagram for obtaining the movement amount of the camera from the parallax. [Figure 4] (a) Original image of the tram line connector, (b) Image showing the shape of the line segments of the extracted tram line connector, (c) Thinned image. [Figure 5] (a) and (b) Images between two frames, and (c) Diagram showing the parallax D. [Figure 6] It is a principle diagram for calculating the position of the intersection point pi that is the corresponding point of both cameras for obtaining three-dimensional coordinates. [Figure 7] It is an image of the tram line connector showing the ratio of the height h to the intersection point pi. [Figure 8] It is an example of three-dimensional coordinates calculated along the shape of the line segments of the tram line connector. [Figure 9] It is a graph of an approximation formula using intermediate variables in each of the three-dimensional directions for the three-dimensional coordinates of the tram line connector. [Figure 10] It is a graph of the curvature corresponding to the image of the tram line connector in the height direction.

Embodiments for Carrying Out the Invention

[0020] The following describes embodiments of the abnormality detection system and method for overhead line connectors according to the present invention. First, the abnormality detection system will be explained using Figure 1.

[0021] As shown in Figure 1, the anomaly detection system 1 includes two cameras 11 (first camera) and 12 (second camera) facing each other, with their optical axes oriented perpendicular to the direction of travel of the vehicle 10 (railway vehicle). Cameras 11 and 12 are mounted on the roof 13 of the vehicle 10, and their optical axes are oriented diagonally upward so as to look up at the overhead wire 20 from both the left and right sides. Cameras 11 and 12 are installed at a predetermined distance from each other. The overhead wire 20 consists of, for example, a trolley wire 21 and a suspension wire 22, and includes a connector 23 (railway line connector) for suspending the trolley wire 21 from the suspension wire 22.

[0022] Cameras 11 and 12 are configured to synchronize with each other and capture images of the overhead wires 20 in the optical axis direction at a predetermined frame rate while the vehicle 10 is in motion. For example, cameras 11 and 12 are connected to an imaging device 6 mounted on the vehicle 10, which controls the frame rate, start and end of imaging, etc. The imaging device 6 also has a built-in storage device 7 that can store the captured image data. After the vehicle 10 has traveled, the image data can be input from the storage device 7 to the analysis device 5 using a storage medium, or the image data can be transmitted to the analysis device 5 while the vehicle is in motion, and the analysis device 5 can perform image analysis. In other words, the analysis device 5 can be installed in a room such as a fixed facility on the ground or mounted on the vehicle, and the installation location can be freely selected. As a result, the anomaly detection system 1 can analyze the images of the connectors 23 captured by cameras 11 and 12 using the analysis device 5. It is also preferable to equip the vehicle 10 with an illumination device that irradiates the overhead wires 20 with light to assist the imaging by cameras 11 and 12.

[0023] Incidentally, as mentioned above, connector 23 may deteriorate and break due to metal fatigue, and the shape of the connector itself greatly influences how easily this metal fatigue progresses. Conventionally, the condition of overhead line connectors has been judged by visually inspecting their shape and other installation conditions. Therefore, we would like to detect abnormalities in the deterioration state, or those that are likely to cause rapid deterioration such as metal fatigue, by measuring and quantitatively evaluating the curvature of the shape. To this end, the abnormality detection system 1 will detect abnormalities when the curvature of connector 23 falls outside a predetermined range. For example, it is possible to determine that an abnormality occurs when the maximum value of the curvature exceeds a threshold. To this end, the abnormality detection system 1 measures the curvature of connector 23 in three-dimensional space using the following method. Note that connector 23 consists of upper and lower fittings 24a and 24b (see Figure 4) for connecting to the trolley wire 21 and the suspension wire 22, and a bent wire lead wire 25 that connects them. Here, "curvature of the connector" refers to the curvature of this lead wire 25.

[0024] In the anomaly detection system 1, stereo measurement is performed using two frames of images obtained from each of the cameras 11 and 12. In normal stereo measurement, images of the same object obtained from each of two cameras are used to measure the depth dimension of that object from the parallax. On the other hand, in the anomaly detection system 1, while the vehicle 10 is moving, images of the same object, the connector 23, are obtained from one camera at two different locations over two frames, creating parallax for the connector 23 between the two frames of images and performing stereo measurement. In particular, since the lead wires 25 of the connector 23 are bent in the direction of travel of the vehicle 10, obtaining images from the left and right directions relative to the direction of travel can increase the curvature of the lead wires 25 in the image (in two dimensions). Therefore, compared to the case where images are taken from the front and rear directions relative to the direction of travel of the vehicle 10, which reduces the curvature in the image, the influence of measurement errors in the image can be reduced, and the curvature in three dimensions can be measured accurately.

