Vehicle accident prediction system, vehicle accident prediction method, storage medium, and learned model generation system

Abstract

A vehicle accident prediction system, a vehicle accident prediction method, a storage medium, and a learned model generation system acquire a learning data set (D3) configured of a feature quantity group data (D1) including a first feature quantity (D11) representing an attribute of a driver of a vehicle, a second feature quantity (D12) representing a state of the vehicle, and a third feature quantity (D13) in which a plurality of second feature quantities (D12) are combined, and accident data (D2) related to an accident of the vehicle, generate a learned model (M) that predicts an accident of the vehicle from the feature quantity group data (D1) by learning using a plurality of the learning data sets (D3) acquired, input the feature quantity group data (D1) that becomes a prediction target, and predict an accident of the vehicle from the input feature quantity group data (D1) using the generated learned model (M).

Description

Technical Field

[0001] This invention relates to a vehicle accident prediction system, a vehicle accident prediction method, a storage medium, and a learned model generation system. Background Technology

[0002] As a technology related to vehicle accident prediction, for example, Patent Document 1 discloses a traffic accident prediction device comprising an accident occurrence pattern learning unit and an accident occurrence prediction unit. The accident occurrence pattern learning unit learns accident occurrence patterns using past traffic data through a prescribed learning algorithm. The accident occurrence prediction unit quantitatively outputs the tendency for traffic accidents to occur based on measured traffic data at the current time or predicted traffic data after the current time and the learning results of the accident occurrence pattern learning unit.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2014-35639 Summary of the Invention

[0006] The technical problem that the invention aims to solve

[0007] However, the traffic accident prediction device described in the aforementioned Patent Document 1 has room for further improvement, for example, in terms of improving the accuracy of accident prediction.

[0008] The present invention was made in view of the above circumstances, and its object is to provide a vehicle accident prediction system, a vehicle accident prediction method, a storage medium, and a learned model generation system capable of appropriately predicting accidents.

[0009] Technical means for solving problems

[0010] To achieve the above objectives, the vehicle accident prediction system of the present invention is characterized by comprising: a preprocessing unit that acquires a learning dataset consisting of feature set data and accident data related to vehicle accidents, wherein the feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values; a model generation unit that uses the multiple learning datasets acquired by the preprocessing unit to generate a learned model that predicts vehicle accidents based on the feature set data; a prediction object input unit that inputs the feature set data that becomes the prediction object; and a prediction unit that uses the learned model generated by the model generation unit to predict vehicle accidents based on the feature set data input by the prediction object input unit.

[0011] In addition, in the aforementioned vehicle accident prediction system, the feature set data includes a fourth feature representing the driving scenario of the vehicle.

[0012] To achieve the above objectives, the vehicle accident prediction method of the present invention is characterized by comprising: a step of acquiring a learning dataset consisting of feature set data and accident data related to vehicle accidents, wherein the feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values; a step of generating a learned model for predicting vehicle accidents based on the feature set data by using the acquired multiple learning datasets; a step of inputting the feature set data as the prediction object; and a step of predicting vehicle accidents based on the input feature set data using the generated learned model.

[0013] To achieve the above objectives, the storage medium of the present invention is characterized in that it stores a vehicle accident prediction program, which causes a computer to perform the following processes: acquiring a learning dataset consisting of feature set data and accident data related to vehicle accidents, wherein the feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values; using the acquired learning datasets, generating a learned model that predicts vehicle accidents based on the feature set data; inputting the feature set data that is the prediction object; and using the generated learned model to predict vehicle accidents based on the input feature set data.

[0014] To achieve the above objectives, the learned model generation system of the present invention is characterized by comprising: a preprocessing unit that acquires a learning dataset consisting of feature set data and accident data related to vehicle accidents, wherein the feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values; and a model generation unit that uses the multiple learning datasets acquired by the preprocessing unit to generate a learned model that predicts vehicle accidents based on the feature set data.

[0015] Invention Effects

[0016] The vehicle accident prediction system, vehicle accident prediction method, storage medium, and learned model generation system involved in this invention can effectively perform accident prediction. Attached Figure Description

[0017] Figure 1 This is a block diagram illustrating the general structure of the vehicle accident prediction system involved in the implementation method.

[0018] Figure 2 This is a schematic diagram illustrating the learning phase and the usage phase of the processing circuit of the vehicle accident prediction system according to the embodiment.

[0019] Figure 3 This is a flowchart illustrating an example of the processing performed by the processing circuit of the vehicle accident prediction system according to the embodiment.

[0020] Figure 4 This is a block diagram representing the general structure of the vehicle accident prediction system involved in the modified example.

