Driving accident risk estimation device, driving accident risk estimation method, driving accident risk estimation program, artificial intelligence learning device for driving accident risk estimation device, artificial intelligence learning method for driving accident risk estimation method, and artificial intelligence learning program

Abstract

A driving accident risk estimation device is provided that predicts the risk of a vehicle driving accident. [Solution] A driving accident risk estimation device 101 according to the present disclosure includes an input data receiving unit 15, an estimation unit 19, and an estimated data output unit 25. The input data receiving unit 15 receives input data including time-series data from a predetermined past period, including the heart rate or heart rate variability of a vehicle driver 11, and basic data related to the driver's physical and mental health, including BMI and blood pressure. The estimation unit 19 inputs the input data received by the input data receiving unit 15 to a trained artificial intelligence 23, causing the artificial intelligence 23 to calculate estimated data on the risk of the driver 11 causing an accident while driving. The estimated data output unit 25 outputs the estimated data calculated by the artificial intelligence 23. Furthermore, the time-series data includes data expressed in numerical values ​​based on data measured by a heart rate sensor 1. The basic data includes data expressed in numerical values.

Description

[Technical Field]

[0001] The present invention relates to a driving accident risk estimation device, a driving accident risk estimation method, a driving accident risk estimation program, an artificial intelligence learning device for a driving accident risk estimation device, an artificial intelligence learning method for a driving accident risk estimation method, and an artificial intelligence learning program. [Background technology]

[0002] There are known technologies for predicting health status by detecting physiological characteristics of the body, such as heart rate, using a sensor and processing them using a computer, such as AI (artificial intelligence). For example, Patent Document 1 discloses a technology for obtaining a stress index from heart rate variability data and further evaluating mental health. Patent Document 2 discloses a technology for calculating a fatigue index from heart rate data and further generating product and service information. Patent Document 3 discloses a technology for predicting disabilities and poor health from heart rate. Furthermore, Patent Document 4, filed by the same applicant as the present application, discloses a technology for estimating an evaluation score for activities of daily living based on the Functional Independence Measure (FIM), which is useful for occupational therapy evaluations, based on sleep data, such as the number of turns in bed, obtained by a sleep sensor, and basic data, such as age and height.

[0003] Based on their successful experience using artificial intelligence to estimate evaluation scores for activities of daily living (ADL) from sleep data, the inventors of this application suspected that the risk of driving accidents (i.e., the likelihood of an accident occurring) involving vehicles such as taxis, buses, trucks, private cars, motorcycles, bicycles, and trains, which are generally considered accidental, could be predicted based on physiological features, suggesting a correlation between the two. Published research papers have also shown a correlation between changes in heart rate and impaired driving performance. For example, Non-Patent Document 1 reveals that increased cognitive load increases physiological arousal, leading to increased heart rate and blood pressure. Meanwhile, Non-Patent Document 2 reveals that unconscious signs of drowsiness while driving are manifested by a decrease in heart rate.

[0004] Regarding heart rate variability, there are known research papers that show a correlation with driving performance. For example, Non-Patent Document 3 reveals that measuring heart rate variability is promising for real-time monitoring of stress and cognitive load while driving. Furthermore, Non-Patent Document 4 reveals that heart rate variability (e.g., SDNN and RMSSD) increases when a driver is fatigued.

[0005] Regarding the meaning of "heart rate variability," Non-Patent Document 5 states that heart rate variability (HRV) refers to the fluctuation of the interval between R waves (RR interval; RRI) on the order of milliseconds in an electrocardiogram (ECG). While it would be most effective to use the R wave that shows the highest peak as an index, in the present invention, it is generally understood to refer to the fluctuation of the heart rate cycle.

[0006] The inventors of the present application thought that if such a correlation exists, it might be possible to realize a useful device for predicting the risk of vehicle driving accidents by collecting past accident data, using this as training data to train an artificial intelligence, and then using the trained artificial intelligence. Note that neither Patent Documents 1 to 4 nor Non-Patent Documents 1 to 5 mention technology for estimating the risk of driving accidents using artificial intelligence. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2017-213278 [Patent Document 2] Patent No. 6717913 [Patent Document 3] Patent No. 7432878 [Patent Document 4] Patent No. 6994262 [Non-patent literature]

[0008] [Non-Patent Document 1] Emma J. Nilsson, Jonas Baergman, Mikael Ljung Aust, Gerald Matthews, Bo Svanberg Let Complexity BringClarity: A Multidimensional Assessment of Cognitive Load Using Physiological Measures Frontiers inNeuroergonomics Frontiers Media 2022; 3: 787295 [Non-patent document 2] Masatoshi Yanagidaira and Mitsuo Yasushi, "Development of Driving State Estimation Technology - Detection of Drowsiness State by Heart Rate Analysis -" Pioneer R&D (Pioneer Corporation) 2004; 14(3): 17-27 [Non-patent document 3] Karina Rollandovna Arutyunova et al., “Heart rate dynamics for cognitive load estimation in a driving simulation task” Sci Rep. 2024 Dec 30;14:31656. doi (https: / / pmc.ncbi.nlm.nih.gov / articles / PMC11685601 / #:~:text=aim%20of%20this%20large,and%20mental%20distraction%20during%20highway) [Non-patent document 4] Cheng Yu Tsai et al., “ B ehaviors Using a Dynamic Weighted Moving Average Model With a Long Short-Term Memory Network Based on Heart Rate Variability” Taipei Medical University (TMU) Academic Hub / Pure Experts(https: / / hub.tmu.edu.tw / en / publications / predicting-fatigue-associated-aberrant-driving-behaviors-using-a-#:~:text=relatively%20high%20SHAP%20values,used%20in%20realistic%20driving%20scenarios) [Non-patent document 5] Koichi Fujiwara, "Heart Rate Variability Analysis for Health Monitoring," Systems / Control / Information, Vol. 61, No. 9, pp. 381-386, 2017 Summary of the Invention [Problem to be solved by the invention]

[0009] The present invention has been made based on the above insight of the inventors, and aims to provide a driving accident risk estimation device, a driving accident risk estimation method, and a driving accident risk estimation program that predict the risk of a vehicle driving accident. Another aim of the present invention is to provide an artificial intelligence learning device for the driving accident risk estimation device, an artificial intelligence learning method for the driving accident risk estimation method, and an artificial intelligence learning program. [Means for solving the problem]

[0010] To achieve the above object, a first aspect of the present invention provides a driving accident risk estimation device, comprising an input data receiving unit, an estimation unit, and an estimated data output unit. The input data receiving unit receives input data including time-series data from a predetermined past period, including a vehicle driver's heart rate or heart rate variability, and basic data, which is data related to the driver's mind and body. The estimation unit inputs the input data received by the input data receiving unit into a trained artificial intelligence, causing the artificial intelligence to calculate estimated data on the risk of the driver causing an accident while driving. The estimated data output unit outputs the estimated data calculated by the artificial intelligence. Furthermore, the time-series data includes data expressed in numerical values ​​based on data measured by a heart rate sensor. The basic data includes data expressed in numerical values. The estimated data includes data evaluated in a numerical scale.

