Vehicle side light illumination status detection system and vehicle side light illumination status detection device

Abstract

To provide a vehicle side lamp lighting state detection system capable of accurately detecting a lighting state of a vehicle side lamp by suppressing the reduction in determination accuracy of whether or not the vehicle side lamp is turned on even when a difference between the saturation of a vehicle body color and the saturation of the vehicle side lamp of a railway vehicle is small.SOLUTION: A vehicle side lamp lighting state detection system 100 includes an image pickup unit 10 for photographing a vehicle image 30 in which a vehicle side lamp 3 provided on a side surface 2e of a railway vehicle 2 is included, an image processing unit 11 for acquiring lighting information 20 indicating whether or not the vehicle side lamp 3 included in the vehicle image 30 is turned on by using a learned model 4 which is leaned by machine learning using the image in which the vehicle side lamp 3 is included, and a control unit 12 for detecting the lighting state of the vehicle side lamp 3 based on the lighting information 20.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a side lamp lighting state detection system and a side lamp lighting state detection device that detect the lighting state of a side lamp by determining whether the side lamp provided on a railway vehicle is lit or not.

Background Art

[0002] Conventionally, a side lamp lighting state detection system that detects the lighting state of a side lamp by determining whether the side lamp provided on a railway vehicle is lit or not is known (see, for example, Patent Document 1).

[0003] Patent Document 1 discloses a side lamp lighting detection system including a photographing device that photographs a side lamp and a determination device that determines whether the side lamp is lit or not using a photographed image obtained by the photographing device. In the determination device disclosed in Patent Document 1, in order to suppress a decrease in determination accuracy due to external light, it is determined whether the side lamp is lit or not based on the saturation of the side lamp shown in the photographed image.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the side lamp lighting detection system (side lamp lighting state detection system) described in Patent Document 1, since it is determined whether the side lamp is lit or not based on the saturation of the side lamp shown in the photographed image (vehicle image), for example, when the railway vehicle has a vehicle body color with a small difference from the saturation of the side lamp, there is a disadvantage that the determination accuracy of whether the side lamp is lit or not decreases. In this case, there is a problem that the detection accuracy of the lighting state of the side lamp decreases.

[0006] This invention was made to solve the above-mentioned problems, and one of its objectives is to provide a vehicle side lamp illumination status detection system that can accurately detect the illumination status of vehicle side lamps by suppressing a decrease in the accuracy of determining whether or not the vehicle side lamps are illuminated, even when the difference between the saturation of the vehicle body color and the saturation of the vehicle side lamps is small. [Means for solving the problem]

[0007] To achieve the above objective, the vehicle side lamp illumination status detection system according to the first aspect of this invention includes: an image capturing unit that captures a vehicle image showing the vehicle side lamps installed on the side of a railway vehicle; an image processing unit that acquires illumination information, which is information determining whether or not the vehicle side lamps shown in the vehicle image are illuminated, using a trained model learned by machine learning using the image showing the vehicle side lamps; and a control unit that detects the illumination status of the vehicle side lamps based on the illumination information. The system includes a storage unit that stores information on the number of side lights set in advance, and a control unit that acquires the number of side lights visible in the vehicle image and outputs error information if the acquired number of side lights differs from the set number of side lights. .

[0008] The vehicle side light illumination status detection system according to the first phase described above comprises an image processing unit that acquires illumination information, which is information determining whether or not the vehicle side lights in the vehicle image are illuminated, using a trained model trained by machine learning using an image of the vehicle side lights, and a control unit that determines the illumination status of the vehicle side lights based on the illumination information. Here, it is considered that the trained model trained by machine learning can acquire the boundary between the railway vehicle and the vehicle side lights even when the difference between the saturation of the railway vehicle's body color and the saturation of the vehicle side lights is small. Therefore, by configuring it as described above, the image processing unit acquires illumination information, which is information determining whether or not the vehicle side lights are illuminated, using the trained model, so that it can accurately determine whether or not the vehicle side lights are illuminated even when the difference between the saturation of the railway vehicle's body color and the saturation of the vehicle side lights is small. As a result, even when the difference between the saturation of the railway vehicle's body color and the saturation of the vehicle side lights is small, it is possible to suppress a decrease in the accuracy of determining whether or not the vehicle side lights are illuminated, thereby providing a vehicle side light illumination status detection system that can accurately detect the illumination status of the vehicle side lights.

[0009] In the vehicle side light illumination state detection system according to the first aspect described above, preferably, the image processing unit is configured to acquire learning estimated region information, which is information about the area in the vehicle image in which the vehicle side lights are visible, estimated by a trained model, along with illumination information, and the control unit is configured to detect the illumination state of the vehicle side lights based on the illumination information and the learning estimated region information. With this configuration, the control unit detects the illumination state of the vehicle side lights using the learning estimated region information along with the illumination information, so the position of the vehicle side lights visible in the vehicle image can be obtained using the learning estimated region information. As a result, for example, even if multiple vehicle side lights are visible in a single vehicle image, the illumination state of each individual vehicle side light can be distinguished and detected based on the illumination information and the learning estimated region information.

[0010] In this case, preferably, The memory unit is, The system stores pre-set region information for each railway vehicle, which is the region of the side lights that is larger than the region of the learned estimation region information. death The control unit determines the learning estimation region based on the learning estimation region information acquired in the image processing unit and the setting region information stored in the memory unit. Vehicle images may show side lights. It determines whether or not it is included in the setting area, and the learning estimation area 、 The system is configured to output lighting information if the learning estimation area is included in the setting area, and to output error information if the learning estimation area is not included in the setting area. With this configuration, if the control unit determines that the learning estimation area is included in the setting area, lighting information is output. For example, by outputting lighting information to platform doors, the lighting information of the vehicle's side lights can be used for opening and closing platform doors. Also, if the control unit determines that the learning estimation area is not included in the setting area, error information is output. Therefore, it is possible to prevent platform doors from opening when a railway vehicle other than the pre-set railway vehicle is stopped there.

[0011] In a configuration in which the control unit determines whether or not a learning estimation region is included in a set region based on the learning estimation region information acquired by the image processing unit and the set region information stored in the memory unit, it is preferable that the memory unit is further configured to store set side light count information, which is information on the number of side lights set in advance, and the control unit is configured to acquire the number of side lights that appear in the vehicle image and to output error information if the acquired number of side lights differs from the set side light count information. With this configuration, the control unit can easily determine whether or not a railway vehicle stopped at the platform is a pre-set railway vehicle by comparing the number of side lights acquired from the learning estimation region information with the set side light count information.

[0012] In the vehicle side light illumination state detection system according to the first aspect described above, preferably, the image acquisition unit is configured to capture images of multiple vehicle side lights, and the trained model is configured to output illumination information for each vehicle side light when multiple vehicle side lights are captured in a single image. With this configuration, when multiple vehicle side lights are captured in a single image, illumination information is output for each vehicle side light, so even if multiple vehicle side lights are captured by a single image acquisition unit, illumination information for each individual vehicle side light can be obtained. As a result, for example, there is no need to arrange an image acquisition unit for each vehicle side light, as in the case where one vehicle side light is captured by one image acquisition unit, so it is possible to suppress an increase in the number of parts and suppress an increase in the complexity of the system configuration.