[0025] In this embodiment, stereo measurement involves measuring the depth dimension, which is the distance from the line through which the camera's focal point moves between two frames to a specific point on the image. The three-dimensional coordinates of this specific point are then determined from the depth dimension. In other words, the three-dimensional coordinates of the space on the vehicle 10 are determined from the depth dimension for a specific point defined on the connector 23. At this time, the amount of movement dx of cameras 11 and 12 during shooting between the two frames (see Figure 3) greatly affects the accuracy of this depth dimension measurement. Therefore, it is desirable to determine the amount of movement dx of cameras 11 and 12 as accurately as possible. Accordingly, in this embodiment, the amount of movement dx of cameras 11 and 12 is accurately determined using the parallax of the images between the two frames.

[0026] Based on the above, a method for determining the curvature of the connector 23 using the anomaly detection system 1 will be explained with reference to Figures 2 to 10.

[0027] As shown in Figure 2, first, based on the control of the imaging device 6, the same connector 23 is imaged with both cameras 11 and 12 while the vehicle 10 is in motion (imaging step: S1). At this time, the imaging device 6 causes both cameras 11 and 12 to take images at a predetermined frame rate while the vehicle 10 is in motion, so that a large number of synchronized images can be obtained from both cameras. The obtained images are sequentially stored in the storage device 7.

[0028] Next, the image captured from the storage device 7 is input to the analysis device 5 (image input step: S2), and image analysis is performed.

[0029] First, images to be analyzed are selected from the numerous images obtained. Here, two frames capturing the same connector 23 are selected. In other words, a total of four images are selected for analysis, two frames each from camera 11 and camera 12. However, the selected images must be captured by camera 11 and camera 12 at the same position (i.e., at the same time) relative to the direction of travel of the vehicle 10. This image selection can be automated based on the relationship between the vehicle's speed and frame rate. For example, after determining the interval between two frames based on the speed and frame rate, one image containing the connector 23 within a predetermined range of the field of view is extracted from the numerous images captured by one camera and designated as the first frame. The second frame is then selected accordingly, and images captured at the same time are selected from the numerous images captured by the other camera. In this way, four images to be analyzed are selected. The four images to be analyzed may also be selected by an operator.

[0030] The four images used for image analysis can be freely selected by defining two frames in the direction of travel, but it is preferable to select frames that allow for accurate calculation of curvature. For example, the position of connector 23 within the field of view can be significantly different left to right between the first and second frames, that is, frames that include connector 2 near the left and right edges of the field of view. This allows for a larger amount of camera movement between these two frames, which can improve the accuracy of subsequent calculations. On the other hand, selecting an image that includes connector 23 near the edge of the field of view makes the image more susceptible to distortion that tends to occur at the edges of the image, as well as the vertical movement of the connector between the two frames. Taking these factors into account, the two frames are determined in a way that allows for accurate calculation of curvature.

[0031] This ensures that, within the field of view, the same connector 23 is captured in the first frame on the front side in the direction of travel, and in the second frame on the rear side in the direction of travel. Furthermore, as described above, the two cameras 11 and 12 are synchronized, and images are obtained from both cameras 11 and 12 at the same position relative to the direction of travel of the vehicle 10. Therefore, cameras 11 and 12 move the same distance (movement amount dx; see Figure 3) between these two frames.

[0032] Once the images to be analyzed are selected, a reference point is determined that is captured in common to all four images (two frames each) obtained from the two cameras 11 and 12, and then the amount of movement of cameras 11 and 12 between the two frames is calculated (movement amount calculation step: S3).

[0033] The reference point p (see Figure 3) is visible in all four images described above and is determined to correspond to a single point in space. For example, it is preferable to determine the reference point p to correspond to a metal fitting fixed to the overhead wire, such as a hanger, which can be easily identified by image recognition. If it is an overhead wire fitting, its position can be easily extracted based on the known shape of the fitting using template matching or image recognition using a trained AI. As will be described later, the vertical positions of the images from camera 11 and camera 12 will be correlated. For this reason, the positions of both the upper end fitting 24a and the lower end fitting 24b of the connector 23 are extracted here.

