Driver abnormality prediction method and apparatus

Abstract

To provide a method and device for detecting a sign of an abnormality of a driver, which can detect a sign of an abnormality of a driver while achieving both high responsiveness and high reliability.SOLUTION: A method for detecting a sign of an abnormality of a driver comprises: Step B of estimating an interval BI of heartbeat of a driver, on the basis of heartbeat detected by a camera of a driver; Step C of obtaining time-series data about intervals of heartbeat of the driver in a predetermined time; Step D of standardizing the time-series data about the intervals of heartbeat; Step E of calculating a probability PA at which the standardized intervals of heartbeat are included within a predetermined range (an A range) with 0 as a center, in the standardized time-series data about the intervals of heartbeat CG-BIz; and Step F of determining that a sign of an abnormality of the driver has been detected, when the probability PA is larger than a predetermined abnormality sign threshold TA.SELECTED DRAWING: Figure 7

Description

Technical Field

[0001] The present invention relates to a driver abnormality prediction detection method and apparatus for detecting signs of driver abnormality while driving a vehicle.

Background Art

[0002] Conventionally, a technique for detecting signs of driver abnormality based on changes in the driver's driving behavior is known. However, in order to detect signs of driver abnormality from changes in driving behavior, it is necessary to monitor changes in driving behavior for at least about several minutes, and it is impossible to deal with cases where the driver's driving ability is lost in a short time (for example, epilepsy, heart attack, stroke, etc.).

[0003] Therefore, it has been considered to detect signs of driver abnormality earlier than when changes in driving behavior appear by detecting changes in biological functions that reduce the driver's driving ability. For example, the autonomic nervous system, which consists of the sympathetic nervous system and the parasympathetic nervous system, regulates involuntary biological functions independently of human consciousness. When an abnormality occurs in the autonomic nervous system due to some disease, the influence appears in biological functions controlled by the autonomic nervous system, such as circulation, sweating, pupils, etc. Therefore, it is considered that by measuring changes in biological information representing the states of these biological functions, such as changes in heart rate, sweating amount, pupil diameter, etc., an abnormality in the autonomic nervous system can be detected and used to estimate signs of driver abnormality.

[0004] In particular, from the viewpoints that the heart rate can be measured non - contact among biological information and that the in - vehicle camera mounted on the vehicle can be utilized as a sensor, detecting an abnormality in the autonomic nervous system based on the driver's heart rate has been studied.

[0005] Beyond drivers, technologies for detecting biological abnormalities using heart rate data have also been considered. For example, a technology has been proposed to detect signs of epileptic seizures using heart rate variability indices (SDNN, etc.) analyzed in the time domain and heart rate variability indices (HF, LF, LF / HF, etc.) analyzed in the frequency domain (see, for example, Patent Document 1). Furthermore, technologies have been proposed to determine sudden changes in physical condition using biological signals in the VLF band and ULF band as heart rate variability indices analyzed in the frequency domain (see, for example, Patent Document 2), and to detect the state of the heart by nonlinearly analyzing fluctuations in heart rate variability signals (see, for example, Patent Document 3).

[0006] Furthermore, it has been proposed to use the non-Gaussian parameter λ (non-Gaussian index) of the probability density distribution (PDF) as another heart rate variability index obtained through nonlinear analysis (see, for example, Non-Patent Document 1). In this method, the probability density function of heart rate variability and its non-Gaussian index λ are determined using a 24-hour continuously recorded Holter electrocardiogram of chronic heart failure patients. It has been reported that there was a significant difference in the non-Gaussian index λ obtained from the acquired data between survivors and non-survivors during the observation period after data acquisition. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2015-112423 [Patent Document 2] Japanese Patent Publication No. 2020-130264 [Patent Document 3] Japanese Patent Publication No. 2010-184041 [Non-patent literature]

[0008] [Non-Patent Document 1] Ken Seino, "Non-Gaussian Statistics of Intermittent Fluctuations and Their Application to Heart Rate Variability Analysis," Journal of the Japan Society for Precision Engineering, Vol. 77, No. 2, 2011, pp. 153-157. [Overview of the project] [Problems that the invention aims to solve]

[0009] However, conventional methods, such as those described in Patent Document 1, have good responsiveness but suffer from susceptibility to noise. That is, while it is possible to determine abnormalities based on relatively short-term heart rate data, the reliability is low. Furthermore, conventional methods, such as those described in Patent Document 2, require longer data to obtain biosignals in the VLF and ULF bands, resulting in poor responsiveness. Moreover, conventional methods, such as those described in Patent Document 3, produce unstable and unreliable results when using short-term scales. To improve reliability, it is necessary to use long-term scales or acquire short-term scale data multiple times over a long period, which in either case is time-consuming. In addition, conventional methods, such as those described in Non-Patent Document 1, require the acquisition of long-term data, such as 24 hours, which also results in poor responsiveness.

[0010] Therefore, it is difficult to use the conventional techniques described above for detecting abnormal signs in drivers, which require both high responsiveness and high reliability.

[0011] This invention was made to solve these problems, and aims to provide a driver abnormality prediction method and apparatus that can detect driver abnormality signs while achieving both high responsiveness and high reliability. [Means for solving the problem]

[0012] To solve the above-mentioned problems, the present invention provides a driver abnormality prediction detection method for detecting signs of abnormality in a driver operating a vehicle, comprising: step A, a pulse detection device detecting the driver's pulse; step B, a processor estimating the driver's pulse interval based on the pulse detected by the pulse detection device; and a processor calculating the driver's pulse interval. Based on time-series data, the average value of the time-series data for the relevant pulse interval is shifted to 0, and the coarse-grained pulse interval is calculated by coarsening it over the most recent predetermined time scale. The process C involves the processor, coarse graining Pulse interval By dividing by the standard deviation of the coarse-grained pulse interval, standardized Calculate the coarse-grained pulse interval.Process D and the processor standardized coarse graining In the time-series data of pulse intervals, the values ​​are standardized to a predetermined range centered around 0. coarse graining The process includes step E, which calculates the probability of a pulse interval occurring, and step F, which determines that the processor has detected a driver malfunction if the probability is greater than a predetermined abnormality prediction threshold. The predetermined range centered at 0 is the range of standardized coarse-grained pulse intervals in the region where the peak of the probability distribution of standardized coarse-grained pulse intervals in the abnormal group (standardized by dividing the coarse-grained pulse intervals obtained from the group of subjects with autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulse intervals) protrudes higher than the peak of the probability distribution of standardized coarse-grained pulse intervals in the normal group (standardized by dividing the coarse-grained pulse intervals obtained from the group of subjects with normal autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulse intervals). .

