Accident analysis system and accident analysis method

Abstract

To reduce the time and effort required for analyzing an accident and to reproduce the accident more accurately. [Solution] In an accident analysis system (1) that analyzes accident data, the accident data is binary data recorded in an event data recorder (EDR) installed in one or more target vehicles when an accident involving the target vehicles occurs, and the system is equipped with a data receiving unit (11) that receives the binary accident data recorded in the EDR and the location data of the target vehicles via a wireless communication network, a data conversion unit (12) that converts the received binary accident data into numerical data using a learning model that has learned by machine learning data formats that differ for each automobile manufacturer, and an accident reconstruction unit (13) that recreates the circumstances under which the accident occurred for the target vehicles from the converted numerical data and the received location data.

Description

[Technical Field]

[0001] The present disclosure relates to an accident analysis system and an accident analysis method. [Background technology]

[0002] Vehicles equipped with event data recorders (EDRs) and related technologies are becoming more common. Patent Document 1 discloses a system for collecting and managing event data and the like stored by the event data recorder. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2023-519510 Summary of the Invention [Means for solving the problem]

[0004] An accident analysis system according to one embodiment of the present disclosure is an accident analysis system that analyzes accident data, wherein the accident data is binary data recorded in an event data recorder (EDR) installed in one or more target vehicles when an accident involving the target vehicles occurs, and the system is equipped with: a data receiving means that receives the binary accident data recorded in the EDR and location data of the target vehicles via a wireless communication network; a data conversion means that converts the received binary accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; and an accident reproduction means that reproduces the accident occurrence circumstances of the target vehicles from the converted numerical data and the received location data.

[0005] An accident analysis system according to one embodiment of the present disclosure includes an acquisition means for acquiring numerical data obtained by converting accident data, the accident data being data recorded in an event data recorder (EDR) installed in one or more target vehicles when an accident involving the target vehicles occurs, and further includes a simulation execution means for generating reproduction data of the accident occurrence situation of the target vehicles by simulation based on the numerical data, and a correction means for correcting the reproduction data by at least referring to the numerical data.

[0006] An accident analysis method according to one embodiment of the present disclosure is an accident analysis method executed by one or more processors for analyzing accident data, wherein the accident data is binary data recorded in an event data recorder (EDR) installed in one or more target vehicles when an accident involving the target vehicles occurs, and includes: a data reception step for receiving the binary accident data recorded in the EDR and location data of the target vehicles via a wireless communication network; a data conversion step for converting the received binary accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; and an accident reproduction step for reproducing the circumstances under which the accident occurred of the target vehicles from the converted numerical data and the received location data.

[0007] An accident analysis method according to one aspect of the present disclosure is an accident analysis method executed by one or more processors, and includes an acquisition step of acquiring numerical data obtained by converting accident data, where the accident data is data recorded in an event data recorder (EDR) installed in one or more target vehicles when an accident involving the target vehicles occurs, and further includes a simulation execution step of generating reproduction data of the accident occurrence situation of the target vehicles by simulation based on the numerical data, and a correction step of correcting the reproduction data by at least referring to the numerical data. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a block diagram showing a functional configuration of an accident analysis system according to the present disclosure. [Figure 2] FIG. 1 is a flow diagram showing an example of a processing flow by an accident analysis system according to the present disclosure. [Figure 3] FIG. 2 is a diagram for explaining processing by the accident analysis system according to the present disclosure. [Figure 4] FIG. 2 is a diagram for explaining processing by the accident analysis system according to the present disclosure. [Figure 5] FIG. 2 is a diagram for explaining information generated by the accident analysis system according to the present disclosure. [Figure 6] FIG. 2 is a block diagram illustrating an example of a functional configuration of an accident reconstruction unit according to the present disclosure. [Figure 7] FIG. 10 is a flowchart showing an example of a processing flow by an accident reconstruction unit according to the present disclosure. [Figure 8] FIG. 10 is a block diagram illustrating another example of the functional configuration of the accident reconstruction unit according to the present disclosure. [Figure 9] FIG. 10 is a flow diagram showing an example of the flow of an accident report creation process by the accident analysis system according to the present disclosure. [Figure 10] FIG. 10 is a diagram illustrating an example of a display screen generated by the accident analysis system according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0009] [Embodiment 1] Hereinafter, an embodiment of the present disclosure will be described in detail with reference to the drawings. For ease of understanding, the background and problems of the present disclosure will be described first, and then the details of the present disclosure will be described.

[0010] <Background and Issues> Vehicles equipped with event data recorders (EDRs) and related technologies are becoming more common. When an accident involving a vehicle equipped with an EDR occurs, the event data (also called accident data) recorded in the EDR at the time of the accident is used to analyze the accident.

[0011] The accident data is in binary format, and the format generally varies depending on the automobile manufacturer. This has been one of the reasons for the increased effort required to analyze an accident by referring to the accident data. Furthermore, even if an accident is analyzed by referring to the accident data, there is a problem in that it is not easy to accurately reproduce the accident.

[0012] An object of one aspect of the present disclosure is to provide a technology that can reduce the effort required to analyze an accident by referring to accident data and can more accurately reproduce the accident.

[0013] <Overview of Accident Analysis System 1> The accident analysis system 1 according to this embodiment is an accident analysis system that analyzes accident data, as will be described below. The accident data is binary data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a data receiving means for receiving the binary accident data recorded in the EDR and the target vehicle's location data via a wireless communication network; a data conversion means for converting the received binary format accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; Accident reproducing means for reproducing the circumstances under which an accident occurred in the target vehicle from the converted numerical data and the received position data; It is equipped with:

[0014] In this way, the accident analysis system 1 according to this embodiment uses a learning model that has learned different data formats for each automobile manufacturer to convert binary accident data into numerical data, thereby reducing the effort required to analyze an accident. Furthermore, the numerical data obtained by this conversion and the position data of the target vehicle are referenced to reproduce the circumstances under which the accident occurred, allowing for more accurate reproduction of the accident.

[0015] (Configuration of information processing system 100) The configuration of an information processing system 100 according to this embodiment, including an accident analysis system 1, will be described with reference to FIG. 1. FIG. 1 is a block diagram showing the configuration of the information processing system 100 according to this embodiment. As shown in FIG. 1, the information processing system 100 includes the accident analysis system 1, one or more vehicles (vehicles 50A, 50B, etc. in the example of FIG. 1) communicatively connected to the accident analysis system 1 via a network N, and a terminal device 60. Here, the network N may be configured, for example, by a mobile communication system such as 3G, 4G, 5G, or 6G, but is not limited thereto, and may also include Wi-Fi (registered trademark) or an in-house LAN. Note that the accident analysis system 1 or the information processing system 100 may further include one or more servers (not shown).

[0016] (Vehicle 50A) 1, a vehicle 50A (also referred to as vehicle A) includes a communication unit 51A, an event data recorder EDR_A, and a location identification unit 52A. -Operations operated by the driver (steering, brakes, accelerator, etc.) A drive unit (engine, transmission, wheels, etc.) for driving the vehicle in response to the driver's operation received via the operation unit, and -Control unit (ECU, etc.) that controls the drive unit However, these components are not shown in the drawings. The vehicle 50A may be an autonomous vehicle that performs autonomous driving according to a predetermined algorithm.

[0017] The event data recorder EDR_A is a recorder that records data related to the behavior of the vehicle 50A and is configured to meet predetermined requirements or specifications. Vehicle 50A speed data Steering data for vehicle 50A Brake data for vehicle 50A Acceleration data of vehicle 50A (also called the collision vector in the event of an accident) Data on the operation of safety features (airbags, etc.) of vehicle 50A The various types of data stored in the event data recorder EDR_A are collectively referred to as event data. Event data including data at the time of an accident is also referred to as accident data. The event data recorder EDR_A records, for example, - Overwrite while retaining data for a specified period - Locks data recording due to sudden changes in acceleration With the above configuration, the event data can be recorded at the moment when an accident involving the vehicle 50A occurs, and for a predetermined period before and after the accident.

