WO2016104496A1 - Drowsiness estimation device and drowsiness estimation program - Google Patents

Abstract

[Problem] To estimate drowsiness from various directions, and to cope with individual differences and estimate drowsiness more accurately. [Solution] The present invention is provided with: an RRI acquisition means (22) for acquiring RRI data, i.e. data related to the R-R interval, from a signal obtained from an RRI sensor which detects signals corresponding to R waves in an electrocardiogram signal; an index-value calculation means (23) which calculates, on the basis of a result obtained by statistically processing the RRI data, and a result of an spectral analysis of the RRI data, index values for a plurality of types of activity indices related to the activity of autonomic nerves; and a drowsiness estimation means (24) which, on the basis of drowsiness estimation rules formed by estimation functions evaluated using the change state and/or a threshold value related to each of the activity indices, evaluates the index values calculated by the index-value calculation means (23), to estimate drowsiness.

Description

Sleepiness estimation apparatus and sleepiness estimation program

The present invention relates to a sleepiness estimation apparatus and a sleepiness estimation program.

In recent years, drowsiness has attracted attention from the viewpoint of health management or from the viewpoint of preventing automobile accidents. In particular, it is important to estimate that sleepiness is occurring in the case of driving a car, manipulating / operating equipment, desk work, and the like.

As a conventional technique for capturing drowsiness, Patent Document 1 discloses a device that detects the drowsiness level of a driver using an FFT processing result of a heartbeat signal. More specifically, the wakefulness index band α is set centering on the peak frequency of the spectrum signal that is the result of the FFT processing of the heartbeat signal, and the sleepiness is centered on a predetermined ratio (65 to 90%) with respect to the peak frequency during wakefulness. The degree index band β is set. The parameter Sp (= βp / (αp + βp)) is calculated using the intensity αp, βp of the spectrum signal belonging to the arousal degree index band α and the sleepiness degree index band β in a predetermined time zone after the start of driving, and this parameter Sp Based on this, the degree of sleepiness is evaluated.

Also, in Cited Document 2, based on the driver's heartbeat and the like, when specifying and outputting a chronically dangerous position for driving such as a dark place, a place with poor visibility, or a place with many on-street parking, It is described that the dangerous position is accurately identified by identifying the influence of the fluctuation and obtaining an appropriate heartbeat fluctuation.

Also, in Cited Document 3, heart rate interval data in which RR intervals are time-sequentially obtained from a heart rate waveform is obtained, and this heart rate interval data is frequency-analyzed to obtain a power spectrum density (PSD) and an autonomic nerve total power (TP). Based on the frequency analysis result of heart rate fluctuations including the initial state, the sleepiness scale with the origin of the sleepiness position and the origin of the awakening position set based on the estimated sleepiness position and the estimated awakening position included in the initial state It is described that the sleepiness determination is performed using this sleepiness scale.

Further, in Cited Document 4, a heartbeat interval is detected from a heartbeat signal, a spectral density with respect to fluctuation of the heartbeat interval is obtained, a maximum spectral density corresponding to the maximum frequency and the maximum frequency is calculated, and the calculated maximum frequency and maximum frequency are calculated. It is described that it is determined whether or not the user is in an arousal state based on the fluctuation tendency of the maximum spectral density corresponding to.

JP 2004-350773 A JP 2013-205965 A JP 2014-12042 A International Publication 2008/065724 Pamphlet

However, the conventional processes related to sleepiness described above are based on the RR interval (hereinafter referred to as RRI), but often use the spectral density as an index as it is. In Cited Document 3, although spectral density is used as an index together with the result of statistical processing of RRI data, spectral density is used as an index as it is.

Depending on the conventional techniques as described above, there is no index for sufficiently estimating sleepiness, and it is highly likely that the detection of incomplete sleepiness will end. In particular, there are cases in which changes are unlikely to occur in some indicators due to individual differences, and there are cases in which changes are seen in other indicators, and conventional methods have not been able to accurately estimate sleepiness. .

The present invention has been made as a solution to the conventional problems related to sleepiness estimation as described above, and its purpose is to provide a sleepiness estimation apparatus capable of estimating sleepiness with higher accuracy corresponding to individual differences and the like, and It is to provide a drowsiness estimation program.

An apparatus for estimating sleepiness according to the present invention includes an RRI acquisition unit that acquires RRI data that is data of an RR interval from a signal obtained by an RRI sensor that detects a signal corresponding to an R wave of an electrocardiogram signal; Based on the result of statistical processing and the result of spectrum analysis of the RRI data, an index value calculation means for calculating an index value for a plurality of types of activity indexes related to autonomic nerve activity, a threshold value and / or a fluctuation state for each activity index And drowsiness estimation means for evaluating the index value calculated by the index value calculation means and estimating drowsiness based on a drowsiness estimation rule constituted by an estimation function evaluated by the above.