[0021] Symbol Explanation

[0022] 1.1A Vehicle Accident Prediction System

[0023] 10, 110, 210 input devices

[0024] 20, 120, 220 output devices

[0025] 30, 130, 230 storage circuits

[0026] 40, 140, 240 processing circuits

[0027] 41, 141 Pre-processing Section

[0028] Model Generation Departments 42 and 142

[0029] 43, 243 Prediction Object Input Section

[0030] 44, 244 Forecasting Department

[0031] 45, 245 Output Section

[0032] 100 Learned Model Generation System

[0033] 200 Vehicle Accident Prediction Device

[0034] AL machine learning algorithm

[0035] D0 Raw Data

[0036] D1 Feature set data

[0037] D11 First characteristic quantity

[0038] D12 Second characteristic quantity

[0039] D13 Third characteristic quantity

[0040] D14 Fourth characteristic quantity

[0041] D2 Accident Data

[0042] D3 Learning Dataset

[0043] D4 Predicted object data

[0044] D5 Prediction Results Data

[0045] M learned model Detailed Implementation

[0046] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to these embodiments. Furthermore, the constituent elements in the following embodiments include constituent elements that can be easily substituted by those skilled in the art or substantially the same constituent elements.

[0047] [Implementation Method]

[0048] Figure 1 The vehicle accident prediction system 1 of this embodiment shown is a system for predicting vehicle accidents. In the vehicle accident prediction system 1, as... Figure 2 As shown, there is a learning phase for generating a learned model M for predicting vehicle accidents, and a usage phase for using the learned model M to predict vehicle accidents. The vehicle accident prediction system 1 is implemented, for example, via various computer devices such as personal computers, workstations, and tablet terminals. Hereinafter, refer to... Figure 1 , Figure 2 The structure of the vehicle accident prediction system is described in detail.

[0049] Specifically, the vehicle accident prediction system 1 includes an input device 10, an output device 20, a storage circuit 30, and a processing circuit 40. The input device 10, output device 20, storage circuit 30, and processing circuit 40 are communicatively connected to each other via a network.

[0050] Input device 10 is a device for receiving various inputs from the vehicle accident prediction system 1. Input device 10 may be, for example, an operation input device that receives various operational inputs from the user, or a data input device that receives data (information) inputs from devices other than the vehicle accident prediction system 1. Operation input devices may be implemented using, for example, a mouse, keyboard, trackball, switch, button, joystick, touchpad, touchscreen, contactless input circuit, or voice input circuit. Data input devices may be implemented using, for example, a communication interface that allows various data to be sent and received between devices via communication without being restricted by wired or wireless connections, or a storage medium interface that reads various data from storage media such as floppy disks (FD), magnetic disks (Magneto-Optical disks), CD-ROMs, DVDs, USB storage devices, SD card storage devices, or flash memory.

[0051] Output device 20 is a device that provides various outputs from vehicle accident prediction system 1. Output device 20 may be implemented as a display that outputs and shows various image information, a speaker that outputs sound information, or a data output device that outputs data (information) to other devices besides vehicle accident prediction system 1. Data output devices may be implemented as communication interfaces that transmit and receive various data between devices via communication without being restricted by wired or wireless connections, or storage medium interfaces that write various data to the same storage medium as described above. Data input devices and data output devices may also be used in combination in some or all of the structure.

[0052] The storage circuit 30 (storage medium) is a circuit that stores various types of data. The storage circuit 30 is implemented, for example, by semiconductor memory elements such as RAM (Random Access Memory), flash memory, hard disks, optical disks, etc. The storage circuit 30 stores, for example, programs used by the vehicle accident prediction system 1 to perform various functions. The programs stored in the storage circuit 30 include programs for enabling the input device 10 to function, programs for enabling the output device 20 to function, programs for enabling the processing circuit 40 to function, etc. Furthermore, the storage circuit 30 stores various types of data, such as the raw data D0 input via the input device 10, the data required for various processes in the processing circuit 40, the learning dataset D3 used in learning the learned model M, the learned model M, and the prediction result data D5 output via the output device 20. The storage circuit 30 reads these various types of data as needed by the processing circuit 40, etc. Alternatively, the storage circuit 30 can also be implemented via a cloud server or the like connected to the vehicle accident prediction system 1 via a network.

[0053] The processing circuit 40 is a circuit that implements various processing functions in the vehicle accident prediction system 1. The processing circuit 40 is implemented, for example, by a processor. A processor can be, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), an ASIC (Application Specific Integrated Circuit), or a FPGA (Field Programmable Gate Array). The processing circuit 40 implements these processing functions, for example, by executing a program read from the storage circuit 30.