[0018] The above requirements or specifications that the event data recorder must satisfy do not limit the present embodiment, but examples include requirements stipulated by the Ministry of Land, Infrastructure, Transport and Tourism in Japan, standards stipulated by the National Highway Traffic Safety Administration in the United States, and standards established by United Nations regulations in Europe. These requirements or specifications stipulate the type of data to be stored and the recording interval, but do not necessarily stipulate a specific data format. Therefore, the format of the event data (accident data) recorded in the event data recorder is not necessarily standardized, and various formats may be used by different automobile manufacturers. Specific examples of event data (accident data) will be described later.

[0019] The communication unit 51A is an interface for transmitting and receiving data via the network N. As an example, the communication unit 51A transmits the event data stored in the event data recorder EDR_A to the data receiving unit 11 provided in the accident analysis system 1. The communication unit 51A may transmit the event data at predetermined time intervals, or may transmit the event data when an accident occurs, or after a predetermined time has elapsed since the accident occurred.

[0020] The position identifying unit 52A identifies the position of the vehicle 50A and transmits position data indicating the identified position to the accident analysis system 1 via the communication unit 51A. The specific configuration of the position identifying unit 52A is not limited to this embodiment, but as an example, the position (coordinates) of the vehicle 50A near the ground surface may be detected by a Global Navigation Satellite System (GNSS). When such a configuration is adopted, the position data may also be referred to as GNSS data. The communication unit 51A may transmit the position data at predetermined time intervals, or may transmit the data when an accident occurs or after a predetermined time has elapsed since the accident occurred.

[0021] It should be noted that the destination of the data transmitted by the communication unit 51A is not necessarily limited to the accident analysis system 1. The communication unit 51A may be configured to transmit the data to at least one of one or more servers (not shown) provided in the accident analysis system 1 or the information processing system 100.

[0022] (Vehicle 50B) 1, vehicle B is equipped with a communication unit 51B, an event data recorder EDR_B, and a location identification unit 52B, similar to vehicle A. The basic functions of the communication unit 51B, the event data recorder EDR_B, and the location identification unit 52B are similar to those of the communication unit 51A, the event data recorder EDR_A, and the location identification unit 52A, and therefore redundant explanations will be omitted. However, the vehicle 50B may be a different model or automobile manufacturer from the vehicle 50A.

[0023] (Terminal device 60) The terminal device 60 is, for example, a terminal device operated by a user of the accident analysis system 1, and includes a communication unit 61, a control unit 62, and an input / output unit 63, as shown in FIG.

[0024] The control unit 62 controls each unit included in the terminal device 60. The input / output unit 63 is an interface with an input device that accepts data input and an output device that outputs data. Examples of input devices include, but are not limited to, a microphone, a camera, an eye-gaze input device, a keyboard, and a touchpad. Examples of output devices include, but are not limited to, a speaker and an LCD display. The input / output unit 63 may also be referred to as a presentation unit. As an example, the input / output unit 63 displays a display screen including an accident report generated by the accident analysis system 1, and accepts input from the user on the display screen.

[0025] The communication unit 61 is an interface for transmitting and receiving data via the network N. As an example, the communication unit 61 receives an accident report from the accident analysis system 1. The communication unit 61 also transmits the user's input received via the input / output unit 63 to the accident analysis system 1.

[0026] (Accident Analysis System 1) As shown in FIG. 1, the accident analysis system 1 includes a data receiving unit 11, a data converting unit 12, an accident reconstruction unit 13, a report creating unit 14, a transmitting unit 15, and a storage unit 20. Some of these components included in the accident analysis system 1 may be connected to each other via a wired path such as a bus, or may be connected to each other via the above-mentioned network N or other wireless path. The above units may also be configured to be distributed and located on different physical servers or virtual servers. The above units may also be configured to be included in a single device. In this case, the accident analysis system 1 is also referred to as an accident analysis device 1.

[0027] In the following description, one or more vehicles involved in the accident among one or more vehicles 50A, 50B, etc. included in the information processing system 100 may be referred to as target vehicle 50. Also, the event data recorder provided in the target vehicle 50 may be referred to as event data recorder EDR. Also, the location identification unit provided in the target vehicle 50 may be referred to as location identification unit 52.

[0028] The data receiving unit 11, data conversion unit 12, accident reconstruction unit 13, report creation unit 14, transmission unit 15, and memory unit 20 provided in the accident analysis system 1 are examples of the data receiving means, data conversion means, accident reconstruction means, report creation means, transmission means, and memory means according to this embodiment.

[0029] (Storage unit 20) The storage unit 20 stores various data referenced by each unit of the accident analysis system 1 and various data generated by each unit of the accident analysis system 1. As an example, the storage unit 20 stores the following data as shown in FIG. Accident data CD Location data PD One or more machine-learned models (ML) Numerical Data ND Reproduction Data RD Corrected reproduction data ARD Language Model LL Accident Report CR is stored.

[0030] (Accident data CD) The accident data CD is event data stored in the event data recorder EDR provided in one or more target vehicles 50 related to the accident, and is binary data. More specifically, the accident data CD includes event data stored in the event data recorder EDR at the time the accident occurred and for a predetermined period before and after the accident.

[0031] (Position data PD) The position data PD is data indicating the positions of one or more target vehicles 50 involved in the accident. As an example, the position data PD includes a time series of position data of the target vehicles 50 identified by the position identification unit 52 provided in one or more target vehicles 50 involved in the accident. More specifically, the position data PD includes, of the position data identified by the position identification unit 52, position data at the time of the accident and for a predetermined period before and after the accident.

[0032] (Learning model ML) Each of the one or more learning models ML is a model that has learned a data format that differs for each automobile manufacturer. As an example, the multiple learning models ML include: Learning model ML_A that learned the data format used by automobile manufacturer A, which manufactured target vehicle 50A Learning model ML_B that learned the data format used by automobile manufacturer B, which manufactured target vehicle 50B The specific data format does not limit the present embodiment, but will be described later as an example. As the learning model ML, for example, a rule-based model or a deep learning model may be used.

[0033] (Numerical data ND) The numerical data ND is data obtained by converting the binary format accident data CD using a data conversion unit 12 (described later). The numerical data ND is, for example, data composed of numeric values ​​expressed in decimal notation. Specific examples of the numerical data ND will be described later.

[0034] (Reproduced Data RD) The reproduction data RD is data generated by the accident reproduction unit 13, which will be described later, by referring to at least one of the numerical data ND and the position data PD, and is data that expresses the circumstances under which an accident occurred for the target vehicle 50. The reproduction data RD includes, as an example, data on the trajectories (position, speed, direction, etc. at each time) of one or more target vehicles 50 related to the accident at the time of the accident and for a predetermined period before and after the accident.

[0035] (corrected reproduction data ARD) The corrected reproduction data ARD is data obtained by correcting the above-mentioned reproduction data RD by the accident reproduction unit 13, which will be described later. The corrected reproduction data ARD is also data that expresses the circumstances under which an accident occurred for the target vehicle 50, and includes, for example, data on the trajectories (position, speed, direction, etc. at each time) of one or more target vehicles 50 related to the accident at the time of the accident and for a predetermined period before and after the accident.

[0036] (Language model LL, accident report CR) The language model LL is a language model used by the report creation unit 14 (described later) to create an accident report CR relating to the circumstances of the accident occurrence. A large language model (LLM) may be used as the language model LL. The language model LL may also be stored in one or more servers other than the storage unit 20.

[0037] (Data receiving unit 11) The data receiving unit (data receiving means) 11 receives the binary accident data CD recorded in the event data recorder EDR of the target vehicle 50 and the position data PD of the target vehicle 50 via the wireless communication network N. Here, the data receiving unit 11 may receive these data from the target vehicle 50, or may be configured to receive them from at least one of one or more servers (not shown) provided in the accident analysis system 1 or the information processing system 100. Furthermore, the accident data CD includes event data recorded in the event data recorder EDR of the target vehicle 50 when an accident occurs involving one or more vehicles including the target vehicle 50. The types of data included in the event data have been described above, so a repeated description will be omitted. The accident data CD and position data PD received by the data receiving unit 11 are stored in the memory unit 20, for example.

