Railway track detection device and railway track detection method

Abstract

To more accurately and easily determine an opened / closed state of a switch present in a railway track from a captured image.SOLUTION: A railway track detection device in an embodiment comprises: a railway track detection unit that detects a railway track in a captured image captured by a capturing device that captures an area around a railway vehicle and outputs the railway track detection image, by using a railway track detection model, which is a trained model for inputting an image and outputting the railway track detection image as a detection result of a railway track composed of a pair of rails; a switch detection unit that, by applying image processing to the railway track detection image, detects a pair of triangular areas in the railway track detection image, each of which has a triangular shape between the rails, and, when the pair of triangular areas is detected, outputs a specific image in which each vertex of the pair of triangular areas is specified as a point of a switch for dividing the railway tracks into a plurality of railway tracks or for merging the plurality of railway tracks; and a switch state determination unit that determines an open / closed state of the switch from the specified image, by using a switch state determination model which is a trained model for inputting an image including the points of the switch and outputting the opened / closed state of the switch.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] An embodiment of the present invention relates to a rail detection device and a rail detection method. [Background technology]

[0002] In recent years, there has been a demand for automated driving of railway vehicles. To realize automated driving of railway vehicles, a more accurate and detailed map database (for example, a database including branching and merging positions) is required. In addition, frequent maintenance of existing map databases is required.

[0003] Here, a railway facility has multiple tracks, and switches are installed along the tracks to branch the tracks into multiple tracks or to merge the multiple tracks. When a railway vehicle moves to another track, it is necessary to switch the opening and closing of a point indicating a switch, which is the part at the switch that actually branches the tracks. In order to determine whether the opening and closing direction of such a switch matches the traveling direction of the railway vehicle, a technology is required that processes images taken by an on-board camera mounted on the railway vehicle and accurately determines the state of the opening and closing direction of the switch. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6855712 [Patent Document 2] Patent Publication No. 2021-157486 [Patent Document 3] Patent No. 7068026 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the running environment of railway vehicles, the illuminance fluctuates greatly depending on the environment such as weather and season, and the time of day, and the appearance of the rails changes greatly. In addition, the shooting angle changes greatly due to the swaying and tilting of the railway vehicle while it is running, making it difficult to always stably detect the tracks in the captured image. Therefore, it is even more difficult to detect the switch from the track and accurately determine the open / closed state of the switch switch.

[0006] One of the objects of this embodiment is to provide a track detection device and a track detection method that can more accurately and easily determine the open / closed state of a switch located on a track from a captured image. [Means for solving the problem]

[0007] A track detection device according to an embodiment is a track detection device mounted on a railway vehicle, and includes: a track detection unit that detects tracks from an image captured by an imaging device that captures an image of the area around the railway vehicle using a track detection model, which is a trained model that inputs an image and outputs a track detection image that is the detection result of a track consisting of a pair of rails, and outputs the track detection image; a switch detection unit that applies image processing to the track detection image to detect a pair of triangular areas, where the shape between the rails is triangular, from the track detection image, and when the pair of triangular areas are detected, outputs a specific image in which each of the vertices of the pair of triangular areas is identified as a switch point for branching the track into multiple tracks or merging the multiple tracks; a switch state determination unit that determines the open / closed state of the switch from the specific image using a switch state determination model, which is a trained model that inputs an image including the switch points and outputs the open / closed state of the switch; and an output unit that outputs the determination result of the open / closed state of the switch. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating an example of an overall configuration of a vehicle system according to an embodiment. [Diagram 2]FIG. 2 is a block diagram illustrating an example of a functional configuration of the railroad detection device according to the embodiment. [Diagram 3] FIG. 3 is a schematic diagram illustrating an example of a branching device according to an embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of a branch case among output results of the determination unit according to the embodiment. [Diagram 5] FIG. 5 is a diagram showing an example of an output from the determination unit when two railroad tracks merge in the embodiment. [Figure 6] FIG. 6 is a diagram for explaining an example of a method for determining the range of a branch region in the embodiment. [Figure 7] FIG. 7 is a diagram for explaining an example of a method for determining the range of a branch region in the embodiment. [Figure 8] FIG. 8 is a schematic diagram illustrating an example of a data structure of a layer of information regarding railroad branching in a map DB according to the embodiment. [Figure 9] FIG. 9 is a diagram illustrating an example in which the collation result is normal in the embodiment. [Figure 10] FIG. 10 is a diagram illustrating an example of a case where the collation result is abnormal in the embodiment. [Figure 11] FIG. 11 is a diagram showing an example in which a turnout is detected from a location that is not registered in the map DB in the embodiment. [Figure 12] FIG. 12 is a flowchart illustrating an example of a procedure of a railroad detection process according to the embodiment. [Figure 13] FIG. 13 is a diagram illustrating an example of an image generated in the course of the railroad detection process according to the embodiment. [Figure 14] FIG. 14 is a flowchart illustrating an example of a procedure of the matching process according to the embodiment. [Figure 15] FIG. 15 is a diagram showing an example of a determination result of an open / closed state of a branch switch when a main line is branched into two lines in the embodiment. [Figure 16] FIG. 16 is a diagram showing an example of a determination result of a switch state when two lines merge into one line in the embodiment. [Figure 17] FIG. 17 is a diagram showing an example of a determination result of an open / closed state of a switch when both a branch and a merge exist simultaneously in one captured image in the embodiment. [Figure 18] FIG. 18 is a diagram showing an example of a complex branch in the embodiment. [Figure 19] FIG. 19 is a diagram illustrating an example of a captured image including a plurality of tracks where a branch of a sequencer crossing exists in the embodiment. [Figure 20] FIG. 20 is a diagram for explaining an example of setting the range of a branching region in the case of a complex branching structure in the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, the present embodiment will be described with reference to the accompanying drawings.

[0010] (Embodiment) (Overall configuration of vehicle system 1) FIG. 1 is a diagram showing an example of an overall configuration of a vehicle system 1 according to an embodiment. As shown in FIG. 1, the vehicle system 1 according to the present embodiment includes a railroad vehicle RV and a central device 200 provided outside the railroad vehicle RV. The railroad vehicle RV runs on rails R. The railroad vehicle RV includes a camera 10, an antenna 50, a display device 30, and a track detection device 100. The track detection device 100 and the central device 200 are connected by a wireless network. The antenna 50 and the display device 30 are each connected to the track detection device 100 by wire or wirelessly.