[0054] The overall structure of the vehicle accident prediction system 1 according to this embodiment has been described above. Based on this structure, the processing circuit 40 according to this embodiment has the function of performing various processes to generate a learned model M during the learning phase, and the learned model M is used to predict vehicle accidents. In addition, the processing circuit 40 according to this embodiment has the function of using the learned model M to predict vehicle accidents during the usage phase.

[0055] To achieve the various processing functions described above, the processing circuit 40 of this embodiment is functionally configured to include a preprocessing unit 41, a model generation unit 42, a prediction object input unit 43, a prediction unit 44, and an output unit 45. The processing circuit 40, for example, executes a program read from the storage circuit 30, thereby implementing the processing functions of the preprocessing unit 41, the model generation unit 42, the prediction object input unit 43, the prediction unit 44, and the output unit 45.

[0056] The preprocessing unit 41 is a component that performs various preprocessing operations on the data used to train the learned model M during the learning phase. In this embodiment, the preprocessing unit 41 is capable of performing the processing to acquire a learning dataset D3 consisting of feature set data D1 and incident data D2.

[0057] The learning dataset D3, acquired by the preprocessing unit 41, is teacher data used when generating the learned model M through machine learning. The learning dataset D3 is constructed by associating feature data D1 related to vehicle driving with accident data D2 related to accidents involving the vehicle at the driving state specified by the feature data D1. Furthermore, the learning dataset D3 consists of the feature data D1 quantified as explanatory variables and the accident data D2 quantified as target variables.

[0058] The feature quantity group data D1 typically includes data containing various feature quantities related to vehicle operation. In this embodiment, the feature quantity group data D1 includes a first feature quantity D11, a second feature quantity D12, a third feature quantity D13, and a fourth feature quantity D14. The first feature quantity D11, the second feature quantity D12, the third feature quantity D13, and the fourth feature quantity D14 are feature quantities that influence the occurrence of vehicle accidents and are respectively related to the occurrence of vehicle accidents. For example, the first feature quantity D11, the second feature quantity D12, the third feature quantity D13, and the fourth feature quantity D14 are set as feature quantities correlated with the risk of accident occurrence based on insights obtained by analyzing data from multiple vehicle accidents. Hereinafter, an example of the first feature quantity D11, the second feature quantity D12, the third feature quantity D13, and the fourth feature quantity D14 will be described.

[0059] The first feature quantity D11 is a "driver attribute feature quantity" representing the attributes of the vehicle's driver, and is a value obtained by quantifying the attributes of the vehicle's driver. The first feature quantity D11 may include, for example, values ​​obtained by quantifying the driver's attendance information. As an example, the first feature quantity D11 may include values ​​obtained by quantifying the driver's number of days registered with the vehicle management operator, the driver's past driving distance / time, the driver's working hours during past prescribed periods, and the number of days elapsed since the driver's last working day. The first feature quantity D11 may be set based on insights such as the impact of the vehicle's driver attributes (e.g., recent attendance status) on the occurrence of vehicle accidents.

[0060] The second feature quantity D12 is a "vehicle state feature quantity" representing the vehicle's state, and is a value obtained by quantifying the vehicle's state. The second feature quantity D12 may include, for example, values ​​obtained by quantifying detection results from various onboard devices, sensors, cameras, position measuring devices, etc., installed in the vehicle. Typically, the second feature quantity D12 processes these detection results as individual data. As an example, the second feature quantity D12 may include values ​​obtained by quantifying the vehicle's speed (maximum, minimum, average, variance), vehicle acceleration, vehicle deceleration, the rotational speed of the vehicle's power source, and the amount of change in the vehicle's direction. The second feature quantity D12 may be set based on insights such as the vehicle's state influencing the occurrence of vehicle accidents.

[0061] The third feature quantity D13 is a "combined feature quantity" formed by combining multiple second feature quantities D12, and is a value obtained by quantifying the combination of multiple second feature quantities D12. For example, the third feature quantity D13 may include a value obtained by quantifying multiple detection results detected by various on-board equipment, sensors, cameras, position measuring devices, etc., mounted in the vehicle. Typically, the third feature quantity D13 is processed as composite data composed of multiple second feature quantities D12 in a manner representing a specific driving condition (stage), where each second feature quantity D12 is obtained by processing the aforementioned detection results as separate data. As an example, the third feature quantity D13 may include values ​​obtained by quantifying the acceleration distribution, deceleration distribution, average acceleration / deceleration time, directional change distribution, directional change time, driving power source speed distribution, and directional change distribution for each speed segment. The third characteristic quantity D13 is set based on insights such as: even if the distribution of the first second characteristic quantity D12 (e.g., acceleration) is the same, the accident rate of the vehicle may be different if the second second characteristic quantity D12 (e.g., speed range) is different.

