A method for creating a boarding / alighting monitoring system, a boarding / alighting monitoring method, and a dedicated AI image analysis algorithm.

Abstract

This invention provides a technology for monitoring passengers getting on and off boarding, which enables image analysis with good accuracy in various environments. [Solution] The passenger boarding and alighting monitoring system of the present invention comprises a monitoring camera that acquires images of passengers boarding and alighting from a railway vehicle; an environmental condition detection unit that detects environmental conditions such as weather conditions, time conditions, or conditions related to individual stations when images of passengers are acquired; an environmental category selection unit that automatically selects an environmental category corresponding to the environmental conditions; and an image analysis unit that uses a dedicated AI image analysis algorithm specialized for the automatically selected environmental category to detect dangerous events from the images of passengers, such as passengers in wheelchairs, passengers with white canes, and passengers approaching a vehicle just before or after the vehicle doors close. In addition, the dedicated AI image analysis algorithm may be trained using image data stored in an image data storage unit with dangerous events annotated on it as training data.

Description

Technical Field

[0001] The present invention relates to an on-off monitoring system, an on-off monitoring method, and a method for creating a dedicated AI image analysis algorithm.

Background Art

[0002] In recent years, the use of an on-off monitoring system that monitors whether a dangerous event occurs during boarding and alighting using images from monitoring cameras installed on the sides of railway vehicles has been on the rise. FIG. 1 is a schematic diagram showing an example of a state in which a monitoring camera in a conventional on-off monitoring system is arranged on a railway vehicle. FIG. 1(a) is a view of vehicle 1 equipped with monitoring camera 2 from above, and FIG. 1(b) is a view of one vehicle 1 from the side. In order to monitor the vicinity of vehicle door 3 during boarding and alighting, a pair of monitoring cameras 2 are provided on both sides of vehicle 1, one facing forward in the vehicle traveling direction and the other facing backward in the vehicle traveling direction. The images captured by monitoring camera 2 are displayed, for example, on a monitor installed in the driver's cab to monitor the situation of passengers boarding and alighting near vehicle door 3, and are used to ensure safety and improve services for passengers.

[0003] For example, Patent Document 1 discloses a technique for supporting passengers who require assistance, such as wheelchair passengers and passengers with white canes, using such an on-off monitoring system. In Patent Document 1, a video of the vicinity of the vehicle door during boarding and alighting captured by a monitoring camera is analyzed by an information processing device to detect passengers who require assistance, and this is notified to station staff to enable prompt support.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Conventional boarding and alighting monitoring systems analyze captured images using an information processing device without considering environmental conditions such as weather or time of day to detect persons requiring assistance. However, in rainy weather, for example, raindrops can enter the image as noise, or distortion may occur, requiring different measures than in sunny weather to maintain the accuracy of image analysis. Similarly, at night, the contrast between the brightness of the object being analyzed and the background changes, requiring different measures than during the day to maintain the accuracy of image analysis. Thus, there is a need to analyze dangerous events with good accuracy in various environments. Patent Document 1 does not necessarily disclose this recognition of the problem.

[0006] Therefore, the present invention aims to provide a technology related to passenger boarding and alighting monitoring that enables image analysis with good accuracy in various environments. [Means for solving the problem]

[0007] To solve the above problems, one representative passenger boarding and alighting monitoring system of the present invention comprises a monitoring camera that acquires images of passengers boarding and alighting from a railway vehicle, an environmental condition detection unit that detects the environmental conditions when the images of passengers are acquired, an environmental category selection unit that automatically selects an environmental category corresponding to the environmental conditions, and an image analysis unit that detects dangerous events from the images of passengers using a dedicated AI image analysis algorithm specialized for the automatically selected environmental category. [Effects of the Invention]

[0008] According to the present invention, it is possible to provide a technology for monitoring passenger boarding and alighting that enables image analysis with good accuracy in various environments. Issues, structures, and effects other than those mentioned above will be clarified by the following explanation of the implementation methods. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is a schematic diagram showing an example of how surveillance cameras in a conventional passenger boarding / alighting monitoring system are positioned on a railway vehicle. [Figure 2] Figure 2 shows an example of the configuration of a passenger boarding and alighting monitoring system according to one embodiment of the present invention. [Figure 3] Figure 3 is a functional block diagram of the analysis server. [Figure 4] Figure 4 is a diagram showing an example of an environmental classification table. [Figure 5] Figure 5 is a table illustrating dedicated AI image analysis algorithms corresponding to each environmental category in Figure 4(b). [Figure 6] Figure 6 shows an example of the processing flow in the boarding / alighting monitoring system of this embodiment. [Modes for carrying out the invention]

[0010] Embodiments of the present invention will be described below with reference to the drawings. However, the present invention is not limited by this embodiment. In addition, the same parts are denoted by the same reference numerals in the drawings.