[0034] Once a reference point p is established, the parallax d1 of the two frames obtained by camera 11 with respect to the reference point p and the parallax d2 of the two frames obtained by camera 12 with respect to the reference point p are measured. In other words, the difference in the left-right position of the reference point p on the images between the two frames is determined. Note that since cameras 11 and 12 have their optical axes pointed in a plane perpendicular to the direction of travel of the vehicle 10, that is, in a direction perpendicular to the direction of travel, a parallax occurs in the left-right direction corresponding to the direction of travel on the images between the two frames.

[0035] Referring to Figure 3, the amount of movement dx of cameras 11 and 12 due to the movement of vehicle 10 is determined using the parallax d1 from camera 11 and the parallax d2 from camera 12. Specifically, let L be the distance between cameras 11 and 12 (the distance between their focal points), and let f be the focal lengths of cameras 11 and 12. Assume that the coordinates of the point in space corresponding to the reference point p from which the parallax was obtained are (x,y). Note that the focal lengths f of the two cameras 11 and 12 are the same. In this coordinate system, the x-axis direction is the direction of travel of vehicle 10, and the y-axis direction is the left-right direction perpendicular to the direction of travel of vehicle 10. The midpoint between camera 11 and camera 12 is assumed to be y=0, and the y-coordinate is assumed to be positive on the side of camera 11. Then the distance in the y-direction from camera 11 to reference point p is L / 2-y. Similarly, the distance in the y-direction from camera 12 to reference point p is L / 2+y.

[0036] Since the triangle with the parallax d1 of the image from camera 11 as its base and the height as its focus f is similar to the triangle with the displacement dx as its base and the distance L / 2-y from the focal point of camera 11 to the reference point p as its height, dx = d1 × (L / 2 - y) / f (Equation 1) The same applies to the parallax d2. dx = d² × (L / 2 + y) / f (Equation 2) This is the result.

[0037] Next, multiplying both sides of equation 2 by d1 / d2, dx×d1 / d2=d1(L / 2+y) / f (Formula 3) And if we add the left sides of (Equation 3) and (Equation 1) together, and the right sides together, then dx × (1 + d1 / d2) =d1(L / 2-y) / f+d1(L / 2+y) / f (Equation 4) This is the result. Rearranging equation 4, dx=L / f×d1d2 / (d2+d1) (Formula 5) Equation (5) does not include the coordinates of the reference point p. Therefore, from Equation (5), it can be seen that the amount of movement dx can be determined from the parallaxes d1 and d2, the distance L between cameras 11 and 12, and the focal length f. In other words, by measuring the parallaxes d1 and d2 of cameras 11 and 12 with respect to the reference point p, the amount of movement dx of cameras 11 and 12 can be determined regardless of the coordinates of the reference point p. If there is an alternative means to accurately determine the amount of movement dx, such as using a Doppler sensor, then such means can also be used.

[0038] Once the amount of movement dx of cameras 11 and 12 between two frames is determined, the disparity D between these two frames is used to perform stereo measurement of the connector 23 and determine the 3D coordinates along the line of the connector 23 (3D coordinate determination step: S4).

[0039] In stereo measurement, the depth dimension of the image at a specific point (the intersection point pi, described later) is first determined using the parallax of that point. Below, an example using an image obtained by camera 11 is shown.

[0040] As shown in Figure 4, prior to stereo measurement, image processing is performed first. Here, the image obtained between the two frames described above (Figure (a)) is used to extract the region of the connector 23 on the image, particularly the shape of the lead wires 25, using image recognition with segmentation (Figure (b); shape extraction step: S4-1). For such image recognition, semantic segmentation can be suitably used, for example. The extracted connector 23 on the image is further processed to thin the lines (Figure (c)). This allows for accurate extraction of the shape of the lead wires 25 of the connector 23.

[0041] Next, the 3D coordinates are calculated for each corresponding point along the shape of the lead wire 25 between the two frames of images (coordinate calculation step: S4-2).

[0042] Here, first, as shown in Figure 5, we calculate the parallax D of the two images between the two frames described above. Using the two thinned images, we determine the intersection point pi of the horizontal line H and the lead line 25, which pass through the same height in the images. The difference between the lateral positions A1 and A2 of the respective intersection points pi in the two images is defined as the parallax D. Note that the vertical displacement within the images between the two frames can be ignored, the height of the horizontal line H is the same in both images, and the corresponding point in space for the intersection point pi is the same.