In the sleepiness estimation apparatus according to the present invention, the activity index as a result of statistical processing of the RRI data includes:
SDRR: (standard deviation of RRI)
RMSSD: (the square root of the root mean square of the difference between adjacent RRIs)
SDSD: (standard deviation of the difference between adjacent RRIs)
pRR50: (Percentage where the difference between adjacent RRIs exceeds 50 (milliseconds))
It is characterized by including at least one of these.

In the sleepiness estimation apparatus according to the present invention, the activity index as a result of spectrum analysis of the RRI data includes:
LF: (PSD (power spectral density function) 0.04 to 0.15 [Hz] power)
HF: (PSD 0.15-0.40 [Hz] power)
HF / (LF + HF)
p _i (i = 0, 1, 2,..., 9): (PSD 0.15 + i × 0.025 to 0.15+ (i + 1) × 0.025 [Hz] power)
It is characterized by including at least one of these.

In the drowsiness estimation apparatus according to the present invention, the index value calculation means calculates an index value for unit time using the RRI data for each first time, collects the index values for unit time for each type of activity index, and The index value vector time series is created by arranging the vectorized index values in a time series, and the sleepiness estimation means performs sleepiness estimation using an estimation function that is evaluated using a threshold vector and the index value vector time series. Features.

In the sleepiness estimation apparatus according to the present invention, the estimation function of the sleepiness estimation rule includes a comparison condition between each of the thresholds corresponding to the activity index and a condition of presence / absence of a predetermined fluctuation state, and is either AND or OR. Or two or more of these AND and OR are used to combine the above conditions.

The drowsiness estimation apparatus according to the present invention is characterized by including data shaping means that performs at least one of trend removal, abnormal value removal, data interpolation, and filter processing, which are removal of a predetermined frequency component.

The sleepiness estimation program according to the present invention is an RRI acquisition unit that acquires RRI data that is data of an RR interval from a signal obtained by an RRI sensor that detects a signal corresponding to an R wave of an electrocardiogram signal. Index value calculation means for calculating an index value for a plurality of types of activity indexes related to autonomic nerve activity based on the result of statistical processing of data and the result of spectrum analysis of the RRI data, threshold value and / or fluctuation for each activity index Based on a drowsiness estimation rule configured by an estimation function that is evaluated according to a state, the index value calculated by the index value calculation unit is evaluated to function as a drowsiness estimation unit that estimates drowsiness.

In the sleepiness estimation program according to the present invention, the activity index as a result of statistical processing of the RRI data includes:
SDRR: (standard deviation of RRI)
RMSSD: (the square root of the root mean square of the difference between adjacent RRIs)
SDSD: (standard deviation of the difference between adjacent RRIs)
pRR50: (Percentage where the difference between adjacent RRIs exceeds 50 (milliseconds))
It is characterized by including at least one of these.

In the sleepiness estimation program according to the present invention, the activity index as a result of spectrum analysis of the RRI data includes:
LF: (PSD (power spectral density function) 0.04 to 0.15 [Hz] power)
HF: (PSD 0.15-0.40 [Hz] power)
HF / (LF + HF)
p _i (i = 0, 1, 2,..., 9): (PSD 0.15 + i × 0.025 to 0.15+ (i + 1) × 0.025 [Hz] power)
It is characterized by including at least one of these.

In the sleepiness estimation program according to the present invention, the index value calculation means calculates an index value for unit time using the RRI data for each first time, collects the index values for unit time for each type of activity index, and The index value vector time series is created by arranging the vectorized index values in a time series, and the sleepiness estimation means performs sleepiness estimation using an estimation function that is evaluated using a threshold vector and the index value vector time series. Features.

In the sleepiness estimation program according to the present invention, the estimation function of the sleepiness estimation rule includes a comparison condition between each of the thresholds corresponding to the activity index and a condition of presence / absence of a predetermined fluctuation state, and is either AND or OR. Or two or more of these AND and OR are used to combine the above conditions.

The sleepiness estimation program according to the present invention causes the computer to function as data shaping means that performs at least one of trend removal, abnormal value removal, data interpolation, and filter processing, which are removal of a predetermined frequency component.

According to the present invention, based on the result of statistical processing of RRI data and the result of spectrum analysis of RRI data, index values are calculated for a plurality of types of activity indexes related to autonomic nerve activity, and threshold values for each activity index and / or The activity index is evaluated and sleepiness is estimated based on the sleepiness estimation rule configured by the estimation function that is evaluated based on the fluctuation state. Therefore, the activity index of the autonomic nerve based on the result of statistical processing of the RRI data and the spectrum analysis of the RRI data are performed. By using a plurality of autonomic nerve activity indexes calculated based on the results, sleepiness is estimated from various directions, and more accurate sleepiness can be estimated corresponding to individual differences and the like.