[0011] The camera 10 is capable of capturing an image in the traveling direction of the railroad vehicle RV. The camera 10 may be a device capable of measuring the distance to an object, such as a stereo camera. The camera 10 is directed to the front of the railroad vehicle RV, that is, in the traveling direction of the railroad vehicle RV. As a result, the camera 10 captures an image in the traveling direction of the railroad vehicle RV. The camera 10 transmits the captured image to the track detection device 100. Note that the image captured by the camera 10 is not limited to the front of the railroad vehicle RV, but may be an image of the periphery of the railroad vehicle RV. For example, in order to grasp the location and open / closed state of a switch for the purpose of inspection and maintenance of the track, the camera 10 may be provided at the rear of the railroad vehicle RV and configured to capture an image of the rear of the railroad vehicle RV. The camera 10 is an example of an imaging device. Note that the camera is not limited to a stereo camera, and may be a monocular camera.

[0012] The antenna 50 is installed near the center of the railcar RV in the width direction and receives radio signals transmitted from positioning satellites. The radio signals include information for measuring the position of the railcar RV.

[0013] In addition, the railway vehicle RV may be equipped with independent sensors such as a three-axis acceleration sensor, a three-axis gyro sensor, a three-axis geomagnetic sensor, a wheel speed sensor, and a tachometer generator (TG: tachograph generator).

[0014] The display device 30 is a display device such as a display unit, and is connected to the track detection device 100 so as to be capable of displaying various types of information.

[0015] The track detection device 100 is a device that detects tracks from an image captured by a camera 10, detects switches in the tracks, and determines the open / closed state of the switches. Here, the tracks are composed of a pair of left and right rails R. A switch is a device for branching a track into multiple tracks or merging the multiple tracks.

[0016] The central unit 200 is an information processing device such as a server device. The central unit 200 communicates with the track detection device 100 to control the running of the railroad vehicle RV. The central unit 200 of this embodiment generates and updates a map database (hereinafter referred to as a "map DB") on which the railroad vehicle RV can run, and transmits the map database to the track detection device 100. The central unit 200 also transmits information on the tracks and switches on which the railroad vehicle RV runs to the track detection device 100 at a predetermined timing. Furthermore, the central unit 200 receives various information from the track detection device 100.

[0017] (Configuration of the track detection device) Next, the railroad detection device 100 will be described in detail. 2 is a block diagram showing an example of a functional configuration of the railroad detection device 100 according to the embodiment. As shown in FIG. 2, the railroad detection device 100 according to the embodiment mainly includes a determination unit 13, a positioning unit 16, a railroad information generation unit 14, a collation unit 15, a display processing unit 18, a notification unit 20, a communication unit 19, and a storage unit 17.

[0018] The positioning unit 16 locates the current position of the railcar RV. Specifically, the positioning unit 16 can measure the position of the railcar RV based on radio signals received by an antenna 50 from an artificial satellite. The position of the railcar RV is position information about the vicinity of the center of the left and right rails of the track. The positioning unit 16, for example, constitutes a Global Navigation Satellite System (GNSS), performs positioning (satellite positioning) of the railcar RV based on radio waves received by the antenna 50, and outputs position information (also referred to as positioning information) as a measurement result.

[0019] Here, GNSS is a general term for satellite positioning systems such as the US GPS, Japan's Quasi-Zenith Satellite System (QZSS), Russia's GLONASS (GLO), the European Union's Galileo (GAL), and China's Beidou (BDS). Therefore, the receiver that constitutes the antenna 50 and the positioning unit 16 is configured to be capable of performing positioning using any of these satellite positioning systems.

[0020] The positioning unit 16 can measure the current position of the railway vehicle RV by using a detection result from an inertial sensor (not shown) in addition to or separately from the positioning result based on the radio waves received by the antenna 50. For example, the positioning unit 16 can measure the current position by integrating an output value from a wheel tachometer generator of an inertial sensor (not shown).

[0021] The determination unit 13 detects tracks and switches and determines the open / closed states of switch switches from the captured images acquired by the camera 10. As shown in Fig. 2, the determination unit 13 includes a track detection unit 131, a switch detection unit 132, and a switch state determination unit 133. Hereinafter, the open / closed state of the switch switch may also be referred to as the "open / closed state of the switch."

[0022] The track detection unit 131 receives an image captured by the camera 10 of the forward direction of the rail vehicle RV, detects tracks from the captured image, and outputs a track detection image. The track detection image is an image of the track detection result. The track detection image is classified into three classes: a pair of left and right rails that constitute the track, a gauge indicating the gap between the pair of left and right rails, and other pixels.

[0023] Specifically, the track detection unit 131 includes a track detection model 1311, and uses this track detection model 1311 to detect one or more tracks from a captured image and output a track detection image.

[0024] The track detection model 1311 is a trained model that inputs an image and outputs a track detection image that is a detection result of a track consisting of a pair of rails. The track detection model 1311 is generated by machine learning on an arbitrary computer using training data in advance. The learning algorithm used to generate the track detection model 1311 can be, for example, semantic segmentation in deep learning, but is not limited thereto.

[0025] The track detection model 1311 is trained in advance using a large number of learning images of a pair of rails and a switch that constitute a track, and annotations in which each pixel indicates a pair of rails, a gauge, or other pixels that are assigned in advance as correct answer data to each learning image. That is, the parameters of the learning model are updated so that the output value obtained by inputting each learning image into a learning model modeled by a neural network or the like matches the correct answer data corresponding to the learning image, thereby generating the track detection model 1311.

[0026] The track detection unit 131 performs semantic segmentation on the captured image from the camera 10 using this track detection model 1311, and is able to output a track detection image in which the captured image is classified into a pair of left and right track rails, gauge, and other areas.

[0027] Fig. 3 is a schematic diagram showing an example of a turnout in an embodiment, in which a pair of rails R1 is turned into a pair of rails R1 and a pair of rails R2. Fig. 4 is a diagram showing an example of a branch case among the output results of the determination unit 13 according to the embodiment. Fig. 4(a) shows an example of a track detection image output by the track detection unit 131 from a captured image of a track including the switch shown in Fig. 4. As shown in the example of the track detection image in Fig. 4(a), the left rails R1L, R2L and right rails R1R, R2R of the track and each class of gauge are shown with different hatching, and other classes are shown without hatching. The track detection model 1311 may be configured to classify the left rails R1L, R2L and the right rails R1R, R2R into the same class, and then further classify them into left rails R1L, R2L and right rails R1R, R2R in post-processing such as image processing.

