Abnormality detection system, abnormality detection device, abnormality detection method, and program

Abstract

To provide a technique for providing robustness against vibrations in a train pitch direction when an abnormality of a line structure ahead of a train in an advancing direction is detected using a three-dimensional LiDAR device mounted on the train.SOLUTION: The abnormality detection system comprises: point group acquisition means for acquiring a point group corresponding to a line structure ahead of a train in an advancing direction; a point group correction means for dividing the point group in a lattice shape in a planar view, and correcting the point group so as to replace, for each cell, height values of a plurality of ranging points belonging to the cell with a representative value of the height values of the ranging points to reduce an error of a height value of the point group caused by vibrations in a pitch direction of the train; and abnormality detection means for detecting an abnormality of the line structure on the basis of the point group.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to an anomaly detection system, an anomaly detection device, an anomaly detection method, and a program. [Background technology]

[0002] Patent Document 1 discloses a monitoring device that uses a three-dimensional LiDAR device mounted on a train to determine the presence or absence of an obstacle ahead. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-203634 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, when using a 3D LiDAR device to detect an anomaly in a track structure several hundred meters before the anomaly, there is a problem that the detection accuracy is reduced due to vibration in the pitch direction of the train, because the elevation value of the measurement point generated by the 3D LiDAR device deviates significantly from the true value in the vertical direction.

[0005] An object of the present disclosure is to provide a technology that is robust against vibrations in the pitch direction of a train when detecting abnormalities in track structures ahead of the train's direction of travel using a point cloud generating device mounted on a train. [Means for solving the problem]

[0006] A point cloud acquisition means for acquiring a point cloud corresponding to a track structure ahead in a traveling direction of a train; a point cloud correction means for dividing the point cloud into a lattice shape in a planar view, and correcting the point cloud for each cell by replacing the elevation values ​​of a plurality of ranging points belonging to the cell with a representative value of the elevation values ​​of the plurality of ranging points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; an abnormality detection means for detecting an abnormality in the railway structure based on the point cloud; Including, An anomaly detection system is provided.

[0007] A point cloud acquisition means for acquiring a point cloud corresponding to a track structure ahead in a traveling direction of a train; a point cloud correction means for dividing the point cloud into a lattice shape in a planar view, and correcting the point cloud for each cell by replacing the elevation values ​​of a plurality of ranging points belonging to the cell with a representative value of the elevation values ​​of the plurality of ranging points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; an abnormality detection means for detecting an abnormality in the railway structure based on the point cloud; Including, An anomaly detection apparatus is provided.

[0008] The computer A point cloud corresponding to the track structure ahead of the train in the direction of travel is acquired. The point cloud is divided into a grid in a planar view, and the point cloud is corrected so that, for each cell, the elevation values ​​of a plurality of measurement points belonging to the cell are replaced with a representative value of the elevation values ​​of the plurality of measurement points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; detecting an abnormality in the railway structure based on the point cloud; A method for detecting anomalies is provided. Effect of the Invention

[0009] According to the present disclosure, a technology is realized that is robust against vibrations in the pitch direction of the train when detecting abnormalities in track structures ahead of the train in the direction of train travel using a point cloud generating device mounted on a train. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram of an anomaly detection system. [Diagram 2] FIG. 2 is a block diagram of a train. [Diagram 3] FIG. [Figure 4] FIG. 11 is a diagram for explaining the variation of distance measurement points. [Diagram 5] This is a histogram of the elevation values ​​of the measurement points. [Figure 6] FIG. 13 is a diagram showing how a point cloud corresponding to a rail structure is divided into a grid shape. [Figure 7] FIG. [Figure 8] FIG. 13 is a diagram showing a difference pattern. [Figure 9] 4 is a control flow of the abnormality detection device. [Figure 10] FIG. 2 is a block diagram of a train. [Figure 11] FIG. 13 is a diagram showing a luminance pattern. [Figure 12] 4 is a control flow of the abnormality detection device. [Figure 13] FIG. 2 is a block diagram of a train. [Figure 14] FIG. 13 is a diagram showing a point cloud missing region. [Figure 15] FIG. 13 is a diagram showing a point cloud density pattern. [Figure 16] 4 is a control flow of the abnormality detection device. [Figure 17] FIG. 2 is a block diagram of a train. [Figure 18] FIG. 1 is a diagram illustrating dimensional compression. [Figure 19] 1 is a graph showing the relationship between distance from a train and altitude value. [Figure 20] 1 is a graph showing the relationship between distance from a train and altitude value. [Figure 21] FIG. 13 is a diagram showing a frequency analysis result. [Figure 22] FIG. 13 is a diagram showing a frequency analysis result. [Diagram 23] 4 is a control flow of the abnormality detection device. [Figure 24] FIG. 2 is a block diagram of a train. [Diagram 25] 1 is a graph showing the relationship between distance from a train and altitude value. [Figure 26] 1 is a graph showing the relationship between distance from a train and altitude value. [Figure 27] 4 is a control flow of the abnormality detection device. [Figure 28] FIG. 2 is a block diagram of a train. [Figure 29] 1 is a graph showing the relationship between distance from a train and altitude value. [Diagram 30] 4 is a control flow of the abnormality detection device. [Diagram 31] FIG. 2 is a block diagram of a train. [Diagram 32] 1 is a diagram illustrating a case where a processing circuit included in an abnormality detection device is configured with a processor and a memory. [Diagram 33] 1 is a diagram illustrating a case where a processing circuit included in an abnormality detection device is configured with dedicated hardware. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] (Summary of the Disclosure) Hereinafter, an overview of the present disclosure will be described with reference to the drawings. Fig. 1 is a block diagram of an anomaly detection system. As shown in Fig. 1, the anomaly detection system 100 includes a point cloud acquisition unit 101, a point cloud correction unit 102, and an anomaly detection unit 103.

[0012] The point cloud acquisition means 101 acquires a point cloud corresponding to a track structure ahead in the traveling direction of the train.

[0013] The point cloud correction means 102 divides the point cloud into a grid in a planar view, and corrects the point cloud so that, for each cell, the elevation values ​​of the multiple distance measurement points belonging to that cell are replaced with a representative value of the elevation values ​​of the multiple distance measurement points. In this way, the point cloud correction means 102 reduces errors in the elevation values ​​of the point cloud caused by vibrations in the pitch direction of the train.

[0014] The abnormality detection means 103 detects an abnormality in the railway structure based on the point cloud.

[0015] According to the above configuration, a technology is realized that is robust against vibrations in the pitch direction of the train when detecting abnormalities in track structures ahead of the train in the direction of train travel using a point cloud generating device mounted on the train.

[0016] (First embodiment) Next, a first embodiment of the present disclosure will be described. Fig. 2 is a block diagram of a train 1. The train 1 runs along a track structure 2.

[0017] The train 1 includes a three-dimensional LiDAR device 3 (hereinafter simply referred to as the LiDAR device 3), a self-location estimation device 4, a camera 5, an anomaly detection device 6, a driving device 7, a braking device 8, and wheels 9.

[0018] The LiDAR device 3 is a specific example of a point cloud generating device that generates a point cloud by sensing the front of the train 1 in the traveling direction. The LiDAR device 3 of this embodiment is a direct ToF (Time of Flight) type. That is, the LiDAR device 3 generates a three-dimensional point cloud in the forward traveling direction of the train 1 by emitting laser light in the forward traveling direction of the train 1 and measuring the time required to receive the reflected light. However, instead of this, the LiDAR device 3 may be an FMCW (Frequency Modulated Continuous Wave) type that generates the above three-dimensional point cloud based on the frequency difference between the laser light emitted in the forward traveling direction of the train 1 and the reflected light. In addition, the LiDAR device 3 may be an indirect ToF type that generates the above three-dimensional point cloud based on the phase difference between the laser light emitted in the forward traveling direction of the train 1 and the reflected light. The LiDAR device 3 outputs the generated three-dimensional point cloud (hereinafter also simply referred to as a point cloud) to the abnormality detection device 6.