[0011] (Dangerous Events) In this disclosure, a hazardous event is an event related to passengers that may affect the operation of the railway and that requires special attention from the train driver or station staff to ensure safety. This includes not only events related to passengers themselves, such as passengers in wheelchairs or with white canes, passengers with strollers, passengers with large / long luggage, and groups of infants, but also events related to passenger behavior, such as passengers rushing to board or alight just before the doors close or passengers approaching the train after the doors have closed. The scope of events to be considered hazardous events can be appropriately set by the system user according to needs, system load, and costs.

[0012] (environmental conditions) In the present disclosure, the environmental conditions refer to the situations during boarding and alighting, which are conditioned by events that can affect the accuracy of image analysis. In the case of weather, environmental conditions such as sunny, rainy, snowy, etc. can be considered, and in the case of time zones, environmental conditions such as morning, afternoon, night can be considered. In addition to the environmental conditions common to each station such as weather and time zone, it is also possible to set individual stations as environmental conditions (for example, when the ambient brightness is significantly different from other stations due to the structure of the station, etc., the station-specific situation can be considered as an environmental condition), and it is possible to appropriately set according to the needs of system users.

[0013] [System Configuration] FIG. 2 is a schematic diagram showing a configuration example of a boarding / alighting monitoring system according to an embodiment of the present invention. The railway vehicle is composed of a leading vehicle 10a, one or more intermediate vehicles 10b, and a trailing vehicle 10c. The boarding / alighting monitoring system in this embodiment includes a monitoring camera 11, a hub 12, a monitor 13, an image data storage unit 15, an environmental condition detection unit 14, a control unit 16, and an analysis server 17. Each functional unit constituting the boarding / alighting monitoring system is connected via the hub 12 to form a network.

[0014] ​​​​​​​​​​​​​The hub 12 constitutes the network of the entire vehicle. The hub 12 of each vehicle is connected to the hub 12 of an adjacent vehicle by a network cable. The network cable is connected to a splicing box (not shown) via under the floor of the vehicle, and the splicing boxes between adjacent vehicles are connected by jumper wires.

[0017] (Monitor) The monitor 13 is a display device that visually notifies the driver of dangerous events detected by the boarding and alighting monitoring system. In FIG. 2, the monitor 13 is provided in the driver's cabs of the leading vehicle 10a and the trailing vehicle 10c. In addition to the driver's cab, the monitor 13 may be installed in the station premises or the central operation command room.

[0018] The display screen of the monitor 13 is divided, and the images captured by each monitoring camera 11 are assigned to and displayed on their respective display screens. The manner of displaying the detected dangerous event is not particularly limited, and it may be highlighted by being surrounded by a red frame, attention may be called by text, or it may be overlaid and displayed by other appropriate means. However, it is desirable to consider that if the display becomes too complicated, it will impose a burden on the monitor and it will take time to take necessary actions. The display screen of the monitor 13 may be always divided according to each monitoring camera 11, or may be automatically divided in response to the detection of a dangerous event. Alternatively, without dividing the display screen of the monitor 13, a list of dangerous events and their detection locations may be displayed in a list format on a single display screen. If information regarding the location where a dangerous event is detected is unnecessary, this may be omitted, and it is also possible to simply notify the detection of a dangerous event.

[0019] The means for notifying the driver of a dangerous event may use buzzer voice means etc. in addition to or instead of the monitor 13 according to the content of the information to be notified.

[0020] (Environmental condition detection unit) The environmental condition detection unit 14 has a function of detecting a predetermined environmental condition (such as sunny weather etc.) from data regarding the environment acquired during boarding and alighting. The means of acquiring data used to detect environmental conditions are not particularly limited. For example, in the case of weather-related environmental conditions, a general-purpose algorithm that determines weather conditions such as sunny or rainy from surveillance camera image data may be used, or data from external sensors such as rain sensors may be used. It is also possible to use data related to the operation of the equipment when the surveillance camera's autofocus automatically adjusts due to distortion of the subject's edges caused by raindrops. Furthermore, environmental conditions related to time of day (time of day conditions) can be detected using GPS time. GPS time may be received from devices mounted on the vehicle that operate in synchronization with GPS time, such as the surveillance camera 11 or the control unit 16, or it may be obtained directly from GPS. Furthermore, if individual stations are determined by environmental conditions, they can be detected using a station information database installed on the train. Alternatively, station name information transmitted from outside the train may be used for detection.