[0028] The switch detection unit 132 applies image processing to the track detection image output by the track detection unit 131 to detect a pair of triangular areas where the shape between the rails is triangular from the track detection image, and when a pair of triangular areas is detected, outputs a specific image in which each vertex of the pair of triangular areas is specified as a switch point. Here, a point refers to a switch.

[0029] When multiple tracks are detected in the track detection image, the switch detection unit 132 identifies the track on which the rail vehicle RV runs from the track detection image, detects one or more pairs of triangular areas in front of the identified running track, and outputs a specific image in which the vertices of the triangles of the one or more pairs of triangular areas are identified as points of one or more switch. Note that when a single track is detected in the track detection image, the switch detection unit 132 identifies the track on which the rail vehicle RV runs.

[0030] As shown in the example of a switch in Fig. 3, the switch is composed of a point section 101 and a lead section 102. In order to correctly determine the open / closed state of the switch of the switch, it is necessary to detect the point section 101 with high accuracy.

[0031] The switch detection unit 132 performs an expansion process on the track detection image to expand the width of each rail in the track detection image, and detects a pair of triangular regions from the image after the expansion process. Here, for example, morphological processing can be used as the expansion process.

[0032] Fig. 4(b) shows an example of the track detection image shown in Fig. 4(a) subjected to expansion processing. The switch detection unit 132 performs expansion processing on each of the left rails R1L, R2L and the right rails R1R, R2R in the track detection image shown in Fig. 4(a) to expand the width of each rail. As a result, as shown in Fig. 4(b), the right rail R1R comes into contact with the right rail R2R, and the left rail R1L comes into contact with the left rail R2L.

[0033] As a result, as shown in Fig. 4(b), the shape between the rails converted from the point section to the lead section of the turnout becomes triangular. This triangular part is called a triangular area. A pair of left and right triangular areas appears corresponding to the two left and right rails. For this reason, the turnout detection unit 132 deletes the rail portion from the image of Fig. 4(b) after the expansion process, and detects a pair of left and right triangular areas 401a, 401b from the gauge area. Fig. 4(c) shows an example in which a pair of triangular areas 401a, 401b has been extracted.

[0034] 4(c), the vertices of the pair of triangular regions 401a and 401b correspond to the ends of the point of the junction. Therefore, the junction detection unit 132 identifies the vertices of the triangular regions as the ends of the point of the junction, sets a predetermined range including the vertices as a junction region 403 indicating the range of the point of the junction, and outputs the specified image.

[0035] 4 shows an example where the tracks are branched at a branch point, but FIG. 5 shows an example of the output of the determination unit 13 when two tracks merge in the embodiment. In the case of a merger, the track detection unit 131 and the switch detection unit 132 also perform the same processing as in the example shown in FIG.

[0036] Here, the range of the branch region 403 will be described. The turnout detection unit 132 sets the width, which is the length in a direction intersecting the traveling direction of the railcar RV in the turnout area 403, to be longer than the interval between the pair of rails.

[0037] The switch detection unit 132 sets the vertical width, which is the length in the traveling direction of the railcar, based on the horizontal width in the switch area 403. For example, the switch detection unit 132 can set the vertical width to be the same length as the horizontal width, or the vertical width to be 1 / 2 the length of the vertical width.

[0038] The turnout detection unit 132 can also set the vertical width of the turnout area 403 based on the horizontal width and internal and external parameters related to the camera 10. The vertical width of the turnout area 403 may be set longer as the distance from the current position of the railcar RV to the turnout point increases. Furthermore, the turnout detection unit 132 can adjust the vertical width of the turnout area according to the detected angle of the apex of the triangular area (i.e., the angle between the main rail and the tongue rail). As an example, when the angle of the apex of the triangular area is small, the point portion becomes longer, so the turnout detection unit 132 may be configured to set the vertical width of the turnout area to be longer.

[0039] 6 and 7 are diagrams for explaining an example of a method for determining the range of the branching area 403 in the embodiment. Fig. 6 shows a case where the angle of the apex of the triangular area indicating the branching point is small. Fig. 7 shows a case where the angle of the apex of the triangular area indicating the branching point is larger than that in Fig. 6.

[0040] As shown in Fig. 6, when the angle of the apex of the triangular area is small, the angle change in the running direction of the railcar RV is small and the point portion of the switch is long, so the switch detection unit 132 sets the vertical width of the switch area to be large. On the other hand, as shown in Fig. 7, when the angle of the apex of the triangular area is large, the angle change in the running direction of the railcar RV is large and the point portion of the switch is short, so the switch detection unit 132 sets the vertical width of the switch area to be small.

[0041] Returning to FIG. 2, the switch state determination unit 133 uses the switch state determination model 1331 to determine the open / closed state of the switch from the specific image output by the switch detection unit 132. The switch state determination model 1331 is a trained model that inputs an image including points of the switch and outputs the open / closed state of the switch. The switch state determination model 1331 is generated by machine learning using training data in advance by an arbitrary computer. The learning algorithm used to generate the switch state determination model 1331 is, for example, class classification in deep learning, but is not limited thereto.

[0042] The switch state determination model 133 is trained in advance for a specific image including a given switch point using annotations indicating the results of multi-class classification such as left switch, right switch, left merging, right merging, and going straight, which are given in advance as correct answer data for each learning image. That is, the parameters of the learning model are updated so that the output value obtained by inputting each learning image into a learning model modeled by a neural network or the like matches the correct answer data corresponding to the learning image, and the switch state determination model 1331 is generated. Here, left switch, right switch, left merging, right merging, going straight, etc. indicate the open / closed state of the switch.

[0043] The switch state determination unit 133 performs class classification on the specific image including the switch point output by the switch detection unit 132 using this switch state determination model 1331, and can classify the output specific image into left branch, right branch, left merging, right merging, and straight ahead based on the open / closed state of the switch switch.

[0044] The track information generating unit 14 generates track information based on the processing results of the determining unit 13, i.e., the switch detection result which is the output result of the switch detection unit 132, the determination result of the open / closed state of the switch which is the output result of the switch state determining unit 133, and the own vehicle position of the railway vehicle RV measured by the positioning unit 16. As an example, the track information generating unit 14 can detect a switch 200 m (longitude and latitude information) ahead in the traveling direction of the own vehicle position of the railway vehicle RV, and generate that the state of the switch is a left branch.