[0019] The point cloud generating device is not limited to the LiDAR device 3. Any device capable of sensing the area ahead of the train 1 in the traveling direction may be used as the point cloud generating device. For example, the point cloud generating device may be configured to generate a three-dimensional point cloud using a radar device (Radio Detection And Ranging), an ultrasonic sensor, a stereo camera, or a combination of these. The point cloud generating device may also be configured to generate a three-dimensional point cloud by Structure from Motion (SfM) from a plurality of two-dimensional images obtained by capturing an image of the area ahead of the train 1 in the traveling direction.

[0020] The self-location estimation device 4 estimates the self-location of the train 1 in a global coordinate system. The self-location estimation device 4 is typically configured with an IMU (Inertial Measurement Unit), a GNSS (Global Navigation Satellite System), or a combination of these. However, instead of this, the self-location estimation device 4 may estimate the self-location of the train 1 in the global coordinate system by SLAM (Simultaneous Localization and Mapping) using a three-dimensional point cloud output from the LiDAR device 3. The self-location estimation device 4 outputs self-location information indicating the estimated self-location to the anomaly detection device 6.

[0021] The camera 5 captures an image of the area ahead in the traveling direction of the train 1 to generate a forward image. The camera 5 outputs the generated forward image to the abnormality detection device 6.

[0022] The driving device 7 includes a motor and a motor driver. The driving device 7 drives the wheels 9 in accordance with a control signal from the abnormality detection device 6.

[0023] The braking device 8 is composed of a brake and a hydraulic circuit. The braking device 8 brakes the wheels 9 in accordance with a control signal from the abnormality detection device 6.

[0024] The abnormality detection device 6 uses the LiDAR device 3 mounted on the train 1 to detect an abnormality in the track structure 2 ahead in the traveling direction of the train 1. Here, the track structure 2 is typically composed of a subgrade, a ballast layered on the subgrade, a number of sleepers arranged on the ballast, and two rails supported by the number of sleepers.

[0025] The abnormality detection device 6 includes a point cloud acquisition unit 15, a point cloud extraction unit 16, a vibration correction unit 17, an abnormality detection unit 18, a control unit 19, a first point cloud storage unit 20, a second point cloud storage unit 21, and a third point cloud storage unit 22.

[0026] The point cloud acquisition unit 15 acquires a point cloud from the LiDAR device 3. Here, the point cloud output by the LiDAR device 3 to the anomaly detection device 6 is defined in the local coordinate system of the LiDAR device 3. Therefore, the point cloud acquisition unit 15 corrects the point cloud acquired from the LiDAR device 3 based on the self-position information acquired from the self-position estimation device 4. In addition, the point cloud acquisition unit 15 further corrects the point cloud acquired from the LiDAR device 3 based on the mounting position and mounting attitude of the LiDAR device 3 on the train 1. As a result, the point cloud is defined in the global coordinate system. The point cloud acquisition unit 15 stores the corrected point cloud in the first point cloud storage unit 20. The point cloud acquisition unit 15 executes the above acquisition, correction, and storage every time the LiDAR device 3 outputs a point cloud to the anomaly detection device 6 while the train 1 is running. Therefore, for example, when the LiDAR device 3 outputs a point cloud to the anomaly detection device 6 every 10 msec, the point cloud acquisition unit 15 executes the above acquisition, correction, and storage every 10 msec.

[0027] The point cloud extraction unit 16 extracts a point cloud corresponding to the track structure 2 ahead in the traveling direction of the train 1 from the point cloud stored in the first point cloud storage unit 20. The point cloud extraction unit 16 stores the extracted point cloud in the second point cloud storage unit 21. In Fig. 3, the point cloud stored in the first point cloud storage unit 20 is indicated by the symbol a, and the point cloud corresponding to the track structure 2 ahead in the traveling direction of the train 1 is indicated by the symbol b.

[0028] Various methods may be adopted as a method by which the point cloud extraction unit 16 extracts a point cloud corresponding to the track structure 2 ahead in the traveling direction of the train 1 from the point cloud stored in the first point cloud storage unit 20. For example, the point cloud extraction unit 16 can identify a point cloud corresponding to the track structure 2 based on map information including the position coordinates of the track structure 2 and self-location information output from the self-location estimation device 4. In addition, the point cloud extraction unit 16 can identify a point cloud corresponding to the track structure 2 based on a forward image output from the camera 5 using an object detection technology. Examples of the object detection technology include Faster R-CNN (Recurrent Neural Network), YOLO (You Only Look Once), and SSD (Single Shot multibox Detector).

[0029] The vibration correction unit 17 corrects the point cloud to reduce errors in the elevation values ​​of the point cloud caused by vibrations in the pitch direction of the train 1. The correction mechanism of the vibration correction unit 17 is as follows.

[0030] FIG. 4 shows the variation in elevation values ​​of the point cloud caused by vibration in the pitch direction of the train 1. That is, the train 1 repeatedly vibrates in the pitch direction due to rail deflection caused by the train 1's own weight and rail joints. Under this condition, when the LiDAR device 3 mounted on the train 1 measures the distance to an object c, the measurement point d corresponding to the object c will vary upward and downward from the true measurement point d0. For example, when the train 1 pitches so that the front of the train 1 rises, the LiDAR device 3 will output a measurement point d1 that is lower than the true measurement point d0. On the other hand, when the train 1 pitches so that the front of the train 1 falls, the LiDAR device 3 will output a measurement point d2 that is higher than the true measurement point d0. Note that the point cloud is a collection of multiple measurement points.

[0031] Here, although the multiple ranging points d corresponding to the object c will certainly vary up and down due to vibration in the pitch direction of the train 1, if the number of multiple ranging points d corresponding to the object c is sufficiently large, the distribution of the elevation values ​​of the multiple ranging points d can be considered to follow a normal distribution centered on the true ranging point d0, as shown in FIG. 5. In FIG. 5, the horizontal axis is the elevation value and the vertical axis is the degree. Therefore, the vibration correction unit 17 corrects the elevation values ​​of the multiple ranging points d shown in FIG. 4 so as to replace the elevation values ​​of the multiple ranging points d with a representative value of the elevation values ​​of the multiple ranging points d. Here, the representative value is typically the mean. However, instead of this, it may be the median or the mode.

[0032] The correction mechanism of the vibration correction unit 17 has been described above. Next, a procedure for applying the above correction mechanism to the point cloud stored in the second point cloud storage unit 21 will be described. FIG. 6 is a plan view of the track structure 2. As described above, the track structure 2 is composed of a roadbed, a trackbed 2b stacked on the roadbed, a plurality of sleepers 2c arranged on the trackbed, and two rails 2d supported by the plurality of sleepers 2c. Although FIG. 6 does not show a point cloud in a plan view of the track structure 2, a large number of point clouds actually exist. As shown in FIG. 6, the vibration correction unit 17 first divides the point cloud stored in the second point cloud storage unit 21 into a lattice shape in a plan view. In other words, the vibration correction unit 17 forms a plurality of cells e arranged in a lattice shape in a plan view. In this embodiment, each cell e is a square in a plan view, and the length of one side is set to be sufficiently smaller than the width dimension 2f of the head of the rail 2d. The shape of each cell e may be a rectangle instead of a square. Since each cell e is formed by dividing the cell into a lattice in plan view, it has an infinite length in the vertical direction. Then, the vibration correction unit 17 corrects the point cloud stored in the second point cloud storage unit 21 so as to replace the elevation values ​​of the multiple distance measurement points d belonging to each cell e with a representative value of the elevation values ​​of the multiple distance measurement points d. In this way, according to the above-mentioned correction mechanism, it is possible to reduce the error in the elevation values ​​of the point cloud caused by the vibration in the pitch direction of the train 1 for each cell e. The vibration correction unit 17 stores the point cloud corrected by the above-mentioned correction mechanism in the third point cloud storage unit 22.

[0033] 2, the abnormality detection unit 18 detects an abnormality in the railway track structure 2 based on the point cloud stored in the third point cloud storage unit 22. Specifically, the abnormality detection unit 18 includes a railway track structure analysis unit 23, a difference pattern generation unit 24, a fluctuation detection unit 25, and a fourth point cloud storage unit 26.