The block diagram of 1st Embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which shows the electrocardiogram waveform for demonstrating RRI in an electrocardiogram signal. The block diagram which shows the internal structure of the system which comprised 1st Embodiment of the sleepiness estimation apparatus which concerns on this invention centering on the terminal. The flowchart which shows the process by the sleepiness estimation processing program of 1st Embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the 1st structural example of the data shaping means which is the principal part in embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the 2nd structural example of the data shaping means which is the principal part in embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the 3rd structural example of the data shaping means which is the principal part in embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the 4th structural example of the data shaping means which is the principal part in embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the structural example of the index value calculation means which is the principal part in embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which shows an example of the RRI data taken in into embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which shows an example of the data comprised by the RRI data taken in by embodiment of the sleepiness estimation apparatus which concerns on this invention, and its taking-in time. The figure which shows an example of RRI data in case loop count of the loop process is 0 in embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which shows an example of RRI data in case loop count of the loop process is 3 in embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which shows an example of an index name, an index variable, and its threshold value about several activity index used in embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which is RRI data processed in embodiment of the drowsiness estimation apparatus which concerns on this invention, Comprising: RRI data (before shaping) 4000 seconds after index calculation start time T0 (cnt = 400). The figure which shows the data after shaping the waveform by the trend removal part of embodiment of the sleepiness estimation apparatus which concerns on RRI data shown in FIG. The figure which shows the data after carrying out waveform shaping by the abnormal value removal part of embodiment of the sleepiness estimation apparatus which concerns on RRI data shown in FIG. The figure which shows the data after waveform shaping by the data interpolation part of embodiment of the sleepiness estimation apparatus which concerns on RRI data shown in FIG. The figure which shows the filter characteristic of the filter process part of embodiment of the sleepiness estimation apparatus which concerns on this invention. The figure which shows the data after carrying out waveform shaping by the filter process part of embodiment of the sleepiness estimation apparatus which concerns on RRI data shown in FIG. The figure which shows PSD obtained by the FFT direct method. The figure which shows PSD calculated | required by the maximum entropy method (burg method). The figure which shows the area which calculates | requires power p _i of the specific frequency band which is the parameter | index which embodiment of the sleepiness estimation apparatus which concerns on this invention uses with respect to PSD obtained by the FFT direct method shown in FIG. Shows respect PSD obtained by the maximum entropy method shown in FIG. 22 (burg method), a section for obtaining the power p _i of the specific frequency band, which is an embodiment uses an indication of drowsiness estimation apparatus according to the present invention. The block diagram of 2nd Embodiment of the sleepiness estimation apparatus which concerns on this invention. The flowchart which shows the process by the sleepiness estimation program of 2nd Embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram of 3rd Embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the internal structure of the system which comprised 3rd Embodiment of the sleepiness estimation apparatus which concerns on this invention centering on the terminal and the server. The flowchart which shows the process by the sleepiness estimation program of 3rd Embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram of 4th Embodiment of the sleepiness estimation apparatus which concerns on this invention. The block diagram which shows the internal structure of the system which comprised 4th Embodiment of the sleepiness estimation apparatus which concerns on this invention centering on the terminal and the server.

Hereinafter, embodiments of a sleepiness estimation apparatus and a sleepiness estimation program according to the present invention will be described with reference to the accompanying drawings. In each figure, the same components are denoted by the same reference numerals and redundant description is omitted. The first embodiment of the drowsiness estimation apparatus is configured as shown in FIG. In the present embodiment, a heart rate sensor can be used as the RRI sensor 10 that detects a signal corresponding to the R wave of the electrocardiogram signal. In addition to the heart rate sensor, the RRI sensor 10 may use a configuration of a part for taking out an electrocardiogram signal of an electrocardiograph or a pulse wave sensor.

The RRI sensor 10 is provided on a living body, detects an electrocardiogram signal as shown in FIG. 2B by wireless or wired, and outputs an RRI (before shaping) 1001 as shown in FIG.

Here, the ECG signal will be described. As shown in FIG. 2A, a QRS wave having an R peak can be observed in the electrocardiogram signal. QRS waves are assumed to be generated when the entire ventricle is rapidly excited. Further, the P wave in front thereof is a wave when excitement occurs in the sinoatrial node and the atrium contracts. Further, the T wave appearing behind the QRS wave is a wave generated when the excitement of the ventricle is recovered. For the analysis of the heart, the PQ time, QRS time, QT time and the like obtained from the P wave, QRS wave, and T wave are used as important parameters. In the electrocardiogram signal, as shown in FIG. 2B, R waves appear at predetermined intervals according to the pulsation, so that RRI is also an important parameter and is known to be related to autonomic nerve activity. . In this embodiment, this RRI data is used for sleepiness estimation.

As shown in FIG. 1, in the sleepiness estimation apparatus of the present embodiment, the terminal 20 includes an RRI acquisition unit 21, a data shaping unit 22, an index value calculation unit 23, a sleepiness estimation unit 24, and an output unit 25.