[0011] According to this configuration, it is possible to predict the risk of a vehicle driving accident. In the present invention, the term "heartbeat" is used to include "pulsebeat."

[0012] A second aspect of the present invention is a driving accident risk estimation device according to the first aspect, wherein the estimated data of the risk is estimated data of the risk that the driver will cause a driving accident of different severity levels while driving. This configuration makes it possible to predict the risk of driving accidents at multiple levels of severity.

[0013] A third aspect of the present invention is a driving accident risk estimation device according to the first or second aspect, wherein the time series data includes a variable that expresses the prominence of the characteristic that the heart rate rises once and then falls within the specified past period.

[0014] According to this configuration, since the input data includes data that is suitable for the calculation of the estimated data, the calculation of the estimated data can be performed with higher accuracy, and the learning of the artificial intelligence can be performed more efficiently.

[0015] A fourth aspect of the present invention is the driving accident risk estimation device according to any one of the first to third aspects, wherein the basic data includes BMI (Body Mass Index) or blood pressure.

[0016] According to this configuration, since the input data includes data that is suitable for the calculation of the estimated data, the calculation of the estimated data can be performed with higher accuracy, and the learning of the artificial intelligence can be performed more efficiently.

[0017] A fifth aspect of the present invention is a driving accident risk estimation method, comprising: (A) a computer receiving input data including time-series data from a predetermined period in the past, including a vehicle driver's heart rate or heart rate variability, and basic data, which is data related to the driver's mind and body; (B) the computer inputs the received input data into a trained artificial intelligence, thereby causing the artificial intelligence to calculate estimated data on the risk of the driver having an accident while driving; and (C) the computer outputting the estimated data calculated by the artificial intelligence. Further, the time-series data includes data expressed in numerical values ​​based on data measured by a heart rate sensor. The basic data includes data expressed in numerical values. The estimated data includes data evaluated in a numerical scale.

[0018] This configuration makes it possible to predict the risk of a vehicle driving accident.

[0019] A sixth aspect of the present invention is a driving accident risk estimation method according to the fifth aspect, wherein the estimated data of the risk is estimated data of the risk that the driver will cause a driving accident of different severity levels while driving.

[0020] This configuration makes it possible to predict the risk of driving accidents at multiple levels of severity.

[0021] A seventh aspect of the present invention is a driving accident risk estimation method according to the fifth or sixth aspect, wherein the time series data includes a variable that expresses the prominence of the characteristic that the heart rate rises once and then falls within the specified past period.

[0022] According to this configuration, since the input data includes data that is suitable for the calculation of the estimated data, the calculation of the estimated data can be performed with higher accuracy, and the learning of the artificial intelligence can be performed more efficiently.

[0023] An eighth aspect of the present invention is the driving accident risk estimation method according to any one of the fifth to seventh aspects, wherein the basic data includes BMI (Body Mass Index) or blood pressure. According to this configuration, since the input data includes data that is suitable for the calculation of the estimated data, the calculation of the estimated data can be performed with higher accuracy, and the learning of the artificial intelligence can be performed more efficiently.

[0024] A ninth aspect of the present invention is a driving accident risk estimation program which, when read by a computer, causes the computer to execute the driving accident risk estimation method according to any one of the fifth to eighth aspects.

[0025] This configuration makes it possible to predict the risk of a vehicle driving accident.

[0026] A tenth aspect of the present invention provides an artificial intelligence learning device for a driving accident risk estimation device, which trains the artificial intelligence used in the driving accident risk estimation device according to the first aspect, and includes an input data receiving unit, a teacher data receiving unit, and a learning unit. The input data receiving unit receives input data including time-series data from a predetermined past period, including a vehicle driver's heart rate or heart rate variability, and basic data, which is data related to the driver's mind and body. The teacher data receiving unit receives teacher data, which is actual data on the driver's accidents and no accidents while driving, corresponding to the input data. The learning unit trains the artificial intelligence to estimate the teacher data from the input data by inputting the input data received by the input data receiving unit and the teacher data received by the teacher data receiving unit to the artificial intelligence. Furthermore, the time-series data includes data expressed in numerical values ​​based on data measured by a heart rate sensor. The basic data includes data expressed in numerical values. The actual data includes data expressed in numerical steps.

[0027] According to this configuration, a trained artificial intelligence is obtained for use in the driving accident risk estimation device according to the first aspect. The estimated data output by the trained artificial intelligence corresponds to estimated data on the risk of a driver causing an accident while driving.

[0028] An eleventh aspect of the present invention is an artificial intelligence learning device according to the tenth aspect, wherein the input data accepted by the input data accepting unit and the teacher data accepted by the teacher data accepting unit have an increased ratio of positive data to negative data.

[0029] This configuration enables the artificial intelligence to estimate the risk of driving accidents with a higher degree of accuracy.

[0030] A twelfth aspect of the present invention is an artificial intelligence learning device according to the tenth aspect, in which the learning unit processes the input data received by the input data receiving unit and the teacher data received by the teacher data receiving unit to increase the ratio of positive data to negative data, and then inputs the data to the artificial intelligence.

[0031] This configuration enables the artificial intelligence to estimate the risk of driving accidents with a higher degree of accuracy.

[0032] A thirteenth aspect of the present invention is an artificial intelligence learning device according to the tenth aspect, wherein the historical data on accidents and no accidents while the driver is driving is historical data on driving accidents and no accidents at multiple levels of different severity.

[0033] According to this configuration, a trained artificial intelligence (AI) can be obtained for use in the driving accident risk estimation device according to the second aspect. The estimated data output by the trained AI corresponds to estimated data on the risk of a driver causing a driving accident at multiple stages of different severity while driving.

[0034] A 14th aspect of the present invention is an artificial intelligence learning device according to aspect 13, wherein the input data accepted by the input data accepting unit and the teacher data accepted by the teacher data accepting unit have an increased ratio of positive data to negative data for each of the multiple stages of different severity.

[0035] This configuration enables the artificial intelligence to estimate the risk of driving accidents with even greater accuracy.

[0036] A 15th aspect of the present invention is an artificial intelligence learning device according to the 13th aspect, in which the learning unit performs processing to increase the ratio of positive data to negative data for the input data received by the input data receiving unit and the teacher data received by the teacher data receiving unit for each of the multiple levels of severity, and then inputs the data to the artificial intelligence.

[0037] This configuration enables the artificial intelligence to estimate the risk of driving accidents with even greater accuracy.

[0038] A 16th aspect of the present invention is an artificial intelligence learning device according to any one of the 10th to 15th aspects, wherein the time series data includes a variable that expresses the prominence of the characteristic that the heart rate rises once and then falls within the specified past period.

[0039] According to this configuration, a trained artificial intelligence used by the driving accident risk estimation device according to the third aspect is obtained.

[0040] A seventeenth aspect of the present invention is the artificial intelligence learning device according to any one of the tenth to sixteenth aspects, wherein the basic data includes BMI (Body Mass Index) or blood pressure.