[0062] The fourth feature quantity D14 is a "scene feature quantity" representing the vehicle's driving scenario, and is a value obtained by quantifying the vehicle's driving scenario. The fourth feature quantity D14 may include, for example, values ​​obtained by quantifying various driving scenarios based on the external environment, climate, terrain, driver's psychological state, etc., surrounding the vehicle during driving. The fourth feature quantity D14 may be quantified using detection results from various onboard devices, sensors, cameras, position measuring devices, etc., or other values. For example, the fourth feature quantity D14 may include values ​​obtained by quantifying driving scenarios during periods of high traffic volume, driving scenarios after a rest, driving scenarios later than the predicted arrival time at the destination, driving scenarios when entering narrow roads, and driving scenarios in inclement weather. The fourth feature quantity D14 is set based on the insight that even if the first feature quantity D11, the second feature quantity D12, and the third feature quantity D13 are the same, the accident rate may differ depending on the driving scenario.

[0063] Accident data D2 is data related to vehicle accidents. Accident data D2 includes information about accidents involving the vehicle while it is in motion, as specified by the associated feature data group D1. Here, as an example, accident data D2 includes at least information indicating whether an accident occurred, and may also include information indicating the accident location (latitude and longitude), cause of the accident, type of accident, amount of damage, etc.

[0064] The preprocessing unit 41 acquires a learning dataset D3 formed by associating the aforementioned feature quantity group data D1 and the aforementioned accident data D2 corresponding to the feature quantity group data D1 as a set. For example, the preprocessing unit 41 can directly acquire the pre-generated learning dataset D3 from a device other than the vehicle accident prediction system 1 via a data input device constituting the input device 10. Alternatively, the preprocessing unit 41 can also create and acquire the learning dataset D3 by performing various preprocessing operations on the raw data D0 input from other devices other than the vehicle accident prediction system 10. For example, the preprocessing unit 41 can either preprocess the raw data D0 every time it is input, or preprocess the raw data D0 at appropriate times based on the user's operation of the operation input device constituting the input device 10.

[0065] In this case, the raw data D0 preprocessed by the preprocessing unit 41 can be input from devices other than the vehicle accident prediction system 1 via the data input device constituting the input device 10, or it can be input by the user via the operation input device constituting the input device 10. This raw data D0 may include, for example, vehicle system data, operator data, accident statistics data, and external data. Vehicle system data may include, for example, vehicle signals, data detected by vehicle-mounted devices such as driver recorders and digital dashcams, sensors, cameras, and position measuring devices, and may also include information such as vehicle speed (maximum, minimum, average, variance), vehicle acceleration, vehicle deceleration, vehicle power source speed, and vehicle directional change. Operator data may include, for example, data held by operators such as transportation companies and bus companies, and may also include information such as operator ID, vehicle ID, driver ID, attendance records, in-vehicle and out-of-vehicle animations, and driver vital signs. Accident statistics may include, for example, data held by damage insurance companies, and may also include information such as the operator ID of the accident, the vehicle ID of the accident, the date and time of the accident, the latitude and longitude of the accident, the type of accident, and the amount of damage. External data includes data held by other external devices or databases, and may also include information such as maps (road categories, building / facility categories), traffic congestion, weather, pedestrian distribution, and population density.

[0066] Furthermore, the preprocessing performed on the raw data D0 by the preprocessing unit 41 includes, for example, the following: collecting and combining the raw data D0; extracting feature quantity group data D1, such as the first feature quantity D11, the second feature quantity D12, the third feature quantity D13, and the fourth feature quantity D14, from the raw data D0 and quantifying them as explanatory variables; extracting accident data D2 from the raw data D0 and quantifying it as the target variable; and establishing a correlation between the quantified feature quantity group data D1 and the quantified accident data D2 and combining them.

[0067] The preprocessing unit 41 stores the multiple learning datasets D3 obtained as described above in the storage circuit 30.

[0068] The model generation unit 42 is a part capable of performing the following processing: generating a learned model M that predicts vehicle accidents based on feature set data D1 through machine learning during the learning phase. In this embodiment, the model generation unit 42 can perform the following processing: generating a learned model M through machine learning using multiple learning datasets D3 obtained by the preprocessing unit 41. For example, the model generation unit 42 performs the process of learning and generating the learned model M at appropriate times based on the operations of the user operating the input device constituting the input device 10.