[0013] In this case, preferably, the control unit is After the vehicle image is acquired by the image acquisition unit, and before the lighting information is acquired by the image processing unit, In the case of multiple vehicle images where the side lights overlap, the system is configured to exclude the area containing the overlapping side lights from the determination of whether or not the side lights are illuminated. This configuration prevents the system from determining whether or not a side light that appears overlapping in multiple vehicle images is illuminated. As a result, even if a single side light appears in multiple vehicle images, the illumination status of the side light can be accurately detected.

[0014] In a configuration in which the control unit excludes areas where overlapping side lights are visible from the determination target for determining whether or not the side lights are illuminated, preferably the control unit is configured to exclude such areas from the determination target for determining whether or not the side lights are illuminated. After the vehicle image is acquired by the image acquisition unit, and before the lighting information is acquired by the image processing unit, The system is configured to exclude areas where overlapping side lights are visible from the determination of whether or not the side lights are on, by performing a masking process on those areas. With this configuration, the control unit performs a masking process on areas where overlapping side lights are visible, making it easy to exclude those areas from the determination of whether or not the side lights are on.

[0015] In the vehicle side light illumination state detection system according to the first phase described above, preferably, the system further includes a stopping position information acquisition unit that acquires stopping position information, which is information about the stopping position of the railway vehicle, and the control unit is configured to detect the illumination state of the vehicle side lights based on the stopping position information and illumination information acquired by a trained model. With this configuration, even if there is a discrepancy in the stopping position of the railway vehicle, the setting area included in the setting area information can be corrected based on the stopping position information. As a result, even if there is a discrepancy in the stopping position of the railway vehicle, a decrease in the detection accuracy of the illumination state of the vehicle side lights can be suppressed.

[0016] In the vehicle side light illumination status detection system according to the first phase described above, preferably, the trained model is a deep neural network that has been trained to determine whether or not the vehicle side lights are illuminated from a vehicle image. With this configuration, by using a deep neural network, which is a trained model suitable for object detection, the vehicle side lights visible in the vehicle image can be detected with high accuracy.

[0017] The vehicle side lamp illumination status detection device according to the second aspect of this invention includes an image acquisition unit that acquires a vehicle image showing the vehicle side lamps installed on the side of a railway vehicle, an image processing unit that acquires illumination information, which is information determining whether or not the vehicle side lamps shown in the vehicle image are illuminated, using a trained model learned by machine learning using the image showing the vehicle side lamps, and a control unit that detects the illumination status of the vehicle side lamps based on the illumination information. The system includes a storage unit that stores information on the number of side lights set in advance, and a control unit that acquires the number of side lights visible in the vehicle image and outputs error information if the acquired number of side lights differs from the set number of side lights..

[0018] In the vehicle-side lamp lighting state detection device according to the second aspect, as described above, an image processing unit that acquires lighting information, which is information for determining whether or not a vehicle-side lamp appearing in a vehicle image is lit, by using a learned model learned by machine learning using an image in which the vehicle-side lamp appears, and a control unit that detects the lighting state of the vehicle-side lamp based on the lighting information are provided. Thereby, similar to the vehicle-side lamp lighting state detection system according to the first aspect, even when the difference between the saturation of the vehicle body color of the railway vehicle and the saturation of the vehicle-side lamp is small, it is possible to provide a vehicle-side lamp lighting state detection device that can accurately detect the lighting state of the vehicle-side lamp by suppressing a decrease in the determination accuracy of whether or not the vehicle-side lamp is lit.

Advantages of the Invention

[0019] According to the present invention, as described above, even when the difference between the saturation of the vehicle body color of the railway vehicle and the saturation of the vehicle-side lamp is small, it is possible to provide a vehicle-side lamp lighting state detection system that can accurately detect the lighting state of the vehicle-side lamp by suppressing a decrease in the determination accuracy of whether or not the vehicle-side lamp is lit.

Brief Description of the Drawings

[0020] [Figure 1] It is a schematic diagram showing the overall configuration of the vehicle-side lamp lighting state detection system according to the first embodiment. [Figure 2] It is a schematic diagram for explaining a vehicle image. [Figure 3] It is a schematic diagram for explaining the masking process performed by the control unit according to the first embodiment on a vehicle image. [Figure 4] It is a schematic diagram for explaining a configuration for generating a learned model. [Figure 5] It is a schematic diagram for explaining a configuration in which the image processing unit according to the first embodiment acquires lighting information and learned estimation region information. [Figure 6] It is a schematic diagram for explaining a configuration for setting a set region for a vehicle image. [Figure 7]This is a block diagram illustrating the configuration in which the control unit according to the first embodiment detects the illumination status of the vehicle's side lights. [Figure 8] This is a schematic diagram illustrating the configuration in which the control unit according to the first embodiment detects the illumination status of the vehicle's side lights. [Figure 9] This is a flowchart illustrating the process by which the vehicle side light illumination status detection device according to the first embodiment generates a learned model. [Figure 10] This is a flowchart illustrating the process by which the vehicle side light illumination status detection device according to the first embodiment outputs illumination information. [Figure 11] This is a schematic diagram showing the overall configuration of the vehicle side lamp illumination status detection system according to the second embodiment. [Figure 12] This is a schematic diagram illustrating a configuration in which a control unit according to the second embodiment corrects the set area based on stop position information. [Figure 13] This is a flowchart illustrating the process by which the vehicle side light illumination status detection device according to the second embodiment outputs illumination information. [Modes for carrying out the invention]

[0021] The following describes embodiments of the present invention based on the drawings.

[0022] [First Embodiment] (Overall configuration of the vehicle side light illumination status detection system) Referring to Figures 1 to 8, the configuration of a vehicle side lamp illumination status detection system 100 equipped with a vehicle side lamp illumination status detection device 1 according to the first embodiment of the present invention will be described. The vehicle side lamp illumination status detection system 100 is a system that detects the illumination status of a vehicle side lamp 3 provided on a railway vehicle 2. The vehicle side lamp 3 is a side lamp provided on the side 2e of the railway vehicle 2 to indicate the open / closed state of the door 2f (see Figure 2) of the railway vehicle 2. In other words, the vehicle side lamp 3 is a door-closed vehicle side lamp.

[0023] Furthermore, in the example shown in Figure 1, the railway vehicle 2 is a four-car train consisting of cars 1 through 4 (cars 2a through 2d). Each car of the railway vehicle 2 is equipped with one side light 3. That is, cars 1 through 4 (cars 2a through 2d) are each equipped with side lights 1 through 4 (cars 3a through 3d). Note that the number of cars in the railway vehicle 2 can be any number.

[0024] The vehicle side light illumination status detection device 1 detects the illumination status of multiple vehicle side lights 3 installed on the railway vehicle 2. The vehicle side light illumination status detection device 1 is also configured to output illumination information 20, which is information determining whether or not the vehicle side lights 3 are illuminated, to a higher-level device 101. The higher-level device 101 includes, for example, a control device for platform doors installed on a platform.