[0013] According to the present invention configured in this way, the processor estimates the driver's pulse interval based on the pulse detected by the pulse detection device, acquires time-series data of the driver's pulse interval for a predetermined period, standardizes the time-series data of the pulse interval, and determines that a driver malfunction has been detected if the probability that the standardized pulse interval falls within a predetermined range centered on 0 in the standardized time-series data is greater than a predetermined abnormality prediction threshold. Therefore, even from time-series data of pulse intervals of a short period of time, such as several tens of seconds, it is possible to acquire time-series data of the standardized pulse interval, calculate the probability that the standardized pulse interval falls within a predetermined range centered on 0, and detect an abnormality. This makes it possible to detect driver malfunctions with both high responsiveness and high reliability. Furthermore, it is possible to determine abnormality warning signs in areas where there is a clear difference between cases with and without autonomic nervous system abnormalities. This can improve the reliability of driver abnormality warning detection.

[0014] In the present invention, preferably, a predetermined range centered on 0 is The standardized coarse-grained pulse interval is in the range of -0.03 to 0.03. .

[0016] In the present invention, preferably, the abnormality prediction threshold is Normal group standardized coarse graining If the value obtained by integrating the probability distribution of pulse intervals over a predetermined range centered on 0 is greater than the value obtained by integrating it, Abnormal group standardized coarse graining The value obtained by integrating the probability distribution of pulse intervals over a predetermined range centered on 0 is smaller than the value obtained by integrating it.

[0017] According to the present invention configured in this way, the boundary between cases where the autonomic nervous system is abnormal and cases where it is normal within a predetermined range centered on 0 can be used as the abnormality prediction threshold to determine whether an abnormality is present. This improves the reliability of the driver's abnormality prediction detection.

[0018] In the present invention, preferably, steps C, D, E, and F are executed each time the pulsation interval is estimated in step B.

[0019] According to the present invention configured as described above, when a sign of an abnormality in the autonomic nervous system of the driver occurs, it can be detected promptly. Thereby, it is possible to detect a sign of abnormality of the driver while achieving both high responsiveness and high reliability.

[0020] From another aspect, the present invention is a driver abnormality prediction detection device that detects a sign of abnormality of a driver who drives a vehicle, and includes a pulsation detection device that detects the pulsation of the driver, a memory that stores a program, and a processor that executes the program. The processor estimates the pulsation interval of the driver based on the pulsation detected by the pulsation detection device, obtains time-series data of the pulsation interval of the driver for a predetermined time, Acquired Time-series data of the pulsation interval Based on this, the average value of the time series data of the pulse interval is shifted to 0, and the coarse-grained pulse interval is calculated by coarsening it on the most recent predetermined time scale. is By dividing by the standard deviation of the coarse-grained pulse interval, normalized The coarse-grained pulse interval was calculated. and, in the time-series data of the normalized pulsation interval, calculates the probability that the normalized pulsation interval falls within a predetermined range centered on 0, and determines that a sign of abnormality of the driver has been detected when the probability is greater than a predetermined abnormality prediction threshold value. coarse graining is configured to coarse graining The predetermined range centered at 0 is the range of standardized coarse-grained pulse intervals in which the peak of the probability distribution of standardized coarse-grained pulse intervals in the abnormal group (standardized by dividing the coarse-grained pulse intervals obtained from the group of subjects with autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulse intervals) protrudes higher than the peak of the probability distribution of standardized coarse-grained pulse intervals in the normal group (standardized by dividing the coarse-grained pulse intervals obtained from the group of subjects with normal autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulse intervals). .

[0021] Also according to the present invention configured as described above, even from time-series data of the pulsation interval for a short time of about several tens of seconds, it is possible to obtain time-series data of the normalized pulsation interval, calculate the probability that the normalized pulsation interval falls within a predetermined range centered on 0, and detect a sign of abnormality. Thereby, it is possible to detect a sign of abnormality of the driver while achieving both high responsiveness and high reliability.

[0022] In the present invention, preferably, the pulsation detection device is a driver camera that photographs the body surface of the driver.

[0023] According to the present invention configured in this way, driver malfunction prediction can be performed at low cost and without burdening the driver, without using special equipment for pulse detection. [Effects of the Invention]

[0024] According to the driver abnormality prediction method and apparatus of the present invention, it is possible to detect driver abnormalities while achieving both high responsiveness and high reliability. [Brief explanation of the drawing]

[0025] [Figure 1] This is a block diagram showing a schematic configuration of a vehicle equipped with a driver abnormality prediction device according to an embodiment of the present invention. [Figure 2] This is a block diagram showing the functional configuration of a driver state estimation device according to an embodiment of the present invention. [Figure 3] This is a flowchart of the driver state estimation process according to an embodiment of the present invention. [Figure 4] This is a flowchart of the first parameter learning process according to an embodiment of the present invention. [Figure 5] This is a flowchart of the second parameter learning process according to an embodiment of the present invention. [Figure 6] This is a flowchart of the coarse-graining process for pulsation intervals according to an embodiment of the present invention. [Figure 7] This is a flowchart of the abnormality prediction process according to an embodiment of the present invention. [Figure 8] This graph shows the probability distribution of the standardized coarse-grained pulse interval (CG-BIz) in a group of subjects with a normal autonomic nervous system and a group of subjects with an abnormal autonomic nervous system. [Figure 9] This is a flowchart of the abnormality prediction process according to the second embodiment of the present invention. [Figure 10] This is a time chart showing the time evolution of the pulsation interval before an epileptic seizure occurs, the coarse-grained pulsation interval, and the probability that the coarse-grained pulsation interval falls within region A and region B, respectively. [Figure 11]This is a time chart showing the time evolution of the pulse interval, coarse-grained pulse interval, and the probability that the coarse-grained pulse interval falls within region A and region B, respectively, in a state of sympathetic nervous system dysfunction. [Figure 12] This is a time chart showing the time evolution of the pulse interval, coarse-grained pulse interval, and the probability that the coarse-grained pulse interval falls within region A and region B, respectively, in a state of parasympathetic nervous system dysfunction. [Figure 13] This is a time chart showing the time evolution of the pulse interval, coarse-grained pulse interval, and the probability that the coarse-grained pulse interval falls within region A and region B, respectively, in a healthy individual. [Modes for carrying out the invention]

[0026] Hereinafter, with reference to the attached drawings, a driver abnormality prediction method and apparatus according to embodiments of the present invention will be described.