The RRI acquisition means 21 acquires RRI data sent by the RRI sensor 10. The RRI sensor 10 may output an electrocardiogram signal. In this case, the RRI acquisition unit 21 creates RRI data based on the electrocardiogram signal. The data shaping means 22 has a configuration for performing at least one of trend removal, abnormal value removal, data interpolation, and filter processing, which is removal of a predetermined frequency component. These configurations will be described in detail later as a trend removal unit 222, an abnormal value removal unit 223, a data interpolation unit 224, a filter processing unit 225, and the like.

The index value calculation means 23 calculates the index value of the autonomic nerve based on the RRI data. For example, based on the result of statistical processing of the RRI data and the result of spectrum analysis of the RRI data, a plurality of types For the autonomic nerve activity index, the index value can be calculated. The index value calculation means 23 calculates an index value for unit time using the RRI data for each first time, collects the index values for unit time for each type of activity index, and vectorizes the index values. Create an index value vector time series in a series.

The sleepiness estimation means 24 estimates sleepiness by evaluating the activity index calculated by the index value calculation means 23 on the basis of a sleepiness estimation rule constituted by an estimation function that is evaluated based on a threshold value related to an activity index and / or a fluctuation state. To do.

The output means 25 outputs the estimation result by the sleepiness estimation means 24 and is used for generating an alarm, stopping the operation of the device, warning, or the like. The terminal 20 can be configured by a smartphone, a tablet terminal, a mobile terminal, or the like.

The cloud storage 30 is connected to the terminal 20. The cloud storage 30 includes sensor characteristic information 1101 and sleepiness estimation rules 1102 in advance. The sensor characteristic information 1101 is used when the data shaping unit 22 executes data shaping. The sleepiness estimation rule 1102 is used when the sleepiness estimation means 24 estimates sleepiness. The cloud storage 30 indicates that the acquired RRI data acquired in the terminal 20 may be stored as acquired RRI 1012X of history data, and this acquired RRI 1012X is not shown in the present embodiment. It may be used for updating the sleepiness estimation rule 1102.

Specifically, the sleepiness estimation of the above configuration has the configuration shown in FIG. The RRI sensor 10 described above, the terminal 20, and the cloud storage 30 are included.

The terminal 20 performs processing under the control of the CPU, and includes a memory that temporarily stores data being processed and the like, and a nonvolatile memory that stores various processing data. A storage device. The terminal 20 includes a communication unit that performs communication via a telephone line or communication via a network, an input / output unit that can input information and commands via a touch screen, and display various data and images, and time data. Has a timer to retrieve. Communication with the cloud storage 30 can be performed by the communication unit. Further, the storage device includes various application programs, application software, various browsers, a data processing program for performing data processing such as data acquisition and data shaping performed in the present embodiment, and a sleepiness estimation program for performing sleepiness estimation processing. Etc., parameters and programs for sleepiness estimation processing are provided.

The cloud storage 30 performs processing under the control of the CPU, and includes a memory that temporarily stores data being processed, a non-volatile memory that stores various types of processing data, and the like. And a communication unit that performs communication via the network (here, mainly communication with the terminal 20). The storage device also includes various parameters and programs for sleepiness estimation processing.

In the sleepiness estimation apparatus having the above configuration, the CPU performs processing corresponding to the flowchart shown in FIG. 4 in the terminal 20 using the data processing program and the sleepiness estimation program. As shown in FIGS. 1 and 4, the cloud storage 30 includes sensor characteristic information 1101 and sleepiness estimation rules 1102. The sensor characteristic information 1101 is used for the purpose of correcting the RRI data in the data shaping unit 22 shown in FIGS. 1 and 4 in order to absorb the characteristic difference due to the difference in the sensor model. For example, the peak of the QRS wave is used. Is corrected by the correction value when the characteristic is not sharp or is sharp but has a large error. When the sensor characteristic information 1101 does not exist in the cloud storage 30, it is not necessary to perform correction particularly for the purpose of absorbing differences in sensor models.

The sleepiness estimation rule 1102 is configured by an estimation function that is evaluated based on a threshold and / or a fluctuation state regarding each activity index. The estimation function of the sleepiness estimation rule 1102, which will be described in detail later, includes a comparison condition between the activity index and the corresponding threshold value, and a condition of presence / absence of a predetermined fluctuation state, and uses either AND or OR. Alternatively, two or more of these AND and OR can be used to combine the above conditions.

In the processing shown in FIG. 4, the start processing unit 201 is first executed, the loop start unit 202 is then started, and the RRI acquisition unit 21 and the subsequent steps are executed. The processing after the RRI acquisition unit 21 sandwiched between the loop start unit 202 and the loop end determination unit 203 is loop processing, and the loop processing is continued until the loop end determination unit 203 determines the end. The termination in this case can be executed by the user performing a predetermined key operation from the terminal 20, for example. In the loop processing, the loop counter cnt is incremented every time the processing returns to the RRI acquisition means 21. Here, cnt = 0, 1, 2,... (Starting value of cnt is zero).