[0025] As shown in Figure 1, the vehicle side light illumination status detection system 100 comprises a vehicle side light illumination status detection device 1 and an image acquisition unit 10.

[0026] The vehicle side light illumination status detection device 1 comprises an image processing unit 11, a control unit 12, a storage unit 13, and an image acquisition unit 14.

[0027] The image capturing unit 10 is configured to capture a vehicle image 30 that shows the side lights 3 provided on the side 2e of the railway vehicle 2. The image capturing unit 10 is provided on a platform and is configured to capture images of the stationary railway vehicle 2. In this embodiment, the side light illumination status detection system 100 comprises a plurality of image capturing units 10. In the example shown in Figure 1, the side light illumination status detection system 100 comprises a first image capturing unit 10a and a second image capturing unit 10b. The image capturing unit 10 (first image capturing unit 10a and second image capturing unit 10b) is, for example, a camera capable of capturing color images.

[0028] In this embodiment, the image capturing unit 10 is configured to capture images of multiple vehicle side lights 3. In the example shown in Figure 1, each of the first image capturing unit 10a and the second image capturing unit 10b is configured to capture images of multiple vehicle side lights 3. Specifically, the first image capturing unit 10a is configured to capture images of the first vehicle side lights 3a to the third vehicle side lights 3c provided on the first vehicle 2a to the third vehicle 2c, as shown by the dashed lines 50 and 51.

[0029] Furthermore, the second image capturing unit 10b is configured to capture images of the second side lights 3b to the fourth side lights 3d installed on the second vehicle 2b to the fourth vehicle 2d, as shown by the dashed lines 52 and 53. The dashed lines 50 to 53 are for convenience to indicate the imaging range of each image capturing unit 10.

[0030] The image processing unit 11 is configured to acquire illumination information 20, which is information determining whether or not the vehicle side lights 3 in the vehicle image 30 are lit, using a trained model 4 that has been trained by machine learning using images showing the vehicle side lights 3. The image processing unit 11 includes, for example, a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory). Details of how the image processing unit 11 acquires illumination information 20 using the trained model 4 will be described later.

[0031] The control unit 12 is configured to detect the illumination status of the vehicle side lights 3 based on the illumination information 20. The control unit 12 includes, for example, a CPU, ROM, and RAM. Details of how the control unit 12 detects the illumination status of the vehicle side lights 3 will be described later.

[0032] The memory unit 13 is configured to store the learned model 4. The memory unit 13 is also configured to store the setting area information 22 and the set number of side lights information 23, which will be described later. In this embodiment, the memory unit 13 stores multiple sets of setting area information 22 and multiple sets of side lights information 23. Specifically, the memory unit 13 stores the setting area information 22 and the set number of side lights information 23 in association (linked) with each railway vehicle 2 that is stopped at the platform. The memory unit 13 is a non-volatile storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0033] The image acquisition unit 14 is configured to acquire a vehicle image 30 that shows the side lights 3 provided on the side 2e of the railway vehicle 2. The image acquisition unit 14 is configured to output the acquired vehicle image 30 to the image processing unit 11. The image acquisition unit 14 is, for example, an input / output interface.

[0034] (Vehicle image) Figure 2 is a schematic diagram of the vehicle image 30 captured by the image acquisition unit 10 (see Figure 1).

[0035] Vehicle image 30 shows the railway vehicle 2 (see Figure 1). Vehicle image 30 also shows the side lights 3 (see Figure 1) installed on each vehicle. In the example shown in Figure 2, vehicle image 30 shows the first vehicle 2a to the third vehicle 2c, and the first side lights 3a to the third side lights 3c installed on the first vehicle 2a to the third vehicle 2c.

[0036] (Masking process) Here, the third side marker light 3c visible in the vehicle image 30 is captured by both the first image capturing unit 10a (see Figure 1) and the second image capturing unit 10b (see Figure 1). In this case, if the third side marker light 3c visible in both the vehicle image 30 captured by the first image capturing unit 10a and the vehicle image 30 captured by the second image capturing unit 10b is determined for each image, then a number of side marker lights 3 (see Figure 1) greater than the number of side marker lights 3 (see Figure 1) installed on the railway vehicle 2 (see Figure 1) will be determined.

[0037] Therefore, in this embodiment, as shown in Figure 3, the control unit 12 (see Figure 1) is configured to exclude the region 30a (see Figure 2) in which the overlapping side lights 3 (see Figure 1) are visible from the determination target for determining whether or not the side lights 3 are lit, for one of the multiple vehicle images 30 (see Figure 2) in which the side lights 3 (see Figure 1) are overlapping. Specifically, the control unit 12 is configured to exclude the region 30a in which the overlapping side lights 3 are visible from the determination target for determining whether or not the side lights 3 are lit by performing a masking process on the region 30a in which the overlapping side lights 3 are visible. The vehicle image 31 shown in Figure 3 is the vehicle image 30 after the masking process has been performed. For example, the control unit 12 is configured to black out the region 30a in which the overlapping side lights 3 are visible, as shown in region 31a, as part of the masking process. Furthermore, based on user input, the control unit 12 performs masking on the area 30a in which the smaller side lights 3 are visible among the multiple vehicle images 30 that overlap.

[0038] (Generating a pre-trained model) Next, with reference to Figure 4, we will describe the configuration for generating the trained model 4 (see Figure 1).

[0039] The trained model 4 is generated by training the training model 5 to output lighting information 20 and training estimated region information 21, using training images 32 and 33 as training data. The training estimated region information 21 is information about the region in which the side lights 3 are visible in multiple vehicle images 30.

[0040] Teacher image 32 and teacher image 33 are images in which a vehicle side light area 40, which is the area where the vehicle side light 3 is visible, is set, and a label is set indicating whether or not the vehicle side light 3 is lit. Teacher image 32 is an image in which the vehicle side light area 40 and a label indicating whether or not the vehicle side light 3 is lit are set on the vehicle image 30 without any masking process. Teacher image 33 is an image in which the vehicle image 31, which has undergone masking, has a label indicating whether or not the vehicle side light area 40 and a label indicating whether or not the vehicle side light 3 is lit are set.

[0041] As shown in Figure 4, the training image 32 is obtained by setting labels on the vehicle side light area 40 and the vehicle side light 3 for multiple vehicle images 30 captured by multiple image capturing units 10, indicating whether or not the vehicle side light area 40 and the vehicle side light 3 are lit. The label indicating whether or not they are lit is, for example, a numerical value. That is, if the vehicle side light 3 is lit, the label is set to "1". If the vehicle side light 3 is off, the label is set to "0". In the example shown in Figure 4, in order to distinguish whether or not the vehicle side light 3 is lit, the vehicle side light 3 and the vehicle side light area 40 are shown with solid lines in images where the vehicle side light 3 is lit, and the vehicle side light 3 and the vehicle side light area 40 are shown with dashed lines in images where the vehicle side light 3 is off. That is, training image 32a and training image 33a are images where the vehicle side light 3 is off. Also, training image 32b and training image 33b are images where the vehicle side light 3 is lit.