[0043] Once the parallax D is obtained, the distance from the focal line of camera 11 to the point in space above vehicle 10 corresponding to the intersection point pi, i.e., the depth dimension Z, is determined using the stereo measurement formula (Equation 6) below. Here, the focal length f of camera 11 and the travel distance dx of camera 11 are used as the baseline length. Z=dx×f / D (Formula 6)

[0044] In this way, the depth dimension Z is obtained sequentially at one-pixel intervals or at predetermined intervals from the upper end to the lower end of the connector 23 on the image. As described above, the positions of the upper and lower fittings 24a and 24b have been extracted, so the processing can be carried out sequentially from top to bottom or bottom to top at predetermined intervals in the vertical direction between fittings 24a and 24b. It is preferable to calculate the parallax D after calculating a moving average using multiple locations A1 or A2 in the vertical direction for positions A1 and A2 on the image. This reduces the effect of noise and smooths the change in parallax D obtained in response to movement of the measurement position in the vertical direction.

[0045] Then, as shown in Figure 6, the depth dimension Z1 from camera 11 and the depth dimension Z2 from camera 12 are used to determine the corresponding intersection point pi, and the coordinates of the corresponding point in space (the corresponding point of intersection point pi) are determined. At this time, since the distance L between camera 11 and camera 12 and the depth dimensions Z1 and Z2 form a triangle, the coordinates of the corresponding point of intersection point pi can be determined using the method of triangulation.

[0046] As shown in Figure 7, the intersection point pi of the images obtained by camera 11 and camera 12 is matched using the height positions of the upper and lower fittings 24a and 24b. The positions of fittings 24a and 24b are extracted when determining the position of the reference point p, as described above. Therefore, by determining the intersection point pi of the images from camera 11 and camera 12 so that the ratio of the height h to the intersection point pi to the dimension W between fittings 24a and 24n (h / W) is the same, the intersection point pi of the two images can be made to match.

[0047] As described above, the three-dimensional coordinates along the lead wire 25 of the connector 23 can be determined. The lead wire 25 has a shape in which its height gradually increases from bottom to top, even while bending. This is also true in the images taken by cameras 11 and 12, which are viewed from a diagonal angle below. In other words, if the position in the height direction is determined in the image, the point on the lead wire 25 is determined to be a single point. Therefore, the depth dimension and coordinates described above can be determined along the lead wire 25 by sequentially processing the images so that the height of the intersection point pi changes from top to bottom or bottom to top. In other words, the three-dimensional coordinates can be determined for each corresponding point of the intersection point pi in the images from both cameras 11 and 12, which follow the shape of the lead wire 25.

[0048] Here, if the elevation angle of camera 11 (or 12) is known, the 3D coordinates of the corresponding point of intersection pi can be determined using only the image from one of the cameras 11 (or 12). For example, the intersection pi is represented by pixel coordinates with the center of the image as the origin, and the image is projected onto a virtual plane that passes through the corresponding point of intersection pi and is perpendicular to the optical axis. Then, the coordinates of the corresponding point of intersection pi on the coordinate axis with the optical axis as the origin on the virtual plane can be obtained by multiplying the pixel coordinates of intersection pi by the size of the pixel and Z1 / f (or Z2 / f). Then, a rotation matrix is ​​obtained from the camera's elevation angle, a coordinate transformation is performed, and the 3D coordinates of the corresponding point of intersection pi are obtained from the coordinates of the virtual plane.

[0049] As described above, the three-dimensional coordinates (x, y, z) of the lead wire 25 can be obtained as shown in Figure 8.

[0050] Next, the curvature of the connector 23 is calculated by applying an approximation formula to each of the three-dimensional directions, using the vertical pixel position of the image as a parameter (curvature calculation step: S4-3).

[0051] In detail, first, as shown in Figure 9, an approximation formula is applied to each of the three-dimensional directions (x, y, z) of the obtained lead wire 25, i.e., the x, y, and z directions, using the pixel position t in the vertical direction (height direction) of the image as a parameter. As mentioned above, since the lead wire 25 has a shape that gradually increases in the height direction, such an approximation formula can also be determined by using the pixel position t in the height direction as a parameter.

[0052] Furthermore, the equation of the three-dimensional curve with t as a parameter is defined as shown in (Equation 7) below.

number

[0053] In this case, the curvature k can be expressed as shown in (Equation 8) below using the first and second derivatives of (Equation 7). Note that these first and second derivatives may also be obtained, for example, by filtering the 3D coordinates of the lead wire 25 and then calculating the difference between adjacent data points. In this case, the approximation formula described above becomes unnecessary.

number

[0054] The details of how to derive (Equation 8) from (Equation 7) are publicly known, as shown in "Differential Geometry of Curves and Surfaces (Revised Edition)" by Shoichi Kobayashi, published by Shokabo Co., Ltd., and are therefore omitted from this explanation.