[0027] [composition] First, the configuration of the driver abnormality prediction detection device according to this embodiment will be described with reference to Figures 1 and 2. Figure 1 is a block diagram showing the schematic configuration of a vehicle equipped with the driver abnormality prediction detection device, and Figure 2 is a block diagram showing the functional configuration of the driver state estimation device.

[0028] As shown in Figure 1, the vehicle 1 includes a driver state estimation device 10 configured to estimate the state of the driver of the vehicle 1, a driver camera 20 for photographing the driver, a vehicle control device 30 configured to control the vehicle 1 based on information output from the driver state estimation device 10, a display 31 and a warning device 32 that output information in response to control by the driver state estimation device 10, a vehicle actuator 40 that operates the power source, brakes, steering system, etc. of the vehicle 1 in response to control by the vehicle control device 30, and lighting devices 41 (including headlights, brake lights, and turn signals) and a horn 42 that operate in response to control by the vehicle control device 30. The driver camera 20 photographs the driver's body surface (more specifically, the face surface) and can detect pulse waves (pulses) in blood vessels near the body surface from changes in brightness of visible light and near-infrared light in the obtained image.

[0029] The driver state estimation device 10 mainly comprises one or more processors 10a, such as CPUs, which perform various processes, and one or more memories 10b (such as ROM, RAM, or hard disks) that store programs to be executed by the processors 10a, and various data necessary for the execution of these programs.

[0030] Next, as shown in Figure 2, the processor 10a of the driver state estimation device 10 functions as a head posture estimation unit 11, a driver abnormal posture determination unit 12, and an abnormality prediction detection device 100. The abnormality prediction detection device 100 includes a pulse interval estimation unit 101, a driver personal authentication unit 102, a personal parameter selection unit 103, a first parameter learning unit 106, a second parameter learning unit 107, a pulse interval coarse-graining processing unit 108, a buffer 109, and a driver abnormality prediction determination unit 110. In addition, the memory 10b of the driver state estimation device 10 stores an abnormal posture threshold 13, a driver-specific first parameter 104, a driver-specific second parameter 105, and an abnormality prediction threshold 111.

[0031] Specifically, the head posture estimation unit 11 estimates the driver's head posture based on the driver's image captured by the driver camera 20. The driver abnormal posture determination unit 12 determines whether the driver's posture is abnormal based on the driver's head posture estimated by the head posture estimation unit 11 and the abnormal posture threshold 13 stored in the memory 10b. Based on the determination result, the vehicle control device 30 controls various parts of the vehicle 1 and outputs warnings, etc., via the display 31 and warning device 32.

[0032] The pulse interval estimation unit 101 estimates the driver's pulse interval based on the driver's video captured by the driver camera 20. The driver personal authentication unit 102 identifies the driver using facial recognition technology based on the driver's video captured by the driver camera 20.

[0033] The personal parameter selection unit 103 obtains the first and second parameters corresponding to the driver identified by the driver personal authentication unit 102 from the memory 10b. The pulse interval coarse-graining processing unit 108 uses the first and second parameters obtained by the personal parameter selection unit 103 to perform coarse-graining processing on the driver's pulse interval estimated by the pulse interval estimation unit 101.

[0034] The first parameter learning unit 106 learns driver-specific first parameters based on the driver's pulse interval estimated by the pulse interval estimation unit 101 and stores them in memory 10b. The second parameter learning unit 107 learns driver-specific second parameters based on the coarse-grained pulse interval calculated by the pulse interval coarse-graining processing unit 108 and stores them in memory 10b.

[0035] The driver abnormality prediction unit 110 determines whether or not signs of a driver abnormality have been detected based on the coarse-grained pulse interval calculated by the pulse interval coarse-graining processing unit 108 and the abnormality prediction threshold 111 stored in the memory 10b, and outputs the determination result to the driver abnormal posture determination unit 12. If the driver abnormality prediction unit 110 detects signs of a driver abnormality, the driver abnormality prediction unit 110 outputs an alarm or the like via the display 31 or alarm device 32, and the driver abnormal posture determination unit 12 changes the abnormal posture threshold 13 stored in the memory 10b in a direction that makes it more likely that the driver posture will be judged as abnormal, and determines whether or not the driver posture is abnormal.

[0036] [process] Next, the processes performed in the driver state estimation device 10 described above will be explained with reference to Figures 3 to 8. Figure 3 is a flowchart of the driver state estimation process according to this embodiment. Figure 4 is a flowchart of the first parameter learning process according to this embodiment. Figure 5 is a flowchart of the second parameter learning process according to this embodiment. Figure 6 is a flowchart of the coarse-graining process for the pulse interval according to this embodiment. Figure 7 is a flowchart of the abnormality prediction process according to this embodiment. Figure 8 is a graph showing the probability distribution of the occurrence of the standardized coarse-grained pulse interval CG-BIz for a group of subjects with a normal autonomic nervous system and a group of subjects with an abnormality. In explaining the flowcharts in Figures 3 to 7, although the processes included in these flowcharts are actually executed by the processor 10a of the driver state estimation device 10, for the sake of ease of understanding, they will be explained as being executed by each functional unit shown as a block in Figure 2.

[0037] The driver state estimation process shown in Figure 3 starts when the power to vehicle 1 is turned ON and ends when the power to vehicle 1 is turned OFF or after the processing of the final step has been completed.

[0038] First, in step S1, the driver personal authentication unit 102 identifies the driver using known facial recognition technology based on the driver's video input from the driver camera 20. The facial data of each driver used for facial recognition is acquired in advance by the driver camera 20 and stored in the memory 10b.

[0039] Next, in step S2, the pulse interval estimation unit 101 estimates the driver's pulse interval based on the driver's video input from the driver camera 20. Specifically, the pulse interval estimation unit 101 detects pulse waves from the video of the driver's body surface captured by the driver camera 20 and estimates the time interval between adjacent peaks in the pulse wave as the pulse interval BI (Beat Interval).

[0040] Next, in step S3, the pulse interval coarse-graining processing unit 108 determines whether the first and second parameters used for pulse interval coarse-graining processing have been learned, or more specifically, whether the first and second parameters of the driver identified in step S1 are stored in memory 10b.

[0041] As a result, if the first and second parameters have not been learned (step S3: NO), in step S4, the first parameter learning unit 106 performs the first parameter learning process, and the second parameter learning unit 107 performs the second parameter learning process.