[0042] The trained model 4 (see Figure 1), generated by training the learning model 5, outputs lighting information 20 and trained estimated region information 21 when a vehicle image 30 is input. Specifically, the trained model 4 outputs a numerical value as lighting information 20 indicating whether the side lights 3 are lit or not. More specifically, the trained model 4 outputs "1" if the side lights 3 are lit and "0" if the side lights 3 are off.

[0043] Furthermore, the trained model 4 (see Figure 1) outputs information about the area of ​​the side lights 3 (side light area 40) visible in the vehicle image 30 as trained estimated area information 21. Specifically, the trained model 4 outputs the coordinate values ​​of the area of ​​the side lights 3 visible in the vehicle image 30. More specifically, the trained model 4 outputs the coordinate values ​​of two diagonally opposite points in the area of ​​the side lights 3. Note that if multiple side lights 3 are visible in a single image, the trained model 4 is configured to output lighting information 20 for each side light 3. Also, if multiple side lights 3 are visible in a single image, the trained model 4 is configured to output trained estimated area information 21 for each side light 3.

[0044] In other words, the training image 32 and training image 33 are the training input data, and the information of the area of ​​the side lights 3 (side light area 40) shown in the vehicle image 30, and the label indicating whether or not the side lights 3 are lit are the training output data.

[0045] In this embodiment, the trained model 4 (see Figure 1) is a deep neural network that has been trained to determine whether or not the side lights 3 are illuminated from multiple vehicle images 30.

[0046] (Acquisition of lighting information and learning estimation area information) Next, with reference to Figure 5, the configuration in which the image processing unit 11 acquires lighting information 20 and learning estimation area information 21 will be described. As shown in Figure 5, the image capturing unit 10 acquires a vehicle image 30 of the railway vehicle 2 (see Figure 1). The image capturing unit 10 outputs the acquired vehicle image 30 to the image acquisition unit 14.

[0047] The image acquisition unit 14 outputs the vehicle image 30 input from the image imaging unit 10 to the image processing unit 11.

[0048] The image processing unit 11 acquires lighting information 20 and learned estimation area information 21 based on the vehicle image 30 input from the image acquisition unit 14 and the trained model 4 read from the storage unit 13. In other words, the image processing unit 11 (see Figure 1) is configured to acquire learned estimation area information 21, which is information about the areas in which the side lights 3 are visible in multiple vehicle images 30, as estimated by the trained model 4, along with the lighting information 20. The image processing unit 11 outputs the acquired lighting information 20 and learned estimation area information 21 to the control unit 12.

[0049] The control unit 12 is configured to detect the illumination state of the vehicle side lamp 3 based on the illumination information 20 and the learning estimation area information 21.

[0050] Here, various types of railway vehicles 2 (see Figure 1) stop at the platform. Also, the train configuration differs depending on the type of railway vehicle 2. Therefore, if the control unit 12 were to output lighting information 20 to the higher-level device 101 (see Figure 1) based solely on the illumination status of the side lights 3 without determining the type of railway vehicle 2, there is a possibility that platform doors would be opened at a location where no railway vehicle 2 is stopped. For this reason, in this embodiment, the control unit 12 is configured to determine whether or not the railway vehicle 2 stopped at the platform is a vehicle that has been pre-set to stop there.

[0051] (Settings area) To determine whether a railway vehicle 2 (see Figure 1) stopped at a platform is a vehicle that has been pre-set to stop, a setting area 41 is set for each railway vehicle 2 in the vehicle image 30 (see Figure 1) and vehicle image 31 captured by the image capturing unit 10 (see Figure 1). The example shown in Figure 6 shows the setting area 41 set for the vehicle image 31. The setting area 41 is the area in the vehicle image 30 and vehicle image 31 in which the vehicle side lights 3 may be visible. The setting area 41 is set to a size that takes into account the deviation of the stopping position of the railway vehicle 2. Specifically, the setting area 41 is the area of ​​the vehicle side lights 3 that is larger than the area of ​​the learning estimation area information 21 (see Figure 5) (learning estimation area 42 (see Figure 8)) that has been pre-set for each railway vehicle 2 (see Figure 1). In this embodiment, the storage unit 13 (see Figure 5) is configured to store setting area information 22 (see Figure 5), which is information about the setting area 41.

[0052] Here, since the image capturing unit 10 (see Figure 1) is fixed to the platform, the number of side lights 3 visible in the vehicle image 31 captured by the image capturing unit 10 does not change if the type of railway vehicle 2 (see Figure 1) is the same. Therefore, the setting area 41 is set for each type of railway vehicle 2 for the vehicle image 30 (vehicle image 31) captured by the image capturing unit 10.

[0053] Furthermore, the setting area 41 is set for each vehicle side light 3 visible in the vehicle image 31. In the example shown in Figure 6, two vehicle side lights 3 (first vehicle side light 3a and second vehicle side light 3b) are visible in the vehicle image 31. Therefore, in the example shown in Figure 6, the setting area 41 includes a first setting area 41a set for the first vehicle side light 3a and a second setting area 41b set for the second vehicle side light 3b.

[0054] (Determination of whether or not a train car is scheduled to stop in advance) Next, referring to Figures 7 and 8, we will describe the configuration in which the control unit 12 determines whether or not the railway vehicle 2 (see Figure 1) is a vehicle that is pre-set to stop.

[0055] As shown in Figure 7, the control unit 12 determines whether the railway vehicle 2 (see Figure 1) is a vehicle that has been set to stop in advance, based on the learning estimation area information 21 acquired by the image processing unit 11 and the setting area information 22 stored in the storage unit 13.

[0056] The control unit 12 acquires learning estimation area information 21 from the image processing unit 11. The control unit 12 also acquires setting area information 22 from the storage unit 13. The control unit 12 also acquires lighting information 20 from the image processing unit 11.

[0057] The control unit 12 determines, based on the learning estimation area information 21 and the setting area information 22, whether or not the railway vehicle 2 (see Figure 1) is a vehicle that has been set to stop in advance.

[0058] Specifically, as shown in Figure 8, the control unit 12 (see Figure 7) determines whether the learning estimation region 42 is included in the setting region 41 based on the learning estimation region information 21 (see Figure 7) acquired by the image processing unit 11 (see Figure 7) and the setting region information 22 (see Figure 7) stored in the storage unit 13 (see Figure 7).

[0059] As shown in images 34a to 34c of Figure 8, the control unit 12 superimposes both the setting area 41 and the learning estimation area 42. Images 34a to 34c in Figure 8 are images in which the setting area 41 and the learning estimation area 42 are superimposed on vehicle image 30 (see Figure 2) and vehicle image 31 (see Figure 3). Images 34a to 34c are included for convenience to illustrate the configuration in which the control unit 12 determines whether or not the learning estimation area 42 is included in the setting area 41, and therefore do not show the railway vehicle 2 (see Figure 1) and the vehicle side light 3 (see Figure 1). Furthermore, the control unit 12 does not necessarily have to generate images 34a to 34c.

[0060] The control unit 12 determines whether all learning estimation areas 42 are included in the setting area 41. The control unit 12 acquires the setting area information 22 for each vehicle stored in the memory unit 13 (see Figure 7), and determines whether all learning estimation areas 42 are included in the setting area 41 included in the setting area information 22 for each vehicle. Furthermore, for the learning estimation areas 42 to be included in the setting area 41 means that the learning estimation areas 42 are contained within the setting area 41.