Loop processing (between the loop start unit 202 and the loop end determination unit 203) is performed every index value calculation time interval DT [seconds] determined by the start processing unit 201. Therefore, if the measurement start time by the RRI sensor 10 is T0, and the RRI data length handled in one loop process is LT [seconds], the cnt-th loop process has time T0 + cnt × DT to time T0 + cnt × DT + LT. RRI data acquired during LT [seconds] is processed.

The start processing unit 201 sets parameters for each process performed in this process. The parameter is the time interval DT [seconds] (index value calculation time interval) of reprocessing by the subsequent loop processing (between the loop start unit 202 and the loop end determination unit 203). RRI data handled in one loop processing The data includes parameters for the long time LT [seconds], data shaping processing parameters and constants in the subsequent data shaping means 22, and heart rate analysis parameters and constants in the subsequent index value calculation means 23. Normally, a default value prepared in advance can be used as the parameter, but some candidate parameters may be prepared and changed by executing a change procedure at a predetermined timing or the like.

Although it overlaps with the contents already described in the description of the cloud storage 30, the start processing unit 201 accesses the cloud storage 30 and reads if there is sensor characteristic information 1101 corresponding to the RRI sensor model. When the sensor characteristic information 1101 does not exist in the cloud storage 30, no correction is performed particularly for the purpose of absorbing differences between sensor models.

Further, the start processing unit 201 reads the drowsiness estimation rule 1102 connected to the cloud storage 30 and updated for improving the estimation accuracy. When the updated sleepiness estimation rule 1102 does not exist in the cloud storage 30, the sleepiness estimation rule 1102 used in the previous sleepiness estimation process is applied.

Following the processing of the start processing unit 201, the loop start unit 202 is activated, and the processing of the RRI acquisition unit 21 is executed. The RRI acquisition unit 21 sequentially executes the following using the measurement start time T0, the RRI time length LT, and the index value calculation time interval DT. That is, the RRI (before shaping) 1001 is acquired from the RRI sensor 10 in real time. For example, data as shown in FIG. 10 is acquired. In the example of FIG. 10, “680” (unit: [milliseconds]) is first fetched from the RRI sensor 10, and thereafter “710”, “593”, “827”,... ing.

Next, the RRI acquisition unit 21 acquires the time 1002 when one piece of RRI data is generated from a timer built in the RRI sensor 10 or a timer provided in the terminal 20. Data associating the time 1002 with the RRI (before shaping) 1001 is added and stored in a register (not shown) as RRI2 (before shaping) 1012. This data is, for example, the data shown in FIG. 11. In FIG. 11, the first data acquired from the RRI sensor 10 is the time of 15:09:19 seconds and 600 milliseconds, and in FIG. It indicates that it is data at the indicated time.

The data from the predetermined time to the predetermined time in the RRI2 (before shaping) 1012 of the register is extracted to generate the RRI3 (before shaping) 1013 shown in FIG. After the index value calculation start time T0, in the cnt-th loop of the processing according to the flowchart of FIG. 4 (cnt = 0, 1, 2,. Handled RRI.

In the subsequent interpolation processing of the data shaping means 22, one piece of data before and after the time T0 + cnt × DT to the time T0 + cnt × DT + LT is also required, and therefore acquired in LT seconds from the time T0 + cnt × DT to the time T0 + cnt × DT + LT. RRI3 (before shaping) 1013 includes the RRI that has been performed and one RRI before and after the RRI.

For example, when the index value calculation start time T0 is 15:10:20, LT = 300 [seconds] (= 5 [minutes]), DT = 10 seconds, and cnt = 0, the time is 15:10:20. The data up to 15:15:20 includes the data immediately before 15:10:20 and the data after 15:15:20, as shown in FIG. .

When cnt = 3, the data from 15:10:50 to 15:15:50, the data before 15:10:50, and the data after 15:15:50 The data as shown in FIG.

Following the processing by the RRI acquisition means 21, the processing by the data shaping means 22 is performed. For the data shaping means 22, a required one is selected from a plurality of configurations shown in FIGS. Here, four types of data shaping means 22 (FIG. 5), data shaping means 221A (FIG. 6), data shaping means 221B (FIG. 7), and data shaping means 221C (FIG. 8) will be described.

The data shaping means 22 includes a trend removing unit 222, an abnormal value removing unit 223, a data interpolation unit 224, and a filter processing unit 225. In this order, the data shaping unit 22 performs shaping processing on the RRI 3 (before shaping) 1013, and finally RRI (after shaping) 1014 is generated. The trend removing unit 222 removes RRI ultra-low frequency components that are not handled in the present embodiment. The householder method is applied to RRI3 (before shaping) 1013 to calculate the least square estimation amount. Next, the least square curve is removed from the RRI 3 (before shaping) 1013. When handling RRI data of about 5 minutes, the order of the Householder method is about 6th (Mino Hino, “Spectrum Analysis”, Asakura Shoten (2009)).