[0042] Now, the first parameter learning process will be explained with reference to Figure 4. First, in step S21, the first parameter learning unit 106 stores the pulse interval BI estimated in step S2 into the memory 10b.

[0043] Next, in step S22, the first parameter learning unit 106 determines whether the time-series data of the pulse interval BI has been stored in memory 10b for a predetermined time W1 (for example, 45 seconds) or longer. If the result is that the time-series data of the pulse interval BI has not been stored in memory 10b for a predetermined time W1 or longer (step S22: NO), the process returns to step S21. Thereafter, steps S21 and S22 are repeated until the time-series data of the pulse interval BI has been stored in memory 10b for a predetermined time W1 or longer.

[0044] On the other hand, if the time-series data of the pulse interval BI is accumulated in memory 10b for a predetermined time W1 or longer (step S22: YES), in step S23, the first parameter learning unit 106 calculates the average value of the pulse interval BI accumulated in memory 10b and stores it in memory 10b as the first parameter of the driver identified in step S1, i.e., the average pulse interval L1. After that, the process returns to the main routine of the driver state estimation process.

[0045] Next, the second parameter learning process will be explained with reference to Figure 5. First, in step S31, the second parameter learning unit 107 causes the pulse interval coarse-graining processing unit 108 to execute the coarse-graining process for the pulse interval BI described later, from steps S41 to S49, and stores the coarse-grained pulse interval CG-BI in memory 10b.

[0046] Next, in step S32, the second parameter learning unit 107 determines whether the time-series data of the coarse-grained pulse interval CG-BI has been stored in memory 10b for a predetermined time W2 (e.g., 45 seconds) or longer. If the result is that the time-series data of the coarse-grained pulse interval CG-BI has not been stored in memory 10b for a predetermined time W2 or longer (step S32: NO), the process returns to step S31. Thereafter, steps S31 and S32 are repeated until the time-series data of the coarse-grained pulse interval CG-BI has been stored in memory 10b for a predetermined time W2 or longer.

[0047] On the other hand, if the time-series data of the coarse-grained pulse interval CG-BI is accumulated in memory 10b for a predetermined time W2 or longer (step S32: YES), in step S33, the second parameter learning unit 107 calculates the standard deviation of the coarse-grained pulse interval CG-BI accumulated in memory 10b and stores it in memory 10b as the second parameter L2 of the driver identified in step S1. After that, the process returns to the main routine of the driver state estimation process.

[0048] Returning to Figure 3, if the first and second parameters have been learned in step S3 (step S3: YES), or if the first parameter learning process and the second parameter learning process have been performed in step S4, then in step S5, the pulse interval coarse-graining processing unit 108 performs coarse-graining processing on the pulse interval.

[0049] Here, the coarse-graining process for the pulse interval will be explained with reference to Figure 6. First, in step S41, the pulse interval coarse-graining processing unit 108 stores the pulse interval BI estimated in step S2 and the acquisition time of the pulse interval BI (for example, the time when the later peak was detected among the adjacent peaks of the pulse wave used to estimate the pulse interval BI) in the memory 10b, relating them to each other.

[0050] Next, in step S42, the pulse interval coarse-graining processing unit 108 determines whether the time-series data of the pulse interval BI has been stored in the memory 10b for a predetermined time of 2 × S or more. Here, S is the coarse-graining time scale used by the pulse interval coarse-graining processing unit 108 when coarsing the time-series data of the pulse interval BI, and is predetermined and stored in the memory 10b. In this embodiment, for example, S = 25 seconds, in which case 2 × S = 50 seconds.

[0051] If the result of the determination in step S42 is that the time-series data of the pulse interval BI has not been accumulated in memory 10b for a predetermined time of 2 × S or more (step S42: NO), the process returns to step S41. Thereafter, steps S41 and S42 are repeated until the time-series data of the pulse interval BI has been accumulated in memory 10b for a predetermined time of 2 × S or more.

[0052] On the other hand, if the time-series data of the pulse interval BI is accumulated in memory 10b for a predetermined time of 2 × S or more (step S42: YES), in step S43, the pulse interval coarse-graining processing unit 108 uses the most recent pulse interval BI for a predetermined time of 2 × S (for example, 50 seconds) accumulated in memory 10b and its acquisition time to perform spline interpolation of the time-series data of the pulse interval BI. Here, if U is the acquisition time of the latest pulse interval BI accumulated in memory 10b (i.e., the latest detection time of the pulse), then spline interpolation will be performed on the time-series data of the pulse interval BI from time U - 2 × S to time U.

[0053] Next, in step S44, the pulse interval coarse-graining processing unit 108 resamples the time series data of the pulse interval BI that was spline-interpolated in step S43 at equal time intervals, thereby obtaining the time series data of the pulse interval BIr after resampling (from time U-2×S to time U).

[0054] Next, in step S45, the pulse interval coarse-graining processing unit 108 calculates time-series data of pulse interval BIr0 (from time U-2×S to time U) by shifting the average value of the resampled pulse interval BIr to 0, by subtracting the average pulse interval L1 (the first parameter of the driver) learned in the first parameter learning process and stored in memory 10b from the resampled pulse interval BIr.

[0055] Next, in step S46, the pulse interval coarse-graining processing unit 108 calculates the integrated time series data IS, which is the value obtained by integrating the time series data of the pulse interval Bir0, whose average value has been shifted to 0, from time U-2×S to time i (U-2×S≦i≦U).

[0056] Next, in step S47, the pulse interval coarse-graining processing unit 108 calculates the trend component ISt of the integral time series data IS by fitting the integral time series data IS (from time U-2×S to time U) calculated in step S46 to a cubic curve using the least squares method.

[0057] Next, in step S48, the pulse interval coarse-graining processing unit 108 calculates integrated time series data IS0 (from time U-2×S to time U) from which the trend component has been removed by subtracting the trend component ISt calculated in step S47 from the integrated time series data IS.

[0058] Next, in step S49, the pulse interval coarse-graining processing unit 108 calculates the difference CG-BI between the last value of the most recent predetermined time S (i.e., the value IS0(U) at time U) and the first value (i.e., the value IS0(US) at time US) in the integral time series data IS0 calculated in step S48, and stores it in memory 10b.