[0045] The storage unit 17 is a storage medium such as a hard disk drive (HDD) or a solid state drive (SSD). The storage unit 17 of this embodiment stores a map DB 171. The map DB 171 is an example of map information.

[0046] The map DB 171 is information on a map of an area including the travel route of the railroad vehicle RV. The map DB 171 is composed of multiple layers. For example, it is composed of a layer in which basic map data at a 1 / 2500 level from the Geospatial Information Authority of Japan is registered, a layer in which detection area sections and detection area ranges are registered, a layer in which information on civil engineering structures such as tunnels and bridges is registered, etc. One of these multiple layers includes a layer in which information on railroad branching is registered.

[0047] 8 is a schematic diagram illustrating an example of a data structure of a layer of information about railroad junctions in the map DB 171 according to the embodiment. As illustrated in FIG. 8, the layer of information about junctions in the map DB 171 associates position information of a junction on the railroad, an open / closed state of the junction for each time period, and management information.

[0048] The position information of the switch is the position coordinates of the switch on the map. The open / close state of the switch is the correct open / close state of the switch switch for each time period. The management information indicates a processing method when the switch and the open / close state of the switch determined by the determination unit 13 do not match the registered contents of the map DB 171 or do not match the notification contents from the central device 200.

[0049] The image storage designates whether or not to store the detected image in the memory unit 17 when the switch and the open / closed state of the switch determined by the determination unit 13 do not match the registered contents of the map DB 171. ON means that the image is stored, and OFF means that the image is not stored.

[0050] The map update specifies whether or not to transmit a request to update the map DB 171 to the central device 200 when the switch and the open / closed state of the switch determined by the determination unit 13 do not match the registered contents of the map DB 171. When it is on, it means that the request to update the map DB 171 is transmitted to the central device 200, and when it is off, it means that the request to update the map DB 171 is not transmitted to the central device 200.

[0051] The fault detection notification specifies whether or not to notify of a fault when the switch and the open / closed state of the switch determined by the determination unit 13 do not match the notification from the central device 200. When it is on, it means that a fault is notified, and when it is off, it means that a fault is not notified.

[0052] The above management information of the map DB 171 is all set to ON by default and can be changed by the administrator. Furthermore, these pieces of information in the map DB 171 are merely examples, and the map DB 171 may not have some of these pieces of information, or may have other pieces of information.

[0053] 2, the collation unit 15 collates the position information of the switch detected by the switch detection unit 132 and the open / closed state of the switch determined by the switch state determination unit 133 with the position information of the switch in the map DB 171 and the open / closed state of the switch for a time period including the current time. If the collation results are not consistent, the collation unit 15 simultaneously records the traveling position of the railroad vehicle RV (host vehicle) measured by the positioning unit 16 and the captured image captured by the camera 10 in the storage unit 17 if image storage in the management information of the map DB 171 is set to on.

[0054] In addition, if the comparison results do not match, the comparison unit 15 determines whether the open / closed state of the switch determined by the switch state determination unit 133 matches the traveling direction of the railway vehicle RV received by the communication unit 19 described later.

[0055] The display processing unit 18 causes the display device 30 to display the position information of the switch detected by the switch detection unit 132, the open / closed state of the switch determined by the switch state determination unit 133, and the comparison result by the comparison unit 15.

[0056] The communication unit 19 transmits and receives various information and notifications to and from the central device 200 via wireless communication. Specifically, when the collation result by the collation unit 15 with the map DB 171 does not match, the communication unit 19 according to this embodiment transmits an instruction to correct the map DB 171 to the central device 200. Note that the instruction to correct the map DB 171 by the communication unit 19 is issued when map update in the management information of the map DB 171 is set to ON.

[0057] Furthermore, as described above, if the comparison result does not match, and image storage in the management information of the map DB 171 is set to on, the comparison unit 15 simultaneously records the traveling position of the railway vehicle RV (host vehicle) and the captured image captured by the camera 10 in the memory unit 17, thereby improving the reliability of the map DB 171 update process or correction process in the central device 200. Furthermore, the communication unit 19 receives information relating to the traveling direction of the railcar RV from the central device 200.

[0058] If the collation result by the collation unit 15 is no match, the notification unit 20 outputs to the display device 30 a message indicating that the collation result is no match. Furthermore, when the open / closed state of the switch determined by the switch state determination unit 133 differs from the traveling direction of the railway vehicle RV received by the communication unit 19, the notification unit 20 outputs an alarm to the display device 30 to the effect that the switch has a malfunction. The output of the alarm by the notification unit 20 is performed when the malfunction detection notification in the management information of the map DB 171 is set to ON. The notification unit 20 and the communication unit 19 may be configured to transmit a notice to the central device 200 that a switch has a failure. The notification unit 20, the display processing unit 18, and the display device 30 are an example of an output unit.

[0059] An example of a screen display of the result of the matching process will be described below. 9 is a diagram showing an example of a case where the collation result is normal in the embodiment. The position and open / closed state of the switch detected by the determination unit 13 match the registered information in the map DB 171, and the collation result is normal. In this case, the function of saving images serving as evidence, the function of updating the map DB, and the function of notifying fault detection are turned off by default. The setting of the image saving function can be changed by an administrator or the like.

[0060] FIG. 10 is a diagram showing an example of a case where the collation result is abnormal in the embodiment. FIG. 10 shows an example of a case where the position information of the switch detected by the determination unit 13 matches the registered information of the map DB 171, but the switch open / close state of the detected switch is different from the open / close state of the switch registered in the map DB 171. When the collation unit 15 detects an abnormality, it has a possibility of leading to a serious accident, so the function of saving images as evidence is turned on. In this case, the administrator or the like can change (set to off) the image saving function. In addition, it is also possible to set the update function of the map DB 171 and the fault detection notification function to on / off.

[0061] Fig. 11 is a diagram showing an example of a case where, in the embodiment, a switch is detected from a location that is not registered in the map DB 171. In this case, as shown in Fig. 11, all of the default settings of the function for saving images that serve as evidence, the function for updating the map DB 171, and the function for notifying failure detection are turned on.

[0062] (Railway detection processing) Next, a track detection process performed by the track detection device 100 of this embodiment configured as above will be described. FIG. 12 is a flowchart illustrating an example of a procedure of a railroad detection process according to the embodiment. FIG. 13 is a diagram illustrating an example of an image generated in the course of the railroad detection process according to the embodiment. First, the track detection unit 131 acquires a captured image from the camera 10 (S101). Fig. 13(a) is an example of the captured image. Then, the track detection unit 131 detects a track area from the captured image using a track detection model 1311, and outputs a track detection image (S102). Fig. 13(b) is an example of a track detection image detected from the captured image of Fig. 13(a).