[0041] According to this configuration, a trained artificial intelligence used by the driving accident risk estimation device according to the fourth aspect is obtained.

[0042] An 18th aspect of the present invention provides an artificial intelligence learning method for a driving accident risk estimation device, which trains the artificial intelligence used by the driving accident risk estimation device according to the first aspect, comprising: (a) a computer receiving input data including time-series data from a predetermined past period, including a vehicle driver's heart rate or heart rate variability, and basic data, which is data related to the driver's mind and body; (b) the computer receiving teacher data, which is historical data on the driver's accidents and no accidents while driving, corresponding to the input data; and (c) the computer inputs the received input data and the received teacher data into the artificial intelligence, thereby training the artificial intelligence to estimate the teacher data from the input data. Further, the time-series data includes data expressed in numerical values ​​based on data measured by a heart rate sensor. The basic data includes data expressed in numerical values. The historical data includes data expressed in numerical steps.

[0043] According to this configuration, a trained artificial intelligence is obtained for use in the driving accident risk estimation device according to the first aspect. The estimated data output by the trained artificial intelligence corresponds to estimated data on the risk of a driver causing an accident while driving.

[0044] A 19th aspect of the present invention is an artificial intelligence learning method according to the 18th aspect, wherein the ratio of positive data to negative data is increased in the received input data and the received teacher data.

[0045] This configuration enables the artificial intelligence to estimate the risk of driving accidents with a higher degree of accuracy. A twentieth aspect of the present invention is an artificial intelligence learning method according to the eighteenth aspect, in which the artificial intelligence is trained by processing the received input data and the received teacher data to increase the ratio of positive data to negative data, and then inputting the data into the artificial intelligence.

[0046] This configuration enables the artificial intelligence to estimate the risk of driving accidents with a higher degree of accuracy.

[0047] A 21st aspect of the present invention is an artificial intelligence learning method according to the 18th aspect, in which the historical data on accidents and no accidents while the driver is driving is historical data on driving accidents and no accidents at multiple levels of different severity.

[0048] According to this configuration, a trained artificial intelligence (AI) can be obtained for use in the driving accident risk estimation device according to the second aspect. The estimated data output by the trained AI corresponds to estimated data on the risk of a driver causing a driving accident at multiple stages of different severity while driving.

[0049] A 22nd aspect of the present invention is an artificial intelligence learning method according to the 21st aspect, in which the ratio of positive data to negative data is increased for the input data and the teacher data that are accepted for each of the multiple stages of different severity.

[0050] This configuration enables the artificial intelligence to estimate the risk of driving accidents with even greater accuracy.

[0051] A 23rd aspect of the present invention is an artificial intelligence learning method according to the 21st aspect, in which the artificial intelligence is trained by processing the received input data and the received teacher data to increase the ratio of positive data to negative data for each of the multiple stages of different severity, and then inputting the data into the artificial intelligence.

[0052] This configuration enables the artificial intelligence to estimate the risk of driving accidents with even greater accuracy.

[0053] A 24th aspect of the present invention is an artificial intelligence learning method according to any one of the 18th to 23rd aspects, wherein the time series data includes a variable that expresses the prominence of the characteristic that the heart rate rises once and then falls within the specified past period.

[0054] According to this configuration, a trained artificial intelligence used by the driving accident risk estimation device according to the third aspect is obtained.

[0055] A 25th aspect of the present invention is the artificial intelligence learning method according to any one of the 18th to 24th aspects, wherein the basic data includes BMI (Body Mass Index) or blood pressure.

[0056] According to this configuration, a trained artificial intelligence used by the driving accident risk estimation device according to the third aspect is obtained.

[0057] A 26th aspect of the present invention is an artificial intelligence learning program which, when read by a computer, causes the computer to execute an artificial intelligence learning method according to any one of the 18th to 25th aspects.

[0058] According to this configuration, a trained artificial intelligence used by the driving accident risk estimation device according to any one of the first to fourth aspects is obtained. [Effects of the Invention]

[0059] As described above, the present invention provides a driving accident risk estimation device, a driving accident risk estimation method, and a driving accident risk estimation program for predicting the risk of a vehicle driving accident.The present invention also provides an artificial intelligence learning device for the driving accident risk estimation device, an artificial intelligence learning method for the driving accident risk estimation method, and an artificial intelligence learning program. [Brief explanation of the drawings]

[0060] [Figure 1] 1 is a diagram illustrating an example of the configuration of a driving accident risk estimation system including a driving accident risk estimation device according to an embodiment of the present invention. [Figure 2] 2 is a block diagram illustrating the configuration of the driving accident risk estimation device of FIG. 1. FIG. [Figure 3] 3 is a table showing an example of input data and output data of the driving accident risk estimation device shown in FIG. 2. FIG. [Figure 4] This is a schematic diagram showing the pattern of increases and decreases in heart rate over the past few days when driving accidents are more likely to occur. [Figure 5] 3 is a schematic diagram illustrating the conceptual configuration of the artificial intelligence of the driving accident risk estimation device illustrated in FIG. 2. FIG. [Figure 6] 3 is a flowchart illustrating the flow of processing of a driving accident risk estimation method implemented by the driving accident risk estimation device illustrated in FIG. 2. [Figure 7] 3 is a flowchart illustrating the flow of processing of an artificial intelligence learning method implemented by the driving accident risk estimation device illustrated in FIG. 2. DETAILED DESCRIPTION OF THE INVENTION

[0061] 1 is a diagram illustrating the configuration of a driving accident risk estimation system including a driving accident risk estimation device according to one embodiment of the present invention. In addition to a driving accident risk estimation device 101, this driving accident risk estimation system 100 includes a heartbeat sensor 1, a network 5, and servers 7 and 9. The heartbeat sensor 1, the network 5, and the servers 7 and 9 are devices connected to the driving accident risk estimation device 101 and cooperate with the driving accident risk estimation device 101.

[0062] The driving accident risk estimation device 101 is a device that contributes to preventing driving accidents by the vehicle driver 11, who is the subject of the driving accident risk estimation, by outputting driving accident risk estimation data based on heart rate data and basic data of the vehicle driver 11. In the illustrated example, the driving accident risk estimation device 101 is incorporated into the computer 10. That is, by installing and running a specific application in the computer 10, a processing device (processor) such as a central processing unit (CPU) of the computer 10 functions as the driving accident risk estimation device 101. In the illustrated example, it is assumed that the vehicle driver 11 is a taxi driver, and that the computer 10 is managed by a taxi company.

[0063] The heart rate sensor 1 is a device that detects the heartbeat of a subject and measures the heart rate. Heart rate means the number of times the heart beats in a certain period of time (for example, one minute). The heart rate sensor 1 may be a device that detects heartbeats and measures heart rate variability. As already mentioned, "heart rate variability" means fluctuations in the heart rate cycle. In the following explanation, a device that measures the heart rate will be exemplified as the heart rate sensor 1. Also, as already mentioned, in this application, "heart rate" is used as a term that also includes "pulse." Therefore, the heart rate sensor 1 may be a device that detects the "pulse" from the subject's wrist or the like and measures the "pulse rate" and "pulse rate variability (fluctuations in the pulse rate cycle)."