[0061] Image 34a shows the first setting region 41a, the second setting region 41b, the first learning estimation region 42a, and the second learning estimation region 42b. In Image 34a, the first learning estimation region 42a and the second learning estimation region 42b are contained within the first setting region 41a and the second setting region 41b, respectively.

[0062] Image 34b shows the third setting region 41c, the fourth setting region 41d, the third learning estimation region 42c, and the fourth learning estimation region 42d. In Image 34b, the third learning estimation region 42c and the fourth learning estimation region 42d are contained within the third setting region 41c and the fourth setting region 41d, respectively.

[0063] Image 34c shows the fifth setting region 41e and the fifth learning estimation region 42e. In image 34c, the fifth learning estimation region 42e is contained within the fifth setting region 41e. In other words, the example shown in Figure 8 indicates that the train car 2 (see Figure 1) stopped at the platform is a train car that has been pre-set to stop.

[0064] Referring again to Figure 7, in this embodiment, the control unit 12 is configured to output lighting information 20 to the host device 101 if the learning estimation area 42 (see Figure 8) is included in the setting area 41 (see Figure 8). The control unit 12 is also configured to output error information 24 to the host device 101 if the learning estimation area 42 is not included in the setting area 41.

[0065] Here, if the number of learning estimation regions 42 is less than the number of setting regions 41, all learning estimation regions 42 may be included in the setting regions 41. However, if the number of learning estimation regions 42 and the number of setting regions 41 are different, the train car 2 (see Figure 1) stopped at the platform is not considered to be a train car that has been pre-set to stop there.

[0066] Therefore, the control unit 12 is configured to determine whether or not the railway vehicle 2 is a vehicle that is pre-set to stop, based on the number of side lights 3. In this embodiment, the storage unit 13 is configured to store pre-set side light count information 23, which is information on the number of side lights 3 that are pre-set. The control unit 12 is configured to acquire the number of side lights 3 that are visible in multiple vehicle images 30 (see Figure 1). Specifically, the control unit 12 acquires the number of side lights 3 that are visible in multiple vehicle images 30 by acquiring the number in the learning estimation area 42 (see Figure 8). Then, the control unit 12 determines whether or not the railway vehicle 2 is a vehicle that is pre-set to stop by comparing the number of side lights 3 that are visible in the multiple vehicle images 30 with the pre-set side light count information 23. The control unit 12 is configured to output error information 24 to the higher-level device 101 if the acquired number of side lights 3 and the pre-set side light count information 23 are different. Furthermore, if the acquired number of vehicle side lights 3 matches the set number of vehicle side lights information 23, the control unit 12 outputs lighting information 20 to the higher-level device 101.

[0067] (Process for generating a pre-trained model) Next, referring to Figure 9, we will explain the process by which the control unit 12 (see Figure 1) generates the trained model 4 (see Figure 1).

[0068] In step S1, the control unit 12 acquires a vehicle image 30 (see Figure 1). In this embodiment, the control unit 12 acquires multiple vehicle images 30.

[0069] In step S2, the control unit 12 performs a masking process on a pre-set vehicle image 30 from among the multiple vehicle images 30.

[0070] In step S3, the control unit 12 sets the area 40 (see Figure 4) of the side light 3 (see Figure 4) and a label indicating whether the side light 3 is lit or not on the vehicle image 30, based on the user's input. If the side light 3 is lit, the control unit 12 sets the label to "1". If the side light 3 is off, the control unit 12 sets the label to "0".

[0071] In step S4, the control unit 12 trains the learning model 5 (see Figure 4) using a vehicle image 30 on which the area 40 of the side lights 3 and a label indicating whether or not the side lights 3 are lit are set. That is, in step S4, the control unit 12 generates a trained model 4 (see Figure 1) by training the learning model 5 using the training image 32 (see Figure 4) and the training image 33 (see Figure 4).

[0072] In step S5, the control unit 12 stores the generated trained model 4 in the storage unit 13 (see Figure 1). After that, the process ends.

[0073] (Lighting information output processing) Next, referring to Figure 10, the process by which the vehicle side light illumination status detection device 1 (see Figure 1) outputs illumination information 20 (see Figure 1) will be explained.

[0074] In step S10, the image processing unit 11 (see Figure 5) acquires vehicle images 30 (see Figure 5). Specifically, the image processing unit 11 acquires multiple vehicle images 30 captured by each of the multiple image capturing units 10 (see Figure 5) via the image acquisition unit 14 (see Figure 5).

[0075] In step S11, the image processing unit 11 acquires the trained model 4 (see Figure 5). Specifically, the image processing unit 11 reads the trained model 4 from the storage unit 13 (see Figure 5).

[0076] In step S12, the image processing unit 11 acquires lighting information 20 (see Figure 5) and learned estimation area information 21 (see Figure 5). Specifically, the image processing unit 11 acquires lighting information 20 and learned estimation area information 21 by inputting multiple vehicle images 30 into the trained model 4.

[0077] In step S13, the control unit 12 acquires the setting area information 22 (see Figure 1) and the set number of side lights information 23 (see Figure 1). Specifically, the control unit 12 reads the setting area information 22 and the set number of side lights information 23 from the storage unit 13.

[0078] In step S14, the control unit 12 determines whether or not there is a learning estimation area 42 (see Figure 8) that is not included in the setting area 41 (see Figure 8). If there is a learning estimation area 42 that is not included in the setting area 41, the process proceeds to step S15. If there is no learning estimation area 42 that is not included in the setting area 41, the process proceeds to step S16.

[0079] In step S15, the control unit 12 outputs error information 24 (see Figure 7) to the host device 101 (see Figure 7). After that, the process ends.

[0080] If the process proceeds from step S14 to step S16, in step S16, the control unit 12 obtains the number of vehicle side lights 3 (see Figure 1). Specifically, the control unit 12 obtains the number of vehicle side lights 3 by obtaining the number of learning estimation areas 42.

[0081] In step S17, the control unit 12 determines whether the number of side lights 3 matches. Specifically, the control unit 12 determines whether there is set side light count information 23 that matches the information on the number of side lights 3 obtained in step S16. If the number of side lights 3 matches, the process proceeds to step S18. If the number of side lights 3 does not match, the process proceeds to step S15.

[0082] In step S18, the control unit 12 outputs lighting information 20 to the host device 101. After that, the process ends.

[0083] (Effects of the first embodiment) In the first embodiment, the following effects can be obtained.