[0021] (Image data storage section) The image data storage unit 15 continuously stores image data captured by the surveillance camera 11. The image data stored in the image data storage unit 15 is used when annotating dangerous events (described later).

[0022] (Control Unit) The control unit 16 has the function of controlling the operation of the entire passenger boarding and alighting monitoring system. Based on operational information such as the opening and closing of both doors and passenger boarding / alighting information of the vehicle doors (whether or not it is the vehicle door on the side that will be used as a boarding / alighting exit at the next station) from a higher-level operation control device (not shown) that manages the overall operation of railway vehicles, such as a TMS (TRAIN MANAGEMENT SYSTEM), the control unit 16 controls each component of the passenger boarding and alighting monitoring system to perform the necessary processing in a timely manner (for example, by activating the monitoring camera on the side of the vehicle door that opens and closes at a designated station). Furthermore, the control unit 16 also has the function of acquiring data for detecting environmental conditions from GPS, various sensors, and the operation of surveillance cameras. The data acquired for detecting these environmental conditions is sent to the analysis server.

[0023] (Analysis server) Figure 3 is a functional block diagram of the analysis server. The analysis server 17 includes an environment classification selection unit 18 and an image analysis unit 19. The environmental classification selection unit 18 stores an environmental classification table, which will be described later. When the environmental conditions detected by the environmental condition detection unit 14 during boarding and alighting are input, it has the function of automatically selecting an environmental classification corresponding to the environmental conditions using the environmental classification table and outputting it to the image analysis unit 19.

[0024] The image analysis unit 19 stores a dedicated AI image analysis algorithm corresponding to each environmental category. Based on the environmental category automatically selected by the environmental category selection unit 18, it performs image analysis of the image data captured by the surveillance camera 11 using a dedicated AI image analysis algorithm specialized for that environmental category. If a dangerous event is detected as a result of the image analysis, it is displayed on the monitor 13 (described above). Furthermore, when detecting an approaching person as a dangerous event, in addition to person detection by the image analysis unit 19, detection should be performed in conjunction with other information that indicates danger when a person is near the vehicle doors, such as the time just before departure, the departure buzzer sound, and door closing detection.

[0025] In terms of hardware configuration, the environmental condition detection unit 14, control unit 16, and analysis server 17 execute their respective functions through program processing by a computer processor such as a CPU. These functional units also include random access semiconductor memory, memory devices, or storage media (either volatile or non-volatile) for storing programs (algorithms) and data. When realizing different functions through program processing by the processor and memory, they can be arranged as separate hardware components, or a single piece of hardware can be shared by switching or selecting functions using software.

[0026] (Environment classification table and algorithm) Next, we will explain the environment classification table and the algorithms corresponding to each environment classification. Figure 4 is a diagram showing an example of an environmental classification table. Figure 4(a) is an example where only weather is used as an environmental condition, and three conditions are set: "sunny," "rainy," and "snowy." And "ALG * 1", ALG * 2", ALG * The environmental classification table is composed of three environmental classifications, "3". In this case, if the environmental condition detection unit 14 detects "rainy weather" as an environmental condition, the environmental category will be "ALG * 2 will be automatically selected. The environmental table in Figure 4(a) could be used in cases such as railway vehicles operating during specific time periods, or lines where platform brightness remains nearly constant regardless of the time of day.

[0027] Figure 4(b) shows an example where weather and time of day are used as environmental conditions, with three weather conditions: "sunny," "rainy," and "snowy," and three time of day conditions: "morning," "afternoon," and "night." The environmental classification table is then composed of nine environmental classifications, "ALG1," "ALG2," ... "ALG9." In this case, if the environmental condition detection unit 14 detects "rainy weather" as a weather condition and "afternoon" as a time of day, then environmental category "ALG5" will be automatically selected. It should be noted that the types and number of environmental conditions are not limited to this example, and it is possible that a combination of three or more environmental conditions may result in the creation of a three-dimensional or higher-dimensional environmental classification table.