[0069] The model generation unit 42 uses multiple learning datasets D3 as teacher data and performs machine learning based on various machine learning algorithms AL to generate a learned model M. Examples of the machine learning algorithms AL used include well-known algorithms such as Deep Learning, Neural Network, Logistic Regression, Ensemble Learning, Support Vector Machine, Random Forest, and Naive Bayes. The model generation unit 42 uses the feature set data D1 from the learning dataset D3 as explanatory variables and the accident data D2 as the target variable to perform machine learning on the learned model M. As a result of this machine learning, the model generation unit 42 generates a learned model M that has undergone machine learning for predicting vehicle accidents based on the feature set data D1.

[0070] The learned model M is implemented, for example, through a neural network. In this case, the model generation unit 42 generates the learned model M by performing machine learning using multiple learning datasets D3 to learn the learning weighting coefficients used as weights in the neural network.

[0071] The learned model M generated by the model generation unit 42 is a model that takes the feature set data D1 as input and outputs the value obtained by quantifying the prediction of vehicle accidents. That is, the learned model M has the following functions: receiving the feature set data D1 as input and outputting the value obtained by quantifying the prediction of vehicle accidents based on the feature set data D1. More specifically, the learned model M enables the computer to perform the following functions: performing calculations on the feature set data D1 input to the input layer of the neural network based on the learned weighting coefficients in the neural network, and outputting the value obtained by quantifying the prediction of accidents from the output layer of the neural network.

[0072] In other words, the value obtained by quantifying the accident prediction of a vehicle from the learned model M is equivalent to the value obtained by predicting the accident risk of that vehicle. As an example, the value obtained by quantifying the accident prediction of a vehicle is the value obtained by quantifying whether an accident has occurred, but it could also be the value obtained by quantifying the cause of the accident, the type of accident, the amount of damage, etc.

[0073] The model generation unit 42 stores the learned model M generated as described above in the storage circuit 30. At this time, if the previously generated learned model M has already been stored in the storage circuit 30, the model generation unit 42 replaces the stored learned model M with the newly generated learned model M.

[0074] The prediction object input unit 43 is a part that performs the following processing: during the usage phase, it inputs feature quantity group data D1 that becomes the prediction object. Here, the feature quantity group data D1 that becomes the prediction object is sometimes referred to as "prediction object data (input data) D4". The prediction object data D4 can be input from other devices besides the vehicle accident prediction system 1 via the data input device constituting the input device 10, or it can be input by the user via the operation input device constituting the input device 10. The prediction object input unit 43 of this embodiment can perform the processing of inputting the prediction object data D4 received via the input device 10 to the prediction unit 44. For example, the prediction object input unit 43 can input the prediction object data D4 in real time in conjunction with the vehicle's movement, or it can input the prediction object data D4 after the vehicle's movement has ended at an appropriate time. The prediction object input unit 43 can also temporarily store the input prediction object data D4 in the storage circuit 30.

[0075] The prediction unit 44 is a part that has the function of performing the following processing: predicting vehicle accidents using the learned model M during the usage phase. The prediction unit 44 of this embodiment can perform the following processing: using the learned model M generated by the model generation unit 42, predicting vehicle accidents based on the feature data group D1, i.e., the prediction object data D4, which is input by the prediction object input unit 43.

[0076] The prediction unit 44 takes the prediction object data D4 input by the prediction object input unit 43 as input data to the learned model M generated by the model generation unit 42, and outputs a value obtained by quantifying the prediction of vehicle accidents from the learned model M based on the input data. Thus, the prediction unit 44 predicts the vehicle accident under the driving state defined by the prediction object data D4 (which becomes the feature set data D1 of the prediction object). Here, as an example, the prediction unit 44 outputs the value obtained by quantifying whether a vehicle accident has occurred as the value obtained by quantifying the prediction of vehicle accidents, and predicts whether a vehicle accident has occurred. The prediction unit 44 stores the value obtained by quantifying the output prediction of vehicle accidents as prediction result data (output data) D5 in the storage circuit 30.

[0077] The output unit 45 is a part capable of performing the following processing: outputting the prediction result of the vehicle accident based on the prediction result of the prediction unit 44. In this embodiment, the output unit 45 can perform the processing of outputting the prediction result data D5 predicted by the prediction unit 44 via the output device 20. The prediction result data D5 can be output as image information via the display constituting the output device 20, or as sound information via the speaker constituting the output device 20. Furthermore, the prediction result data D5 can also be output to other devices besides the vehicle accident prediction system 1 via the data output device constituting the output device 20. For example, the output unit 45 can output the prediction result data D5 in real time according to the vehicle's movement, or it can output the prediction result data D5 at appropriate times according to the user's operation of the operation input device constituting the input device 10.

[0078] Next, refer to Figure 3 The flowchart illustrates the processing sequence of the vehicle accident prediction method in vehicle accident prediction system 1.