[0038] (Data conversion unit 12) The data conversion unit (data conversion means) 12 converts the binary accident data CD received by the data receiving unit 11 into numeric data ND using a learning model (machine-learned model) ML that has learned different data formats for each automobile manufacturer through machine learning. As described above, the numeric data ND is, for example, data composed of numeric values ​​expressed in decimal notation.

[0039] As an example, the data conversion unit 12 converts accident data CD_A recorded in the event data recorder EDR_A of the target vehicle 50A into numerical data ND_A using a learning model ML_A that has learned the data format used by the automobile manufacturer A that manufactured the target vehicle 50A.

[0040] Similarly, the data conversion unit 12 converts the accident data CD_B recorded in the event data recorder EDR_B of the target vehicle 50B into numerical data ND_B using a learning model ML_B that has learned the data format used by the automobile manufacturer B that manufactured the target vehicle 50B. The numerical data ND (ND_A and / or ND_B) obtained by the conversion by the data conversion unit 12 is stored in the storage unit 20, for example.

[0041] (Accident Reenactment Part 13) The accident reconstructing unit (accident reconstructing means) 13, for example, reconstructs the accident occurrence situation of the target vehicle 50 from the numerical data ND converted by the data converting unit 12 and the position data PD. Here, the reconstructing process of the accident occurrence situation by the accident reconstructing unit 13 includes the following steps: A process of generating reproduction data RD of the accident occurrence situation by referring to the numerical data ND. A process of generating corrected reproduction data ARD by correcting the reproduction data RD with reference to the position data. Here, as described above, the reproduction data RD (corrected reproduction data ARD) includes, for example, data on the trajectories (position, speed, direction, etc. at each time) of one or more vehicles related to the accident at the time of the accident and for a predetermined period before and after the accident. The reproduction data RD (corrected reproduction data ARD) may also be configured to further include position data of one or more objects (traffic lights, crosswalks, center lines, guardrails, etc.) at the accident site where the accident occurred.

[0042] The specific configuration of the accident reconstruction unit 13 is not limited to this embodiment, but as an example, the accident reconstruction unit 13 may be configured to include a simulation execution unit 131 and a correction unit 132, as shown in Fig. 1. Here, the simulation execution unit 131 may be configured to generate reconstruction data RD of the accident occurrence situation by a simulation with reference to the numerical data ND. Furthermore, the correction unit 132 may be configured to correct the reconstruction data RD with reference to the numerical data ND and GNSS data as the position data PD.

[0043] (Report Writing Section 14) The report creation unit (report creation means) 14 creates a report (accident report CR) on the accident occurrence circumstances based on the accident occurrence circumstances recreated by the accident reconstruction unit 13. As an example, the report creation unit 14 creates a report (accident report CR) including sentences on the accident occurrence circumstances using a language model LL different from the learning model ML used by the above-mentioned data conversion unit 12. The accident report CR created by the report creation unit 14 is stored in the memory unit 20, as an example, and is transmitted to the terminal device 60 via the transmission unit 15. Specific examples of the accident report CR will be described later.

[0044] As described above, the accident analysis system 1 according to this embodiment is an accident analysis system that analyzes the accident data CD, The accident data CD is binary data recorded in an event data recorder (EDR) mounted on one or more target vehicles 50 when an accident involving the target vehicles 50 occurs, a data receiving unit 11 that receives the binary accident data CD recorded in the EDR and the position data PD of the target vehicle 50 via a wireless communication network N; a data conversion unit 12 that converts the received binary format accident data CD into numerical data ND using a learning model ML that has learned a data format that differs for each automobile manufacturer; an accident reconstruction unit 13 that reconstructs the accident occurrence situation of the target vehicle 50 from the converted numerical data ND and the received position data PD; It is equipped with:

[0045] In this way, the accident analysis system 1 according to this embodiment uses the learning model ML that has learned different data formats for each automobile manufacturer to convert the binary format accident data CD into numerical data ND, thereby reducing the effort required to analyze an accident. Furthermore, the numerical data ND obtained by the conversion and the position data PD of the target vehicle 50 are referenced to reproduce the accident occurrence circumstances of the target vehicle 50, allowing the accident to be reproduced more accurately.

[0046] (Processing flow in accident analysis system 1) FIG. 2 is a flow diagram showing an example of the flow of processing in the accident analysis system 1.

[0047] (Step S11) In step S11, the data receiving unit (data receiving means) 11 receives the binary accident data CD recorded in the event data recorder EDR provided in the target vehicle 50 and the position data PD of the target vehicle 50 via the wireless communication network N. The left side of FIG. 3 shows accident data (binary data) CD1, which is an example of the accident data CD received by the data receiving unit 11 in this step. Here, the accident data CD1 is binary data representing the engine speed of the target vehicle 50. As shown in FIG. 3, the accident data CD1 is data recorded at 0.01 second intervals.

[0048] The left side of Fig. 4 shows accident data (binary data) CD2, which is another example of the accident data CD received by the data receiving unit 11 in this step. Here, the accident data CD2 is binary data that represents the steering angle of the target vehicle 50. As shown in Fig. 4, the accident data CD2 is also data recorded at 0.01 second intervals.

[0049] (Step S12) Next, in step S12, the data conversion unit (data conversion means) 12 converts the binary accident data CD received by the data receiving unit 11 into numerical data ND using a learning model (machine-learned model) ML that has learned a data format that differs for each automobile manufacturer. The right side of Figure 3 shows numerical data ND1, which is an example of the numerical data ND obtained by the conversion by the data conversion unit 12 in this step. Here, the numerical data ND1 is decimal numerical data representing the engine speed of the target vehicle 50, obtained by converting the binary data CD1 representing the engine speed. As shown in Figure 3, the numerical data ND1 is also data at 0.01 second intervals.

[0050] The right side of Fig. 4 shows numeric data ND2, which is another example of the numeric data ND obtained by the conversion by the data conversion unit 12 in this step. Here, the numeric data ND2 is decimal numeric data representing the steering angle of the target vehicle 50, obtained by converting the binary data CD2 representing the steering angle. As shown in Fig. 4, the numeric data ND2 is also data at 0.01 second intervals.

[0051] (Step S13) Subsequently, in step S13, the accident reconstructing unit (accident reconstructing means) 13 reconstructs, for example, the accident occurrence situation of the target vehicle 50 from the numerical data ND converted by the data converting unit 12 and the position data PD. As described above, the accident reconstructing unit 13 performs the process of reconstructing the accident occurrence situation. A process of generating reproduction data RD of the accident occurrence situation by referring to the numerical data ND. A process of generating corrected reproduction data ARD by correcting the reproduction data RD with reference to the position data. As described above, the reproduction data RD (corrected reproduction data ARD) may include, for example, data on the trajectories (position, speed, direction, etc. at each time) of one or more vehicles related to the accident at the time of the accident and for a predetermined period before and after the accident. The reproduction data RD (corrected reproduction data ARD) may further include at least one of one or more objects (traffic lights, crosswalks, center lines, guardrails, etc.) at the accident site where the accident occurred and position data of the objects.

[0052] The upper part of Fig. 5 shows an example of the reproduction data RD (corrected reproduction data ARD) generated by the accident reproduction unit 13 in this step. As shown in the upper part of Fig. 5, the reproduction data RD (corrected reproduction data ARD) includes the following: The trajectory of the target vehicle 50A related to the accident at the time of the accident and a predetermined period before and after the accident is represented by 50A (t=0.00), 50A (t=2.00), and 50A (t=4.00). The direction (azimuth) of the vehicle 50A at each of the above positions As shown in the upper part of FIG. 5, the reproduced data RD (corrected reproduced data ARD) includes the following: The position of the vehicle 50B indicated by 50B(t=4.00) as part of the trajectory of the subject vehicle 50B related to the accident at the time of the accident and for a predetermined period before and after the accident. The direction (azimuth) of the vehicle 50B at the above time As shown in the upper part of FIG. 5, the reproduced data RD (corrected reproduced data ARD) contains the following: One or more objects at the accident scene where the accident occurred include a crosswalk, a center line, etc.

[0053] (Step S14) Next, in step S14, the report creation unit (report creation means) 14 creates a report (accident report CR) on the accident occurrence situation based on the accident occurrence situation recreated by the accident reconstruction unit 13. As an example, the report creation unit 14 creates a report (accident report CR) including sentences on the accident occurrence situation using a language model LL different from the learning model ML used by the data conversion unit 12 described above.