[0064] In the illustrated example, the heart rate sensor 1 has a communication function that automatically acquires heart rate data of the vehicle driver 11, who is the subject of measurement, and transmits the acquired data wirelessly or otherwise via a network 5 to a driving accident risk estimation device 101. The heart rate sensor 1 may be connected to the network 5 via a user terminal (not shown) such as a smartphone, or when close to the driving accident risk estimation device 101, may communicate directly with the driving accident risk estimation device 101 via short-range wireless. In the illustrated example, the heart rate sensor 1 is a wristwatch-type wearable sensor, and Fitbit (registered trademark) is one example.

[0065] If the vehicle driver 11 is a taxi driver, it is desirable to measure the heart rate while the vehicle driver 11 is sleeping or at work, when the exercise load is approximately constant. In the illustrated example, the heart rate of the vehicle driver 11 is measured while the driver is at work. If the heart rate is measured while the driver is sleeping, a sleep sensor that can be placed under the bed can also be used as the heart rate sensor 1. The heart rate sensor 1 may be constantly connected to the driving accident risk estimation device 101 and continuously transmit the heart rate data, or the heart rate data that has been temporarily accumulated may be transmitted collectively, for example, on a daily basis, or transferred via a connector, memory device, or the like. The time-series data of the heart rate obtained by the heart rate sensor 1 is input to the driving accident risk estimation device 101.

[0066] In the illustrated example, the network 5 is the Internet. The server 7 is owned by a facility such as a hospital and stores basic data such as medical records of the vehicle driver 11. The server 7 is connected to the network 5. The server 7 may be owned by an external provider and used by the facility such as a hospital. The driving accident risk estimation device 101 can access the server 7 to acquire basic physical and mental data of the vehicle driver 11, such as the BMI (Body Mass Index), blood pressure, and overall health checkup results. Information leakage can be prevented by requiring the input of, for example, an identification code (ID) and a password in communication between the driving accident risk estimation device 101 and the server 7. The basic data may be manually input by an administrator or operator of the driving accident risk estimation device 101, rather than through communication with the server 7.

[0067] The server 9 is connected to the network 5 and has constructed an artificial intelligence that can be used through the network 5. The driving accident risk estimation device 101 uses the artificial intelligence to estimate the risk of the vehicle driver 11 having an accident while driving, based on the heart rate data and basic data of the vehicle driver 11. The artificial intelligence may be constructed in the computer 10 as part of the driving accident risk estimation device 101, or may be constructed in the computer 10 separately from the driving accident risk estimation device 101 so as to be accessible by the driving accident risk estimation device 101, or may be artificial intelligence external to the computer 10, such as artificial intelligence provided by the server 9.

[0068] 2 is a block diagram illustrating the configuration of a driving accident risk estimation device 101. The driving accident risk estimation device 101 includes an interface 13, an input data receiving unit 15, a teacher data receiving unit 17, an estimation unit 19, a learning unit 21, an artificial intelligence 23, and an estimated data output unit 25. The interface 13 is a device part that enables communication between the driving accident risk estimation device 101 itself and external devices in accordance with a predetermined protocol for each external device. Communication between the driving accident risk estimation device 101 and the heart rate sensor 1, servers 7 and 9, an input device 27 such as a keyboard, an output device 29 such as a printer or display, and a storage medium 31 such as a USB memory or CD-ROM is performed via the interface 13.

[0069] The input data receiving unit 15 receives input data including the heart rate data and basic data of the vehicle driver 11. The estimation unit 19 inputs the input data received by the input data receiving unit 15 to the artificial intelligence 23, causing the artificial intelligence 23 to calculate estimated data on the risk of the vehicle driver 11 causing an accident while driving. If the artificial intelligence 23 has already learned, it outputs highly accurate estimated data on the driving accident risk. The estimated data output unit 25 outputs the estimated data calculated by the artificial intelligence 23.

[0070] The estimated data output by the estimated data output unit 25 is transmitted to, for example, the output device 29 via the interface 13. This allows, for example, a taxi company that manages the vehicle driver 11 to obtain estimated data on the risk of a driving accident. Based on the received estimated data, the taxi company can prevent driving accidents by the vehicle driver 11 by urging the driver 11 to be careful, recommending that the driver take a rest, or changing the work schedule.

[0071] As already mentioned, the computer 10 (see FIG. 1) incorporating the driving accident risk estimation device 101 may be a device managed by an organization such as a taxi company that manages the vehicle driver 11, or may be a device managed by an individual if the vehicle driver 11 is an individual who drives a private car. Alternatively, the computer 10 may be managed by a service company that provides a driving accident risk estimation service to many users. As already mentioned, communication between the heart rate sensor 1 and the computer 10 may be performed via a user terminal such as a smartphone managed by the vehicle driver 11.

[0072] The artificial intelligence 23 is able to output highly accurate estimation data through machine learning. The driving accident risk estimation device 101 has a teacher data receiving unit 17 and a learning unit 21, which makes it possible for the driving accident risk estimation device 101 to train the artificial intelligence 23 by itself, without using an external artificial intelligence learning device. In other words, the driving accident risk estimation device 101 also has a built-in artificial intelligence learning device that trains the artificial intelligence 23 through machine learning.

[0073] When the driving accident risk estimation device 101 performs machine learning, the input data receiving unit 15 receives input data including heart rate data and basic data, and the teacher data receiving unit 17 receives teacher data, which is correct driving accident risk data corresponding to the input data. The learning unit 21 inputs the input data received by the input data receiving unit 15 and the teacher data received by the teacher data receiving unit 17 to the artificial intelligence 23, thereby training the artificial intelligence 23 to estimate the teacher data from the input data. By inputting a large number of pairs of input data and teacher data that are associated with each other into the driving accident risk estimation device 101, the learning of the artificial intelligence 23 progresses and the accuracy of estimation improves.

[0074] By correlating the heart rate data and basic data collected in the past for various vehicle drivers 11 with the driving accident risk data obtained as actual results corresponding to these data and recording them, for example, in a storage medium 31, the input data receiving unit 15 and the teacher data receiving unit 17 can sequentially read out a large amount of data required for learning from the storage medium 31, and the learning unit 21 can repeatedly learn the artificial intelligence 23 for each of the read data. In this way, the driving accident risk estimation device 101 can switch between two operating modes: an estimation mode in which estimation data is calculated and output using the artificial intelligence 23, and a learning mode in which the artificial intelligence 23 undergoes machine learning. The switching of the operating modes can be instructed, for example, by the input device 27.