[0079] Figure 3 The vehicle accident prediction method of the vehicle accident prediction system 1 shown includes an acquisition step (step S1), a generation step (step S2), an input step (step S3), a prediction step (step S4), and an output step (step S5). Here, the processing related to each of the above steps is performed by the processing circuit 40 of the vehicle accident prediction system 1.

[0080] First, the preprocessing unit 41 of the processing circuit 40 performs the step of acquiring a learning dataset D3 consisting of feature set data D1 including first feature D11, second feature D12, third feature D13, and fourth feature D14, and accident data D2 (step S1). In this case, the preprocessing unit 41 can directly acquire the learning dataset D3 from a device other than the vehicle accident prediction system 1 via the input device 10, or it can create and acquire the learning dataset D3 by performing various preprocessing operations on the raw data D0 input from a device other than the vehicle accident prediction system 1 via the input device 10. The preprocessing unit 41 stores the acquired multiple learning datasets D3 in the storage circuit 30.

[0081] Next, the model generation unit 42 of the processing circuit 40 uses the multiple learning datasets D3 acquired in the acquisition step (step S1) to perform a generation step (step S2) to generate a learned model M through machine learning. The model generation unit 42 stores the generated learned model M in the storage circuit 30. At this time, if the previously generated learned model M is already stored in the storage circuit 30, the model generation unit 42 replaces the stored learned model M with the newly generated learned model M.

[0082] Next, the prediction object input unit 43 of the processing circuit 40 performs the input step (step S3) of inputting the feature quantity group data D1, i.e., the prediction object data D4, which is the prediction object, to the prediction unit 44 of the processing circuit 40. In this case, the prediction object input unit 43 can input the prediction object data D4 received from a device other than the vehicle accident prediction system 1 via the input device 10, or it can input the prediction object data D4 received through user operation via the input device 10. In addition, the prediction object input unit 43 can also temporarily store the input prediction object data D4 in the storage circuit 30.

[0083] Next, the prediction unit 44 of the processing circuit 40 performs a prediction step (step S4): using the learned model M generated in the generation step (step S2), it predicts a vehicle accident based on the prediction object data D4 (feature set data D1 of the prediction object) input in the input step (step S3). In this case, the prediction unit 44 inputs the prediction object data D4 to the learned model M and outputs a value obtained by quantifying the prediction of the vehicle accident from the learned model M based on the object data D4. Thus, the prediction unit 44 predicts the vehicle accident in the driving state as defined by the prediction object data D4. The prediction unit 44 stores the value obtained by quantifying the output prediction of the vehicle accident as prediction result data D5 in the storage circuit 30.

[0084] Next, the output unit 45 of the processing circuit 40 executes the output step (step S5) of outputting the prediction result data D5 of the vehicle accident predicted in the prediction step (step S4), thus ending the processing of this flowchart. In this case, the output unit 45 can output the prediction result data D5 as image information and sound information via the output device 20, or it can output it to other devices other than the vehicle accident prediction system 1 via the output device 20.

[0085] The above-described vehicle accident prediction method can be implemented by executing a pre-prepared vehicle accident prediction program on a computer such as a personal computer or workstation. This vehicle accident prediction program causes the computer to perform each of the above-described steps: acquisition (step S1), generation (step S2), input (step S3), prediction (step S4), and output (step S5).

[0086] The vehicle accident prediction system 1, vehicle accident prediction method, and vehicle accident prediction program described above can generate a learned model M with high prediction accuracy by using not only the second feature quantity D12 representing the vehicle's state, but also the first feature quantity D11 representing the driver's attributes, and the third feature quantity D13 obtained by combining multiple second feature quantities D12, etc., based on the high level of the input feature quantities themselves, to perform vehicle accident prediction. As a result, the vehicle accident prediction system 1, vehicle accident prediction method, and vehicle accident prediction program can perform accident prediction appropriately. Therefore, the vehicle accident prediction system 1, vehicle accident prediction method, and vehicle accident prediction program can, for example, predict accident risks more precisely.

[0087] Here, in the vehicle accident prediction system 1, vehicle accident prediction method, and vehicle accident prediction program described above, in addition to the first feature quantity D11, the second feature quantity D12, and the third feature quantity D13, a learned model M is generated based on the fourth feature quantity D14 representing the driving scenario of the vehicle, and vehicle accident prediction is performed. Therefore, higher-order accident prediction corresponding to the driving scenario can be performed.

[0088] Furthermore, the vehicle accident prediction system 1, vehicle accident prediction method, and vehicle accident prediction program can provide more accurate accident prediction results as described above for various purposes, such as providing real-time driving warnings to operators and drivers, evaluating driving skills / habits / accident risks, visualizing improvement points, providing guidance / education for accident risk mitigation actions, and creating safe operation plans. Additionally, the vehicle accident prediction system 1, vehicle accident prediction method, and vehicle accident prediction program can also output accident prediction results on a per-run basis, per driver basis, per vehicle basis, and per operator basis.