[0054] The lower part of Fig. 5 shows an example of the accident report CR created by the report creation unit 14 in this step. As shown in the lower part of Fig. 5, the accident report CR includes, for example, Date and time of the accident ·Accident classification Analysis results based on EDR data and location data - Estimated percentage of fault The information is generated using the language model LL with reference to the accident occurrence circumstances recreated by the accident reconstruction unit 13. The accident report CR is transmitted to the terminal device 60, for example, and viewed by the user via the input / output unit 63. More specific processing by the report creation unit 14 will be described later.

[0055] In addition, as shown in the bottom of Figure 5, the accident report CR includes the following: -UI button BT1 for viewing accident reconstruction data (for example, corrected reconstruction data ASD) UI button BT2 to view more detailed reports based on EDR and location data UI button BT3 for viewing data referenced by the accident reconstruction department 13 and the report creation department 14 UI button BT4 to view similar accident cases -UI button BT5 for inputting feedback from users on accident report CR When the user presses the UI button BT1, for example, the transmission unit 15 transmits the accident reproduction data shown in the upper part of FIG.

[0056] Furthermore, when the user presses the UI button BT2, for example, the transmission unit 15 transmits a more detailed accident report created by the report creation unit 14 to the terminal device 60 and presents it to the user via the input / output unit 63.

[0057] Also, when the user presses the UI button BT3, as an example, the transmission unit 15 transmits the data referenced by the accident reconstruction unit 13 and the report creation unit 14 to the terminal device 60 and presents it to the user via the input / output unit 63.

[0058] Also, when a user presses the UI button BT4, as an example, the transmission unit 15 acquires data relating to accident cases similar to the accident in question from the memory unit 20, transmits the acquired data to the terminal device 60, and presents it to the user via the input / output unit 63.

[0059] Furthermore, when the user selects the UI button BT5 to input feedback, the accident reconstruction unit 13 or the report creation unit 14 refers to the feedback and corrects the accident occurrence situation or the accident report CR.

[0060] (An example of a specific configuration of the accident reconstruction unit 13) 6 is a block diagram showing an example of a specific configuration of the accident reconstruction unit 13 included in the accident analysis system 1. As shown in FIG. 6, the accident reconstruction unit 13 according to this configuration example includes an acquisition unit 130, a simulation execution unit 131, a correction unit 132, and a storage unit 133.

[0061] FIG. 6 also shows a more specific example of some of the data stored in the storage unit 20. As shown in FIG. 20, the data stored in the storage unit 20 can be broadly divided into a target data group and a reference data group. The target data group can be expressed as data related to the target accident that the accident reconstruction unit 13 is attempting to reconstruct. On the other hand, the reference data group can be expressed as past data that is referenced to reconstruct the target accident. In the following description, the subscript T may be added to data included in the target data group, and the subscript R may be added to data included in the reference data group.

[0062] (Acquisition part 130) The acquisition unit (acquisition means) 130 Numerical data ND_T obtained by converting accident data CR acquired from the event data recorder EDR of the target vehicle 50 related to the target accident Position data PD_T acquired from the position identification unit 52 of the target vehicle 50 is acquired. Here, as described above, the accident data CR_T is data recorded in the event data recorder EDR mounted on one or more target vehicles 50 when an accident involving the target vehicle 50 occurs. Also, as described above, the numerical data ND_T is decimal data obtained by converting the accident data CR by the data conversion unit 12. The numerical data ND_T is time-series data made up of data at each time t. The numerical data ND_T (in other words, the time series of the numerical data ND_T) may be expressed as X(t).

[0063] (Simulation execution unit 131) The simulation execution unit (simulation execution means) 131 executes a simulation with reference to the numerical data X(t) to generate reproduction data RD_T of the accident occurrence situation of the target vehicle 50. Here, the reproduction data RD_T is made up of data at each time t, and the reproduction data RD_T may also be expressed as Y(t). Although a specific example of the simulation executed by the simulation execution unit 131 is not limited to this example, one example is a simulation using a physical model with reference to the numerical data X(t).

[0064] (correction unit 132) The correction unit (correction means) 132 generates corrected reproduction data ASD by correcting the reproduction data RD with at least reference to the numerical data ND_T. The correction unit (correction means) 132 may also be configured to generate corrected reproduction data ASD by correcting the reproduction data RD with further reference to position data PD_T. As shown in FIG. 6 , the correction unit 132 according to this example includes a waveform generator 1321, a graph generator 1322, a correction data generator 1323, a correction data regenerator 1324, and a corrected reproduction data generator 1325.

[0065] (Waveforming unit 1321) The waveform generator 1321 generates a waveform from the numerical data X(t), which is time-series data, using a first model. As an example, the waveform generator 1321 may be configured to include the first model. However, this is not intended to limit the present example. Alternatively, one or more servers included in the information processing system 100 may include the first model, and the waveform generator 1321 may provide the numerical data X(t) to the first model via the transmitter 15.

[0066] The processing by the waveform generator 1321 can also be expressed as a process of extracting data in X(t) in time series using the first model (feature extraction process). Changes in acceleration over time Whether or not the brakes were applied at each point The waveform data WD_T generated by the waveform generator 1321 includes data or feature extraction related to the behavior of the target vehicle 50, as shown in the figure. The waveform data WD_T generated by the waveform generator 1321 is, for example, a waveform in which vehicle speed and acceleration are arranged as waveforms with "horizontal axis: time, vertical axis: numerical value." The waveform data WD_T can be used to categorize the "movements" of the vehicle. In other words, the waveform data WD_T generated by the waveform generator 1321 can be used as tag data for categorizing the "movements" of the vehicle. While the specific configuration of the first model is not limited to this example, it may include, for example, a one-dimensional convolutional network (Conv1D). The term "waveform generation" in the above description does not limit this embodiment. The "waveform generation" described above may also be expressed as "profiling," "graphing," or "visual data generation."

[0067] (Graphing unit 1322) The graphing unit 1322 graphs the position data PD_T (GNSS data), which is time-series data, using a second model. As an example, the graphing unit 1322 can be configured to include the second model. However, this is not intended to limit the present example. The graphing unit 1322 may be configured such that one or more servers included in the information processing system 100 include the second model, and the graphing unit 1322 provides the position data PD_T to the second model via the transmission unit 15.

[0068] The process by the graphing unit 1322 can also be expressed as a process of graphing the position data PD_T and converting it into numerical features (graph embedding). As an example of such a conversion process, Sharp curves ·Height difference The graph data GD_T generated by the graphing unit 1322 includes data related to the longitude of the target vehicle 50 and conversion into numerical features, as shown in the figure. The graph data GD_T generated by the graphing unit 1322 includes nodes (points, positioning points) and edges (lines, roads), and the graph can categorize "what route the vehicle took." In other words, the graph data GD_T generated by the graphing unit 1322 can be used as tag data for categorizing "what route the vehicle took." The specific configuration of the second model is not limited to this example, but can include a graph attention network (GAT), as an example.

[0069] (Correction data generation unit 1323) The correction data generation unit 1323 corrects the reproduction data Y(t) by referring to the waveform data WD_T generated by the waveform generation unit 1321 and the graph data GD_T generated by the graph generation unit 1322. In other words, the correction data generation unit 1323 generates correction data AD_T for correcting the reproduction data Y(t) generated by the simulation execution unit 131 by referring to the waveform data WD_T generated by the waveform generation unit 1321 and the graph data GD_T generated by the graph generation unit 1322. The correction data AD_T may also be expressed as ΔY(t). The processing by the correction data generation unit 1323 can also be expressed as processing (residual estimation) that estimates a deviation ΔY(t) from the reproduction data Y(t) based on the waveform data WD_T and the graph data GD_T.

[0070] The correction data generation unit 1323 may be configured to use at least one of the input waveform data WD_T and graph data GD_T as tag data to identify data similar to at least one of the input waveform data WD_T and graph data GD_T in the reference data group, and then use ΔY(t) associated with the similar data as ΔY(t) corresponding to the input waveform data WD_T and graph data GD_T.