[0063] Next, the switch detection unit 132 determines the travel lane of the host vehicle from one or more tracks in the track detection image (S103).

[0064] Here, the switch detection unit 132 may determine the traveling lane of the vehicle by a method of judging the traveling wheels based on internal parameters and external parameters of the camera 10. Alternatively, the switch detection unit 132 may be configured to use a method of using the results of the GPS signal received by the positioning unit 16. Alternatively, the switch detection unit 132 may be configured to analyze a track detection image to determine the traveling lane.

[0065] Then, as described above, the switch detection unit 132 detects a switch point ahead of the track of the determined travel lane (S104). Fig. 13(c) is a diagram showing an example in which a switch point is detected.

[0066] When a switch point is detected ahead of the track of the travel lane (S104: Yes), the switch detection unit 132 extracts a triangular area that is the switch area from the track detection image as described above, and outputs a specific image (S106). Fig. 13(d) is a diagram showing an example of the specific image.

[0067] Next, the switch state determination unit 133 determines the open / closed state of the switch from the specific image using the switch state determination model 1331 (S107). Next, the track information generation unit 14 generates and outputs track information including the switch (S108). Then, the process returns to S103, and the processes from S106 to S108 are repeatedly executed until a switch point is not detected ahead of the track of the travel lane.

[0068] In S104, if a switch point is not detected ahead of the track of the travel lane (S104: No), the collation unit 15 performs a collation process (S105). When the collation process is completed, the process ends.

[0069] Next, the collation process in S105 will be described. FIG. 14 is a flowchart illustrating an example of a procedure of the matching process according to the embodiment. The collation unit 15 acquires information such as the position information of a switch and the open / close state of the switch from the map DB 171 (S201). Next, the collation unit 15 checks whether or not the detected position information of the switch and the open / close state of the switch, which are the judgment results of the judgment unit 13, match the position information of the switch and the open / close state of the switch acquired from the map DB 171 in S201 (S202).

[0070] If the comparison shows that the two match (S202: Yes), the process ends and returns to the caller. On the other hand, if the comparison result shows that the two do not match (S202: No), the communication unit 19 transmits an instruction to update the map DB 171 to the central device 200 (S203). The notification unit 20 also outputs a notice of the failure detection together with an alarm to the display device 30 (S204). Then, the process returns to the caller.

[0071] Fig. 15 is a diagram showing an example of a determination result of an open / close state of a switch switch when a main line is branched into two lines in an embodiment. Note that in Fig. 15, the background other than the line and the switch is omitted for convenience of explanation.

[0072] Fig. 15(a) shows the detection result of the rails and gauge of the track. In the case of a branch, vertices 1701, 1702 of the triangular area of ​​the gauge corresponding to the branch point face downward (toward the viewer in Fig. 15(a)). The branch detection unit 132 detects the lower vertices 1701, 1702 of the pair of left and right triangular areas, and since the lower part of the vertex 1701 is the branch point indicating a junction, the detected branch point is set as the branch area 403 in Fig. 15(b).

[0073] 16 is a diagram showing an example of a result of judging the state of a switch when two tracks merge into one track in an embodiment. In FIG. 16, the background other than the tracks and the switch is omitted for convenience of explanation.

[0074] In Fig. 16(a), the detection result is also the rails and gauge of the track. In the case of merging, vertices 1801 and 1802 of the triangular area of ​​the gauge corresponding to the switch point face upward (the back of Fig. 16(a)). The switch detection unit 132 detects the upper vertices of the pair of left and right triangular areas, and since the upper parts of the vertices become the switch point indicating merging, the detected switch point is set as the switch area 403 in Fig. 16(b).

[0075] Fig. 17 is a diagram showing an example of the determination result of the open / closed state of a switch when both a branch and a merge exist simultaneously in one captured image in the embodiment. Fig. 17 shows an example of the detection result of the switch point and the determination result of the open / closed state of the switch. As shown in Fig. 17, even if the switch detection unit 132 detects multiple switch with different open / closed states of the switch from one captured image, the switch state determination unit 133 can accurately determine the position of each switch and the open / closed state of the switch.

[0076] In the above, a relatively simple example was described in which one or two tracks were detected in one captured image. However, in the present embodiment, by using the above-described processing, it is possible to accurately detect switches and determine the open / closed state of switches even in captured images in which there are multiple tracks and a complex branching structure.

[0077] FIG. 18 is a diagram showing an example of a complex branch in an embodiment. FIG. 18 shows the most complex branch structure of a sequencer crossing. Here, the branch structure of a sequencer crossing is a track structure in which two sets of crossovers are provided in an X-shape between two adjacent tracks. A sequencer crossing is composed of four sets of switches and one crossing. By passing through the sequencer crossing, trains running on either track can cross over to the adjacent track without backing up.

[0078] The following will explain the branching structure of sequencer crossing shown in FIG. 18 as an example. FIG. 19 is a diagram illustrating an example of a captured image including a plurality of tracks where a branch of a sequencer crossing exists in the embodiment. In this embodiment, the track detection unit 131 outputs a track detection image in which multiple tracks are detected from the captured image shown in Fig. 19. As described above, the switch detection unit 132 determines the travel lane of the vehicle from among the multiple tracks. Then, the switch detection unit 132 determines the state of the switch with the travel lane as the center. The determined travel lane in Fig. 19 is indicated by an arrow 2005. Then, the switch detection unit 132 detects the point of the switch ahead of the travel lane. The detection method is as described above.

[0079] 19, two pairs of triangular areas 2001a, 2001b are detected between a pair of rails ahead of the travel lane.The turnout detection unit then sets turnout areas 403a, 403b for each of the pairs.

[0080] An example of the determination results of the branch regions 403a and 403b is as follows. The result of the state determination of the switch in the branching area 403a corresponding to the triangular area 2001a is a junction from the left, and the result of the state determination of the switch in the branching area 403b corresponding to the triangular area 2001b is a branch, enter to the right.