[0075] In the example of FIG. 2, the artificial intelligence 23 is incorporated into the computer 10 as part of the driving accident risk estimation device 101. Alternatively, as illustrated by the dotted line in FIG. 2, an artificial intelligence constructed on an external server 9 or the like may be used. In this case, the estimation unit 19, the learning unit 21, and the estimated data output unit 25 operate the external artificial intelligence via a network 5 or the like. The estimated data output unit 25 causes the external artificial intelligence to output estimated data, receives the estimated data via the interface 13 via the estimated data output unit 25, and further outputs the received estimated data to an output device 29 or the like via the interface 13 via the estimated data output unit 25. When an external artificial intelligence is used, the artificial intelligence 23 constituting part of the driving accident risk estimation device 101 becomes unnecessary.

[0076] FIG. 3 is a table diagram illustrating input data and output data of the driving accident risk estimation device 101. The input data 150 in the illustrated example includes heart rate data 51 and basic data 53. The heart rate data 51 is time-series data for a predetermined period of time in the past. The input data 150 is input to the trained artificial intelligence 23 (or 9), whereby driving accident risk estimation data 160 is calculated. Below, an example is given of how each piece of data can be expressed for handling by the driving accident risk estimation device 101. This is merely an example, and it is obvious that other ways of expressing it are also possible.

[0077] FIG. 4 is a schematic diagram showing an outline of the pattern of increases and decreases in heart rate over the past few days on days when driving accidents are likely to occur. Combining the descriptions in Non-Patent Documents 1 and 2, as shown in the example, it is predicted that an accident is likely to occur when the heart rate initially increases due to increased tension over the past few days before the accident, then decreases as fatigue reaches its limit, and then, as signs of drowsiness appear, the heart rate drops below normal levels. For this reason, it is desirable to include time-series data for, for example, the three days prior to the accident in the input data 150 as heart rate data 51. For example, a data sequence of heart rates per minute may be used as heart rate data 51, but a data sequence of average values ​​per fixed period, for example, per hour, may also be used. Various values ​​statistically processed over a fixed period, such as maximum values, minimum values, and standard deviations, may also be used.

[0078] Alternatively, since it is presumed that there is a strong correlation between the heart rate pattern in FIG. 4 and the risk of a driving accident, parameters (i.e., variables) that represent the prominence of the heart rate pattern in FIG. 4 may be defined, and these parameters may be used as the heart rate data 51. In this case, learning of the artificial intelligence 23 can be performed more efficiently. In either case, the heart rate data 51 is a representation of the change in heart rate over time in a specific format, and is still time-series heart rate data.

[0079] For example, assuming that the average daily heart rate increases by N bpm from the normal value two days before the current day and decreases by Mbps the next day (one day before the current day), these parameters N and M may be used as the heart rate data 51. Alternatively, a new parameter expressing whether the parameter N is 10 bpm or more and the parameter M is 5 bpm or more may be used as the heart rate data 51. In this case, the heart rate data 51 may be expressed as a binary value, for example, "0" or "1." Alternatively, the heart rate data 51 may be obtained by combining these various heart rate-related feature values. Data from the heart rate sensor 1, for example, minute-by-minute heart rate data, may be input to the driving accident risk estimation device 101 without processing, and a secondary feature value may be calculated, for example, in the input data receiving unit 15 or the estimation unit 19 of the driving accident risk estimation device 101. Alternatively, already processed feature values ​​may be input to the driving accident risk estimation device 101.

[0080] In the illustrated example, the basic physical and mental data 53 includes BMI, blood pressure, overall health checkup results, aptitude test results, age, and gender. Of these, BMI, blood pressure, and overall health checkup results can be acquired from a server 7 managed by a hospital, for example. Alternatively, an organization such as a taxi company that manages the vehicle driver 11 may have the vehicle driver 11 undergo a health checkup and obtain the results directly from a health checkup institution or by having the vehicle driver 11 bring the results.

[0081] The results of the aptitude test can be obtained, for example, by an organization that manages the vehicle driver 11, which conducts an aptitude test as part of its personnel management. Age, gender, and the like are matters that an organization that manages the vehicle driver 11 can naturally grasp as part of its personnel management. These data included in the basic data 53 can be stored, for example, in a memory that the computer 10 has or an external memory, and can be input to the input data receiving unit 15.

[0082] These data included in the basic data 53 are preferably digitized so as to be suitable for processing by the artificial intelligence 23. For example, BMI, blood pressure, and age, which are originally numerical attributes, can be expressed by their respective numerical values. Alternatively, each of these may be replaced with a numerical value in multiple stages to improve the efficiency of processing by the artificial intelligence 23. For example, BMI may be divided into three stages: less than 18.5, 18.5 to 35, and 35 or more, and each may be replaced with the numerical values ​​"1," "2," and "3." The comprehensive health checkup result and the aptitude test result can also be expressed by multiple numerical values ​​such as "1," "2," and "3" depending on the degree of goodness. Gender can also be expressed by numerical values ​​such as "1" and "2," for example.

[0083] The inventors of the present application have confirmed that, among the illustrated data included in the basic data 53, BMI and blood pressure are important data that affect the driving accident risk estimation data 160. Therefore, it is desirable to include at least one of BMI and blood pressure as the basic data 53. "Age" is known to be correlated with declines in cognitive function, changes in eyesight and visual field, cardiovascular disease and lifestyle-related diseases, etc., and changes in these health conditions are known to be factors that increase accident risk. Therefore, it is meaningful to include "age" in the basic data 43 as an average indicator of physical and mental health conditions.

[0084] In the illustrated example, the driving accident risk estimation data 160 is data that estimates the risk of the vehicle driver 11 causing a driving accident of four levels of severity while driving. The four levels of accident risk in the illustrated example are as follows: Risk 4: Personal injury accident or negligence rate of 80% or more Risk 3: Fault ratio: 70-30% Risk 2: 20-10% fault ratio Risk 1: No accidents Risks 1 to 4 can be expressed by, for example, numbers "1" to "4" or "0" to "3", respectively.

[0085] FIG. 5 is a schematic diagram illustrating the conceptual configuration of the artificial intelligence 23 used by the driving accident risk estimation device 101. The artificial intelligence provided by the server 9 also has a similar configuration, for example. The illustrated artificial intelligence 23 is a neural network, and has an input layer 33 in which nodes that receive data input are arranged, an output layer 37 in which nodes that output data resulting from calculations are arranged, and an intermediate layer 35 in which nodes connecting the input layer 33 and the output layer 37 are arranged. In the illustrated example, there is only one intermediate layer 35, but multiple intermediate layers may be used. The value of the previous node is transmitted to the next node, reflecting the parameters assigned to each node, i.e., the weight and bias value of each node.

[0086] The input layer 33 receives input data 150 received by the input data receiving unit 15, i.e., a set of heart rate data 51 and basic data 53. The input data is transmitted to the output layer 37 via the intermediate layer 35 while reflecting the parameters of each node. The data transmitted to the output layer 37 becomes driving accident risk estimation data 160. The estimation unit 19 (see FIG. 2) inputs a set of the heart rate data 51 and basic data 53 of the vehicle driver 11 to the input layer 33 of the artificial intelligence 23, and causes the output layer 37 to generate driving accident risk estimation data 160 of the vehicle driver 11. The estimation data output unit 25 outputs the generated estimation data after performing conversion such as rounding off decimal places, or without conversion. For example, when risks 1 to 4 are represented by the numbers "1" to "4," respectively, if the number "1.2" is output as the driving accident risk estimation data 160, it can be determined that there is a possibility of a driving accident occurring corresponding to "risk 2: fault ratio 20% to 10%."