[0089] Furthermore, in the embodiments described above, an example of using a system to perform both the learning phase and the usage phase was given as vehicle accident prediction system 1, but the embodiments are not limited to this.

[0090] For example, Figure 4 The vehicle accident prediction system 1A involved in the illustrated variant differs from the vehicle accident prediction system 1 described above in that it is configured as a learned model generation system 100 that performs each process in the learning phase and a vehicle accident prediction device 200 that performs each process in the usage phase.

[0091] The learned model generation system 100 includes an input device 110, an output device 120, a storage circuit 130, and a processing circuit 140, and performs processing to generate a learned model M using machine learning with a learning dataset D3. To realize the various processing functions described above, the processing circuit 140 is conceptually configured to include a preprocessing unit 141 and a model generation unit 142.

[0092] Similar to the preprocessing unit 41 described above, the preprocessing unit 141 is capable of performing the processing of acquiring a learning dataset D3 consisting of feature group data D1 including first feature D11, second feature D12, third feature D13, and fourth feature D14, and accident data D2. The preprocessing unit 141 stores the acquired learning datasets D3 in the storage circuit 130.

[0093] Similar to the model generation unit 42 described above, the model generation unit 142 is capable of performing the process of generating a learned model M using multiple learning datasets D3 obtained by the preprocessing unit 141 through machine learning. The model generation unit 142 stores the generated learned model M in the storage circuit 130.

[0094] The vehicle accident prediction device 200 includes an input device 210, an output device 220, a storage circuit 230, and a processing circuit 240, and uses a learned model M to predict vehicle accidents. In order to realize the above-mentioned processing functions, the processing circuit 240 is functionally configured to include a prediction target input unit 243, a prediction unit 244, and an output unit 245.

[0095] Similar to the prediction object input unit 43 described above, the prediction object input unit 243 is capable of performing processing on the input as feature quantity group data D1, i.e., prediction object data D4, which becomes the prediction object.

[0096] The prediction unit 244, like the prediction unit 44 described above, is capable of performing the following processing: using the learned model M, it predicts vehicle accident processing based on the prediction object data D4 input by the prediction object input unit 243. In this case, the prediction unit 244 can, for example, use the output device 120 of the learned model generation system 100 and the learned model M pre-stored in the storage circuit 230 via the input device 210 of the vehicle accident prediction device 200. This learned model M is a model generated by the learned model generation system 100 as described above.

[0097] The output unit 245, like the output unit 45 described above, is capable of performing the processing of outputting the prediction result data D5 predicted by the prediction unit 44 via the output device 220.

[0098] The other structures of input devices 110 and 210, output devices 120 and 220, storage circuits 130 and 230, and processing circuits 140 and 240 are substantially the same as those of input device 10, output device 20, storage circuit 30, and processing circuit 40 described above.

[0099] Even in this case, the vehicle accident prediction system 1A, the learned model generation system 100, and the vehicle accident prediction device 200 can perform accident predictions as appropriately as the vehicle accident prediction system 1 described above, for example, they can predict accident risks more precisely.

[0100] Furthermore, in this modified example, the learned model M used in the vehicle accident prediction device 200 is not limited to the model generated by the learned model generation system 100 as described above, but can also be a learned model M generated by other systems.

[0101] Furthermore, the vehicle accident prediction system, vehicle accident prediction method, vehicle accident prediction program, and learned model generation system involved in the above-described embodiments of the present invention are not limited to the above-described embodiments, and various modifications can be made within the scope of the claims.

[0102] In the above description, it was stated that the feature set data D1 includes the first feature D11, the second feature D12, the third feature D13, and the fourth feature D14, but it is not limited to this. For example, the feature set data D1 may include the first feature D11 and the second feature D12 but exclude the third feature D13 and the fourth feature D14, or it may include the first feature D11 and the third feature D13 but exclude the second feature D12 and the fourth feature D14, or it may include the first feature D11 and the fourth feature D14 but exclude the second feature D12 and the third feature D13, or it may be a combination of these.

[0103] The processing circuit 40 described above is based on the case where each processing function is implemented by a single processor, but it is not limited to this. The processing circuit 40 can also implement each processing function by combining multiple independent processors and having each processor execute a program. Furthermore, the processing functions of the processing circuit 40 can be appropriately distributed or combined in one or more processing circuits. Additionally, the processing functions of the processing circuit 40 can be implemented entirely or partially by a program, or they can be implemented as hardware based on wiring logic, etc.