[0028] Next, we will explain the dedicated AI image analysis algorithms (sometimes simply referred to as "algorithms") corresponding to each environmental category. Figure 5 is a table illustrating the dedicated AI image analysis algorithms corresponding to each environmental category in Figure 4(b). For example, if the environmental conditions during boarding and alighting are detected as "sunny" and "morning," environmental category ALG1 is automatically selected, and the surveillance camera image is analyzed by a dedicated AI image analysis algorithm (Algorithm 1) that has been specifically trained for that environmental category. Algorithm 1 consists of pre-trained algorithms 1a to 1c, each annotated for a predetermined hazardous event. For example, algorithm 1a is a dedicated AI image analysis algorithm trained specifically for wheelchair users, while algorithms 1b and 1c are dedicated AI image analysis algorithms trained specifically for white canes and strollers, respectively. Therefore, when algorithm 1 is applied to an image, each of algorithms 1a to 1c is executed as a package, and wheelchair users, passengers with white canes, or passengers pushing strollers are analyzed and displayed on the monitor. Each hazardous event may be displayed superimposed or individually. However, since the selection of hazardous events to be detected depends on the system user, it is not always necessary to perform analysis using all the algorithms included in the package. For example, if stroller detection is not required, you can set a flag for algorithm 1c to indicate that it is not needed, and the system will not execute that algorithm during image analysis. Furthermore, there are no particular limitations on the number or types of algorithms (selection of hazardous events to be detected) to be included in the package.

[0029] [Processing flow] Figure 6 shows an example of the processing flow in the boarding / alighting monitoring system of this embodiment. Steps S11 to S15 of the processing flow relate to the notification of dangerous events, and steps S21 to S24 relate to annotation and algorithm learning. First, let me explain the flow of notification regarding dangerous events.

[0030] (Flowchart for reporting dangerous events) Below, as an example, we will explain the processing flow when the hazardous events to be analyzed are "wheelchairs" and "white canes," and the environmental conditions are "weather" and "time of day."

[0031] When the train arrives at the station and the train doors open, the passenger boarding and alighting monitoring system is activated. Images of passengers boarding and alighting are then captured by the monitoring camera 11 on the boarding and alighting side (step S11). The monitoring camera 11 may start recording in synchronization with the activation of the passenger boarding and alighting monitoring system, or it may be configured to record continuously during the train's operation, with only the images taken during the opening and closing of the train doors at stations being sent for analysis by the passenger boarding and alighting monitoring system.

[0032] Next, the environmental condition detection unit 14 detects the environmental conditions at the time of boarding and alighting (step S12). For example, "rainy weather" is detected as the weather condition and "afternoon" as the time of day condition.

[0033] Next, based on the detected environmental conditions, the environmental category selection unit 18 automatically selects an environmental category (step S13). If the weather condition is "rainy" and the time of day condition is "afternoon," "ALG5" is automatically selected as the environmental category from the environmental category table (see Figure 4(b)).

[0034] Next, in the image analysis unit 19, a dedicated AI image analysis algorithm corresponding to the automatically selected environment category is executed to perform image analysis of the surveillance camera image data (step S14). In this example, "Algorithm 5," which is specialized for environment category "ALG5," is used, and algorithms 5a and 5b, which are specialized for wheelchairs and white canes from the algorithm package of algorithm 5, are executed.

[0035] Next, if the image analysis unit 19 detects a dangerous event, it will notify the system (step S15). In this example, if the image analysis performed by "Algorithm 5" detects a "passenger in a wheelchair" or a "passenger with a white cane," these will be displayed on the monitor 13 in a predetermined format.

[0036] This concludes the flowchart for reporting dangerous events. According to this workflow, image analysis is performed using a dedicated AI image analysis algorithm specifically tailored to the environment when passengers board and alight, allowing for more accurate detection of dangerous events than before.

[0037] (Flowchart related to annotation and algorithm learning) Next, we will explain the flow related to annotation and algorithm training. Image data of passengers getting on and off the train, captured by surveillance cameras, is stored in the image data storage unit 15 (step S21).

[0038] Next, at an appropriate time, training image data is extracted from the stored image data (step S22). The method of extraction is not particularly limited; images may be extracted regularly, randomly, or suitable images may be selected manually. It is preferable that the extraction is performed when trains are not running.

[0039] Next, the extracted training image data is annotated with hazardous events (step S23). In this example, the image data is annotated with hazardous events such as "passengers in wheelchairs" and "passengers with white canes." Annotation can be done manually by first sorting the image data by environment category.

[0040] Furthermore, in the flow for reporting hazardous events, information such as environmental conditions ("rainy weather," "afternoon") (step S12), environmental classification ([ALG5]) (step S13), and image analysis results (step S14) can be obtained as profile data. Therefore, it is also possible to link this profile data with the image data stored in step S21 and record it in the image data storage unit 15. In that case, when extracting training image data in step S22, it becomes possible to automatically extract data for each environmental classification. Also, annotation in step S23 can be performed automatically using the results of the image analysis in step S14.