[0087] In order for the driving accident risk estimation data 160 appearing in the nodes of the output layer 37 to be an estimate of the driving accident risk with high accuracy, it is necessary to train the artificial intelligence 23 using the actually measured driving accident risk. Learning is performed by inputting a set of heart rate data 51 and basic data 53 of a certain vehicle driver 11 accepted by the input data accepting unit 15 to the input layer 33, and inputting teacher data for the same vehicle driver 11 accepted by the teacher data accepting unit 17, i.e., the actually measured driving accident risk data, as teacher data to the output layer 37. The learning unit 21 (see FIG. 2 ) inputs such data to the artificial intelligence 23.

[0088] The artificial intelligence 23 calculates estimated data 160 of driving accident risk based on the input heart rate data 51 and basic data 53, generates it in the output layer 37, and calculates the error between the generated estimated data 160 and the driving accident risk data input as training data. The artificial intelligence 23 then changes the parameters of each node from the output layer 37 to the input layer 33, for example, using a well-known error backpropagation algorithm, so that error-free estimated data 160 is generated. This function is provided within the artificial intelligence 23 itself. By preparing many pairs of input data and training data and repeating learning, the artificial intelligence 23 can generate highly accurate estimated data 160. When training the artificial intelligence 23, it is also possible to adjust the number of intermediate layers 35 and the number of nodes in each layer to optimal values. Such techniques are also well known.

[0089] As already mentioned, the teacher data received by the teacher data receiving unit 17 is actually measured driving accident risk data. That is, for cases in which accidents (or no accidents) of driving accident levels corresponding to, for example, risks 1 to 4 have actually occurred in the past, the teacher data is the driving accident risk data (for example, "1", "2", etc.) corresponding to the driving accident level. The estimated data output by the artificial intelligence 23 that has been trained using the teacher data prepared in this way estimates which of risks 1 to 4 the driving accident risk is, i.e., which of the driving accident levels corresponding to risks 1 to 4 is at risk of occurring.

[0090] The past actual data collected to prepare training data contains overwhelmingly more data without accidents (e.g., risk 1) (so-called "majority data" or "negative data") than data with accidents (e.g., risks 2 to 4) (so-called "minority data" or "positive data"). Several methods are already widely known as techniques for efficiently training artificial intelligence23 based on such so-called "imbalanced data," and these can be adopted.

[0091] For example, known techniques include "undersampling," which reduces a large amount of negative data for training, "oversampling," which conversely increases a small amount of positive data for training, and "bagging" (Bootstrap AGGregatING), which randomly selects multiple training datasets, trains multiple AIs 23 using those datasets, and determines the final estimated data 160 by majority voting of the results obtained by the trained AIs 23. It is also possible to combine these techniques, for example, combining "undersampling" and "bagging," and this combination is known to be particularly effective.

[0092] Even if these techniques are used, the learning of the artificial intelligence 23 is still performed by inputting a set of heart rate data 51 and basic data 53 of a vehicle driver 11 received by the input data receiving unit 15 into the input layer 33, as already mentioned, and inputting teacher data for the same vehicle driver 11 received by the teacher data receiving unit 17, i.e., actually measured driving accident risk data, into the output layer 37 as teacher data.

[0093] The learning unit 21 (see FIG. 2 ) may perform operations, such as “undersampling” or “oversampling,” on a pair of input data received by the input data receiving unit 15 and teacher data received by the teacher data receiving unit 17, to increase the ratio of data pairs in which accidents actually occurred, i.e., “positive data,” to data pairs in which no accidents occurred, i.e., “negative data,” before inputting the data pairs to the artificial intelligence 23. Alternatively, the data pairs manipulated in this manner may be recorded, for example, in a storage medium 31, so that the input data receiving unit 15 and the teacher data receiving unit 17 can accept the manipulated data. By having the artificial intelligence 23 learn using data that has undergone such operations, the artificial intelligence 23 can estimate driving accident risks with higher accuracy.

[0094] The distinction between "positive data" and "negative data" may be made for each level of driving accident with different severity. For example, to estimate the risk of a driving accident occurring at a level corresponding to risk 3, data in which a driving accident at a level corresponding to risk 3 has occurred may be treated as "positive data" and other data as "negative data," and the ratio of positive data may be increased, and the set of input data and training data may then be input to the artificial intelligence 23. By having the artificial intelligence 23 learn using data that has undergone such an operation, the artificial intelligence 23 can estimate the risk of a driving accident with even higher accuracy.

[0095] 5, other types of artificial intelligence may be used for the artificial intelligence 23, such as a decision tree-based LGBM (Light GBM; manufactured by Microsoft). LGBM has the advantage of being able to easily analyze which variables in the input data 150 play an important role in the estimated data 160 to be output, and is therefore particularly useful in the process of building the artificial intelligence 23.

[0096] The driving accident risk estimation device 101 (FIG. 2) realizes a driving accident risk estimation method and an artificial intelligence learning method. FIGS. 6 and 7 show examples of the processing procedures of the driving accident risk estimation method and the artificial intelligence learning method.

[0097] FIG. 6 is a flowchart illustrating the processing flow of the driving accident risk estimation method realized by the driving accident risk estimation device 101. When the processing starts, the input data receiving unit 15 receives input data (S1). Next, the estimation unit 19 inputs the input data to the trained artificial intelligence 23, causing the artificial intelligence 23 to calculate estimated data (S3). Next, the estimated data output unit 25 outputs the estimated data calculated by the artificial intelligence 23 (S5). Next, when the driving accident risk estimation device 101 should repeat the processing based on a user instruction or the like (Yes in S7), the processing returns to S1. As a result, the input data receiving unit 15 receives new input data. When the driving accident risk estimation device 101 should not repeat the processing (No in S7), it ends the processing.

[0098] FIG. 7 is a flowchart illustrating the flow of processing of the artificial intelligence learning method implemented by the driving accident risk estimation device 101. When the processing starts, the input data receiving unit 15 receives input data (S21). Furthermore, the teacher data receiving unit 17 receives teacher data (S23). Either of the processing steps S21 and S23 may be performed first, or they may be performed simultaneously. Next, the learning unit 21 inputs the input data and the teacher data to the artificial intelligence 23, thereby training the artificial intelligence to estimate the teacher data from the input data (S25). As already mentioned, the learning unit 21 may perform an operation on the sets of input data and teacher data to increase the ratio of "positive data" to "negative data" before inputting the data sets to the artificial intelligence 23. Alternatively, such an operation may already be performed on the input data received by the input data receiving unit 15 in processing S21 and the sets of teacher data received by the teacher data receiving unit 17 in processing S23.