[0034] The track structure analysis unit 23 extracts point clouds corresponding to the trackbed 2b and the rail 2d based on the point cloud corresponding to the track structure 2. Various methods may be adopted to extract the point clouds corresponding to the trackbed 2b and the rail 2d. Typically, the track structure analysis unit 23 can identify the point clouds corresponding to the trackbed 2b and the rail 2d by clustering the point clouds corresponding to the track structure 2 based on the elevation value. Alternatively, the track structure analysis unit 23 may identify the point clouds corresponding to the trackbed 2b and the rail 2d using an object detection technology represented by PointNet. The track structure analysis unit 23 may first extract the point cloud corresponding to the trackbed 2b, and then identify the point cloud that is higher than the upper surface of the trackbed 2b by a predetermined amount or more as the point cloud of the rail 2d. The track structure analysis unit 23 stores the point clouds corresponding to the trackbed 2b and the rail 2d in the fourth point cloud storage unit 26.

[0035] FIG. 7 shows a cross-sectional view of the track structure 2. The differential pattern generation unit 24 calculates the difference dz in the height direction between the top surface 27 of the head of the rail 2d and the top surface 28 of the track bed 2b based on the point cloud corresponding to the track bed 2b and the rail 2d. The differential pattern generation unit 24 performs the above calculation at a point 50.0 meters away from the train 1, a point 50.1 meters away, a point 50.2 meters away, and so on. That is, the differential pattern generation unit 24 performs the above calculation at a plurality of points defined along the track structure 2. In other words, the differential pattern generation unit 24 performs the above calculation at a plurality of points where the distance from the train 1 gradually increases. As a result, the differential pattern generation unit 24 generates a differential pattern 29 as shown in FIG. 8. In FIG. 8, the horizontal axis is the distance from the train 1, and the vertical axis is the above difference dz. The differential pattern 29 is data showing the variation of the above difference according to the distance from the train 1.

[0036] The fluctuation detection unit 25 detects an abnormal fluctuation in the difference dz based on the difference pattern 29 generated by the difference pattern generation unit 24. Here, the abnormal fluctuation in the difference dz can be defined from various viewpoints. That is, the abnormal fluctuation in the difference dz can be defined as the difference dz being out of a predetermined range. The abnormal fluctuation in the difference dz can be defined as the local fluctuation range of the difference dz exceeding a predetermined amount. In this embodiment, the abnormal fluctuation in the difference dz is defined as the local fluctuation range of the difference dz exceeding a predetermined amount.

[0037] That is, the fluctuation detection unit 25 detects an abnormal fluctuation in the difference dz by detecting that the difference pattern 29 has dropped locally as shown in Fig. 8. Specifically, the fluctuation detection unit 25 detects an abnormal fluctuation in the difference dz by detecting that the amount of local change in the difference pattern 29 has exceeded a predetermined amount as shown in Fig. 8. In this case, it is considered that a local distortion, crack, wear, or defect has occurred in the rail 2d. In addition, the fluctuation detection unit 25 detects an abnormal fluctuation in the difference dz by detecting that the difference pattern 29 has jumped locally. In this case, it is considered that a local distortion has occurred in the rail 2d, or a foreign object such as a stone exists on the rail 2d.

[0038] In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting an abnormal fluctuation in the difference dz. In other words, the abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting a local loss of continuity in the difference dz. The continuity of the difference dz means that the fluctuation range of the difference dz is extremely small and the difference dz maintains a substantially constant value. The continuity of the difference dz can be rephrased as the continuity of the elevation values ​​of the point cloud corresponding to the rail 2d.

[0039] 2, when the abnormality detection unit 18 detects an abnormality in the track structure 2, the control unit 19 outputs an abnormality detection signal to the braking device 8. The braking device 8 brakes the wheels 9 in accordance with the abnormality detection signal output from the control unit 19.

[0040] Next, a description will be given of the control flow of the abnormality detection device 6. FIG.

[0041] First, the point cloud acquisition unit 15 acquires a point cloud from the LiDAR device 3 (S100).

[0042] Next, the point cloud acquisition unit 15 determines whether an abnormality detection timing has arrived (S110). The abnormality detection timing is typically set to arrive every 100 milliseconds. If the point cloud acquisition unit 15 determines that the abnormality detection timing has arrived (S110: YES), the point cloud acquisition unit 15 advances the process to step S120. If the point cloud acquisition unit 15 determines that the abnormality detection timing has not arrived (S110: NO), the point cloud acquisition unit 15 returns the process to S100.

[0043] Next, the point cloud extraction unit 16 extracts a point cloud corresponding to the track structure 2 ahead in the traveling direction of the train 1 from the point cloud acquired by the point cloud acquisition unit 15 (S120).

[0044] Next, the vibration correction unit 17 corrects the point cloud to reduce errors in the elevation values ​​of the point cloud caused by vibrations in the pitch direction of the train 1 (S130).

[0045] Next, the track structure analysis unit 23 of the abnormality detection unit 18 extracts a point cloud corresponding to the track bed 2b and the rails 2d based on the point cloud corresponding to the track structure 2 (S140).

[0046] Next, the differential pattern generation unit 24 of the abnormality detection unit 18 calculates the differences dz in the height direction between the top surface 27 of the head of the rail 2d and the top surface 28 of the trackbed 2b at multiple points where the distance from the train 1 gradually increases. As a result, the differential pattern generation unit 24 generates a differential pattern 29 (S150).

[0047] Next, the fluctuation detection unit 25 of the abnormality detection unit 18 detects an abnormal fluctuation in the difference pattern 29 (S160). If the fluctuation detection unit 25 detects an abnormal fluctuation in the difference pattern 29 (S160: YES), the fluctuation detection unit 25 advances the process to step S170. On the other hand, if the fluctuation detection unit 25 does not detect an abnormal fluctuation in the difference pattern 29 (S160: NO), the fluctuation detection unit 25 returns the process to step S100. In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting an abnormal fluctuation in the difference pattern 29.

[0048] Next, the abnormality detection unit 18 determines whether the abnormality of the track structure 2 detected in step S160 is caused by an ancillary structure (S170). The ancillary structure is typically a railroad crossing or a branch point. That is, if the point cloud corresponding to the track structure 2 includes a point cloud of an ancillary structure, the difference pattern 29 will fluctuate significantly at the point of the ancillary structure. If the abnormality detection unit 18 detects an abnormality every time the train approaches an ancillary structure, the smooth operation of the train 1 will be hindered. Therefore, in step S170, if the abnormality of the track structure 2 detected in step S160 is caused by an ancillary structure, it is determined that the abnormality is not caused by the track structure 2. That is, if the abnormality detection unit 18 determines that the abnormality of the track structure 2 detected in step S160 is caused by an ancillary structure (S170: YES), the abnormality detection unit 18 returns the process to step S100. On the other hand, if the abnormality detection unit 18 determines that the abnormality of the track structure 2 detected in step S160 is not caused by an ancillary structure (S170: NO), the control unit 19 outputs an abnormality detection signal to the braking device 8 (S180). Then, the braking device 8 brakes the wheels 9 in accordance with the abnormality detection signal output from the control unit 19. This makes it possible to prevent the train 1 from passing over the abnormal portion of the track structure 2.

[0049] Various methods may be adopted to determine whether the abnormality of the track structure 2 detected in step S160 is caused by an ancillary structure. For example, the abnormality detection unit 18 can perform the above determination based on the position information of the ancillary structure and its own position information. The abnormality detection unit 18 can also perform the above determination by detecting the ancillary structure based on the forward image output from the camera 5.

[0050] The first embodiment of the present disclosure has been described above. The first embodiment has the following features.

[0051] That is, the abnormality detection device 6 (abnormality detection system) includes a point cloud acquisition unit 15, a vibration correction unit 17, and an abnormality detection unit 18. The point cloud acquisition unit 15 acquires a point cloud corresponding to the track structure 2 ahead of the train 1 in the traveling direction. The vibration correction unit 17 divides the point cloud into a lattice shape in a planar view, and corrects the point cloud so that, for each cell e, the elevation values ​​of the multiple measurement points d belonging to the cell e are replaced with a representative value of the elevation values ​​of the multiple measurement points d. In this way, the vibration correction unit 17 reduces errors in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train 1. The abnormality detection unit 18 detects an abnormality in the track structure 2 based on the point cloud. According to the above configuration, when detecting an abnormality in the track structure 2 ahead of the train 1 in the traveling direction using the LiDAR device 3 mounted on the train 1, a technology having robustness against vibration in the pitch direction of the train 1 can be realized.