15 shows the waveform of RRI3 (before shaping) 1013, and FIG. 16 shows the waveform after trend removal. More specifically, FIG. 15 shows a waveform of RRI3 (before shaping) 1013 4000 seconds after the index value calculation start time T0 (cnt = 400). FIG. 16 shows the waveform data after trend removal obtained by applying the processing by the trend removal unit 222 to the RRI 3 (before shaping) 1013 shown in FIG.

The abnormal value removing unit 223 that performs processing following the trend removing unit 222 removes the abnormal value of the sensor based on a mathematical basis. This is a process performed when an abnormal value is mixed, because a correct value cannot be obtained for the index (activity index) calculated by the index value calculation means 23 thereafter.

FIG. 17 shows data after removing abnormal values obtained by applying the processing by the abnormal value removing unit 223 to the data after removing the trend in FIG. In the present embodiment, the data after trend removal shown in FIG. 16 is handled as an abnormal value if it exceeds ± 300 [ms]. In addition to exclusion by an absolute numerical value such as ± 300, for example, when an RRI histogram is created, data exceeding ± 5σ is treated as an abnormal value.

The data interpolation unit 224 that performs processing next to the abnormal value removal unit 223 calculates RRI data at regular time intervals using spline interpolation, linear interpolation, or the like. This corresponds to the preprocessing for calculating the FFT in the index value calculation means 23 that performs processing thereafter. FIG. 18 shows post-interpolation data obtained by applying the processing by the data interpolation unit 224 to the data after the abnormal value removal shown in FIG. In this embodiment, linear interpolation is performed.

In the filter processing unit 225 that performs processing subsequent to the data interpolation unit 224, an FFT filter is applied as appropriate (Mikio Hino, “Spectrum Analysis”, Asakura Shoten (2009)). The processing of the filter processing unit 225 corresponds to preprocessing for FFT calculation in the index value calculation unit 23 as in the processing of the data interpolation unit 224. FIG. 20 shows an RRI (after shaping) 1014 obtained by applying the processing by the filter processing unit 225 to the data after interpolation in FIG.

Note that the characteristics of the filter applied in this embodiment are as shown in FIG. 19, and the purpose is to focus attention (emphasis) on the frequency components included in the latest data in the frequency analysis after waveform shaping. The old data signal (power) is reduced. The filter used in the filter processing unit 225 is not limited to the characteristics shown in this embodiment, and various filters can be used according to the purpose of analysis.

The data shaping unit 221A shown in FIG. 6 has a configuration in which a first stage abnormal value removing unit 226A is provided in the previous stage of the data shaping unit 22 shown in FIG. The data shaping unit 221A includes a first abnormal value removing unit 226A, a trend removing unit 222, a second abnormal value removing unit 223A, a data interpolation unit 224, and a filter processing unit 225. The first abnormal value removing unit 226A performs the shaping process sequentially on the RRI 3 (before shaping) 1013 by the filter processing unit 225, and finally generates the RRI (after shaping) 1014.

The difference between the first abnormal value removing unit 226A and the second abnormal value removing unit 223A is that the first abnormal value removing unit 226A removes known “obvious” abnormal values related to the characteristics of the sensor, etc. The second abnormal value removing unit 223A removes the abnormal value based on a mathematical basis.

7 includes a data interpolation unit 224 and a filter processing unit 225. The data shaping unit 221B illustrated in FIG. This data shaping means 221B is used particularly when a highly accurate heart rate sensor is used. The data shaping unit 221 </ b> C shown in FIG. 8 includes only the data interpolation unit 224. As described above, the data shaping unit 22 of the present embodiment can employ any configuration except that the data interpolation unit 224 is an essential component. Therefore, a configuration other than the configuration shown in the present embodiment may be added to the data shaping unit 22 as appropriate.

In the data shaping unit 22, the data shaping unit 221A, the data shaping unit 221B, and the data shaping unit 221C described above, correction according to the sensor characteristic information 1101 may be performed on the data at a predetermined time. This is to absorb the characteristic difference due to the difference in the sensor model.

Next, the index value calculation means 23 shown in FIG. 9 will be described. The index value calculation means 23 includes an index 1 calculation unit 232-1 to an index m calculation unit 232-m, and a vectorization unit 233. The index 1 calculation unit 232-1 to the index m calculation unit 232-m calculate m autonomic nerve activity indices (plurality) based on heart rate variability using RRI3 (before shaping) 1013 and RRI (after shaping) 1014. To do.