[0059] The CG-BI calculated in step S49 is the pulse interval CG-BI obtained by coarsening the time series data of pulse interval BI from time US to time U using time scale S. By appropriately setting the time scale S (for example, S=25 seconds in this embodiment), it is possible to remove trend components such as local average heart rate changes due to external factors such as exercise or stress, and to detect characteristics of pulse interval fluctuations that reflect abnormalities in the driver's autonomic nervous system.

[0060] Next, in step S50, the pulse interval coarse-graining processing unit 108 calculates a standardized pulse interval CG-BIz by dividing the coarse-grained pulse interval CG-BI calculated in step S49 by the standard deviation L2 (the second parameter of the driver) of the coarse-grained pulse interval CG-BI learned in the second parameter learning process and stored in memory 10b, and stores it in memory 10b. After that, it returns to the main routine of the driver state estimation process.

[0061] Returning to Figure 3, in step S5, the pulse interval coarse-graining processing unit 108 performs coarse-graining processing on the pulse interval, and then in step S6, the driver abnormality prediction unit 110 performs abnormality prediction processing.

[0062] Here, the abnormality prediction determination process will be explained with reference to Figure 7. First, in step S51, the driver abnormality prediction determination unit 110 determines whether or not the time-series data of the standardized coarse-grained pulse interval CG-BIz has been accumulated in the memory 10b for a predetermined time W or longer. Here, W is the time window of the time-series data of the coarse-grained pulse interval CG-BIz that the driver abnormality prediction determination unit 110 refers to when determining driver abnormalities, and is predetermined and stored in the memory 10b. In this embodiment, for example, W = 50 seconds.

[0063] As a result, if the time-series data of the standardized coarse-grained pulse interval CG-BIz has not been accumulated in the memory 10b for a predetermined time W or longer (step S51: NO), in step S52, the pulse interval coarse-graining processing unit 108 is made to perform coarse-graining processing on the pulse interval. This coarse-graining processing on the pulse interval in step S52 is the same as the coarse-graining processing on the pulse interval in step S5 of the main routine of the driver state estimation processing. Thereafter, steps S51 and S52 are repeated until the time-series data of the standardized coarse-grained pulse interval CG-BIz has been accumulated in the memory 10b for a predetermined time W or longer.

[0064] On the other hand, if time-series data of the standardized coarse-grained pulse interval CG-BIz is accumulated in memory 10b for a predetermined time W or longer (step S51: YES), in step S53, the driver abnormality prediction unit 110 retrieves the most recent time-series data of the standardized coarse-grained pulse interval CG-BIz for the predetermined time W from memory 10b.

[0065] Next, in step S54, the driver abnormality prediction unit 110 calculates the probability PA that the CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz extracted in step S53 falls within a predetermined range (area A, or central area) centered on the mean value of 0.

[0066] Here, we will explain how to define region A, referring to Figure 8. Figure 8 is a graph showing the probability distribution of the standardized coarse-grained pulse interval CG-BIz in a group of subjects with a normal autonomic nervous system and a group of subjects with an abnormal autonomic nervous system. In Figure 8, the horizontal axis shows the value of CG-BIz, and the vertical axis shows the probability density of each value of CG-BIz.

[0067] In Figure 8, the dashed line represents the representative value of the probability distribution of CG-BIz obtained by performing the same coarse-graining process on pulse intervals as explained with reference to Figure 6 on the heart rate data of each subject in the group of subjects with a normal autonomic nervous system (normal group). The upward-sloping band extending along this dashed line represents the 1σ interval of the normal group. Also in Figure 8, the solid line represents the representative value of the probability distribution of CG-BIz obtained by performing the same coarse-graining process on pulse intervals as explained with reference to Figure 6 on the heart rate data of each subject in the group of subjects with autonomic nervous system abnormalities (abnormal group). The band of dots extending along this solid line represents the 1σ interval of the abnormal group.

[0068] Comparing the probability distributions of CG-BIz occurrence in the normal and abnormal groups, as shown in Figure 8, the kurtosis of the CG-BIz probability distribution in the abnormal group is greater than that of the normal group. That is, the peak of the CG-BIz probability density in the abnormal group is higher than that of the normal group in the central region containing the mean value of CG-BIz (0), and lower than that of the normal group in the peripheral regions on both sides of that central region and on the outer edges. This is thought to be because, in the normal group, the sympathetic and parasympathetic nervous systems work to balance each other, causing fluctuations in the pulse interval, whereas in the abnormal group, abnormalities occur in either the sympathetic or parasympathetic nervous system, causing an imbalance, or abnormalities occur in both the sympathetic and parasympathetic nervous systems, eliminating changes in biological function, resulting in smaller fluctuations in the pulse interval. Therefore, the inventors of this application hypothesized that abnormal signs of driver dysfunction due to autonomic nervous system dysfunction could be detected based on the difference in the probability distribution of CG-BIz occurrence between the normal group and the abnormal group in the region near the center, including the mean value of CG-BIz of 0.

[0069] Specifically, within the range near the center including the mean value of CG-BIz (0), the A region (central region) is defined as the range where the 1σ interval of the probability distribution of CG-BIz for the normal group (upward-sloping band) and the 1σ interval of the probability distribution of CG-BIz for the abnormal group (band of dots) do not overlap (for example, the range where CG-BIz is between -0.03 and 0.03). The driver abnormality prediction unit 110 determines that a driver abnormality has been detected if the probability PA that CG-BIz, calculated from the driver's pulse interval, falls within the A region is higher than the abnormality prediction threshold TA. The abnormality prediction threshold TA is set to a value higher than the probability that the representative value +1σ of the probability distribution of CG-BIz for the normal group falls within the A region, and lower than the probability that the representative value -1σ of the probability distribution of CG-BIz for the abnormal group falls within the A region.

[0070] In other words, the abnormality prediction threshold TA is set to be greater than the value obtained by integrating the curve representing the representative value +1σ of the probability distribution of CG-BIz occurrence in the normal group (the curve that defines the upper side of the upward-sloping band) over region A (i.e., the range where CG-BIz is between -0.03 and 0.03), and less than the value obtained by integrating the curve representing the representative value -1σ of the probability distribution of CG-BIz occurrence in the abnormal group (the curve that defines the lower side of the dotted band) over region A. For example, TA = 0.09. The numerical range of CG-BIz defining region A and the abnormality prediction threshold TA are pre-set based on the heart rate data of the normal and abnormal groups and stored in memory 10b.

[0071] Returning to Figure 7, in step S55, the driver abnormality prediction unit 110 determines whether the probability PA calculated in step S54 is greater than the abnormality prediction threshold TA stored in memory 10b (i.e., the probability TA that the representative value -1σ of the probability distribution of the occurrence probability distribution of CG-BIz of the abnormal group falls within region A).