[0071] Furthermore, the correction data generating unit 1323 may be configured to use a third model different from both the first model and the second model to generate the correction data ΔY(t). As an example, the correction data generating unit 1323 may input the waveform data WD_T generated by the waveform generating unit 1321 and the graph data GD_T generated by the graph generating unit 1322 to the third model, and use the data output by the third model as the correction data ΔY(t). Here, the specific configuration of the third model is not limited to this example, but as an example, a neural network including a Transformer can be used.

[0072] The third model may be configured to be trained in advance using training data including waveform data, graph data, and correction data. The third model may be trained using various data included in the reference data group, in other words, various data referenced or derived in past analyses by the accident reconstruction unit 13: Numerical data ND_R Position data PD_R Waveform data WD_R Graph data GD_R Reproduction data RD_R Correction data AD_R Corrected reproduction data ARD_R The third model may be configured to be sequentially learned (updated) by referring to the laws of physics. Here, the meaning of each piece of data included in the reference data group is the same as that of each piece of data included in the target data group described above. The third model may also be configured to be sequentially learned (updated) by using the results of processing by the correction data regeneration unit 1324 (described later) that refers to the laws of physics.

[0073] (Correction data regeneration unit 1324) The correction data regeneration unit 1324 regenerates the correction data AD_T by referring to the correction data ΔY(t) generated by the correction data generation unit 1323 and taking physical laws into consideration. The processing by the correction data regeneration unit 1324 can also be expressed as processing for regenerating the correction data AD_T by referring to physical laws. The correction data generated (regenerated) by the correction data regeneration unit 1324 is also referred to as ΔY(t)'.

[0074] The correction data regeneration unit 1324 regenerates various data included in the reference data group, in other words, various data referenced or derived in past analyses by the accident reconstruction unit 13: Numerical data ND_R Position data PD_R Waveform data WD_R Graph data GD_R Reproduction data RD_R Correction data AD_R Corrected reproduction data ARD_R The correction data may be regenerated by further referring to the reference data set. Furthermore, the processing by the correction data regeneration unit 1324 may include a process of causing the correction data generation unit 1323 (third model) to learn. Therefore, the processing by the correction data regeneration unit 1324 may be expressed as a process of causing the correction data generation unit 1323 (third model) to learn (learning calculation) in consideration of the laws of physics, or in consideration of the laws of physics and the reference data group. Alternatively, the processing by the correction data regeneration unit 1324 may be expressed as a process of confirming or ensuring the physics consistency of the correction data generation unit 1323 (third model).

[0075] In this configuration example, the correction data generation unit 1323 and the correction data regeneration unit 1324 are described as separate blocks, but this does not limit the present example. The processes performed by the correction data generation unit 1323 and the correction data regeneration unit 1324 can be collectively expressed as follows: Waveform data WD_T generated by the waveform generator 1321 The graph data GD_T generated by the graphing unit 1322 The correction data (ΔY(t), ΔY(t)') is generated by referring to the above. The model (for example, the third model) used to generate (regenerate) the correction data is - A model that takes into account physical laws such as equations of motion (a model that incorporates physical laws) A model that is sequentially trained (updated) by referring to past data contained in the reference data set. It is also possible to adopt a configuration in which the processing by the correction data regeneration unit 1324 is omitted, that is, a configuration that does not include the correction data regeneration unit 1324, and such a configuration is also included in this embodiment.

[0076] (Corrected reproduction data generation unit 1325) The corrected reproduction data generation unit 1325 generates the corrected reproduction data ARD_T by referring to the correction data ΔY(t)′ generated by the correction data regeneration unit 1324 and the reproduction data Y(t) generated by the simulation execution unit 131. The corrected reproduction data ARD_T may be expressed as ^Y(t) (Y(t) with a hat). As an example, the corrected reproduction data generation unit 1325 generates the corrected reproduction data ^Y(t) by ^Y(t) = Y(t)+ΔY(t)' Generated by:

[0077] The corrected reproduction data ^Y(t)(ARD_T) generated by the corrected reproduction data generating unit 1325 is stored in the storage unit 20 by the accumulation unit 133. In addition, the accumulation unit 133 stores each data derived in the above-mentioned processing. Waveform data WD_T Graph data RD_T Reproduced data RD_T(Y(t)) Correction data AD_T(ΔY(t), ΔY(t)') Corrected reproduction data ARD_T is stored in the memory unit 20. The above data stored by the accumulation unit 133 will be referred to as a reference data group in future processing by the accident reconstruction unit 13 or in learning of the accident reconstruction unit 13 (for example, learning of a third model).

[0078] In this way, the accident reconstruction unit 13 a storage unit (storage unit 133) for storing the reproduction data RD_T(Y(t)), the waveform data WD_T, the graph data GD_T, and the correction data AD_T(ΔY(t), ΔY(t)′); a learning means for executing a learning process for the third model by referring to the data stored in the storage means; Here, the learning means may be, for example, the correction data regeneration unit 1324 or the correction data generation unit 1323. Alternatively, a learning unit that functions as the learning means may be provided separately from the correction data regeneration unit 1324 and the correction data generation unit 1323.

[0079] Note that the processing by the learning means according to this embodiment is not limited to the above-described examples. As an example, the learning means may perform processing to update a portion of the data included in the target data group and the reference data group based on other data or physical laws so as to improve the quality of the portion of the data (for example, the quality as training data). For example, these processing may include deleting data points that do not conform to physical laws, deleting outliers, etc.

[0080] (Effects of this configuration example) As described above, the accident reconstruction unit 13 according to this configuration example: a simulation execution unit 131 that generates reproduction data RD_T(Y(t)) of the accident occurrence situation of the target vehicle 50 by simulation based on the numerical data ND_T(X(t)); and A correction unit 132 that corrects the reproduction data RD_T(Y(t)) by referring to the numerical data ND_T(X(t)) and the position data PD_T. This allows for accurate reproduction of the accident situation.

[0081] Further, the correction unit 132, for example, The first model converts the time series of numerical data ND_T(X(t)) into a waveform, and A second model for graphing GNSS data as position data PD_T Equipped with A process of correcting the reproduction data RD_T(Y(t)) by referring to the waveform data WD_T obtained by the first model and the graph data GD_T obtained by the second model (processing by the correction data generation unit 1323, the correction data regeneration unit 1324, and the corrected reproduction data generation unit 1325). The correction process may be configured to execute the following. A third model is used in the correction process. The correction process may also refer to the laws of physics.

[0082] According to the above configuration, it is possible to execute a more suitable correction process than a configuration in which the reproduction data RD_T(Y(t)) is corrected by directly referring to the numerical data ND_T(X(t)). In other words, it is possible to improve the accuracy of the correction. Also, as partly described above, as the third model, Physical laws such as equations of motion - Historical data included in the reference data set Therefore, it is possible to further improve the accuracy of correction.

[0083] As described above, the numerical data ND_T is data at 0.01 second intervals. In conventional technology, even if event data (accident data) is recorded in the EDR at 0.01 second intervals, it is common to read the data at intervals longer than 0.01 second (for example, 0.25 second intervals) and use the data for analysis. In this embodiment, analysis is performed using the numerical data ND_T at 0.01 second intervals, so that analysis can be performed with higher accuracy than in the past.

[0084] (Accident reconstruction process flow) Next, the flow of the accident reenactment processing by the accident reenactment unit 13 will be described with reference to Fig. 7. Fig. 7 is a flow chart showing an example of the flow of the accident reenactment processing by the accident reenactment unit 13.

[0085] (Step S130) First, in step S130, the acquisition unit 130 Numerical data ND_T(X(t)) obtained by converting accident data CR acquired from the event data recorder EDR of the target vehicle 50 related to the target accident Position data PD_T acquired from the position identification unit 52 of the target vehicle 50 The process performed by the acquisition unit 130 has been described above, so a duplicated description will be omitted.

[0086] (Step S131) Next, in step S131, the simulation execution unit 131 executes a simulation with reference to the numerical data X(t) to generate reproduction data RD_T(Y(t)) of the accident occurrence situation of the target vehicle 50. The processing by the simulation execution unit 131 has been described above, so a duplicated description will be omitted.