[0052] Furthermore, the abnormality detection unit 18 extracts a rail point cloud (rail track point cloud) as a point cloud corresponding to the rail 2d (rail track) from the point cloud. The abnormality detection unit 18 detects an abnormality in the railway structure 2 based on the continuity of the elevation values ​​of the rail point cloud. With the above configuration, an abnormality in the railway structure 2 can be detected with high detection accuracy.

[0053] The first embodiment and the other embodiments described below can be modified as follows.

[0054] The area in which the abnormality detection unit 18 detects an abnormality in the track structure 2 is not limited to the track structure 2 on which the train 1 is currently running. In a double track section in which a plurality of track structures 2 are adjacent to each other, a point cloud corresponding to an adjacent track structure 2 may be obtained, and an abnormality in the adjacent track structure 2 may be detected in the same manner.

[0055] In the above-described first embodiment, when the abnormality detection unit 18 detects an abnormality in the track structure 2, the control unit 19 controls the braking device 8 so that the braking device 8 decelerates the train 1. However, instead of this, the control unit 19 may notify the driver of the train 1 of the abnormality detection, or transmit an abnormality detection signal to a remote management device.

[0056] Furthermore, after the abnormality detection device 6 detects an abnormality in the track structure 2, in order to further increase the accuracy of the detection of the abnormality, the presence or absence of an abnormality at the abnormal location may be determined again by analyzing the forward image.

[0057] Second embodiment Next, a second embodiment of the present disclosure will be described. Below, the differences between the second embodiment and the first embodiment will be mainly described, and overlapping descriptions will be omitted.

[0058] In the above-described first embodiment, an abnormality in the railway structure 2 is detected based on the continuity of the elevation values ​​of the rail point cloud. In contrast, in the present embodiment, an abnormality in the railway structure 2 is detected based on the continuity of the brightness values ​​of the rail point cloud. Specifically, the method is as follows.

[0059] Fig. 10 shows a block diagram of the train 1. As shown in Fig. 10, the abnormality detection unit 18 includes a luminance pattern generation unit 40 instead of the differential pattern generation unit 24. In this embodiment, each of the distance measurement points constituting the point cloud includes luminance value information in addition to the coordinate information of the distance measurement point.

[0060] The luminance pattern generating unit 40 acquires luminance values ​​of the upper surface 27 of the head of the rail 2d at a point 50.0 meters away, a point 50.1 meters away, a point 50.2 meters away, and so on, from the train 1, based on the point cloud corresponding to the rail 2d. That is, the luminance pattern generating unit 40 executes the above acquisition at a plurality of points defined along the track structure 2. In other words, the differential pattern generating unit 24 executes the above acquisition at a plurality of points where the distance from the train 1 gradually increases. As a result, the luminance pattern generating unit 40 generates a luminance pattern 41 as shown in FIG. 11. In FIG. 11, the horizontal axis is the distance from the train 1, and the vertical axis is the above luminance value. The luminance pattern 41 is data showing the variation of the above luminance value according to the distance from the train 1.

[0061] The fluctuation detection unit 25 detects the abnormal fluctuation of the luminance value based on the luminance pattern 41 generated by the luminance pattern generation unit 40. Here, the abnormal fluctuation of the luminance value can be defined from various viewpoints. That is, the abnormal fluctuation of the luminance value can be defined as the luminance value being out of a predetermined range. The abnormal fluctuation of the luminance value can be defined as the range of local fluctuation of the luminance value exceeding a predetermined amount. In this embodiment, the abnormal fluctuation of the luminance value is defined as the range of local fluctuation of the luminance value exceeding a predetermined amount.

[0062] That is, the fluctuation detection unit 25 detects the abnormal fluctuation in the brightness value by detecting a local drop in the brightness pattern 41 as shown in Fig. 11. Specifically, the fluctuation detection unit 25 detects the abnormal fluctuation in the brightness value by detecting that the amount of local change in the brightness pattern 41 exceeds a predetermined amount as shown in Fig. 11. In this case, it is considered that a local distortion has occurred in the rail 2d.

[0063] In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting the abnormal fluctuation in the brightness value. In other words, the abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting a local loss of the continuity of the brightness value. The continuity of the brightness value means that the fluctuation range of the brightness value is extremely small and the brightness value maintains a substantially constant value.

[0064] Next, a description will be given of the control flow of the abnormality detection device 6. Fig. 12 shows the control flow of the abnormality detection device 6. Steps S100 to S140 in this embodiment are the same as those in the first embodiment.

[0065] In step S150, the luminance pattern generation unit 40 of the abnormality detection unit 18 acquires the luminance values ​​of the upper surface 27 of the head of the rail 2d at a plurality of points where the distance from the train 1 gradually increases. As a result, the luminance pattern generation unit 40 generates a luminance pattern 41 (S150).

[0066] Next, the fluctuation detection unit 25 of the abnormality detection unit 18 detects an abnormal fluctuation in the luminance pattern 41 (S160). If the fluctuation detection unit 25 detects an abnormal fluctuation in the luminance pattern 41 (S160: YES), the fluctuation detection unit 25 advances the process to step S170. On the other hand, if the fluctuation detection unit 25 does not detect an abnormal fluctuation in the luminance pattern 41 (S160: NO), the fluctuation detection unit 25 returns the process to step S100. In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting an abnormal fluctuation in the luminance pattern 41.

[0067] The second embodiment of the present disclosure has been described above. The second embodiment has the following features.

[0068] That is, the abnormality detection unit 18 extracts a rail point cloud (rail point cloud) as a point cloud corresponding to the rail 2d (rail track) from the point cloud. The abnormality detection unit 18 detects an abnormality in the railway structure 2 based on the continuity of the luminance values ​​of the rail point cloud. According to the above configuration, an abnormality in the railway structure 2 can be detected with high detection accuracy.

[0069] Third embodiment Next, a third embodiment of the present disclosure will be described. Below, the differences between the third embodiment and the first embodiment will be mainly described, and overlapping descriptions will be omitted.

[0070] In the first embodiment, an abnormality in the track structure 2 is detected based on the continuity of the elevation values ​​of the rail point cloud. That is, in the first embodiment, the existence of the rail point cloud is assumed. However, since the top surface 27 of the head of the rail 2d is generally a glossy flat surface, it is considered that the rail point cloud cannot be acquired in the first place. Therefore, in this embodiment, a point cloud missing area caused by the inability to acquire the rail point cloud is detected, and an abnormality in the track structure 2 is detected based on the continuity of the point cloud missing area. Specifically, it is as follows.

[0071] Fig. 13 shows a block diagram of the train 1. As shown in Fig. 13, the abnormality detection unit 18 includes a point cloud missing area detection unit 50 and a point cloud density pattern generation unit 51 instead of the track structure analysis unit 23, the difference pattern generation unit 24, and the fourth point cloud storage unit 26.

[0072] FIG. 14 shows the distribution of the point cloud density of the point cloud of the track structure 2. As shown in FIG. 14, the point cloud missing area detection unit 50 divides the point cloud stored in the third point cloud storage unit 22 into a grid in plan view. In other words, the point cloud missing area detection unit 50 forms a plurality of cells f arranged in a grid in plan view. In this embodiment, each cell f is a square in plan view, and the length of one side is set to be sufficiently smaller than the width dimension 2f of the head of the rail 2d. Since each cell f is formed by dividing into a grid in plan view, it has an infinite length in the vertical direction. Then, the point cloud missing area detection unit 50 counts the distance measurement points d belonging to each cell f for each cell f, and the counting result is the point cloud density of the cell f. However, instead of this, the counting result may be divided by the area of ​​the cell f in plan view, and the division result may be the point cloud density of the cell f. Since the areas of the multiple cells f formed by dividing the cell into a grid in a planar view are equal to each other in a planar view, the counting result of the ranging points d belonging to each cell f can be technically considered to be the same as the division result of the counting result by the area.