[0084] In the first embodiment, as described above, the vehicle side light illumination status detection system 100 includes an image capturing unit 10 that captures a vehicle image 30 showing the vehicle side lights 3 provided on the side 2e of the railway vehicle 2, an image processing unit 11 that acquires illumination information 20, which is information determining whether or not the vehicle side lights 3 shown in the vehicle image 30 are illuminated, using a trained model 4 learned by machine learning using an image showing the vehicle side lights 3, and a control unit 12 that detects the illumination status of the vehicle side lights 3 based on the illumination information 20. Here, it is considered that the trained model 4 learned by machine learning can acquire the boundary between the railway vehicle 2 and the vehicle side lights 3 even when the difference between the saturation of the body color of the railway vehicle 2 and the saturation of the vehicle side lights 3 is small. Therefore, by configuring the system as described above, the image processing unit 11 acquires illumination information 20, which is information that determines whether or not the side lights 3 are lit using the trained model 4. As a result, even when the difference between the saturation of the body color of the railway vehicle 2 and the saturation of the side lights 3 is small, the system can accurately determine whether or not the side lights 3 are lit. Consequently, even when the difference between the saturation of the body color of the railway vehicle 2 and the saturation of the side lights 3 is small, the system can suppress a decrease in the accuracy of determining whether or not the side lights 3 are lit, thereby providing a side light illumination status detection system 100 that can accurately detect the illumination status of the side lights 3.

[0085] Furthermore, in the first embodiment, the image processing unit 11 is configured to acquire learned estimated region information 21, which is information about the area in the vehicle image 30 in which the side lights 3 are visible, estimated by the trained model 4, along with the illumination information 20. The control unit 12 is configured to detect the illumination state of the side lights 3 based on the illumination information 20 and the learned estimated region information 21. As a result, the control unit 12 detects the illumination state of the side lights 3 using the learned estimated region information 21 along with the illumination information 20, and the position of the side lights 3 visible in the vehicle image 30 can be obtained using the learned estimated region information 21. Consequently, even if multiple side lights 3 are visible in a single vehicle image 30, the illumination state of each individual side light 3 can be distinguished and detected based on the illumination information 20 and the learned estimated region information 21.

[0086] Furthermore, in the first embodiment, the system is further equipped with a storage unit 13 that stores setting area information 22, which is information about the area of ​​the side lights 3 that is larger than the area of ​​the learning estimation area information 21 (learning estimation area 42) that is set in advance for each railway vehicle 2. The control unit 12 determines whether the learning estimation area 42 is included in the setting area 41 based on the learning estimation area information 21 acquired by the image processing unit 11 and the setting area information 22 stored in the storage unit 13. If the learning estimation area 42 is included in the setting area 41, it outputs lighting information 20, and if the learning estimation area 42 is not included in the setting area 41, it outputs error information 24. As a result, when the control unit 12 determines that the learning estimation area 42 is included in the setting area 41, it outputs lighting information 20, so for example, by outputting lighting information 20 to platform doors, the lighting information 20 of the side lights can be used for opening and closing platform doors. Also, when the control unit 12 determines that the learning estimation area 42 is not included in the setting area 41, it outputs error information 24. Therefore, it is possible to prevent the platform doors from opening when a railway vehicle 2 other than the pre-set railway vehicle 2 is stopped at the platform.

[0087] Furthermore, in the first embodiment, the memory unit 13 is configured to further store set side light count information 23, which is information about the number of side lights 3 set in advance, and the control unit 12 is configured to acquire the number of side lights 3 shown in the vehicle image 30 and to output error information 24 if the acquired number of side lights 3 differs from the set side light count information 23. As a result, the control unit 12 can easily determine whether the railway vehicle 2 stopped at the platform is the railway vehicle 2 set in advance by comparing the number of side lights 3 acquired from the learning estimation area information 21 with the set side light count information 23.

[0088] Furthermore, in the first embodiment, the image acquisition unit 10 is configured to capture images of multiple side lights 3, and the trained model 4 is configured to output illumination information 20 for each side light 3 when multiple side lights 3 are captured in a single image. As a result, when multiple side lights 3 are captured in a single image, illumination information 20 is output for each side light 3, so even if multiple side lights 3 are captured by a single image acquisition unit 10, illumination information 20 for each individual side light 3 can be obtained. Consequently, for example, there is no need to arrange an image acquisition unit 10 for each side light 3, as in the case where one side light 3 is captured by one image acquisition unit 10, thus suppressing an increase in the number of parts and preventing the system configuration from becoming complicated.

[0089] Furthermore, in the first embodiment, the control unit 12 is configured to exclude the region 30a in which the overlapping side lights 3 are visible from the determination target for determining whether or not the side lights 3 are lit, for one of the images in which the side lights 3 are overlapping among the multiple vehicle images 30. This prevents the determination of whether or not the side lights 3 that overlap in multiple vehicle images 30 are lit. As a result, even if one side light 3 is visible in multiple vehicle images 30, the illumination status of the side light 3 can be accurately detected.

[0090] Furthermore, in the first embodiment, the control unit 12 is configured to exclude the area 30a in which the overlapping side lights 3 are captured from the determination target for determining whether or not the side lights 3 are lit, by performing a masking process on the area 30a in which the overlapping side lights 3 are captured. As a result, since the control unit 12 performs a masking process on the area 30a in which the overlapping side lights 3 are captured, the area 30a in which the overlapping side lights 3 are captured can be easily excluded from the determination target for determining whether or not the side lights 3 are lit.

[0091] Furthermore, in the first embodiment, the trained model 4 is a deep neural network that has been trained to determine whether or not the vehicle side lights 3 are illuminated from the vehicle image 30. As a result, by using a deep neural network, which is a trained model suitable for object detection, the vehicle side lights 3 that appear in the vehicle image 30 can be detected with high accuracy.

[0092] Furthermore, in the first embodiment, the vehicle side light illumination state detection device 1 includes an image acquisition unit 14 that acquires a vehicle image 30 showing the vehicle side lights 3 provided on the side 2e of the railway vehicle 2, an image processing unit 11 that acquires illumination information 20, which is information determining whether or not the vehicle side lights 3 shown in the vehicle image 30 are illuminated, using a trained model 4 learned by machine learning using the image showing the vehicle side lights 3, and a control unit 12 that detects the illumination state of the vehicle side lights 3 based on the illumination information 20. As a result, similar to the vehicle side light illumination state detection system 100 according to the first embodiment, even when the difference between the saturation of the body color of the railway vehicle 2 and the saturation of the vehicle side lights 3 is small, it is possible to provide a vehicle side light illumination state detection device 1 that can accurately detect the illumination state of the vehicle side lights 3 by suppressing a decrease in the accuracy of determining whether or not the vehicle side lights 3 are illuminated.

[0093] [Second Embodiment] Referring to Figures 11 and 12, the configuration of the side-mounted light illumination status detection system 200 equipped with the side-mounted light illumination status detection device 201 according to the second embodiment will be described. In the figures, parts with the same configuration as in the first embodiment are denoted by the same reference numerals.

[0094] As shown in Figure 11, the vehicle side light illumination status detection system 200 differs from the vehicle side light illumination status detection system 100 (see Figure 1) according to the first embodiment in that it includes a stop position information acquisition unit 15 and a vehicle side light illumination status detection device 201.

[0095] The stopping position information acquisition unit 15 is configured to acquire stopping position information 25, which is information about the stopping position of the railway vehicle 2. The stopping position information acquisition unit 15 may include various position detection sensors. The stopping position information 25 also includes, for example, the amount of deviation of the railway vehicle 2 from a reference stopping position.

[0096] The vehicle side lamp illumination status detection device 201 differs from the vehicle side lamp illumination status detection device 1 (see Figure 1) according to the first embodiment in that it includes a control unit 210 instead of a control unit 12 (see Figure 1).