[0099] Next, when the driving accident risk estimation device 101 needs to repeat the process based on a user instruction or the like (Yes in S27), it returns the process to S21. As a result, the input data receiving unit 15 receives new input data, and the teacher data receiving unit 17 receives new teacher data. When the driving accident risk estimation device 101 does not need to repeat the process (No in S27), it ends the process.

[0100] As already mentioned, in the example shown in FIG. 1, the driving accident risk estimation device 101 (see FIG. 2) is incorporated into the computer 10. By installing and running a specific application, i.e., a program, on the computer 10, the computer 10 functions as the driving accident risk estimation device 101. This program may be supplied via the network 5, or may be supplied by a storage medium 31 (see FIG. 2) such as a CD-ROM.

[0101] (Other embodiments) (1) Fig. 3 shows an example in which the input data 150 includes the heart rate data 51 and the basic data 53. In contrast, the input data 150 can include data other than the heart rate data 51 and the basic data 53 as long as they are effective for the calculation of the driving accident risk estimation data 160. It is also possible to include time series data other than the heart rate data 51. (2) In the above description, as the heart rate sensor 1, a device for measuring the heart rate has been exemplified. As already described, the heart rate sensor 1 may be a device that detects pulsations and measures heart rate variations.

[0102] (Validity test of feature amount) The inventors of the present application have conducted a demonstration test to search for effective feature amounts. The demonstration test is being carried out in a form in which the invention is not known to the participating cooperators. The heart rate data 51 was acquired by having the vehicle driver 11, who is a taxi driver, wear only Fitbit (registered trademark) during driving duties. During the continuation of the test, it was found that a feature amount of "when the average heart rate per day rises by 10 bpm or more above the normal value and then drops by 5 bpm or more on the next day" has a particularly high correlation with the occurrence of an accident. When this "rise → drop" pattern appears continuously, it is estimated that the risk of an accident occurring in a few days may increase, and intensive attention is called to the operation manager. As a result, for the 40 drivers wearing Fitbit (registered trademark), no accident occurred during the test period of about one month. On the other hand, for the 60 non-wearing drivers, 5 accidents were confirmed during the same period, and the effectiveness of such a feature amount was confirmed. This can be said to also support the fact that an accident is likely to occur when the pattern in Fig. 4 appears.

[0103] (Other feature amounts) Hereinafter, specific examples of the statistically processed feature amounts described while referring to Figs. 3 and 4 will be given. <I. Feature amounts created manually> 1. <Basic statistical amounts per day)> Average heart rate per day Maximum heart rate per day Minimum heart rate per day Daily heart rate standard deviation Daily heart rate variation coefficient 2. <Basic statistics for time granularity (5 minutes to 24 hours, moving average, etc.)> Average heart rate for 5 minutes to 24 hours Standard deviation of average heart rate over 5 minutes to 24 hours 5-minute to 24-hour maximum heart rate Minimum heart rate in 5-minute to 24-hour increments Moving average heart rate (selectable window: 5 minutes to 24 hours) Deviation from moving average heart rate 3. Trends and fluctuation patterns Average increase or decrease in heart rate compared to the previous day Average heart rate gradient over three days (slope per day) Heart rate increases for two consecutive days → Flag for decrease the next day Maximum heart rate change from 5 minutes to 24 hours (daily range) Heart rate change rate from 5 minutes to 24 hours 4.<Features based on medical domain knowledge> Total time spent in the unhealthy heart rate zone (e.g., 80-100 bpm) Total time spent in the stress zone (e.g., over 100 bpm) Total time spent in the low heart rate zone (e.g., below 55 bpm) Number of times your heart rate increased suddenly (e.g., by 10 bpm or more in one minute) Number of times your heart rate suddenly drops (e.g., drops of 10 bpm or more in one minute) 5. Long-term / continuous pattern detection Average heart rate trend over 3 days (last 3 days vs. historical average) Average heart rate over three nights (assuming sleep times) 3-day frequent heart rate variability cycle (cycle detection) 6. <Derived features specialized for "rise → fall") Flags "Increase > X" and "Decrease > Y on the next day" for average heart rate over two consecutive days Frequency of "rise → fall pattern" occurrence (how many times it occurred in 3 days) Maximum swing width (highest - lowest) during the rising → falling pattern <II. Features created by automatic feature engineering> 1. <Automatic generation of statistical quantities> Various representative values, variance, and standard deviation Typical summary statistics such as quartiles and range (max - min) Higher - order statistics such as skewness and kurtosis of the distribution Features such as "total energy on the time axis" and "sample entropy". 2. <Time axis - time series pattern system> Time - lag feature Automatically generate past values such as 1 minute ago, 5 minutes ago, 10 minutes ago, etc. as features directly. Automatically calculate "average, maximum, minimum, standard deviation", etc. over the past 3 minutes, 5 minutes, 10 minutes, etc. Quantify the mean value smoothed exponentially over a specified span as a feature. Coefficients extracted from automatic AR model fitting. 3. <Frequency domain system> Extract a certain number of each frequency component (real part, imaginary part) of the discrete Fourier transform (FFT). Representative values such as total power and center frequency summarized from the FFT spectrum. Perform wavelet transform and quantify the energy in a specific scale band as a feature. 4. <Signal event - pattern detection system> The number of locations where the heart rate is at a local peak (hill). The average and standard deviation of the peak height (= how much the heart rate has increased). Statistics of the coefficients in the continuous wavelet transform. Extract how many times the signal passes near zero (or a reference value) as an indicator of variability. 5. <Change detection - interval count system> The number of times the signal exceeds / falls below the daily average (or moving average). How many times is there a timing of "changing by ○ bpm or more per minute"? Detect spikes from the delta value (front - back difference) and count the number of occurrences. The length of the section where there is a monotonically increasing or decreasing trend over a short period of time and the trend are scored. 6.<Features by category and threshold> The total time and number of times your heart rate was in a particular zone (e.g., below 50 bpm, above 120 bpm, etc.). The number of times your heart rate exceeds x bpm (e.g., to determine stress). The number of times your heart rate dropped below y bpm (e.g., due to an unhealthy low heart rate or excessive relaxation). 7. Autocorrelation Systems Autocorrelation coefficients for lag X (1 minute, 5 minutes, 10 minutes later). Coefficient of partial autocorrelation. How many prominent peaks are there in the autocorrelation function? 8. <Signal Shape System> The number of times the signal slope switches from positive to negative and negative to positive. Aggregated and summarized absolute deviation from the moving average. The length of the longest continuous period above (or below) the moving average. Note: Automatic feature engineering tools typically employ a mechanism that generates hundreds to thousands of these features at once and then selects the most effective features from among them (feature selection). [Explanation of symbols]

[0104] 1 Heart rate sensor, 5 Network, 7, 9 Server, 10 Computer, 11 Vehicle driver, 13 Interface, 15 Input data reception unit, 17 Teacher data reception unit, 19 Estimation unit, 21 Learning unit, 23 Artificial intelligence, 25 Estimation data output unit, 27 Input device, 29 Output device, 31 Memory, 33 Input layer, 35 Intermediate layer, 37 Output layer, 51 Heart rate data, 53 Basic data, 100 Driving accident risk estimation system, 101 Driving accident risk estimation device, 150 Input data, 160 Driving accident risk estimation data.