[0081] Then, the branch detection unit 132 determines the next driving lane based on the next determination result. From the result of judging the state of the turnout in the branch area 403b corresponding to the above triangular area 2001b (branch, right entry), the lane next to the lane indicated by the arrow 2005 is the part indicated by the arrow 2006. A pair of triangular areas 2001c is detected from the lane indicated by the arrow 2006, and an example of the branch area judgment result is as follows. The state determination result of the branching area 403c corresponding to the triangular area 2001c is that a junction is coming from the right.

[0082] Then, the switch detection unit 132 cannot detect a switch point further ahead in the travel lane, and the switch detection process ends. In this way, it is possible to accurately detect switches and determine their open / closed states even for captured images of multiple tracks with complex branching structures.

[0083] Here, the setting of the range of the branch area by the branch detector 132 is performed as follows. FIG. 20 is a diagram for explaining an example of setting the range of a branching region in the case of a complex branching structure in the embodiment.

[0084] As described above, the turnout detection unit 132 determines the width of the turnout area based on the distance between the rails (also called the track width). For example, in the example of FIG. 20, when the turnout detection unit 132 detects four triangular areas 2403a, 2403b, 2043c, and 2403d from the four rails ABCD, it selects a pair of triangular areas based on the distance between the rails. When the predetermined track width is the width between AB, two pairs of AB and CD can be detected from the four triangular areas. That is, the pair of triangular area 2403a and triangular area 2403b, and the pair of triangular area 2403c and triangular area 2403d are two pairs. In this way, by determining the width of the turnout area using the track width, it is possible to eliminate incorrect combinations such as AC and BD.

[0085] As an example, the switch detection unit 132 can determine the width of the switch area based on a track width of 1.5. The switch detection unit 132 can also set the width of the switch area to 1.2 or 2 times the track width, taking into account calculation costs and accuracy. The vertical width of the branch region is determined as described above.

[0086] (Overview) In recent years, there has been a demand for autonomous driving of rail vehicles RVs. To realize autonomous driving of rail vehicles RVs, a more accurate and detailed map DB (for example, a database including branching and merging positions) is required. In addition, frequent maintenance of the existing map DB is required.

[0087] Here, a railway facility has multiple tracks, and switches are installed along the tracks to branch the tracks into multiple tracks or to merge the multiple tracks. When a railway vehicle moves to another track, it is necessary to switch the opening and closing of a switch, which is the part at the switch that actually branches the tracks. In order to determine whether the opening and closing direction of such a switch matches the traveling direction of the railway vehicle, a technology is required that processes images taken by an on-board camera mounted on the railway vehicle and accurately determines the state of the opening and closing direction of the switch.

[0088] The technology in Patent Document 1 proposes a technique for detecting a switch from a captured image and judging whether or not it is possible to enter the switch. This conventional technology proposes a method for detecting candidate points on the track using the brightness difference with the inside of the track as a feature, judging the continuity of the track and tongue rails with thin ends used at switches and expansion joints by brightness values, recognizing the open state of the switch, and judging whether or not it is possible to enter the switch when it is open.

[0089] In addition, the technology of Patent Document 2 relates to a method for detecting railway rails using an image, and discloses dividing the image into a nearby area, which is a close area on the image, and a far area, which is a far area on the image, and searching for an extension line by performing template matching through multiple coordinates for each area.

[0090] The technology of Patent Document 3 detects switch gates by performing template matching through advance learning of image feature amounts.

[0091] However, in the running environment of the railway vehicle RV, the illuminance fluctuates greatly depending on the weather, season, and time of day, and the appearance of the rails changes greatly. In addition, the shooting angle changes greatly due to the swaying and tilting of the railway vehicle RV while it is running, making it difficult to always stably detect the rails from the captured images. Therefore, it is even more difficult to detect the turnout from the rails and automatically determine the open / closed state of the turnout switch.

[0092] For example, in the technology of Patent Document 1, in the running environment of a railway vehicle, the illuminance fluctuates greatly due to the environment such as weather and season and the time of day, so the luminance value of the rail does not necessarily fall within a certain range. Also, depending on the environment in which the railway vehicle runs, there may be sleepers, gravel, concrete, snow, etc. on the track, making it even more difficult to recognize the open / closed state of the switch of the turnout based on the difference in luminance value of the track between the left and right rails.

[0093] In addition, the techniques of Patent Documents 2 and 3 detect tracks based on template matching, so it is possible to detect a simple single track, but when multiple tracks intersect, it is difficult to accurately detect the multiple tracks. In addition, because the traveling direction of the vehicle, the shooting angle, the shape of the branching part, etc. are different, there is a problem that the method using template matching as in Patent Documents 2 and 3 has difficulty in detecting a part of a track or a shape pattern of a part of a track.

[0094] In order to solve such problems, in this embodiment, the track detection device 100 is equipped with: a track detection unit 131 that detects tracks from an image captured by a camera 10 that captures an image of the surrounding area, such as the front of a railway vehicle RV, and outputs a track detection image, using a track detection model 1311, which is a trained model that inputs an image and outputs a track detection image that is the detection result of a track consisting of a pair of rails; a switch detection unit 132 that applies image processing to the track detection image to detect a pair of triangular areas, where the shape between the rails is triangular, from the track detection image, and when a pair of triangular areas is detected, outputs a specific image in which each of the vertices of the pair of triangular areas is identified as a switch point for branching the track into multiple tracks or merging multiple tracks; a switch state determination unit 133 that determines the open / closed state of the switch from the specific image, using a switch state determination model 1331, which is a trained model that inputs an image including the switch point, and outputs the open / closed state of the switch; and a display processing unit 18 that displays the determination result of the open / closed state of the switch on the display device 30.

[0095] Thus, according to this embodiment, a switch is detected and its open / closed state is determined by using a combination of a trained model and image processing. Therefore, according to this embodiment, it is possible to detect a track consisting of a pair of rails relatively stably from an image captured by the camera 10 that captures an image of the periphery, such as the front of the rail vehicle RV, regardless of the running environment of the rail vehicle RV. Furthermore, in this embodiment, it is possible to detect the location of the switch point regardless of the luminance value or texture of the gauge, and it is possible to determine the open / closed state of the switch switch from the switch area indicating the switch point. Therefore, according to this embodiment, it is possible to more accurately and easily determine the open / closed state of the switch present on the track from the captured image.

[0096] In this embodiment, the switch detection unit 132 of the track detection device 100 performs an expansion process on the track detection image to expand the width of each rail in the track detection image, and detects a pair of triangular areas from the image after the expansion process. Therefore, according to this embodiment, the pair of triangular areas can be detected more accurately, and the open / closed state of the switch on the track can be determined more accurately and easily from the captured image.