Examples of the autonomic nerve activity index (plurality) based on heart rate variability used in the present embodiment include the following.
-Activity index related to product moment statistics SDRR: RRI standard deviation (sympathetic and parasympathetic activity index)
RMSSD: root mean square of adjacent RRI differences (parasympathetic activity index)
SDSD: Standard deviation of the difference between adjacent RRIs (parasympathetic activity index)
pRR50: Rate at which the difference between adjacent RRIs exceeds 50 [milliseconds] (parasympathetic activity index)

・ Activity index based on spectrum analysis LF: PSD 0.04-0.15 [Hz] power (mainly sympathetic activity index)
HF: (PSD 0.15-0.40 [Hz] power (parasympathetic activity index)
HF / (LF + HF) (parasympathetic activity ratio)
Specific frequency band power p _i (i = 0, 1, 2,..., 9): PSD power of 0.15 + i × 0.025 to 0.15+ (i + 1) × 0.025 [Hz] Activity index of individual frequency band of nerve)

The activity index related to the product moment statistic is given by the following equation when the data of RRI3 (after shaping) 1013 is x _i (1 ≦ i ≦ m).

The calculation of the index based on the spectrum analysis is performed by first applying an FFT, a maximum entropy method, etc. to the RRI (after shaping) 1014 (that is, {x _i } sequence (1 ≦ i ≦ m)) PSD (power spectral density function). ) The FFT calculation is well known, and detailed description thereof is omitted here.

FIG. 21 is a diagram showing PSD obtained by the FFT direct method. FIG. 22 is a diagram showing PSD obtained by the maximum entropy method (burg method). As a method for obtaining the PSD, there is a method for obtaining by PSD and AR model prediction (Yule-Walker method) in addition to the above.

When the power spectral density function at the frequency f is expressed as PSD (f), LF and HF are given by the following equations. There are various theories about the definition of the integration interval of LF and HF, and there is no fixed one. Therefore, the integration interval in the following formula is an example.

In the above “index based on spectrum analysis”, HF / (LF + HF) is shown as an index, but Ratio may be LF / (LF + HF) or HF / LF. The power p _i (i = 0, 1, 2,..., 9) in the specific frequency band is a unique index provided for the first time by the inventor of the present application, and the integration interval of the HF is 0.15 to 0.40. The section at [Hz] is divided into 10 sections, and the power is calculated in units of divided sections. That is, the power is obtained by the processing shown by the following equation.

FIG. 23 is a diagram showing that the above 0.15 to 0.40 [Hz] section of PSD obtained by the FFT direct method is divided into 10 sections from section I to section X, and the power is obtained in each section. It is. Further, FIG. 24 shows that the above 0.15-0.40 [Hz] section of PSD obtained by the maximum entropy method (burg method) is divided into 10 sections from section I to section X. FIG. Of course, the division number 10 is an example, and it may be divided into an arbitrary number of sections of 2 or more.

In the above embodiment, in the cnt-th loop of sleepiness estimation processing, using the RRI acquired during LT seconds from time T0 + cnt × DT to time T0 + cnt × DT + LT, SDRR, RMSSD, SDSD, pRR50, LF, HF , HF / (LF + HF), p ₀ , p ₁ , p ₂ , p ₃ , p ₄ , p ₅ , p ₆ , p ₇ , p ₈ , p ₉ index values are calculated. Is done.

Hereinafter, each index value is expressed as y _j (j = 1, 2, 3,..., M). In particular, y _j , _cnt (j = 1, 2, 3,..., M), (cnt = 0, 1, 2, when representing an index value at time T0 + cnt × DT + LT (depending on the cnt value) , 3, ...). That is, y _{α, β} means the α-th index value calculated from RRI acquired during LT seconds from time T0 + β × DT to time T0 + β × DT + LT.

The vectorization unit 233 provided in the index value calculation unit 23 shown in FIG. 9 vectorizes the index value calculated in the cnt-th loop of sleepiness estimation processing. That is, the index values y _{1, cnt} to y _{m, cnt} generated based on RRI acquired during the LT seconds from time T0 + cnt × DT to time T0 + cnt × DT + LT are collectively vectorized. Thereby, the index value vector 1021
YY _cnt = {y _{1, cnt} , y _{2, cnt} , y _{3, cnt} ,..., _{Ym , cnt} } is generated. In the text, the capital letters are continuously written as YY in the text. Hereinafter, YY _cnt is treated as an index value vector at time T0 + cnt × DT + LT.

Since the index value calculation means 23 of FIG. 4 includes an index value vector time series unit (not shown), the index value vector time series unit will be described below. In the index value vector time series unit, the index value vectors YY _k calculated up to the k-th time (0 ≦ cnt ≦ k) of the sleepiness estimation process are collected and index value vector time series 1051: DD = {YY ₀ , YY ₁ , YY ₂ ,..., YY _k }.