[0104] The program executed by the processor described above is provided by being pre-set in the storage circuit 30, etc. Alternatively, the program can be provided by being stored in a computer-readable storage medium in a form that can be installed on these devices or as an executable file. Furthermore, the program can also be stored on a computer connected to a network such as the Internet and provided or distributed by downloading it via the network.

[0105] The vehicle accident prediction system, vehicle accident prediction method, storage medium, and learned model generation system involved in this embodiment can also be constructed by appropriately combining the constituent elements of the embodiments and variations described above.

Claims

1. A vehicle accident prediction system, characterized in that, have: The preprocessing unit acquires a learning dataset consisting of feature set data and accident data related to vehicle accidents. The feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values. The model generation unit uses multiple learning datasets obtained by the preprocessing unit to generate a learned model through learning, and the learned model predicts the vehicle's accidents based on the feature set data. The prediction object input unit receives the feature set data that becomes the prediction object; and The prediction unit uses the learned model generated by the model generation unit to predict vehicle accidents based on the feature set data input by the prediction object input unit. The second feature is obtained by processing it as a separate data, while the third feature is processed as composite data consisting of multiple different second features combined in a manner that represents the driving condition.

2. The vehicle accident prediction system according to claim 1, characterized in that, The first feature includes a value obtained by quantifying at least one of the driver's days of registration with the operator managing the vehicle, the driver's past driving distance / time, the driver's past working hours during a specified period, and the number of days elapsed since the driver's last operating day.

3. The vehicle accident prediction system according to claim 1, characterized in that, The third characteristic quantity includes a value obtained by quantifying at least one of the following: the acceleration distribution of the vehicle at each speed segment, the deceleration distribution of the vehicle at each speed segment, the average acceleration / deceleration time of the vehicle at each speed segment, the directional change distribution of the vehicle at each speed segment, the directional change time of the vehicle at each speed segment, the rotational speed distribution of the driving power source of the vehicle at each acceleration segment, and the directional change distribution of the vehicle at each deceleration segment.

4. The vehicle accident prediction system according to claim 2, characterized in that, The third characteristic quantity includes a value obtained by quantifying at least one of the following: the acceleration distribution of the vehicle at each speed segment, the deceleration distribution of the vehicle at each speed segment, the average acceleration / deceleration time of the vehicle at each speed segment, the directional change distribution of the vehicle at each speed segment, the directional change time of the vehicle at each speed segment, the rotational speed distribution of the driving power source of the vehicle at each acceleration segment, and the directional change distribution of the vehicle at each deceleration segment.

5. The vehicle accident prediction system according to any one of claims 1 to 4, characterized in that, The feature quantity group data includes a fourth feature quantity representing the vehicle driving scenario. The fourth feature includes a value obtained by quantifying at least one of the following: driving scenarios during periods of high traffic volume, driving scenarios after a rest, driving scenarios later than the predicted arrival time at the destination, driving scenarios when entering narrow roads, and driving scenarios during inclement weather.

6. A method for predicting vehicle accidents, characterized in that, include: The step of obtaining a learning dataset consisting of feature set data and accident data related to vehicle accidents, wherein the feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values. Using the acquired multiple learning datasets, a learned model is generated through learning, and the learned model predicts the vehicle's accidents based on the feature set data. The step of inputting the feature set data that becomes the object of prediction; as well as The step of using the generated learned model to predict vehicle accidents based on the input feature set data. The second feature is obtained by processing it as a separate data, while the third feature is processed as composite data consisting of multiple different second features combined in a manner that represents the driving condition.

7. A storage medium, characterized in that, The storage medium stores a vehicle accident prediction program, which causes a computer to perform the following processes: A learning dataset is obtained, consisting of feature set data and accident data related to vehicles. The feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values. Using the acquired training datasets, a learned model is generated through learning. This learned model predicts vehicle accidents based on the feature set data. The input is the set of feature values ​​that will be used for prediction. Using the generated learned model, predict vehicle accidents based on the input feature set data. The second feature is obtained by processing it as a separate data, while the third feature is processed as composite data consisting of multiple different second features combined in a manner that represents the driving condition.

8. A learned model generation system, characterized in that, have: The preprocessing unit acquires a learning dataset consisting of feature set data and accident data related to vehicle accidents. The feature set data includes a first feature representing the attributes of the driver of the vehicle, a second feature representing the state of the vehicle, and a third feature combining multiple second feature values. as well as The model generation unit uses multiple learning datasets obtained by the preprocessing unit to generate a learned model, which predicts vehicle accidents based on the feature set data. The second feature is obtained by processing it as a separate data, while the third feature is processed as composite data consisting of multiple different second features combined in a manner that represents the driving condition.

Images