[0097] The control unit 210 is configured to detect the illumination state of the vehicle side lamp 3 based on the stop position information 25 and the illumination information 20 acquired by the learned model 4. The control unit 210 according to the second embodiment is configured to correct the position of the setting area 41 (see Figure 12) based on the stop position information 25. The other configurations of the control unit 210 are the same as those of the control unit 12 (see Figure 1) according to the first embodiment.

[0098] (Adjusting the settings area) Next, referring to Figure 12, a configuration will be described in which the control unit 210 (see Figure 11) according to the second embodiment corrects the position of the setting area 41 using the stop position information 25 (see Figure 11).

[0099] The vehicle image 31 shown in Figure 12 is an image in which the setting area 41 has been set. In the example shown in Figure 12, the setting area 41 before its position was corrected is shown with a dashed line. If the stopping position of the railway vehicle 2 shifts, the position of the vehicle side light 3 shown in the vehicle image 31 will also shift.

[0100] Therefore, the control unit 210 is configured to modify the setting area 41 using the stop position information 25. Specifically, as shown in Figure 12, the control unit 210 sets a modified setting area 43 by correcting the position of the setting area 41 based on the stop position information 25. More specifically, the control unit 210 converts the amount of displacement of the stopping position of the railway vehicle 2 into the number of pixels in the vehicle image 31 based on the stop position information 25 and the imaging magnification of the image acquisition unit 10 (see Figure 11). Then, the control unit 210 sets the modified setting area 43 by shifting the position of the setting area 41 by the converted number of pixels. The control unit 210 also obtains the number of pixels in the vertical direction and the number of pixels in the horizontal direction in the vehicle image 31 corresponding to the amount of displacement of the railway vehicle 2, and modifies the position of the setting area 41.

[0101] In the example shown in Figure 12, the first setting area 41a and the second setting area 41b are set. Therefore, the control unit 210 sets the first modified setting area 43a and the second modified setting area 43b, which are modified versions of the first setting area 41a and the second setting area 41b, respectively.

[0102] The control unit 210 uses a modified setting area 43 instead of the setting area 41 to detect the illumination state of the vehicle side lamp 3 (see Figure 11). Except for the use of the modified setting area 43 instead of the setting area 41, the configuration is the same as that of the control unit 12 in the first embodiment for detecting the illumination state of the vehicle side lamp 3, so a detailed explanation is omitted.

[0103] Next, referring to Figure 13, the process by which the control unit 210 (see Figure 11) according to the second embodiment outputs lighting information 20 (see Figure 11) will be described. Note that components similar to those used in the process by which the control unit 12 (see Figure 1) according to the first embodiment outputs lighting information 20 (see Figure 1) are denoted by the same reference numerals, and detailed explanations are omitted.

[0104] In steps S10 and S11, the image processing unit 11 (see Figure 11) acquires multiple vehicle images 30 and performs masking on the pre-set vehicle images 30.

[0105] In step S20, the control unit 210 acquires the stop position information 25 (see Figure 11). Specifically, the control unit 210 acquires the stop position information 25 from the stop position information acquisition unit 15.

[0106] In steps S12 and S13, the image processing unit 11 acquires lighting information 20 and learning estimation area information 21. The control unit 210 also acquires setting area information 22 and setting vehicle side light count information 23.

[0107] In step S21, the control unit 210 modifies the setting area 41 (see Figure 12). Specifically, the control unit 210 sets a modified setting area 43 (see Figure 12) which is the modified setting area 41.

[0108] The process then proceeds to steps S14 and S15, where the control unit 210 outputs error information 24 (see Figure 7) to the host device 101 (see Figure 11), and then terminates. Alternatively, the process proceeds to steps S14 to S18, where the control unit 210 outputs lighting information 20 (see Figure 11) to the host device 101, and then terminates.

[0109] The other configurations of the second embodiment are the same as those of the first embodiment described above.

[0110] (Effects of the second embodiment) In the second embodiment, the following effects can be obtained.

[0111] In the second embodiment, as described above, the vehicle side light illumination state detection system 200 further includes a stop position information acquisition unit 15 that acquires stop position information 25, which is information about the stopping position of the railway vehicle 2, and the control unit 210 is configured to detect the illumination state of the vehicle side light 3 based on the stop position information 25 and the illumination information 20 acquired by the learned model 4. As a result, even if there is a discrepancy in the stopping position of the railway vehicle 2, the setting area 41 included in the setting area information 22 can be corrected based on the stop position information 25. As a result, even if there is a discrepancy in the stopping position of the railway vehicle 2, it is possible to suppress a decrease in the detection accuracy of the illumination state of the vehicle side light 3.

[0112] Furthermore, the other effects of the second embodiment are the same as those of the first embodiment described above.

[0113] [Differentiation] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the present invention is indicated by the claims rather than the description of the embodiments above, and further includes all modifications (modifications) within the meaning and scope equivalent to the claims.

[0114] For example, in the first and second embodiments described above, an example of a configuration in which the control unit 12 (control unit 210) detects the illumination state of the vehicle side lights 3 based on the illumination information 20 and the learning estimation area information 21 was shown, but the present invention is not limited thereto. For example, the control unit may be configured to detect the illumination state of the vehicle side lights 3 using the illumination information 20 without using the learning estimation area information 21. However, if the control unit is configured to detect the illumination state of the vehicle side lights 3 using the illumination information 20 without using the learning estimation area information 21, it becomes difficult to distinguish and detect the illumination state of multiple vehicle side lights 3 captured in a single image. Therefore, it is preferable that the control unit be configured to detect the illumination state of the vehicle side lights 3 based on the illumination information 20 and the learning estimation area information 21.

[0115] Furthermore, in the first and second embodiments described above, an example was shown in which the control unit 12 (control unit 210) determines whether the learning estimation area 42 is included in the setting area 41 and outputs lighting information 20 or error information 24, but the present invention is not limited thereto. The control unit may be configured to output only lighting information 20. However, if the control unit is configured to output only lighting information 20, it cannot output to the higher-level device 101 that the railway vehicle 2 stopped at the platform is not a pre-set vehicle. Therefore, it is preferable that the control unit is configured to determine whether the learning estimation area 42 is included in the setting area 41 and output lighting information 20 or error information 24.

[0116] Furthermore, in the first and second embodiments described above, an example was shown in which the control unit 12 (control unit 210) compares the number of side lights 3 shown in a plurality of vehicle images 30 with the set number of side lights 23 stored in the storage unit 13, and outputs error information 24 when the number of side lights 3 differs from the set number of side lights 23. However, the present invention is not limited to this. For example, the control unit does not have to be configured to compare the number of side lights 3 with the set number of side lights 23. However, if the control unit is not configured to compare the number of side lights 3 with the set number of side lights 23, then error information 24 will not be output when the number of side lights 3 differs from the set number of side lights 23. Therefore, it is preferable that the control unit is configured to compare the number of side lights 3 with the set number of side lights 23, and output error information 24 when the number of side lights 3 differs.