[0087] (Step S1321) On the other hand, in step S1321, the waveform generator 1321 generates a waveform from the numerical data X(t), which is time-series data, using the first model. The processing by the waveform generator 1321 has been described above, so a duplicated description will be omitted.

[0088] (Step S1322) In step S1322, the graphing unit 1322 graphs the position data PD_T (GNSS data), which is time-series data, using the second model. The processing by the graphing unit 1322 has been described above, so a duplicated description will be omitted.

[0089] (Step S1323) Subsequently, in step S1323, the correction data generation unit 1323 corrects the reproduction data Y(t) by referring to the waveform data WD_T generated by the waveform generation unit 1321 and the graph data GD_T generated by the graph generation unit 1322. In other words, in step S1323, the correction data generation unit 1323 generates correction data AD_T for correcting the reproduction data Y(t) generated by the simulation execution unit 131 by referring to the waveform data WD_T generated by the waveform generation unit 1321 and the graph data GD_T generated by the graph generation unit 1322. Specific processing by the correction data generation unit 1323 has been described above, and therefore a repeated description will be omitted.

[0090] (Step S1324) Subsequently, in step S1324, the correction data regeneration unit 1324 regenerates the correction data AD_T by referring to the correction data ΔY(t) generated by the correction data generation unit 1323 and taking the laws of physics into consideration. The processing by the correction data regeneration unit 1324 can also be expressed as processing for regenerating the correction data AD_T by referring to the laws of physics. The specific processing by the correction data regeneration unit 1324 has been described above, so a duplicated description will be omitted.

[0091] (Step S1325) Next, in step S1325, the corrected reproduction data generation unit 1325 generates corrected reproduction data ARD_T as a process for reproducing the accident occurrence situation by referring to the correction data ΔY(t)′ generated by the correction data regeneration unit 1324 and the reproduction data Y(t) generated by the simulation execution unit 131. The process by the corrected reproduction data generation unit 1325 has been described above, so a duplicated description will be omitted.

[0092] Note that each of the processes of steps S1321 to S1325 may be configured to be executed in parallel with step S131, or may be configured to be executed prior to step S131 or after step S131. The accident reconstruction unit 13 achieves each of the above-mentioned effects by executing the above-mentioned processing flow as an example.

[0093] (Another example of the specific configuration of the accident reconstruction unit 13) FIG. 8 is a block diagram showing another example of the specific configuration of the accident reconstruction unit 13 provided in the accident analysis system 1. As shown in FIG. 8, the accident reconstruction unit 13A according to this configuration example has a configuration similar to that shown in FIG. 6. However, this example differs from the configuration example of FIG. 6 in that the target data group and the reference data group include additional data DD_T and additional data DD_R, respectively. Also, this example differs from the configuration example of FIG. 6 in that the acquisition unit 130 also acquires the additional data DD_T. The following will mainly explain the differences from the configuration example of FIG. 6, and similar explanations will be omitted as appropriate.

[0094] (Acquisition part 130) The acquisition unit 130 Numerical data ND_T obtained by converting accident data CR acquired from the event data recorder EDR of the target vehicle 50 related to the target accident Position data PD_T acquired from the position identification unit 52 of the target vehicle 50 In addition, Additional data DD_T Here, the additional data DD_T refers to data different from the numerical data ND_T and the position data PD_T. The term "additional" simply refers to something different from the numerical data ND_T and the position data PD_T, and this term does not limit the present embodiment.

[0095] More specifically, the additional data DD_T is -Various data obtained by investigating the accident site after the accident Various sensing data obtained by various sensors equipped in the target vehicle 50 In this example, the numeric data ND_T and the additional data DD_T are collectively represented as numeric data X(t).

[0096] (Waveforming unit 1321) The waveform generator 1321 generates a waveform from the numerical data X(t) including the additional data DD_T using the first model. Here, except that the numerical data X(t) includes the additional data DD_T, the configuration is the same as that shown in Fig. 6, and therefore a redundant description of the processing by the waveform generator 1321 will be omitted.

[0097] According to the accident reconstruction unit 13A according to the above configuration example, the accident reconstruction process is executed by further referring to the additional data DD_T, and therefore the accuracy of the accident reconstruction can be further improved.

[0098] (Notes regarding teacher assets (teacher data)) As described above, in the accident reconstruction section 13 (13A), Extracted data (waveform data WD) that graphs (waveforms) EDR numerical data ND every 0.01 seconds ·Extracted data (graph data GD) that numerically characterizes (graphs) the position data (GNSS data) PD The waveform data WD is stored and referenced as teacher data (teacher assets) in the learning of the accident reconstruction unit 13 (13A) (for example, learning of the third model). As described above, the waveform data WD is a waveform in which all data items obtainable from the EDR are arranged in 0.01-second increments, with the horizontal axis representing time and the vertical axis representing numerical values. The graph data GD is composed of nodes (points, positioning points) and edges (lines, roads).

[0099] By storing the above data as a reference data group and using it for learning in the accident reconstruction unit 13 (13A), it becomes possible to manage the location and type of accident by associating it with the EDR waveform data WD and the GNSS feature data (graph data). Furthermore, as partially described above, by further associating the correction data (ΔY(t), ΔY(t)') and data derived in the calculation process for calculating the correction data, more suitable learning processing can be performed by referring to the reference data group. For example, the third model can be used to appropriately infer that the EDR event data (accident data) indicates a speed of 40 km / h, but the actual speed drops by approximately 5 km / h at the sharp curve.

[0100] (Additional notes regarding the effects of Accident Analysis System 1) As described above, the data conversion unit 12 and accident reconstruction unit 13 according to this embodiment perform a process of graphing and extracting EDR numerical data every 0.01 seconds. In conventional technology, it is common to read out the data at intervals longer than 0.01 seconds (for example, every 0.25 seconds) and use it for analysis. In this embodiment, analysis is performed using numerical data ND at 0.01 second intervals, which has the advantage of enabling more accurate analysis than conventional techniques, as described above.

[0101] The advantages of the processing related to the effects of the accident analysis system 1 will be described more specifically below. As an example, suppose that the speed graph indicated by the waveform data WD generated at 0.01-second intervals by the waveform generator 1321 includes a smoothly descending slope from 40 km / h to 35 km / h over approximately 0.2 seconds. In other words, suppose that the waveform generator 1321 detects a smoothly descending slope from 40 km / h to 35 km / h over approximately 0.2 seconds in the speed graph generated from the numerical data ND.

[0102] On the other hand, in the reference data group, the curvature is 0.15m -1 Assume that there are multiple cases in which graph data GD including the curve is associated with past data including the slope described above, and that in these cases, an average deviation of -5 km / h (corrected data ΔY(t), ΔY(t)') is associated.

[0103] In such a case, the third model trained by referring to the reference data group is generated by using waveform data WD including the slope and a curvature of 0.15 m. -1 When the graph data GD including the curve is input, correction data ΔY(t)=−5 is calculated for the reproduction data Y(t)=40.

[0104] As another example, suppose that the waveform data WD generated by the waveform generator 1321 at 0.01-second intervals includes a waveform in which the yaw rate increases slightly from 0.3 rad / s to 0.6 rad / s while the vertical acceleration remains unchanged. In addition, suppose that such a waveform is associated in the reference data group with cases in which the steering wheel was turned further due to tire slip, or cases in which speed recovery was delayed as a result.

[0105] In such a case, when the above waveform is input, the third model, which has been trained by referring to the reference data group, calculates correction data ΔY(t) for the reproduction data Y(t) taking into account the delay in speed recovery.

[0106] As shown in these examples, the accident reconstruction unit 13 according to this embodiment generates waveform data at short time intervals from EDR event data, accumulates it, and uses it for inference, thereby enabling more accurate inference to be performed than simulation results (reconstruction data RD) that refer only to EDR data, thereby enabling accident reconstruction to be performed with higher accuracy.

[0107] (Accident report creation process flow) Next, the processing by the report creation unit 14 included in the accident analysis system 1 will be described with reference to Fig. 9. Fig. 9 is a flow chart showing an example of the flow of processing by the report creation unit 14.