[0073] Next, the point cloud missing region detection unit 50 detects a point cloud missing region 52 extending forward in the traveling direction of the train 1. Specifically, the point cloud missing region detection unit 50 extracts a plurality of point cloud missing cells 53 having a point cloud density equal to or less than a predetermined value (e.g., 3) from among a plurality of cells f shown in Fig. 14. Next, the point cloud missing region detection unit 50 detects two point cloud missing regions 52 extending forward in the traveling direction of the train 1 by performing a labeling process on the extracted plurality of point cloud missing cells 53.

[0074] Next, the point cloud density pattern generation unit 51 acquires the point cloud density at a point 50.0 meters away from the train 1, a point 50.1 meters away from the train 1, a point 50.2 meters away from the train 1, etc., based on one of the two point cloud missing areas 52 detected by the point cloud missing area detection unit 50. That is, the point cloud density pattern generation unit 51 executes the above acquisition at a plurality of points defined along the track structure 2. In other words, the point cloud density pattern generation unit 51 executes the above acquisition at a plurality of points where the distance from the train 1 gradually increases. As a result, the point cloud density pattern generation unit 51 generates a point cloud density pattern 54 as shown in FIG. 15. In FIG. 15, the horizontal axis is the distance from the train 1, and the vertical axis is the above point cloud density. The point cloud density pattern 54 is data showing the fluctuation of the above point cloud density according to the distance from the train 1.

[0075] The fluctuation detection unit 25 detects the abnormal fluctuation of the point cloud density based on the point cloud density pattern 54 generated by the point cloud density pattern generation unit 51. Here, the abnormal fluctuation of the point cloud density can be defined from various viewpoints. That is, the abnormal fluctuation of the point cloud density can be defined as the point cloud density being out of a predetermined range. The abnormal fluctuation of the point cloud density can be defined as the local fluctuation range of the point cloud density exceeding a predetermined amount. The abnormal fluctuation of the point cloud density can be defined as the point cloud density locally exceeding a threshold value Th1. In this embodiment, the abnormal fluctuation of the point cloud density is defined as the point cloud density locally exceeding a threshold value Th1.

[0076] That is, the fluctuation detection unit 25 detects the above-mentioned abnormal fluctuation in the point cloud density by detecting that the point cloud density pattern 54 has locally jumped up and exceeded the threshold value Th1 as shown in Fig. 15. In this case, it is considered that a local distortion, crack, wear, or defect has occurred in the rail 2d, or a foreign object such as a stone is present on the rail 2d.

[0077] In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting an abnormal fluctuation in the point cloud density. In other words, the abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting a local loss of continuity in the point cloud density. The continuity of the point cloud density means that the fluctuation range of the point cloud density is extremely small and the point cloud density maintains a substantially constant value, i.e., zero.

[0078] Next, a description will be given of the control flow of the abnormality detection device 6. Fig. 16 shows the control flow of the abnormality detection device 6. Steps S100 to S130 in this embodiment are the same as those in the first embodiment.

[0079] In step S140, the point cloud missing region detection unit 50 of the abnormality detection unit 18 detects the point cloud missing region 52.

[0080] Next, the point cloud density pattern generating unit 51 of the abnormality detecting unit 18 generates a point cloud density pattern 54 based on the point cloud missing region 52 detected by the point cloud missing region detecting unit 50 (S150).

[0081] Next, the fluctuation detection unit 25 of the abnormality detection unit 18 detects an abnormal fluctuation in the point cloud density pattern 54 (S160). If the fluctuation detection unit 25 detects an abnormal fluctuation in the point cloud density pattern 54 (S160: YES), the fluctuation detection unit 25 advances the process to step S170. On the other hand, if the fluctuation detection unit 25 does not detect an abnormal fluctuation in the point cloud density pattern 54 (S160: NO), the fluctuation detection unit 25 returns the process to step S100. In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting an abnormal fluctuation in the point cloud density pattern 54.

[0082] The third embodiment of the present disclosure has been described above. The third embodiment has the following features.

[0083] That is, the abnormality detection unit 18 detects a point cloud missing region 52 extending forward in the traveling direction of the train 1. Then, the abnormality detection unit 18 detects an abnormality in the track structure 2 based on the continuity of the point cloud missing region 52. According to the above configuration, an abnormality in the track structure 2 can be detected with high detection accuracy.

[0084] (Fourth embodiment) Next, a fourth embodiment of the present disclosure will be described. Below, the differences between the fourth embodiment and the first embodiment will be mainly described, and overlapping descriptions will be omitted.

[0085] In the first embodiment, an abnormality in the track structure 2 is detected based on the continuity of the elevation values ​​of the rail point cloud. In contrast, in the present embodiment, an abnormality in the track structure 2 is detected based on the periodicity of the elevation values ​​of the entire point cloud of the track structure 2 in the traveling direction of the train 1. That is, it is considered that the elevation values ​​of the point cloud of the track structure 2 in the traveling direction of the train 1 periodically jump up depending on the distance from the train 1 due to the presence of sleepers 2c that are intermittently present in the traveling direction of the train 1. Therefore, in the present embodiment, the fluctuation of the elevation values ​​of the entire point cloud of the track structure 2 in the traveling direction of the train 1 is calculated, and an abnormality in the track structure 2 is detected based on the periodicity of the elevation values ​​of the entire point cloud of the track structure 2 in the traveling direction of the train 1.

[0086] Fig. 17 shows a block diagram of the train 1. As shown in Fig. 17, the abnormality detection unit 18 includes a dimensionality reduction unit 60 and a frequency analysis unit 61 instead of the track structure analysis unit 23 and the difference pattern generation unit 24.

[0087] FIG. 18 is an explanatory diagram of dimensional compression. As shown in FIG. 18, the dimensional compression unit 60 divides the point cloud stored in the third point cloud storage unit 22 into stripes perpendicular to the traveling direction of the train 1 in a planar view. In other words, the dimensional compression unit 60 forms a plurality of cells g arranged in stripes in a planar view. Each cell g extends in a direction perpendicular to the traveling direction of the train 1. In this embodiment, the thickness of each cell g in a planar view is set to be smaller than the width dimension 2g of the sleeper 2c. Since each cell g is formed by dividing into stripes in a planar view, it has an infinite length in the vertical direction. Then, the dimensional compression unit 60 sets the elevation value of each cell g as a representative value of the elevation values ​​of a plurality of distance measurement points belonging to the cell g. Here, the representative value is typically the mean value. However, instead of this, it may be the median value or the mode value. As a result, the point cloud of the railway structure 2 is dimensionally compressed into a one-variable function (relationship) between the distance from the train 1 and the altitude value of each cell g. The dimension reduction unit 60 stores the one-variable function after the dimension reduction in the fourth point cloud storage unit 26.

[0088] Next, the frequency analysis unit 61 performs frequency analysis on the one-variable function stored in the fourth point group storage unit 26. Specifically, the frequency analysis unit 61 performs frequency analysis on the one-variable function by using Fourier analysis or wavelet analysis.

[0089] FIG. 19 shows an example of the one-variable function. In FIG. 19, the horizontal axis is the distance from the train 1, and the vertical axis is the altitude value of each cell g. A plurality of sleepers 2c are arranged at intervals of 0.5 meters. In FIG. 19, the altitude value periodically jumps up to correspond to a plurality of sleepers 2c arranged at equal intervals in the traveling direction of the train 1.

[0090] Similarly, FIG. 20 shows an example of the one-variable function. In FIG. 20, the horizontal axis is the distance from the train 1, and the vertical axis is the altitude value of each cell g. The sleepers 2c are arranged at 0.5 meter intervals. FIG. 20 was created on the assumption that there is a foreign object 30 centimeters high and 1 meter deep at a point 40 meters away from the train 1. Therefore, in FIG. 20, the altitude value locally jumps up in the distance range from 40.0 meters to 41.0 meters from the train 1.

[0091] Figure 21 is a spectrogram obtained by performing a short-time Fourier transform on a 100-meter section of the one-variable function shown in Figure 19. In Figure 21, the horizontal axis is the distance from train 1, the vertical axis is the spatial frequency, and the shade of color is the intensity of the corresponding spatial frequency. It can be seen from Figure 21 that the intensity distribution of the spatial frequency is uniform regardless of the distance from train 1.