[0041] Next, once a certain number of annotated image data from step S23 have been accumulated, this data is used to train a dedicated AI image analysis algorithm (step S24). For example, once a certain number of annotations have been made for extracted image data classified under the environment category "ALG5" ("rainy weather" x "afternoon"), the algorithm 5 stored on the ground server is trained (updated), and once the latest algorithm 5 is created, it is replaced with the algorithm 5 in the image analysis unit on the vehicle. This replacement may also be performed by an automatic update transmitted from the ground. As described above, this flow allows for timely annotation and learning (updating) of a dedicated AI image analysis algorithm tailored to specific environmental categories. Therefore, image analysis using the updated algorithm (step S14) can detect hazardous events with greater accuracy.

[0042] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the present invention. For example, the embodiment described above shows the main part of the boarding / alighting monitoring system mounted on the vehicle, but it is not limited to this. It is also possible to install some or all of the components of the boarding / alighting monitoring system on the ground and configure it so that necessary information can be exchanged with the vehicle.

[0043] The following describes, but is not limited to, embodiments that may constitute the present invention. (Aspect 1) A passenger boarding and alighting monitoring system that detects dangerous incidents from passengers boarding and alighting from railway vehicles, Surveillance cameras that capture images of passengers getting on and off, An environmental condition detection unit that detects the environmental conditions when the images of the passengers are acquired, An environmental category selection unit that automatically selects an environmental category corresponding to the aforementioned environmental conditions, A passenger boarding and alighting monitoring system comprising: an image analysis unit that detects dangerous events from images of passengers using a dedicated AI image analysis algorithm specialized for the automatically selected environmental category; and an image analysis unit that detects dangerous events from images of passengers boarding and alighting. (Aspect 2) In the boarding / alighting monitoring system described in Embodiment 1, A passenger boarding and alighting monitoring system in which the aforementioned environmental conditions include weather conditions, time of day conditions, or conditions related to individual stations. (Aspect 3) In the boarding / alighting monitoring system described in Embodiment 2, The detection of the aforementioned time period conditions is performed using GPS time, and this is a boarding / alighting monitoring system. (Aspect 4) In the boarding / alighting monitoring system described in any one of the embodiments 1 to 3, The aforementioned environmental classification is divided into weather-related conditions and time-related conditions, and multiple time-related conditions are further divided for each weather-related condition, in a boarding / alighting monitoring system. (Appendix 5) In the boarding and alighting monitoring system according to any one of embodiments 1 to 4, the monitoring camera is mounted on the railway vehicle. (Aspect 6) In the boarding / alighting monitoring system described in any one of embodiments 1 to 5, It includes an image data storage unit that stores images of the aforementioned passengers, The boarding and alighting monitoring system is characterized in that the dedicated AI image analysis algorithm is trained using image data stored in the image data storage unit with dangerous events annotated on it. (Aspect 7) In the boarding / alighting monitoring system described in Embodiment 6, The passenger boarding / alighting monitoring system includes, as an annotation of hazardous events, passengers using wheelchairs, passengers using white canes, and passengers approaching a vehicle just before or after the vehicle doors close. (Pattern 8) In the boarding / alighting monitoring system described in Embodiment 6 or Embodiment 7, A passenger boarding and alighting monitoring system, in which images of passengers stored in the image data storage unit are linked to information regarding the environmental classification. (Aspect 9) A method for monitoring passengers getting on and off railway vehicles to detect dangerous incidents, The steps include obtaining images of passengers getting on and off the train using surveillance cameras, The environmental condition detection unit detects the environmental conditions when the images of the passengers are acquired, The environmental classification selection unit automatically selects an environmental classification corresponding to the aforementioned environmental conditions, A method for monitoring passengers getting on and off a train, comprising the step of detecting dangerous events from images of passengers using a dedicated AI image analysis algorithm specialized for automatically selected environmental categories, provided by an image analysis unit. (Aspect 10) A method for creating a dedicated AI image analysis algorithm used in the boarding / alighting monitoring method described in aspect 9, In the image data storage unit, the steps include saving the images of the passengers, The steps include annotating dangerous events to the image data stored in the aforementioned image data storage unit, A method for creating a dedicated AI image analysis algorithm, comprising the step of training a dedicated AI image analysis algorithm using the annotated image data as training data. (Aspect 11) In the method for creating a dedicated AI image analysis algorithm described in Embodiment 10, A method for creating a dedicated AI image analysis algorithm, wherein the hazardous events to be annotated include passengers in wheelchairs, passengers with white canes, and passengers approaching a vehicle just before or after the vehicle doors close. (Aspect 12) In the method for creating a dedicated AI image analysis algorithm described in Embodiment 10 or Embodiment 11, A method for creating a dedicated AI image analysis algorithm, further comprising the step of saving information relating to the environmental classification in association with images of passengers getting on and off the train. [Explanation of Symbols]