[0108] (Step S141) First, in step S141, the report creation unit 14 acquires the corrected reproduction data ARD (ARD_T, ARD_R) generated by the accident reconstruction unit 13. Note that the report creation unit 14 can execute each process described below even if it acquires the uncorrected reproduction data RD (RD_T, RD_R) instead of the corrected reproduction data ARD (ARD_T, ARD_R). This embodiment also includes such a configuration.

[0109] (Step S142) Next, in step S142, the report creation unit 14 semantically creates the trajectory of the target vehicle 50 included in the corrected reproduction data ARD. As an example, the report creation unit 14 assigns meaning related to traffic infrastructure information to the trajectory (position (x coordinate, y coordinate, z coordinate), speed, direction, etc.) of the target vehicle 50 included in the corrected reproduction data ARD. More specifically, the report creation unit 14 performs a process of associating traffic signal log information, lane information, etc. with the trajectory (or each part of the trajectory) of the target vehicle 50.

[0110] (Step S143) Subsequently, in step S143, the report creation unit 14 labels the behavioral data with reference to the trajectory semantized in step S142. As an example, the report creation unit 14 associates the behavioral data with the trajectory (or each part of the trajectory) semantized in step S142. As an example, the report creation unit 14 associates (labels) behavioral data such as "entered an intersection when the light was red" with the trajectory semantized in step S142.

[0111] (Step S144) Subsequently, in step S144, the report creation unit 14 performs a violation determination by referring to the trajectory labeled with the behavioral data in step S143. As an example, the report creation unit 14 determines whether the trajectory labeled with the behavioral data in step S143 includes any traffic violation behavior such as running a red light, driving the wrong way, or speeding. Here, the report creation unit 14 may be configured to perform the determination process by referring to preset traffic violation information (also referred to as a traffic violation asset).

[0112] (Step S145) Next, in step S145, the report creation unit 14 converts the trajectory referenced in step S144 and the result of the violation determination made in step S144 into data in a predetermined format. As an example, the report creation unit 14 converts this information into data in JSON format.

[0113] (Step S146) Next, in step S146, the report creation unit 14 performs a process of converting the data in the JSON format into a sentence (a process of generating a sentence). As an example, the report creation unit 14 converts the data in the JSON format into the following sentence: Vehicle A entered the intersection at a red light. Vehicle B was traveling at 72km / h, despite the speed limit being 50km / h. A sentence such as the above is generated in bullet points using a template. The processing in this step may be configured to use a language model LL.

[0114] The report creation unit 14, for example, -The above JSON format data - Instructions to convert the above data into bulleted sentences The above conversion (generation) process can be performed by inputting a prompt including the above into the language model LL.

[0115] (Step S147) Subsequently, in step S147, the report creation unit 14 Cases similar to the target accident identified by the converted sentences in step S146 Cases similar to the subject accident in which negligence was determined As an example, the report creation unit 14 can execute the search process by RAG (Retrieval Augmented Generation) processing using a language model LL.

[0116] The report creation unit 14, for example, The sentence converted (generated) in step S146 - Instructions to search for cases similar to the target accident identified by the above sentence, or cases in which negligence was determined among such cases The above search process can be performed by inputting a prompt including the following into the language model LL. Note that in this step, the information source in which the above search process is performed (in other words, how the search target is set) is not limited to this example, but as an example, Driving school teaching materials ·Case Law Collection Accident insurance settlement agreement The above-mentioned case may be searched for from among these information sources. In this case, the above-mentioned prompt may include information such as the contents of these information sources or the storage location of the information sources.

[0117] (Step S148) Next, in step S148, an accident report CR is generated by referring to the sentence converted in step S146, the case searched in step S147, etc. The report creation unit 14 can execute this generation process using a language model LL. The accident report CR generated in this step may include, for example, -Which of the multiple accident categories does the target accident fall into? - Objective evaluation of the target accident Includes:

[0118] The report creation unit 14, for example, The sentence converted (generated) in step S146 and the target incident identified thereby The cases searched in step S147 - Instructions to prepare a report on the target accident by referring to the above case The generation process can be performed by inputting a prompt including the following into the language model LL:

[0119] (Step S15) Subsequently, in step S15, the report creation unit 14 transmits the accident report CR generated in step S148 to the user of the accident analysis system 1 via the transmission unit 15. As an example, the report creation unit 14 transmits the accident report CR to the terminal device 60.

[0120] 10 shows an example of a display screen including the accident report CR generated in step S148 and displayed on the input / output unit 53 of the terminal device 60. The entire display screen can also be configured to be created by the report creation unit 14.

[0121] In the example shown in FIG. 10, the display screen displays: GUI (Graphical User Interface) element E101 for entering queries about accident data GUI button E102 to generate an accident report CR A display area E103 in which information about the owner or insured of a certain target vehicle (vehicle A) and information about the owner or insured of another target vehicle (vehicle B) are displayed. A display area E104 in which the corrected reconstruction data ARD generated by the accident reconstruction unit 13 is displayed. A display area E105 in which the accident report CR created by the report creation unit 14 is displayed. GUI button group E106 for receiving instructions related to the accident report CR from the user Display area E201 showing the date and area of ​​the accident Map of the area surrounding the accident site (accident area) E202 GUI buttons E203 for receiving more specific instructions regarding the target accident The report creation unit 14 can be configured to create an accident report CR that includes some or all of the above elements E103 to E203.

[0122] In addition, when the user presses any of the GUI buttons included in the GUI button group E106, the accident analysis system 1 identifies the user instruction associated with that GUI button, and the report creation unit 14 executes processing based on that instruction.

[0123] Furthermore, when a user presses any of the GUI buttons included in the GUI button group E203, the accident analysis system 1 identifies the user instruction associated with that GUI button, and at least one of the accident reconstruction unit 13 and the report creation unit 14 executes processing based on that instruction.

[0124] The report creation unit 14 that performs the above-described processing can create an appropriate accident report CR based on the accident occurrence circumstances recreated by the accident reconstruction unit 13, and present it to the user.

[0125] [Software implementation example] The functions of the accident analysis system 1 or one or more devices (hereinafter referred to as "devices") that have each function of the system may be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (in particular, the data receiving unit 11, the data conversion unit 12, the accident reconstruction unit 13, the report creation unit 14, and the transmission unit 15).

[0126] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above embodiments.

[0127] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0128] In addition, some or all of the functions of each of the control blocks can be realized by logic circuits. For example, integrated circuits in which logic circuits that function as each of the control blocks are formed are also included in the scope of the present disclosure. In addition, the functions of each of the control blocks can also be realized by, for example, a quantum computer.

[0129] Furthermore, each process described in each of the above embodiments may be executed by AI (Artificial Intelligence). In this case, the AI ​​may run on the control device or on another device (for example, an edge computer or a cloud server).

[0130] <Summary> The present disclosure describes at least the following aspects.

[0131] (Configuration A1) An accident analysis system that analyzes accident data, The accident data is binary data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a data receiving means for receiving the binary accident data recorded in the EDR and the target vehicle's position data via a wireless communication network; a data conversion means for converting the received binary format accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; an accident reproducing means for reproducing the accident occurrence situation of the target vehicle from the converted numerical data and the received position data; An accident analysis system equipped with:

[0132] (Configuration A2) The accident reproduction means a simulation execution means for generating reproduction data of the accident occurrence situation by simulation based on the numerical data; a correction means for correcting the reproduced data by referring to the numerical data and GNSS data as the position data; The accident analysis system according to configuration A1,

[0133] (Configuration A3) The correction means a first model that converts the time series of the numerical data into a waveform; a second model for graphing the GNSS data; Equipped with The reproduced data is corrected by referring to the waveform data obtained by the first model and the graph data obtained by the second model. The accident analysis system according to configuration A2.

[0134] (Configuration A4) a report creation means for creating a report on the accident occurrence circumstances based on the accident occurrence circumstances recreated by said accident reproduction means; The accident analysis system according to any one of configurations A1 to A3, further comprising:

[0135] (Configuration A5) The report creation means The report including the sentences regarding the accident occurrence situation is created using a language model different from the learning model used by the data conversion means. The accident analysis system according to configuration A4.