[0097] In this embodiment, the track detection unit 131 of the track detection device 100 detects multiple tracks from a captured image using a track detection model 1311 and outputs a track detection image, the switch detection unit 132 identifies a track on which the rail vehicle RV runs from the track detection image in which multiple tracks are detected, detects one or more pairs of triangular areas in front of the identified running track, and outputs the specific image in which the vertices of the triangles of the one or more pairs of triangular areas are identified as points of one or more switch, and the switch state determination unit 133 determines the open / closed state of each of one or more switch from the specific image in which the points of one or more switch are identified using the switch state determination model 1331. Therefore, according to this embodiment, when multiple tracks are simultaneously included in the captured image by the camera 10, even if multiple tracks cross, i.e., a branching section or a merging section exists, the open / closed state of the switch present on the track can be more accurately and easily determined from the captured image.

[0098] In this embodiment, the switch detection unit 132 of the track detection device 100 identifies the apex of the triangular area as the end of the switch point, and outputs a specific image in which a predetermined range including the apex is set as a switch area indicating the range of the point. Therefore, according to this embodiment, the switch point can be more accurately identified, and as a result, the open / closed state of the switch present on the track can be more accurately and easily determined from the captured image.

[0099] In this embodiment, the switch detection unit 132 of the track detection device 100 sets the width, which is the length in the direction intersecting the traveling direction of the rail vehicle RV in the switch area, to be longer than the interval between the pair of rails. Therefore, according to this embodiment, the point of the switch can be specified more accurately, and as a result, the open / closed state of the switch present on the track can be determined more accurately and easily from the captured image.

[0100] In this embodiment, the switch detection unit 132 of the track detection device 100 sets the vertical width, which is the length in the traveling direction of the railroad car in the switch area, based on the horizontal width. Therefore, according to this embodiment, the point of the switch can be more accurately identified from the switch area whose vertical width is based on the horizontal width, and as a result, the open / closed state of the switch existing on the track can be more accurately and easily determined from the captured image.

[0101] Furthermore, in this embodiment, the switch detection unit 132 of the track detection device 100 sets the vertical width, which is the length in the traveling direction of the railcar in the switch area, based on the horizontal width and parameters related to the camera 10. Therefore, according to this embodiment, the point of the switch can be more accurately identified from the switch area set within a range based on the parameters related to the camera 10, and as a result, the open / closed state of the switch present on the track can be more accurately and easily determined from the captured image.

[0102] Furthermore, in this embodiment, the switch detection unit 132 of the track detection device 100 sets the length in the traveling direction of the railroad vehicle RV in the switch area to be longer as the distance from the current position of the railroad vehicle RV to the switch point is longer. For this reason, in this embodiment, the vertical width of the switch area is determined taking into consideration that if the distance from the current position of the railroad vehicle RV to the switch point is long, the switch point will appear longer in the captured image, so that the open / closed state of the switch present on the track can be determined more accurately and easily from the captured image.

[0103] Furthermore, in this embodiment, the turnout detection unit 132 of the track detection device 100 sets the length of the turnout area in the traveling direction of the railway vehicle RV to be longer as the angle of the apex of the detected triangular area becomes smaller. Therefore, in this embodiment, the vertical width of the turnout area is determined taking into consideration that if the angle of the apex of the triangular area is small, the turnout point will appear longer in the captured image, so that the open / closed state of the turnout present on the track can be determined more accurately and easily from the captured image.

[0104] In this embodiment, the track detection unit 131 of the track detection device 100 outputs a track detection image in which a captured image is classified into pixels of a pair of rails, pixels of the gauge between the pair of rails, and pixels of an area other than the pair of rails and the gauge, using a track detection model 1311. Therefore, according to this embodiment, track detection can be performed more accurately, and as a result, the open / closed state of a switch present on the track can be determined more accurately and easily from the captured image.

[0105] In this embodiment, the switch state determination unit 133 of the track detection device 100 determines whether the open / closed state of the switch is at least one of left switch, right switch, left merging, right merging, and straight ahead, using the switch state determination model 1331. Therefore, according to this embodiment, the open / closed state of the switch present on the track can be specifically, more accurately, and easily determined from the captured image.

[0106] In the present embodiment, the track detection device 100 further includes a storage unit 17 that stores a map DB 171 that associates position information of a switch on the track with the open / closed state of the switch for each time period, and a comparison unit 15 that compares the position information of the switch detected by the switch detection unit 132 and the open / closed state of the switch determined by the switch state determination unit 133 with the position information of the switch in the map DB 171 and the open / closed state of the switch for a time period including the current time, and the display processing unit 18 further displays the comparison result by the comparison unit 15 on the display device 30. Therefore, according to the present embodiment, it is possible to always check whether the latest state of the track and the switch matches the registered information in the map DB 171 while the railway vehicle RV is traveling.

[0107] Furthermore, in this embodiment, the track detection device 100 is equipped with a communication unit 19 that transmits an instruction to correct the map DB 171 to the central device 200 if the comparison results are not consistent. Therefore, while the railway vehicle RV is traveling, when it is determined that the latest state of the tracks and switches does not match the map DB 171, the map DB can be updated on the central device 200 side, making it possible to update the map DB quickly.

[0108] In this embodiment, the communication unit 19 of the track detection device 100 further receives information on the traveling direction of the railway from the central device 200, and when the comparison results do not match, the comparison unit 15 further judges whether or not the open / closed state of the switch determined by the switch state determination unit 133 matches the traveling direction of the railway vehicle RV received by the communication unit 19, and the notification unit 20 outputs an alarm when the open / closed state of the switch determined by the switch state determination unit 133 differs from the traveling direction of the railway vehicle RV received by the communication unit 19. Therefore, according to this embodiment, it is possible to notify in advance that the current open / closed state of the switch differs from the traveling direction of the railway vehicle RV.

[0109] The track detection device 100 of this embodiment is equipped with a control device such as a CPU, a storage device such as a ROM and a RAM, an external storage device such as a HDD and a CD drive, a display device such as a display device, and input devices such as a keyboard and a mouse, and has a hardware configuration that utilizes a normal computer.

[0110] A track detection program executed by the track detection device 100 of this embodiment is provided in a state that it is pre-installed in a ROM or the like.

[0111] The track detection program executed by the track detection device 100 of this embodiment may be configured to be provided by being recorded on a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, or a digital versatile disk (DVD) in an installable or executable format.