[0117] Furthermore, in the first and second embodiments described above, an example was shown in which the control unit 12 (control unit 210) performs a masking process on the region 30a in which the overlapping side lights 3 are captured, thereby excluding the region 30a in which the overlapping side lights 3 are captured from the determination target for determining whether or not the side lights 3 are lit. However, the present invention is not limited to this. For example, the control unit may exclude the region 30a in which the overlapping side lights 3 are captured from the determination target in any way, as long as the lighting information 20 of the region 30a in which the overlapping side lights 3 are captured is not used in determining whether or not the side lights 3 are lit. For example, instead of performing a masking process, the control unit may be configured to acquire the coordinate values ​​of the region 30a in which the overlapping side lights 3 are captured and exclude the region 30a in which the overlapping side lights 3 are captured from the determination target. Alternatively, even if the control unit acquires the lighting information 20 of the region 30a in which the overlapping side lights 3 are captured, it may be configured to exclude it from the acquisition target for the number of side lights 3.

[0118] Furthermore, while the first and second embodiments described above show examples where the trained model 4 is a deep neural network, the present invention is not limited to this. The type of trained model is not limited as long as it is possible to obtain the illumination information 20 of the vehicle side light 3 and the trained estimation region information 21.

[0119] Furthermore, while the first and second embodiments described above show an example configuration in which the vehicle side light illumination status detection device 1 (vehicle side light illumination status detection device 201) generates a trained model 4 by training the learning model 5, the present invention is not limited thereto. For example, the trained model 4 generated in an image processing device different from the vehicle side light illumination status detection device may be stored in the storage unit 13.

[0120] Furthermore, while the first and second embodiments described above show examples of configurations in which each image capturing unit 10 captures multiple vehicle side lights 3, the present invention is not limited thereto. For example, each image capturing unit 10 may be configured to capture one vehicle side light 3. However, if each image capturing unit 10 captures one vehicle side light 3, it is necessary to provide the same number of image capturing units 10 as there are vehicle side lights 3, which increases the number of parts and complicates the system configuration. Therefore, it is preferable that at least one of the multiple image capturing units 10 is configured to capture multiple vehicle side lights 3.

[0121] Furthermore, while the first and second embodiments described above show an example in which the vehicle side light illumination state detection system 100 (vehicle side light illumination state detection system 200) comprises two image capturing units 10, the present invention is not limited thereto. For example, if one image capturing unit 10 can capture a vehicle image 30 showing all the vehicle side lights 3, the vehicle side light illumination state detection system may comprise one image capturing unit 10. Also, the vehicle side light illumination state detection system may comprise two or more image capturing units 10. The number of image capturing units 10 comprised of the vehicle side light illumination state detection system is not limited.

[0122] Furthermore, although the second embodiment described above shows an example in which the vehicle side light illumination status detection system 200 includes a stop position information acquisition unit 15, the present invention is not limited thereto. For example, if the vehicle side light illumination status detection device 201 is configured to communicate with the railway vehicle 2, it may be configured to acquire stop position information 25 from the railway vehicle 2. [Explanation of Symbols]

[0123] 1. 201 Vehicle side light illumination status detection device 2 Railway vehicles 2e Side view of a railway vehicle 3 Car side lights 4. Pre-trained models 10 Image acquisition unit 11 Image Processing Unit 12, 210 Control Unit 13 Storage section 14 Image acquisition unit 15 Stop position information acquisition section 20 Lighting Information 21 Learning Estimation Area Information 22. Configuration Area Information 23. Information on the number of side lights set on the vehicle 24 Error Information 25 Stop position information 30 Vehicle Images 30a Area where overlapping vehicle side lights are visible 41 Setting Area 42 Learning Estimation Domain 100, 200 Vehicle Side Light Illumination Status Detection System

Claims

1. An image capturing unit that captures images of a train car, including the side lights installed on the side of the train car, An image processing unit that acquires illumination information, which is information determining whether or not the vehicle side lights shown in the vehicle image are illuminated, using a trained model trained by machine learning using the aforementioned vehicle side lights, A control unit that detects the illumination status of the vehicle side lights based on the illumination information, It includes a storage unit that stores information on the number of set side lights, which is information on the number of side lights set in advance, The control unit is configured to acquire the number of side lights visible in the vehicle image and to output error information if the acquired number of side lights differs from the set number of side lights information, thereby providing a side light illumination status detection system.

2. The image processing unit is configured to acquire, together with the illumination information, the learned estimated region information, which is information about the region in the vehicle image in which the vehicle side lights are visible, as estimated by the trained model. The vehicle side lamp illumination state detection system according to claim 1, wherein the control unit is configured to detect the illumination state of the vehicle side lamp based on the illumination information and the learning estimation area information.

3. The storage unit stores setting area information, which is information about the area of ​​the vehicle side lamp that is larger than the area of ​​the learning estimation area information, which is set in advance for each railway vehicle. The control unit is configured to determine whether the learning estimation area is included in the setting area in which the vehicle side lights may be visible in the vehicle image, based on the learning estimation area information acquired by the image processing unit and the setting area information stored in the storage unit, and to output the lighting information if the learning estimation area is included in the setting area, and to output the error information if the learning estimation area is not included in the setting area, as described in claim 2.

4. The image acquisition unit is configured to capture images of multiple vehicle side lights. The vehicle side light illumination status detection system according to claim 1, wherein the trained model is configured to output illumination information for each vehicle side light when multiple vehicle side lights are captured in a single image.

5. The vehicle side lamp illumination status detection system according to claim 4, wherein the control unit is configured to exclude, after the image acquisition unit has acquired the vehicle image and before the image processing unit has acquired the illumination information, the region in which the vehicle side lamp overlaps from the determination target for determining whether or not the vehicle side lamp is illuminated, for one of the images in which the vehicle side lamp overlaps among the plurality of vehicle images.

6. The vehicle side light illumination status detection system according to claim 5, wherein the control unit is configured to perform a masking process on areas where overlapping vehicle side lights are captured after the vehicle image is acquired by the image capturing unit and before the illumination information is acquired by the image processing unit, thereby excluding areas where overlapping vehicle side lights are captured from the determination target for determining whether or not the vehicle side lights are illuminated.

7. The system further includes a unit that acquires stopping position information, which is information about the stopping position of a railway vehicle. The vehicle side lamp illumination state detection system according to claim 2, wherein the control unit is configured to detect the illumination state of the vehicle side lamp based on the stop position information and the illumination information acquired by the learned model.

8. The vehicle side light illumination status detection system according to claim 1, wherein the trained model is a deep neural network trained to determine from the vehicle image whether or not the vehicle side lights are illuminated.

9. An image acquisition unit that acquires vehicle images showing the side lights installed on the side of the railway vehicle, An image processing unit that acquires illumination information, which is information determining whether or not the vehicle side lights shown in the vehicle image are illuminated, using a trained model trained by machine learning using the aforementioned vehicle side lights, A control unit that detects the illumination status of the vehicle side lights based on the illumination information, It includes a storage unit that stores information on the number of set side lights, which is information on the number of side lights set in advance, The control unit is configured to acquire the number of side lights visible in the vehicle image and to output error information if the acquired number of side lights differs from the set number of side lights information, thereby detecting the illumination status of side lights.

Images