[0055] Then, as shown in Figure 10, the curvature of the lead wire 25 (the curvature of the connector 23) can be obtained. In this figure, the image of the lead wire 25 on the left and the graph of curvature on the right are displayed with corresponding height positions in the vertical direction.

[0056] Furthermore, the abnormality detection system 1 can determine that an abnormality exists when the obtained curvature falls outside a predetermined range. For example, a predetermined threshold may be set, and when the maximum value of the obtained curvature exceeds the applied threshold, an abnormality may be determined and a warning issued. In the case of a warning, for example, a warning sign may be displayed on a display device (not shown). It is preferable that such a threshold be set based on a shape that is likely to rapidly deteriorate, such as metal fatigue of the connector 23, due to vibrations associated with the passage of the pantograph. In other words, the deterioration of the connector 23 can be predicted by whether or not the curvature exceeds the threshold. When the abnormality detection system 1 issues a warning, it is preferable to perform maintenance on the overhead line equipment, such as replacing the connector 23.

[0057] As described above, the anomaly detection system 1 and the anomaly detection method described above can analyze images of the overhead line to measure the three-dimensional shape of the connector 23 and provide a prediction of the deterioration of the connector 23.

[0058] Although representative embodiments of the present invention and their variations have been described above, the present invention is not necessarily limited thereto and can be modified as appropriate by those skilled in the art. That is, those skilled in the art will be able to find various alternative embodiments and modifications without departing from the scope of the attached claims. [Explanation of Symbols]

[0059] 1. Anomaly detection system 5 Analysis device 10 vehicles (railway vehicles) 11, 12 Cameras 20 Overhead lines 23 Connectors (Train line connectors)

Claims

1. A system for detecting abnormalities in overhead line connectors from images captured by a first camera and a second camera, respectively, which are mounted on the roof of a railway vehicle in a direction perpendicular to the direction of travel and positioned to look up at the overhead lines from both the left and right sides, An abnormality detection system for overhead line connectors, characterized by determining the amount of movement of the first camera and the second camera in two frames each, which are captured at predetermined intervals from the moving railway vehicle by the first camera and the second camera, and determining the three-dimensional coordinates along the overhead line connector from the parallax between the images.

2. An abnormality detection system for overhead line connectors according to claim 1, characterized in that a reference point captured in common to all of the aforementioned images is determined, and the amount of movement of the first camera and the second camera is determined from the parallax between the aforementioned images.

3. The overhead line connector abnormality detection system according to claim 2, characterized in that the reference point corresponds to a metal fitting fixed to the overhead line, and the reference point is determined in each of the images based on a known shape of the metal fitting.

4. A method for detecting abnormalities in a railway line connector from images captured by a first camera and a second camera, respectively, which are mounted on the roof of a railway vehicle in a direction perpendicular to the direction of travel and positioned to look up at the overhead wires from both the left and right sides, Image input step: Obtaining images of two frames each captured by the first camera and the second camera at predetermined intervals from the moving railway vehicle, A step of determining a reference point captured in all of the aforementioned images, and calculating the amount of movement of the first camera and the second camera from the parallax between the aforementioned images, A method for detecting abnormalities in an overhead line connector, characterized by comprising a three-dimensional coordinate determination step of determining three-dimensional coordinates along the overhead line connector from the parallax between the aforementioned images.

5. The method for detecting abnormalities in an overhead line connector according to claim 4, characterized in that the step of calculating the amount of movement uses a metal fitting fixed to the overhead line as the reference point, and determines the reference point in each of the images based on the known shape of the metal fitting.

6. The three-dimensional coordinate determination step is: A shape extraction step is performed to extract the shape of the wires of the overhead line connector in each of the above images, The method for detecting an anomaly in an overhead line connector according to claim 4, comprising the step of determining the three-dimensional coordinates for each corresponding point along the shape of the wire between the images, assuming that changes in the height position of the overhead line connector can be ignored.

7. The method for detecting abnormalities in an overhead line connector according to claim 6, characterized in that the shape extraction step includes a step of extracting the region of the overhead line connector by semantic segmentation and performing a thinning process to extract the shape of the wires of the overhead line connector.

8. The method for detecting an anomaly in an overhead line connector according to one of claims 4 to 7, characterized in that the three-dimensional coordinate determination step includes a curvature calculation step in which an approximation formula is applied to each of the three-dimensional directions using the vertical pixel position of the image as a parameter to calculate the curvature of the overhead line connector.

Images