[0092] Figure 22 is a spectrogram obtained by performing a short-time Fourier transform on a 100-meter section of the one-variable function shown in Figure 20. In Figure 22, the horizontal axis is the distance from train 1, the vertical axis is the spatial frequency, and the shade of color is the intensity of the corresponding spatial frequency. Figure 22 shows that the intensity distribution of spatial frequencies is locally different from the rest around a point 40 meters away from train 1.

[0093] The fluctuation detection section 25 detects local changes in the intensity distribution based on the result of the frequency analysis by the frequency analysis section 61 .

[0094] The abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting a local change in the intensity distribution by the fluctuation detection unit 25. In other words, the abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting a local loss of continuity in the intensity distribution. The continuity of the intensity distribution means that the intensity distribution is uniform regardless of the distance from the train 1.

[0095] Next, a description will be given of the control flow of the abnormality detection device 6. Fig. 23 shows the control flow of the abnormality detection device 6. Steps S100 to S130 in this embodiment are the same as those in the first embodiment.

[0096] In step S140, the dimensionality reduction unit 60 of the anomaly detection unit 18 divides the point cloud stored in the third point cloud memory unit 22 into stripes perpendicular to the direction of travel of the train 1 in a planar view, thereby reducing the dimensionality of the point cloud of the railway structure 2 into a one-variable function (relationship) of the distance from the train 1 and the elevation value of each cell g (S140).

[0097] Next, the frequency analysis unit 61 of the anomaly detection unit 18 performs frequency analysis on the one-variable function after the dimensionality reduction (S150).

[0098] Next, the fluctuation detection unit 25 of the abnormality detection unit 18 detects a local change in the intensity distribution of the spatial frequency (S160). If a local change in the intensity distribution of the spatial frequency is detected (S160: YES), the fluctuation detection unit 25 advances the process to step S170. On the other hand, if a local change in the intensity distribution of the spatial frequency is not detected (S160: NO), the fluctuation detection unit 25 returns the process to step S100. In this way, the abnormality detection unit 18 detects an abnormality in the line structure 2 by detecting a local change in the intensity distribution of the spatial frequency.

[0099] The fourth embodiment of the present disclosure has been described above. The fourth embodiment has the following features.

[0100] That is, the abnormality detection unit 18 divides the point cloud of the track structure 2 into stripes perpendicular to the traveling direction of the train 1 in a planar view. For each cell g, the abnormality detection unit 18 sets the elevation value of the cell g as a representative value of the elevation values ​​of multiple distance measurement points belonging to the cell g. The abnormality detection unit 18 detects an abnormality in the track structure 2 based on the periodicity of a one-variable function (relationship) between the elevation values ​​of multiple cells g and the distance from the train 1. With the above configuration, an abnormality in the track structure 2 can be detected with high detection accuracy.

[0101] Furthermore, the abnormality detection unit 18 detects an abnormality in the railway structure 2 by performing frequency analysis on a one-variable function of the altitude values ​​of the multiple cells g and the distance from the train 1. According to the above configuration, an abnormality in the railway structure 2 can be detected based on the periodicity of the one-variable function.

[0102] In addition, when the abnormality detection unit 18 detects an abnormality in the railway structure 2, the abnormality detection unit 18 may identify the abnormal location of the railway structure 2 by referring to the spectrogram shown in Figure 22 and intensity modulating a specific spatial frequency component.

[0103] Furthermore, instead of subjecting the dimensionally compressed function of one variable to frequency analysis using short-time Fourier transform, the dimensionally compressed function of one variable may be subject to frequency analysis using wavelet transform.

[0104] Fifth embodiment Next, a fifth embodiment of the present disclosure will be described. Below, the differences between the fifth embodiment and the fourth embodiment will be mainly described, and overlapping descriptions will be omitted.

[0105] In the above-described fourth embodiment, the abnormality detection unit 18 performs frequency analysis on the dimensionally reduced function of one variable, and detects an abnormality in the railway structure 2 based on the analysis result. In contrast, in the present embodiment, the interval between the multiple sleepers 2c is calculated based on the dimensionally reduced function of one variable, and a change in the interval is detected to detect an abnormality in the railway structure 2.

[0106] Fig. 24 shows a block diagram of the train 1. As shown in Fig. 24, the abnormality detection unit 18 includes a sleeper interval calculation unit 70 instead of the frequency analysis unit 61.

[0107] The sleeper interval calculation unit 70 calculates the intervals between the sleepers 2c based on the dimension-reduced one-variable function. FIG. 25 and FIG. 26 show an example of the one-variable function. In FIG. 25 and FIG. 26, the horizontal axis is the distance from the train 1, and the vertical axis is the altitude value of each cell g. The sleepers 2c are arranged at intervals of 0.5 meters. In FIG. 25 and FIG. 26, the altitude value periodically jumps up to correspond to the sleepers 2c arranged at equal intervals in the traveling direction of the train 1. FIG. 26 was created on the assumption that a foreign object having a height of 12 centimeters is present at a point 40.2 meters from the train 1.

[0108] As shown in Fig. 25, the sleeper spacing calculation unit 70 identifies the location of each sleeper 2c in the traveling direction of the train 1 by comparing the one-variable function with the sleeper threshold value Th2. Next, the sleeper spacing calculation unit 70 calculates the spacing h between two adjacent sleepers 2c in the traveling direction of the train 1. In the example of Fig. 25, the spacings h0 to h7 are all approximately the same length. In contrast, in the example of Fig. 26, the spacing h3 is shorter than the other spacings h, and the spacing h4 is longer than the other spacings h.

[0109] The fluctuation detection unit 25 judges whether the intervals between the multiple sleepers 2c calculated by the sleeper interval calculation unit 70 are constant regardless of the distance from the train 1. In the example of FIG. 26, the interval h locally increases and decreases near a point 40 meters away from the train 1. The fact that the interval h3 is shorter than the other intervals h suggests that a foreign object is present near a point 40.2 meters away from the train 1. In contrast, the fact that the interval h4 is longer than the other intervals h suggests that a point cloud of the sleeper 2c located at a point 40.5 meters away from the train 1 could not be acquired due to a foreign object present near a point 40.2 meters away from the train 1.

[0110] In this manner, the abnormality detection unit 18 detects an abnormality in the track structure 2 by the fluctuation detection unit 25 detecting a local fluctuation in the above-mentioned interval. In other words, the abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting that the periodicity of the above-mentioned interval has been locally lost.

[0111] Next, a description will be given of the control flow of the abnormality detection device 6. Fig. 27 shows the control flow of the abnormality detection device 6. Steps S100 to S140 in this embodiment are the same as those in the fourth embodiment.

[0112] In step S150, the sleeper interval calculation unit 70 of the abnormality detection unit 18 calculates the intervals between the multiple sleepers 2c based on the one-variable function after dimensional reduction (S150).

[0113] Next, the fluctuation detection unit 25 of the abnormality detection unit 18 detects a local fluctuation in the above-mentioned interval (S160). If the local fluctuation in the above-mentioned interval is detected (S160: YES), the fluctuation detection unit 25 advances the process to step S170. On the other hand, if the local fluctuation in the above-mentioned interval is not detected (S160: NO), the fluctuation detection unit 25 returns the process to step S100. In this way, the abnormality detection unit 18 detects an abnormality in the track structure 2 by detecting the local fluctuation in the above-mentioned interval.

[0114] The fifth embodiment of the present disclosure has been described above. The fifth embodiment has the following features.

[0115] That is, the abnormality detection unit 18 obtains the interval between the multiple sleepers 2c of the railway structure 2 based on a one-variable function of the altitude values ​​of the multiple cells g and the distance from the train 1. The abnormality detection unit 18 detects an abnormality in the railway structure 2 based on the fluctuation in the interval between the multiple sleepers 2c. According to the above configuration, an abnormality in the railway structure 2 can be detected based on the periodicity of the one-variable function.

[0116] Sixth embodiment Next, a sixth embodiment of the present disclosure will be described. Below, the differences between the sixth embodiment and the fourth embodiment will be mainly described, and overlapping descriptions will be omitted.