[0136] (Configuration B1) 1. An accident analysis system, comprising: an acquisition means for acquiring numerical data obtained by converting the accident data; The accident data is data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a simulation execution means for generating reproduction data of an accident occurrence situation of the target vehicle by simulation based on the numerical data; a correction means for correcting the reproduction data by at least referring to the numerical data; The accident analysis system further comprises:

[0137] (Configuration B2) The acquisition means further acquires position data of the target vehicle, The correcting means further refers to the position data to generate the corrected reproduction data. 10. The accident analysis system according to claim 1,

[0138] (Configuration B3) The correction means a first model that converts the time series of the numerical data into a waveform; a second model for graphing GNSS data as the position data; Equipped with The reproduced data is corrected by referring to the waveform data obtained by the first model and the graph data obtained by the second model. The accident analysis system according to claim B2.

[0139] (Configuration B4) The processing by the correction means includes: a process of generating correction data for the reproduction data by inputting the waveform data obtained by the first model and the graph data obtained by the second model into a third model; a process of generating the corrected reproduction data by referring to the reproduction data and the correction data; Contains The accident analysis system according to configuration B3.

[0140] (Configuration B5) The process by the correction means includes a process of regenerating the correction data by referring to the laws of physics. The accident analysis system according to configuration B4.

[0141] (Configuration B6) a storage means for storing the reproduction data, the waveform data, the graph data, and the correction data; a learning means for executing a learning process for the third model by referring to the data stored in the storage means; Equipped with The accident analysis system according to configuration B4 or B5.

[0142] (Configuration B7) The numerical data is data at 0.01 second intervals. The accident analysis system according to any one of the configurations B1 to B6.

[0143] (Configuration C1) 1. An accident analysis method executed by one or more processors for analyzing accident data, comprising: The accident data is binary data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a data receiving step of receiving the binary accident data recorded in the EDR and the target vehicle's position data via a wireless communication network; a data conversion step of converting the received binary format accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; an accident reproduction step of reproducing an accident occurrence situation of the target vehicle from the converted numerical data and the received position data; Accident analysis methods including:

[0144] (Configuration D1) 1. A method of accident analysis executed by one or more processors, comprising: an acquisition step of acquiring numerical data obtained by converting the accident data; The accident data is data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a simulation execution step of generating reproduction data of an accident occurrence situation of the target vehicle by simulation based on the numerical data; a correcting step of correcting the reproduction data by at least referring to the numerical data; The accident analysis method further includes:

[0145] (Configuration E1) An accident analysis device, an acquisition means for acquiring numerical data obtained by converting the accident data; The accident data is data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a simulation execution means for generating reproduction data of an accident occurrence situation of the target vehicle by simulation based on the numerical data; a correction means for correcting the reproduction data by at least referring to the numerical data; The accident analysis device further comprises:

[0146] (Configuration F1) A program for causing a computer to function as the accident analysis device according to configuration E1, the program causing a computer to function as the acquisition means, the simulation execution means, and the correction means.

[0147] The present disclosure is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present disclosure.

[0148] Furthermore, the present disclosure enables the analysis of accident data to be performed appropriately, which can contribute to the achievement of Goal 9 of the Sustainable Development Goals (SDGs), "Build resilient infrastructure, promote inclusive and sustainable industrialization, and promote innovation and resilience." [Explanation of symbols]

[0149] 100 ···Information Processing System 1. Accident analysis system 11 Data receiving section 12 Data conversion section 13 Accident Reenactment Section 131 Simulation execution unit 132 Correction unit 14 Report Writing Department 15 Transmitter

Claims

1. An accident analysis system that analyzes accident data, the accident data is binary data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a data receiving means for receiving the binary accident data recorded in the EDR and the target vehicle's position data via a wireless communication network; a data conversion means for converting the received binary format accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; an accident reproducing means for reproducing the accident occurrence situation of the target vehicle from the converted numerical data and the received position data; Equipped with The accident reproduction means a simulation execution means for generating reproduction data of the accident occurrence situation by simulation based on the numerical data; a correction means for correcting the reproduced data by referring to the numerical data and GNSS data as the position data; Equipped with The correction means a first model that converts the time series of the numerical data into a waveform; a second model for graphing the GNSS data; Equipped with correcting the reproduced data by referring to the waveform data obtained by the first model and the graph data obtained by the second model; The processing by the correction means includes: a process of generating correction data for the reproduction data by inputting the waveform data obtained by the first model and the graph data obtained by the second model into a third model; a process of generating the corrected reproduction data by referring to the reproduction data and the correction data; Contains Accident analysis system.

2. a report creation means for creating a report on the accident occurrence circumstances based on the accident occurrence circumstances recreated by said accident reproduction means; The accident analysis system of claim 1 further comprising:

3. The report creation means The report including the sentences regarding the accident occurrence situation is created using a language model different from the learning model used by the data conversion means. The accident analysis system according to claim 2 .

4. 1. An accident analysis system, comprising: an acquisition means for acquiring numerical data obtained by converting the accident data; The accident data is data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a simulation execution means for generating reproduction data of an accident occurrence situation of the target vehicle by simulation based on the numerical data; a correction means for correcting the reproduction data by at least referring to the numerical data; Further provided with The acquisition means further acquires position data of the target vehicle, The correction means further refers to the position data to generate the corrected reproduction data. The correction means a first model that converts the time series of the numerical data into a waveform; a second model for graphing GNSS data as the position data; Equipped with correcting the reproduced data by referring to the waveform data obtained by the first model and the graph data obtained by the second model; The processing by the correction means includes: a process of generating correction data for the reproduction data by inputting the waveform data obtained by the first model and the graph data obtained by the second model into a third model; a process of generating the corrected reproduction data by referring to the reproduction data and the correction data; Contains Accident analysis system.

5. The process by the correction means includes a process of regenerating the correction data by referring to the laws of physics. The accident analysis system according to claim 4 .

6. a storage means for storing the reproduction data, the waveform data, the graph data, and the correction data; a learning means for executing a learning process for the third model by referring to the data stored in the storage means; Equipped with The accident analysis system according to claim 4 .

7. The numerical data is data at 0.01 second intervals. The accident analysis system according to any one of claims 4 to 6.

8. 1. An accident analysis method executed by one or more processors for analyzing accident data, comprising: the accident data is binary data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a data receiving step of receiving the binary accident data recorded in the EDR and the target vehicle's position data via a wireless communication network; a data conversion step of converting the received binary format accident data into numerical data using a learning model that has learned a data format that differs for each automobile manufacturer; an accident reproduction step of reproducing an accident occurrence situation of the target vehicle from the converted numerical data and the received position data; Including, The accident reconstruction step includes: a simulation execution step of generating reproduction data of the accident occurrence situation by simulation based on the numerical data; a correcting step of correcting the reproduced data by referring to the numerical data and GNSS data as the position data; Including, The correction step a first model that converts the time series of the numerical data into a waveform; a second model for graphing the GNSS data; and performing processing using correcting the reproduced data by referring to the waveform data obtained by the first model and the graph data obtained by the second model; The process in the correction step includes: a process of generating correction data for the reproduction data by inputting the waveform data obtained by the first model and the graph data obtained by the second model into a third model; a process of generating the corrected reproduction data by referring to the reproduction data and the correction data; Contains Accident analysis method.

9. 1. A method for accident analysis executed by one or more processors, comprising: an acquisition step of acquiring numerical data obtained by converting the accident data; The accident data is data recorded in an event data recorder (EDR) mounted on one or more target vehicles when an accident involving the target vehicles occurs, a simulation execution step of generating reproduction data of an accident occurrence situation of the target vehicle by simulation based on the numerical data; a correcting step of correcting the reproduction data by at least referring to the numerical data; It further includes In the acquiring step, position data of the target vehicle is further acquired, In the correction step, the position data is further referenced to generate the corrected reproduction data. The correction step a first model that converts the time series of the numerical data into a waveform; a second model for graphing GNSS data as the position data; and performing processing using correcting the reproduced data by referring to the waveform data obtained by the first model and the graph data obtained by the second model; The process in the correction step includes: a process of generating correction data for the reproduction data by inputting the waveform data obtained by the first model and the graph data obtained by the second model into a third model; a process of generating the corrected reproduction data by referring to the reproduction data and the correction data; Contains Accident analysis method.

Images