[0072] As a result, if PA is greater than TA (step S55: YES), the driver abnormality prediction unit 110 determines in step S56 that it has detected a driver abnormality. On the other hand, if PA is less than or equal to TA (step S55: NO), the driver abnormality prediction unit 110 determines in step S57 that no driver abnormality is detected. After step S56 or S57, the process returns to the main routine of the driver state estimation process.

[0073] Returning to Figure 3, after executing the abnormality prediction process in step S6, in step S7, the driver abnormal posture determination unit 12 determines whether or not an abnormality in the driver was detected in the abnormality prediction process. If an abnormality in the driver is detected as a result (step S7: YES), in step S8, the driver abnormal posture determination unit 12 changes the abnormal posture threshold stored in memory 10b in a direction that makes it more likely that the driver's posture will be determined to be abnormal. For example, if an abnormality is determined when the angle of the driver's head is greater than or equal to the first abnormal posture threshold and the duration of that state is greater than or equal to the second abnormal posture threshold, the driver abnormal posture determination unit 12 lowers one or both of the first and second abnormal posture thresholds.

[0074] Next, in step S9, the driver abnormal posture determination unit 12 outputs an alarm via the display 31 or alarm device 32 to notify that an abnormality in the driver has been detected.

[0075] After step S9, or if no signs of driver abnormality are detected in step S7 (step S7: NO), in step S10, the driver abnormal posture determination unit 12 determines whether the driver's posture is abnormal by comparing the driver's head posture estimated by the head posture estimation unit 11 with the first abnormal posture threshold.

[0076] Next, in step S11, the driver abnormal posture determination unit 12 determines whether the duration of the abnormal posture is greater than or equal to a second abnormal posture threshold stored in memory 10b, if the driver posture was determined to be abnormal in step S10. If the duration of the abnormal posture is less than the second abnormal posture threshold (step S11: NO), the process returns to step S2. Thereafter, as long as the duration of the abnormal posture is less than the second abnormal posture threshold, the process from steps S2 to S11 is repeated. That is, each time the driver's pulse interval BI is estimated, coarse-graining processing and abnormality prediction processing are performed on the pulse interval.

[0077] On the other hand, if the duration of the abnormal posture is greater than or equal to the second abnormal posture threshold (step S11: YES), in step S12, the driver abnormal posture determination unit 12 outputs an alarm via the display 31 or alarm device 32 to notify the driver that an abnormality has occurred. Furthermore, in step S13, the driver abnormal posture determination unit 12 controls various parts of the vehicle 1 with the vehicle control device 30 and starts an emergency evacuation to move the vehicle 1 to the shoulder of the road or the like. After that, the driver state estimation process ends.

[0078] [Second Embodiment] Next, the abnormality prediction process according to the second embodiment will be described with reference to Figure 9. Figure 9 is a flowchart of the abnormality prediction process according to the second embodiment. Steps S61 to S64 in the abnormality prediction process shown in Figure 9 are the same as steps S51 to S54 in the abnormality prediction process of the embodiment described with reference to Figure 7, so a detailed explanation will be omitted.

[0079] As explained with reference to Figure 8, abnormal signs in drivers due to autonomic nervous system dysfunction can be detected based on the difference in the probability distribution of CG-BIz occurrence between the normal group and the abnormal group in the central region including the mean value of CG-BIz (0). In the second embodiment, the accuracy of detecting abnormal signs is further improved by focusing on the fact that the probability density of CG-BIz occurrence in the abnormal group is lower than that of the normal group in the peripheral region adjacent to the central region including the mean value of CG-BIz (0), and adding this to the conditions for detecting abnormal signs in drivers.

[0080] Specifically, the B region (peripheral region) is defined as the range where the absolute value of CG-BIz is greater than the range near the center including the mean value of CG-BIz (0), and where the 1σ interval of the probability distribution of CG-BIz for the normal group (upward-sloping band) and the 1σ interval of the probability distribution of CG-BIz for the abnormal group (band of dots) do not overlap (for example, the range where CG-BIz is between -0.9 and -0.35, and between 0.35 and 0.9). The driver abnormality prediction unit 110 determines that a driver abnormality has been detected if the probability PA that CG-BIz calculated from the driver's pulse interval falls within the A region is higher than the abnormality prediction threshold TA, and the probability PB that CG-BIz calculated from the driver's pulse interval falls within the B region is lower than the abnormality prediction threshold TB. The abnormality prediction threshold TB is set to a value lower than the probability that the representative value -1σ of the probability distribution of CG-BIz occurrence in the normal group falls within region B, and higher than the probability that the representative value +1σ of the probability distribution of CG-BIz occurrence in the abnormal group falls within region B.

[0081] In other words, the abnormality prediction threshold TB is set to be smaller than the value obtained by integrating the curve representing the representative value -1σ of the probability distribution of CG-BIz occurrence in the normal group (the curve that defines the lower side of the upward-sloping band) over region B (i.e., the range where CG-BIz is between -0.9 and -0.35, and between 0.35 and 0.9), and larger than the value obtained by integrating the curve representing the representative value +1σ of the probability distribution of CG-BIz occurrence in the abnormal group (the curve that defines the upper side of the dotted band) over region B, for example, TB = 0.5. The numerical range of CG-BIz defining region B and the abnormality prediction threshold TB are pre-set based on the heart rate data of the normal and abnormal groups and stored in memory 10b.

[0082] In step S64 of Figure 9, the probability PA that the CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls within area A is calculated. Then, in step S65, the driver abnormality prediction unit 110 calculates the probability PB that the CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls within area B.

[0083] Next, in step S66, the driver abnormality prediction unit 110 determines whether the probability PA calculated in step S64 is greater than the abnormality prediction threshold TA stored in memory 10b (i.e., the probability TA that the representative value -1σ of the probability distribution of occurrence of CG-BIz in the abnormal group falls within area A), and whether the probability PB calculated in step S65 is less than the abnormality prediction threshold TB stored in memory 10b (i.e., the probability TB that the representative value +1σ of the probability distribution of occurrence of CG-BIz in the abnormal group falls within area B).

[0084] As a result, if PA is greater than TA and PB is less than TB (step S66: YES), the driver abnormality prediction unit 110 determines in step S67 that it has detected a driver abnormality. On the other hand, if PA is less than or equal to TA or PB is greater than or equal to TB (step S66: NO), the driver abnormality prediction unit 110 determines in step S68 that no driver abnormality is detected. After step S67 or S68, the process returns to the main routine of the driver state estimation process.