[0117] In the above-described fourth embodiment, the abnormality detection unit 18 performs frequency analysis on the dimensionally reduced function of one variable, and detects an abnormality in the railway structure 2 based on the analysis result. In contrast, in the present embodiment, an abnormality in the railway structure 2 is detected by detecting that the dimensionally reduced function of one variable exceeds a threshold value.

[0118] 28 shows a block diagram of the train 1. As shown in FIG.

[0119] The fluctuation detection unit 25 compares the dimensionally compressed one-variable function with a threshold value. FIG. 29 shows an example of the above-mentioned one-variable function. In FIG. 29, the horizontal axis is the distance from the train 1, and the vertical axis is the altitude value of each cell g. The multiple sleepers 2c are arranged at intervals of 0.5 meters. In FIG. 29, the altitude value periodically jumps up to correspond to the multiple sleepers 2c installed at equal intervals in the traveling direction of the train 1. Also, FIG. 29 was created on the assumption that a foreign object having a height of 12 centimeters is present at a point 40.2 meters from the train 1. The fluctuation detection unit 25 compares the dimensionally compressed one-variable function with a threshold value Th3. The threshold value Th3 is set to be larger than the threshold value Th2 shown in FIG. 25 used to detect the sleeper 2c in the dimensionally compressed one-variable function. The fluctuation detection unit 25 detects that the dimensionally compressed one-variable function locally exceeds the threshold value Th3 through the above comparison.

[0120] In this way, the abnormality detection unit 18 detects an abnormality in the railway structure 2 by the fluctuation detection unit 25 detecting that the one-variable function locally exceeds the threshold value Th3.

[0121] Next, a description will be given of the control flow of the abnormality detection device 6. Fig. 30 shows the control flow of the abnormality detection device 6. Steps S100 to S140 in this embodiment are the same as those in the fourth embodiment.

[0122] In step S150, the fluctuation detection unit 25 of the anomaly detection unit 18 compares the dimensionally reduced one-variable function with a threshold value Th3.

[0123] Next, the fluctuation detection unit 25 of the abnormality detection unit 18 detects that the one-variable function has crossed the threshold value Th3 (S160). When the fluctuation detection unit 25 detects that the one-variable function has crossed the threshold value Th3 (S160: YES), the fluctuation detection unit 25 advances the process to step S170. On the other hand, when the fluctuation detection unit 25 does not detect that the one-variable function has crossed the threshold value Th3 (S160: NO), the fluctuation detection unit 25 returns the process to step S100. In this way, the abnormality detection unit 18 detects that the one-variable function has crossed the threshold value Th3, thereby detecting an abnormality in the track structure 2.

[0124] The sixth embodiment has been described above. The sixth embodiment has the following features.

[0125] That is, the abnormality detection unit 18 divides the point cloud of the track structure 2 into stripes perpendicular to the traveling direction of the train 1 in a planar view. For each cell g, the abnormality detection unit 18 sets the elevation value of the cell g as a representative value of the elevation values ​​of multiple distance measurement points belonging to the cell g. The abnormality detection unit 18 detects an abnormality in the track structure 2 by comparing the elevation values ​​of multiple cells g with a threshold value Th3. According to the above configuration, an abnormality in the track structure 2 can be detected with high detection accuracy.

[0126] Seventh embodiment Next, a seventh embodiment of the present disclosure will be described. Below, the differences between this embodiment and the fourth embodiment will be mainly described, and overlapping descriptions will be omitted.

[0127] In the above-described fourth embodiment, the abnormality detection unit 18 performs frequency analysis on the dimensionally reduced one-variable function, and detects an abnormality in the railway structure 2 based on the analysis result. In contrast, in the present embodiment, the abnormality in the railway structure 2 is detected by comparing the dimensionally reduced one-variable function with a model function created in advance based on the one-variable functions stored up to now.

[0128] Fig. 31 shows a block diagram of the train 1. As shown in Fig. 31, the abnormality detection unit 18 includes a model comparison unit 90 instead of the frequency analysis unit 61 and the fluctuation detection unit 25.

[0129] The model comparing unit 90 stores, as the model function, a model function that is, for example, a one-variable function as shown in Fig. 19. Then, the model comparing unit 90 compares, for example, a dimensionally reduced one-variable function as shown in Fig. 20 with the model function as shown in Fig. 19.

[0130] The abnormality detection unit 18 detects an abnormality in the railway structure 2 based on the comparison result by the model comparison unit 90. Specifically, the abnormality in the railway structure 2 is detected by detecting that the elevation value of the dimensionally reduced one-variable function is larger than the elevation value of the model function.

[0131] Next, a hardware configuration of the abnormality detection device 6 will be described. In the abnormality detection device 6, the point cloud acquisition unit 15, the point cloud extraction unit 16, the vibration correction unit 17, the abnormality detection unit 18, and the control unit 19 are realized by processing circuits. The first point cloud storage unit 20, the second point cloud storage unit 21, and the third point cloud storage unit 22 are realized by storage circuits. The processing circuit may be a processor and memory that executes a program stored in a memory, or may be dedicated hardware.

[0132] FIG. 32 is a diagram showing an example of a case where the processing circuit of the anomaly detection device 6 is configured with a processor and a memory. When the processing circuit is configured with a processor 1000 and a memory 1001, each function of the processing circuit of the anomaly detection device 6 is realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 1001. In the processing circuit, each function is realized by the processor 1000 reading and executing the program stored in the memory 1001. That is, the processing circuit includes the memory 1001 for storing a program that results in the processing of the anomaly detection device 6 being executed. It can also be said that these programs cause a computer to execute the procedures and methods of the anomaly detection device 6.

[0133] Here, the processor 1000 may be a CPU (Central Processing Unit), a processing device, an arithmetic device, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor), etc. Also, the memory 1001 may be, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), an EEPROM (Electrically EPROM), or other non-volatile or volatile semiconductor memory, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disc), etc.

[0134] Fig. 33 is a diagram showing an example in which the processing circuit included in the abnormality detection device 6 is configured with dedicated hardware. When the processing circuit is configured with dedicated hardware, the processing circuit 1002 shown in Fig. 33 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination of these. Each function of the abnormality detection device 6 may be realized by the processing circuit 1002 on a function-by-function basis, or each function may be realized collectively by the processing circuit 1002.

[0135] It should be noted that the functions of the anomaly detection device 6 may be partially realized by dedicated hardware and partially realized by software or firmware. In this manner, the processing circuit can realize each of the above-mentioned functions by dedicated hardware, software, firmware, or a combination of these.

[0136] The averaging process described with reference to Figs. 4 and 5 can be one of the means for reducing error factors. In contrast, the pattern detection described with reference to Figs. 8, 11, 15, 19, 20, 25, 26, and 29 can be one of the methods for detecting abnormalities. In particular, periodic pattern detection is considered to be effective as a means for acquiring a target signal from data containing noise. Although the noise reduction effect of the above-mentioned averaging process is more effective as the point cloud density increases, the noise reduction effect may be insufficient under conditions such as sparse point clouds. In contrast, extracting a target pattern (or removing non-target frequency components) by frequency analysis can be expected to have the effect of increasing robustness against data containing noise. In addition, for abnormal parts (such as Fig. 8) where the value fluctuates locally, it is considered possible to identify the abnormal parts in the same way if other normal parts are considered to be areas where the spatial frequency 0 / m component is prominent (≒ constant pattern). In practice, it is expected that known pattern information calculated from track data previously compiled into a database will be compared with pattern information obtained by dimensionally compressing point cloud information acquired during operation (comparison of power spectra at specific frequencies, determination of anomalies from spectrogram image information, use of CNN, etc.) to detect abnormalities in areas other than normal equipment around the track.

[0137] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by a person skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

[0138] Each drawing is merely an example for describing one or more embodiments. Each drawing may relate not only to one particular embodiment, but to one or more other embodiments. As can be understood by those skilled in the art, various features or steps described with reference to any one drawing may be combined with features or steps shown in one or more other drawings to create an embodiment not explicitly shown or described, for example. Not all features or steps shown in any one drawing are necessary to describe an exemplary embodiment, and some features or steps may be omitted. The order of steps described in any drawing may be changed as appropriate.