[0112] Furthermore, the track detection program executed by the track detection device 100 of this embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the track detection program executed by the track detection device 100 of this embodiment may be provided or distributed via a network such as the Internet.

[0113] The track detection program executed by the track detection device 100 of this embodiment has a modular structure including each of the functional units mentioned above (track detection unit 131, switch detection unit 132, switch status determination unit 133, positioning unit 16, track information generation unit 14, matching unit 15, display processing unit 18, notification unit 20, and communication unit 19), and in terms of actual hardware, a CPU (processor) reads out the obstacle detection program from the ROM and executes it, thereby loading each of the above units onto a main memory device, and each of the functional units (track detection unit 131, switch detection unit 132, switch status determination unit 133, positioning unit 16, track information generation unit 14, matching unit 15, display processing unit 18, notification unit 20, and communication unit 19) is generated on the main memory device.

[0114] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0115] 1 Vehicle Systems 10. Camera 13 Judgment section 14 Track information generation section 15. Matching section 16 Positioning unit 17 Memory section 18 Display processing section 19 Communications Department 20. Information Department 30 Display device 50 Antenna 100 Railroad track detector 171 Map DB 131 Railroad track detector 132 Switch detector 133 Switch status determination unit 401a, 401b, 2001a, 2001b, 2001c 403, 403a, 403b, 403c Branching Area 1311 Track detection model 1331 Switch state judgment model 200 Central unit

Claims

1. A track detection device mounted on a railway vehicle, a track detection unit that detects the tracks from an image captured by an imaging device that captures an image of the area around the railcar and outputs the track detection image using a track detection model that is a trained model that inputs an image and outputs a track detection image that is a detection result of a track consisting of a pair of rails; a switch detection unit that performs image processing on the track detection image to detect a pair of triangular areas from the track detection image, where the shape between the rails is a triangle, and when the pair of triangular areas are detected, outputs a specific image in which each vertex of the pair of triangular areas is identified as a switch point for branching the track into a plurality of tracks or merging the plurality of tracks; a switch state determination unit that determines an open / closed state of the switch from the specific image by using a switch state determination model that is a trained model that inputs an image including a point of the switch and outputs an open / closed state of the switch; an output unit that outputs a determination result of an open / closed state of the turnout; A track detection device comprising:

2. the switch detection unit performs an expansion process on the track detection image to expand the widths of the rails in the track detection image, and detects the pair of triangular regions from the image on which the expansion process has been performed. The track detection device according to claim 1 .

3. the track detection unit detects a plurality of tracks from the captured image using the track detection model, and outputs the track detection image; the switch detection unit identifies a track on which the railcar runs from the track detection image in which the multiple tracks are detected, detects one or more pairs of triangular areas in front of the identified track on which the railcar runs, and outputs the identified image in which vertices of the triangles in the one or more pairs of triangular areas are identified as points of one or more switch. the switch state determination unit determines an open / closed state of each of the one or more switch gates from a specific image in which points of the one or more switch gates are specified, using the switch state determination model; The track detection device according to claim 1 .

4. the branch detection unit identifies a vertex of the triangular region as an end of the point of the branch, and outputs the specified image in which a predetermined range including the vertex is set as a branch region indicating the range of the point. The track detection device according to claim 1 .

5. the turnout detection unit sets a width, which is a length in a direction intersecting a traveling direction of the railway vehicle in the turnout area, to be longer than a distance between the pair of rails; 5. The track detection device according to claim 4.

6. The switch detection unit sets a vertical width, which is a length in a traveling direction of the railway vehicle in the switch area, based on the horizontal width. The track detection device according to claim 5.

7. the switch detection unit sets a vertical width, which is a length in a traveling direction of the railway vehicle in the switch area, based on the horizontal width and a parameter related to the imaging device; The track detection device according to claim 5.

8. the switch detection unit sets a length of the switch area in the traveling direction of the railway vehicle to be longer as the distance from the current position of the railway vehicle to the switch point is longer; The track detection device according to claim 5.

9. the turnout detection unit sets a length of the turnout area in the traveling direction of the railway vehicle longer as the angle of a vertex of the detected triangular area becomes smaller; The track detection device according to claim 5.

10. the track detection unit uses the track detection model to output the track detection image in which the captured image is classified into pixels of the pair of rails, pixels of a gauge between the pair of rails, and pixels of an area other than the pair of rails and the gauge. The track detection device according to claim 1 .

11. The branching state determination unit determines, by using the branching state determination model, that an open / closed state of the branching is at least one of left branching, right branching, left merging, right merging, and straight ahead. The track detection device according to claim 1 .

12. A storage unit that stores location information of the turnout on the railroad tracks, an open / closed state of the turnout for each time period, and map information associated with each other; a collation unit that compares the position information of the switch detected by the switch detection unit and the open / closed state of the switch determined by the switch state determination unit with the position information of the switch in the map information and the open / closed state of the switch for a time period including the current time; The output unit further displays a collation result by the collation unit.

2. The track detection device according to claim 1 .

13. a communication unit that transmits an instruction to correct the map information to a central device connected to a network when the comparison result does not match; The track detection device according to claim 12 , further comprising:

14. The communication unit further receives information regarding a traveling direction of the railcar from the central device, When the comparison result indicates a mismatch, the comparison unit further determines whether or not the open / closed state of the switch determined by the switch state determination unit matches the traveling direction of the railway vehicle received by the communication unit, The output unit further outputs an alarm when the open / closed state of the switch determined by the switch state determination unit differs from the traveling direction of the railway vehicle received by the communication unit. The track detection device according to claim 13.

15. A track detection method executed by a track detection device mounted on a railway vehicle, comprising: a step of detecting the tracks from an image captured by an imaging device that captures an image of the area around the railcar, and outputting the track detection image using a track detection model that is a trained model that inputs an image and outputs a track detection image that is a detection result of a track consisting of a pair of rails; a step of detecting a pair of triangular areas from the track detection image, in which the shape between the rails is triangular, by performing image processing on the track detection image, and when the pair of triangular areas are detected, outputting a specific image in which each vertex of the pair of triangular areas is identified as a switch point for branching the track into a plurality of tracks or merging the plurality of tracks; determining an open / closed state of the switch from the specific image by using a switch state determination model, which is a trained model that inputs an image including a point of the switch and outputs an open / closed state of the switch; outputting a result of the determination of the open / closed state of the turnout; A track detection method comprising:

Images