[0085] Furthermore, in addition to the conditions in step S66 that PA is greater than TA and PB is less than TB, the condition that the determination that probability PB is less than the abnormality prediction threshold TB continues for a period of time PB2 is longer than the abnormality prediction threshold TB2 (for example, 25 seconds) which is set in advance and stored in memory 10b, may be added to detect driver abnormalities.

[0086] [Examples] Next, an example of driver abnormality detection by the abnormality prediction device 100 of this embodiment will be described with reference to Figures 10 to 13. Figures 10 to 13 are time charts showing the time changes of the pulse interval BI, the standardized coarse-grained pulse interval CG-BIz, and the probability PA and PB of the CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falling into area A and area B, respectively. In particular, Figure 10 shows the state before an epileptic seizure occurs, Figure 11 shows the state of sympathetic nervous system dysfunction, Figure 12 shows the state of parasympathetic nervous system dysfunction, and Figure 13 shows the state without autonomic nervous system abnormalities.

[0087] In the epilepsy example shown in Figure 10, an epileptic seizure occurs at approximately 600 seconds, when the pulse interval BI decreases sharply, but parasympathetic nervous system dysfunction occurs before this seizure. As a result of the parasympathetic nervous system dysfunction, the fluctuation of the pulse interval BI decreases, and from approximately 20 seconds, the probability PA that CG-BIz in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region A begins to rise, and from approximately 30 seconds to approximately 80 seconds, the probability PA exceeds the abnormality prediction threshold TA (0.09 in Figure 10). In addition, the probability PB that CG-BIz in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region B is smaller than the abnormality prediction threshold TB (0.8 in Figure 10) at approximately 0 to 120 seconds, approximately 160 to 320 seconds, and approximately 350 to 670 seconds. Therefore, in the example shown in Figure 10, the driver abnormality prediction unit 110 according to the above-described embodiment or the second embodiment detects a driver abnormality at a point in time of approximately 30 to 80 seconds when the probability PA exceeds the abnormality prediction threshold TA.

[0088] Furthermore, in the example of sympathetic nervous system dysfunction shown in Figure 11, the influence of the sympathetic nervous system on the pulse interval BI disappears, and parasympathetic inhibition becomes dominant, resulting in reduced fluctuations in the pulse interval BI. As a result, from approximately 18 seconds to approximately 68 seconds, the probability PA that CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region A exceeds the abnormality prediction threshold TA (0.09 in Figure 11). Also, the probability PB that CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region B is smaller than the abnormality prediction threshold TB (0.8 in Figure 11) from approximately 20 seconds to approximately 68 seconds. Therefore, in the example in Figure 11, the driver abnormality prediction unit 110 according to the above-described embodiment detects a driver abnormality from approximately 18 seconds to approximately 68 seconds, when the probability PA exceeds the abnormality prediction threshold TA. Furthermore, the driver abnormality prediction unit 110 according to the second embodiment detects a driver abnormality at a point between approximately 20 seconds and approximately 68 seconds, when the probability PA exceeds the abnormality prediction threshold TA and the probability PB is less than the abnormality prediction threshold TB.

[0089] Furthermore, in the example of parasympathetic nervous system dysfunction (specifically heart failure) shown in Figure 12, the influence of the parasympathetic nervous system on the pulse interval BI disappears, and sympathetic nervous system facilitation becomes dominant. As a result, the pulse interval BI is divided into relatively large and small fluctuations, and intermediate fluctuations become smaller. Consequently, from 0 seconds to 100 seconds, the probability PA that CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region A exceeds the abnormality prediction threshold TA (0.09 in Figure 12). Also, the probability PB that CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region B is smaller than the abnormality prediction threshold TB (0.8 in Figure 12) from approximately 35 seconds to approximately 95 seconds. Therefore, in the example in Figure 12, the driver abnormality prediction unit 110 according to the above embodiment detects a driver abnormality from 0 seconds to 100 seconds, when the probability PA exceeds the abnormality prediction threshold TA. Furthermore, the driver abnormality prediction unit 110 according to the second embodiment detects a driver abnormality at a point between approximately 35 seconds and approximately 95 seconds, when the probability PA exceeds the abnormality prediction threshold TA and the probability PB is less than the abnormality prediction threshold TB.

[0090] On the other hand, in the example shown in Figure 13, where there is no abnormality in the autonomic nervous system, the pulse interval BI fluctuates due to the sympathetic and parasympathetic nervous systems working to balance each other. As a result, the probability PA that CG-BIz included in the time-series data of the standardized coarse-grained pulse interval CG-BIz falls into region A does not exceed the abnormality prediction threshold TA (0.09 in Figure 13), and the probability PB that CG-BIz falls into region B is greater than the abnormality prediction threshold TB (0.8 in Figure 13). Therefore, in the example in Figure 13, the driver abnormality prediction unit 110 according to the above-described embodiment or the second embodiment does not detect any driver abnormality.

[0091] [Differentiation] In the embodiment described above, the pulse interval estimation unit 101 was explained to detect pulse waves from images of the driver's body surface captured by the driver camera 20 and estimate the time interval between adjacent peaks in the pulse wave as the pulse interval BI. However, the pulse wave (pulse) may be detected using a device other than the driver camera 20 (for example, a radio wave sensor, a capacitive sensor, a smartwatch with a heart rate detection function, an electrocardiograph, etc.) and the pulse interval BI may be estimated.

[0092] [Effects / Effects] Next, the operation and effects of the driver abnormality prediction method and apparatus of this embodiment described above will be explained.

[0093] The abnormality prediction detection device 100 estimates the driver's pulse interval BI based on the pulse detected by the driver camera 20, acquires time-series data of the driver's pulse interval BI for a predetermined time 2 × S, standardizes the time-series data of the pulse interval BI, calculates the probability PA that the standardized coarse-grained pulse interval CG-BIz falls within a predetermined range (A region) centered on 0 in the time-series data of the standardized coarse-grained pulse interval CG-BIz, and determines that an abnormality in the driver has been detected if the probability PA is greater than a predetermined abnormality prediction threshold TA. Therefore, even from time-series data of pulse interval BI over a short period of time, such as several tens of seconds, it is possible to acquire time-series data of the standardized coarse-grained pulse interval CG-BIz, calculate the probability PA that CG-BIz falls within the A region, and detect an abnormality. This makes it possible to detect driver abnormalities with both high responsiveness and high reliability.