[0139] A part or all of the above-described embodiments can be described as, but is not limited to, the following supplementary notes. (Appendix 1) A point cloud acquisition means for acquiring a point cloud corresponding to a track structure ahead in a traveling direction of a train; a point cloud correction means for dividing the point cloud into a lattice shape in a planar view, and correcting the point cloud for each cell by replacing the elevation values ​​of a plurality of ranging points belonging to the cell with a representative value of the elevation values ​​of the plurality of ranging points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; an abnormality detection means for detecting an abnormality in the railway structure based on the point cloud; Including, Anomaly detection systems. (Appendix 2) The abnormality detection means extracting a rail point cloud as a point cloud corresponding to a rail from the point cloud; detecting an abnormality in the track structure based on continuity of elevation values ​​or brightness values ​​of the track point cloud; 2. The anomaly detection system of claim 1. (Appendix 3) The abnormality detection means Detecting a point cloud missing area extending forward in the direction of travel of the train; detecting an abnormality in the railway structure based on the continuity of the point cloud missing area; 2. The anomaly detection system of claim 1. (Appendix 4) The abnormality detection means The point cloud is divided into stripes perpendicular to the traveling direction in a planar view, and for each cell, an elevation value of the cell is set as a representative value of elevation values ​​of a plurality of distance measurement points belonging to the cell; detecting an abnormality in the track structure based on a periodicity of a relationship between the elevation values ​​of the plurality of cells and the distance from the train; 2. The anomaly detection system of claim 1. (Appendix 5) The abnormality detection means detecting an abnormality in the track structure by performing a frequency analysis on the relationship between the elevation values ​​of the plurality of cells and the distance from the train; 5. The anomaly detection system of claim 4. (Appendix 6) The abnormality detection means determining an interval between a plurality of sleepers provided in the railway structure based on a relationship between the elevation values ​​of the plurality of cells and the distance from the train; detecting an abnormality in the railway structure based on a variation in the intervals between the plurality of sleepers; 5. The anomaly detection system of claim 4. (Appendix 7) The abnormality detection means The point cloud is divided into stripes perpendicular to the traveling direction in a planar view, and for each cell, an elevation value of the cell is set as a representative value of elevation values ​​of a plurality of distance measurement points belonging to the cell; detecting an abnormality in the railway structure by comparing the elevation values ​​of the plurality of cells with a threshold value; 2. The anomaly detection system of claim 1. (Appendix 8) A point cloud acquisition means for acquiring a point cloud corresponding to a track structure ahead in a traveling direction of a train; a point cloud correction means for dividing the point cloud into a lattice shape in a planar view, and correcting the point cloud for each cell by replacing the elevation values ​​of a plurality of ranging points belonging to the cell with a representative value of the elevation values ​​of the plurality of ranging points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; an abnormality detection means for detecting an abnormality in the railway structure based on the point cloud; Including, Anomaly detection device. (Appendix 9) The computer A point cloud corresponding to the track structure ahead of the train in the direction of travel is acquired. The point cloud is divided into a grid in a planar view, and the point cloud is corrected so that, for each cell, the elevation values ​​of a plurality of measurement points belonging to the cell are replaced with a representative value of the elevation values ​​of the plurality of measurement points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; detecting an abnormality in the railway structure based on the point cloud; Anomaly detection methods. (Appendix 10) A program for causing a computer to execute the anomaly detection method described in appendix 9.

[0140] Some or all of the elements (e.g., configurations and functions) described in Supplementary Notes 2 to 7 that are dependent on Supplementary Note 1 may also be dependent on Supplementary Notes 8, 9, and 10 in the same dependent relationship as Supplementary Notes 2 to 7. Some or all of the elements described in any Supplementary Note may be applied to various hardware, software, recording means for recording software, systems, and methods. [Explanation of symbols]

[0141] 1 Train 2 Track structure 2b Road bed 2c sleepers 2d rail 2f Width dimension 2g Width dimension 3 LiDAR device 4 Self-position estimation device 5. Camera 6. Anomaly detection device 7. Drive unit 8 Braking device 9 wheels 15 Point cloud acquisition section 16 Point cloud extraction part 17 Vibration compensation unit 18 Anomaly detection section 19 Control section 20 1st point cloud storage section 21 2nd point cloud storage section 22 Third point cloud storage section 23 Track structure analysis department 24 Differential pattern generation unit 25 Fluctuation detection section 26 4th point cloud storage section 27 Top 28 Top 29 Differential Pattern 40 Luminance pattern generator 41 Brightness Pattern 50 Point cloud missing area detection section 51 Point cloud density pattern generator 52 Point cloud missing area 53 Point Cloud Missing Cells 54 Point Cloud Density Pattern 60 Dimensional compression section 61 Frequency Analysis Section 70 Sleeper Spacing Calculation Section 80 Threshold comparison unit 90 Model Comparison Section c object d Focus point d0 AF point d1 AF point d2 AF point e Cell dz Difference f cell g Cell h interval Th1 threshold Th2 sleeper threshold Th3 threshold

Claims

1. A point cloud acquisition means for acquiring a point cloud corresponding to a track structure ahead in a traveling direction of a train; a point cloud correction means for dividing the point cloud into a lattice shape in a planar view, and correcting the point cloud for each cell by replacing the elevation values ​​of a plurality of ranging points belonging to the cell with a representative value of the elevation values ​​of the plurality of ranging points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; an abnormality detection means for detecting an abnormality in the railway structure based on the point cloud; Including, Anomaly detection systems.

2. The abnormality detection means extracting a rail point cloud as a point cloud corresponding to a rail from the point cloud; detecting an abnormality in the track structure based on continuity of elevation values ​​or brightness values ​​of the track point cloud; The anomaly detection system of claim 1 .

3. The abnormality detection means Detecting a point cloud missing area extending forward in the direction of travel of the train; detecting an abnormality in the railway structure based on the continuity of the point cloud missing area; The anomaly detection system of claim 1 .

4. The abnormality detection means The point cloud is divided into stripes perpendicular to the traveling direction in a planar view, and for each cell, an elevation value of the cell is set as a representative value of elevation values ​​of a plurality of distance measurement points belonging to the cell; detecting an abnormality in the track structure based on a periodicity of a relationship between the elevation values ​​of the plurality of cells and the distance from the train; The anomaly detection system of claim 1 .

5. The abnormality detection means detecting an abnormality in the track structure by performing a frequency analysis on the relationship between the elevation values ​​of the plurality of cells and the distance from the train; The anomaly detection system according to claim 4 .

6. The abnormality detection means determining an interval between a plurality of sleepers provided in the railway structure based on a relationship between the elevation values ​​of the plurality of cells and the distance from the train; detecting an abnormality in the railway structure based on a variation in the intervals between the plurality of sleepers; The anomaly detection system according to claim 4 .

7. The abnormality detection means The point cloud is divided into stripes perpendicular to the traveling direction in a planar view, and for each cell, an elevation value of the cell is set as a representative value of elevation values ​​of a plurality of distance measurement points belonging to the cell; detecting an abnormality in the railway structure by comparing the elevation values ​​of the plurality of cells with a threshold value; The anomaly detection system of claim 1 .

8. A point cloud acquisition means for acquiring a point cloud corresponding to a track structure ahead in a traveling direction of a train; a point cloud correction means for dividing the point cloud into a lattice shape in a planar view, and correcting the point cloud for each cell by replacing the elevation values ​​of a plurality of ranging points belonging to the cell with a representative value of the elevation values ​​of the plurality of ranging points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; an abnormality detection means for detecting an abnormality in the railway structure based on the point cloud; Including, Anomaly detection device.

9. The computer A point cloud corresponding to the track structure ahead of the train in the direction of travel is acquired. The point cloud is divided into a grid in a planar view, and the point cloud is corrected so that, for each cell, the elevation values ​​of a plurality of measurement points belonging to the cell are replaced with a representative value of the elevation values ​​of the plurality of measurement points, thereby reducing an error in the elevation values ​​of the point cloud caused by vibration in the pitch direction of the train; detecting an abnormality in the railway structure based on the point cloud; Anomaly detection methods.

10. A program for causing a computer to execute the anomaly detection method according to claim 9.

Images