Claims

1. An artificial intelligence learning device that trains an artificial intelligence to calculate estimated data on the risk of a vehicle driver causing an accident while driving, from input data including time-series data for a predetermined period in the past, including the driver's heart rate or heart rate variability, and basic data, which is data on the driver's mind and body, comprising: an input data receiving unit that receives input data including time-series data for a predetermined period in the past, including the heart rate or heart rate variability of the vehicle driver, and basic data, which is data related to the driver's mind and body; a teacher data receiving unit that receives teacher data corresponding to the input data, the teacher data being actual data on accidents and no accidents while the driver is driving; a learning unit that inputs the input data received by the input data receiving unit and the teacher data received by the teacher data receiving unit into the artificial intelligence, thereby training the artificial intelligence to estimate the teacher data from the input data, the time-series data includes data expressed in numerical values ​​based on data measured by a sensor, The basic data includes data expressed in numerical values, An artificial intelligence learning device, wherein the performance data includes data expressed in stages using numerical values.

2. 2. The artificial intelligence learning device according to claim 1, wherein the input data received by the input data receiving unit and the teacher data received by the teacher data receiving unit have an increased ratio of positive data to negative data.

3. The artificial intelligence learning device of claim 1, wherein the learning unit performs processing to increase the ratio of positive data to negative data for the input data accepted by the input data accepting unit and the teacher data accepted by the teacher data accepting unit before inputting them to the artificial intelligence.

4. 2. The artificial intelligence learning device according to claim 1, wherein the data on the driver's accidents and no accidents while driving is data on the driver's accidents and no accidents at multiple stages of severity.

5. The artificial intelligence learning device of claim 4, wherein the input data accepted by the input data accepting unit and the teacher data accepted by the teacher data accepting unit have a higher ratio of positive data to negative data for each of the multiple levels of severity.

6. The artificial intelligence learning device described in claim 4, wherein the learning unit performs processing to increase the ratio of positive data to negative data for each of the multiple levels of severity for the input data accepted by the input data accepting unit and the teacher data accepted by the teacher data accepting unit, and then inputs the data to the artificial intelligence.

7. 2. The artificial intelligence learning device according to claim 1, wherein the time series data includes a variable that expresses the prominence of a feature in which the heart rate rises once and then falls within the predetermined past period.

8. The artificial intelligence learning device according to claim 1 , wherein the basic data includes a BMI (Body Mass Index) or a blood pressure.

9. An artificial intelligence learning method for a driving accident risk estimation device, which trains an artificial intelligence to calculate estimated data on the risk of a vehicle driver causing an accident while driving from input data including time series data for a predetermined period in the past, including the heart rate or heart rate variability of the vehicle driver, and basic data, which is data on the driver's mind and body, comprising: A computer receives input data including time-series data of a vehicle driver for a predetermined period in the past, the time-series data including the heart rate or heart rate variability, and basic data relating to the driver's mind and body; The computer receives training data, which is data on the driver's accidents and no accidents while driving, corresponding to the input data; The computer inputs the received input data and the received teacher data into the artificial intelligence, thereby training the artificial intelligence to estimate the teacher data from the input data, the time-series data includes data expressed in numerical values ​​based on data measured by a sensor, The basic data includes data expressed in numerical values, An artificial intelligence learning method, wherein the performance data includes data expressed in stages by numerical values.

10. 10. The artificial intelligence learning method according to claim 9, wherein the accepted input data and the accepted teacher data have an increased ratio of positive data to negative data.

11. The artificial intelligence learning method of claim 9, wherein the training of the artificial intelligence involves processing the received input data and the received teacher data to increase the ratio of positive data to negative data before inputting them to the artificial intelligence.

12. The artificial intelligence learning method according to claim 9, wherein the data on the driver's accidents and no accidents while driving is data on the driver's accidents and no accidents at multiple stages of different severity.

13. The artificial intelligence learning method according to claim 12, wherein the ratio of positive data to negative data is increased in the accepted input data and the accepted teacher data for each of the multiple stages of different severity.

14. The artificial intelligence learning method of claim 12, wherein the training of the artificial intelligence involves processing the received input data and the received teacher data to increase the ratio of positive data to negative data for each of the multiple levels of severity, and then inputting the data to the artificial intelligence.

15. 10. The artificial intelligence learning method according to claim 9, wherein the time series data includes a variable that expresses the prominence of a feature in which the heart rate rises once and then falls within the predetermined past period.

16. The artificial intelligence learning method according to claim 9 , wherein the basic data includes BMI (Body Mass Index) or blood pressure.

17. An artificial intelligence learning program that, when read by a computer, causes the computer to execute the artificial intelligence learning method according to any one of claims 9 to 16.

18. an input data receiving unit that receives input data including time-series data for a predetermined period in the past, including the heart rate or heart rate variability of the vehicle driver, and basic data, which is data related to the driver's mind and body; an estimation unit that inputs the input data received by the input data receiving unit into a trained artificial intelligence to calculate estimated data on the risk of the driver causing an accident while driving; an estimated data output unit that outputs the estimated data calculated by the artificial intelligence, the time-series data includes data expressed in numerical values ​​based on data measured by a sensor, The basic data includes data expressed in numerical values, The estimated data includes data evaluated in stages by numerical values, A driving accident risk estimation device, wherein the time series data includes a variable that expresses the prominence of a characteristic in which the heart rate rises once and then falls within the specified past period.

19. 19. The driving accident risk estimation device according to claim 18, wherein the estimated data of the risk is estimated data of the risk that the driver will cause a driving accident at a plurality of stages of different severity while driving.

20. The driving accident risk estimation device according to claim 18, wherein the basic data includes a BMI (Body Mass Index) or a blood pressure.

21. A computer receives input data including time-series data of a vehicle driver for a predetermined period in the past, the time-series data including the heart rate or heart rate variability, and basic data relating to the driver's mind and body; The computer inputs the received input data into a trained artificial intelligence, thereby causing the artificial intelligence to calculate estimated data on the risk of the driver causing an accident while driving; The computer outputs the estimated data calculated by the artificial intelligence, the time-series data includes data expressed in numerical values ​​based on data measured by a sensor, The basic data includes data expressed in numerical values, The estimated data includes data evaluated in stages by numerical values, A method for estimating a risk of a driving accident, wherein the time series data includes a variable that expresses the prominence of a characteristic in which the heart rate rises once and then falls within the specified past period.

22. The driving accident risk estimation method according to claim 21 , wherein the estimated data of the risk is estimated data of the risk that the driver will cause a driving accident at a plurality of stages of different severity while driving.

23. 22. The method of claim 21, wherein the basic data includes a BMI (Body Mass Index) or a blood pressure.

24. A driving accident risk estimation program that, when read by a computer, causes the computer to execute the driving accident risk estimation method according to any one of claims 21 to 23.

Images