[0094] Furthermore, the predetermined range centered at 0 (region A) is included in the central region where the peak of the probability distribution of the standardized coarse-grained pulse interval CG-BIz obtained from the group of subjects with autonomic nervous system abnormalities is higher than the peak of the probability distribution of the standardized coarse-grained pulse interval CG-BIz obtained from the group of subjects with normal autonomic nervous systems. Therefore, abnormality prediction can be performed in a region where there is a clear difference between cases with and without autonomic nervous system abnormalities. This improves the reliability of the driver's abnormality prediction detection.

[0095] Furthermore, the abnormality prediction threshold TA is greater than the value obtained by integrating the probability distribution of standardized coarse-grained pulse intervals (CG-BIz) obtained from a group of subjects with normal autonomic nervous systems over a predetermined range (region A) centered on 0, and less than the value obtained by integrating the probability distribution of standardized coarse-grained pulse intervals (CG-BIz) obtained from a group of subjects with autonomic nervous system abnormalities over a predetermined range (region A) centered on 0. Therefore, the boundary between cases with and without autonomic nervous system abnormalities in the predetermined range (region A) centered on 0 can be used as the abnormality prediction threshold TA to determine abnormalities. This improves the reliability of the driver's abnormality prediction detection.

[0096] Furthermore, since the coarse-graining processing and abnormality prediction processing for the pulse interval are executed each time the driver's pulse interval BI is estimated, signs of abnormality in the driver's autonomic nervous system can be detected quickly. This enables the detection of driver abnormalities with both high responsiveness and high reliability.

[0097] Furthermore, since the driver's pulse is detected by the driver camera 20 which captures images of the driver's body surface, driver abnormality prediction can be performed at low cost and without burdening the driver, without the need for special equipment for pulse detection. [Explanation of Symbols]

[0098] 1 vehicle 10. Driver state estimation device 10a processor 10b memory 11 Head pose estimation section 12 Driver abnormal posture detection unit 13 Abnormal Posture Threshold 20 Driver Cameras 30 Vehicle control device 31 displays 32 Alarm device 40 Vehicle actuators 41 Lighting devices 42 Horns 100 Anomaly Prediction Detection Device 101 Pulse interval estimation unit 102 Driver Personal Authentication Section 103 Personal Parameter Selection Section 104 Driver-specific first parameter 105 Driver-specific second parameter 106 First Parameter Learning Unit 107 Second Parameter Learning Unit 108 Pulse interval coarse-graining processing unit 109 buffers 110 Driver abnormality prediction unit 111 Anomaly prediction threshold

Claims

1. A driver abnormality prediction method for detecting signs of abnormality in a driver operating a vehicle, Step A of the pulse detection device detecting the pulse of the driver, Step B involves the processor estimating the pulse interval of the driver based on the pulse detected by the pulse detection device, The processor performs a step C in which it calculates a coarsely ground pulse interval by shifting the average value of the time-series data of the pulse interval of the driver to 0 and coarsening it on the most recent predetermined time scale, The processor performs a step D to calculate a standardized coarse-grained pulse interval by dividing the coarse-grained pulse interval by the standard deviation of the coarse-grained pulse interval, The processor performs a step E of calculating the probability that the standardized coarse-grained pulse interval falls within a predetermined range centered on 0 in the time-series data of the standardized coarse-grained pulse interval, Step F of the processor determining that it has detected an indication of an abnormality in the driver when the probability is greater than a predetermined abnormality prediction threshold, It has, A driver abnormality prediction method, wherein the predetermined range centered on 0 is the range of the standardized coarse-grained pulsation intervals in the region where the peak of the probability distribution of the standardized coarse-grained pulsation intervals of the abnormal group, which is standardized by dividing the coarse-grained pulsation intervals obtained from the group of subjects with autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulsation intervals, protrudes higher than the peak of the probability distribution of the standardized coarse-grained pulsation intervals of the normal group, which is standardized by dividing the coarse-grained pulsation intervals obtained from the group of subjects with normal autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulsation intervals.

2. The predetermined range centered on 0 is such that the standardized coarse-grained pulse interval is in the range of -0.03 or more and 0.03 or less. The driver abnormality prediction method according to claim 1.

3. The abnormality prediction threshold is greater than the value obtained by integrating the probability distribution of the standardized coarse-grained pulse intervals of the normal group over a predetermined range centered on 0, and less than the value obtained by integrating the probability distribution of the standardized coarse-grained pulse intervals of the abnormal group over a predetermined range centered on 0. The driver abnormality prediction method according to claim 1.

4. Steps C, D, E, and F are performed each time the pulse interval is estimated in step B. A driver abnormality prediction method according to any one of claims 1 to 3.

5. A driver abnormality prediction device that detects signs of abnormality in a driver operating a vehicle, A pulse detection device for detecting the pulse of the aforementioned driver, Memory to store the program, A processor that executes the aforementioned program, The aforementioned processor, Based on the pulse detected by the pulse detection device, the pulse interval of the driver is estimated. The time-series data of the pulse interval of the driver is acquired over a predetermined period of time. Based on the acquired time-series data of the pulse interval, the average value of the time-series data of the pulse interval is shifted to 0, and the coarse-grained pulse interval is calculated by coarsening it on the most recent predetermined time scale. The standardized coarse-grained pulse interval is calculated by dividing the aforementioned coarse-grained pulse interval by the standard deviation of the said coarse-grained pulse interval. In the time-series data of the standardized coarse-grained pulse interval, the probability that the standardized coarse-grained pulse interval falls within a predetermined range centered on 0 is calculated. If the aforementioned probability is greater than a predetermined abnormality prediction threshold, it is determined that an abnormality in the driver has been detected. It is configured in such a way, The predetermined range centered at 0 is the range of the standardized coarse-grained pulse intervals in which the peak of the probability distribution of standardized coarse-grained pulse intervals in the abnormal group, obtained by dividing the coarse-grained pulse intervals obtained from the group of subjects with autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulse intervals, protrudes higher than the peak of the probability distribution of standardized coarse-grained pulse intervals in the normal group, obtained by dividing the coarse-grained pulse intervals obtained from the group of subjects with normal autonomic nervous system abnormalities by the standard deviation of said coarse-grained pulse intervals. Driver malfunction prediction device.

6. The pulsation detection device is a driver camera that captures images of the driver's body surface. The driver abnormality prediction device according to claim 5.

Images