[0044] 1: Vehicle, 2: Surveillance camera, 3: Vehicle door, 10a: Leading car, 10b: Middle car, 10c: Last car, 11: Surveillance camera, 12: Hub, 13: Monitor, 14: Environmental condition detection unit, 15: Image data storage unit, 16: Control unit, 17: Analysis server, 18: Environmental category selection unit, 19: Image analysis unit

Claims

1. A passenger boarding and alighting monitoring system that detects dangerous incidents from passengers boarding and alighting from railway vehicles, Surveillance cameras that capture images of passengers getting on and off, An environmental condition detection unit that detects the environmental conditions when the images of the passengers are acquired, An environmental category selection unit that automatically selects an environmental category corresponding to the aforementioned environmental conditions, A passenger boarding and alighting monitoring system comprising: an image analysis unit that detects dangerous events from images of passengers boarding and alighting using a dedicated AI image analysis algorithm specialized for the automatically selected environmental category;

2. In the boarding / alighting monitoring system according to claim 1, A passenger boarding and alighting monitoring system in which the aforementioned environmental conditions include weather conditions, time of day conditions, or conditions related to individual stations.

3. In the boarding / alighting monitoring system according to claim 2, The boarding and alighting monitoring system detects the conditions related to the aforementioned time period using GPS time.

4. In the boarding / alighting monitoring system according to claim 2, The aforementioned environmental classification is divided into weather-related conditions and time-related conditions, and multiple time-related conditions are further divided for each weather-related condition, in a boarding / alighting monitoring system.

5. A passenger boarding and alighting monitoring system according to any one of claims 1 to 4, wherein the monitoring camera is mounted on the railway vehicle.

6. In the boarding / alighting monitoring system according to any one of claims 1 to 4, It includes an image data storage unit that stores images of the aforementioned passengers, The boarding and alighting monitoring system is characterized in that the dedicated AI image analysis algorithm is trained using image data stored in the image data storage unit with dangerous events annotated on it as training data.

7. In the boarding / alighting monitoring system according to claim 6, The passenger boarding / alighting monitoring system includes, as an annotation of hazardous events, passengers using wheelchairs, passengers using white canes, and passengers approaching a vehicle just before or after the vehicle doors close.

8. In the boarding / alighting monitoring system according to claim 6, A passenger boarding and alighting monitoring system, in which images of passengers stored in the image data storage unit are linked to information regarding the environmental classification.

9. A method for monitoring passengers getting on and off railway vehicles to detect dangerous incidents, The steps include obtaining images of passengers getting on and off the train using surveillance cameras, The environmental condition detection unit detects the environmental conditions when the images of the passengers are acquired, The environmental classification selection unit automatically selects an environmental classification corresponding to the aforementioned environmental conditions, A method for monitoring passengers getting on and off a train, comprising the step of detecting dangerous events from images of passengers getting on and off a train using a dedicated AI image analysis algorithm specialized for an automatically selected environmental category, provided by an image analysis unit.

10. A method for creating a dedicated AI image analysis algorithm used in the boarding / alighting monitoring method described in claim 9, In the image data storage unit, the steps include saving the images of the passengers, The steps include annotating dangerous events to the image data stored in the aforementioned image data storage unit, A method for creating a dedicated AI image analysis algorithm, comprising the step of training a dedicated AI image analysis algorithm using the annotated image data as training data.

11. In the method for creating a dedicated AI image analysis algorithm according to claim 10, A method for creating a dedicated AI image analysis algorithm, wherein the hazardous events to be annotated include passengers in wheelchairs, passengers with white canes, and passengers approaching a vehicle just before or after the vehicle doors close.

12. In a method for creating a dedicated AI image analysis algorithm according to claim 10 or claim 11, A method for creating a dedicated AI image analysis algorithm, further comprising the step of saving information relating to the environmental classification in association with images of passengers getting on and off the train.

Images