Next, the drowsiness estimation unit 24 that performs processing subsequent to the index value calculation unit 23 will be described. In the drowsiness estimating means 24, the index value vector time series 1051: DD = {YY ₀ , YY ₁ , YY composed of loops 0 to k of the reprocessing loop (loop start unit 202 to loop end determination unit 203). ₂ ,..., YY _k }, and sleepiness is estimated based on the sleepiness estimation rule 1102.

Here, the sleepiness estimation rule 1102 includes a threshold vector RR with a threshold value, an index value vector time series DD, and an estimation function f having the threshold vector RR as arguments. f (DD, RR) means an estimated value. Further, the estimation function f returns a discretized integer value such as an estimated value f (DD, RR) = {1, 0} (1: possibility of sleepiness, 0: no possibility of sleepiness). Designed. In the above, a binary value is shown, but the estimated value f (DD, RR) = {3, 2, 1, 0} (3: high possibility of drowsiness, 2: possible possibility, 1: low possibility, 0: None), and may be quaternized, and is not particularly limited.

The estimation function f is an index value vector time series DD = {YY ₀ , YY ₁ , YY ₂ ,..., YY _k } of at least one element YY _cnt (0 ≦ cnt ≦ k). ) And YY _cnt = {y _{1, cnt} , y _{2, cnt} ,..., Y _{m, cnt} }, and at least one y _{j, cnt} (1 ≦ j ≦ m) can be used. That is, an index value vector corresponding to at least one time among index value vectors corresponding to time k + 1 is referred to, and one or more of m elements of these index value vectors are used. be able to.

A more specific example 1 of the estimation function f will be shown. In Example 1, it can be determined that there is “sleepiness” when the following conditions 1 to 3 are satisfied. This is an example of binary determination.

Condition 1: HF / (LF + HF) ≧ 0.2
Condition 2: p ₅ ≧ 3.8 or p ₆ ≧ 4.2
Condition 3: HF ≦ 200

In the present embodiment, it is assumed that the index name, the index variable, and the threshold value have correspondence as shown in FIG. For example, the variable of HF / (LF + HF) is y ₇ and the threshold is r ₇ (= 0.2). Similarly, the variables of index names p ₅ , p ₆ and HF are y ₁₃ , y ₁₄ and y ₆ , and the threshold values are r ₁₃ (= 3.8), r ₁₄ (= 4.2), r ₆ (= 200). ). If the threshold vector RR corresponding to this is an index number from 1 to 17 in a serial number from the top to the bottom of FIG. 14, RR = {{7, 0.2}, {13, 3.8}, {14 4.2}, {6,200}}, {{index number, threshold}, {index number, threshold}, {index number, threshold}, {index number, threshold},. Can be configured.

Assuming that the latest cnt value meaning the processing time and loop count is k, the estimation function of Example 1 is given as follows.
f (DD, RR): = (y _{7, k} ≧ r ₇ ) ∧ ((y _{13, k} ≧ r ₁₃ ) ∨ (y _{14, k} ≧ r ₁₄ )) ∧ (y _{6, k} ≦ r ₆ )
In the above, ∧ means AND, and ∨ means OR.
It can be set to 1 if the above estimation function f (DD, RR) is TRUE and 0 if it is FALSE.

Note that the processing performed by the estimation function in the present embodiment is that the determination using a general index is Condition 1, and is a characteristic unique to the subject, and is 0.275 to 0.3 particularly in PSD especially during sleepiness. The determination using the tendency that the component in the [Hz] region or the 0.3 to 0.325 [Hz] region increases is the condition 2, and the parasympathetic nerve activity such as meal or talk is activated The determination is Condition 3, and both Condition 1 and Condition 2 are satisfied and Condition 3 is excluded.

The above-mentioned conditions relating to the characteristics unique to the user can be obtained by performing measurement using all the indexes in advance for the user and measuring when sleepiness occurs. From this measurement data, a significant change in each activity index when sleepiness occurs can be obtained and used as an estimation function. In the cloud storage 30, which will be described later, the sleepiness estimation rule 1102 is updated each time the sleepiness estimation process is executed, and the process can be executed in a configuration that improves the estimation accuracy. Thereby, a sleepiness estimation rule specific to the user can be set.

A more specific example 2 of the estimation function f will be described. Example 2 is obtained by adding condition 4 to the estimation function f of the previous example 1, and is as follows.
Condition 1: HF / (LF + HF) ≧ 0.2
Condition 2: p ₅ ≧ 3.8 or p ₆ ≧ 4.2
Condition 3: HF ≦ 200
Condition 4: Condition 1 has been satisfied three times in the past (4 times including the present)

Assuming that the latest cnt value meaning the processing time and loop count is k, the estimation function of this example is given as follows.
f (DD, RR): = (y 7, k ≧ r 7) ∧ ((y 13, k ≧ r 13) ∨ (y 14, k ≧ r 14))
∧ (y _{6, k} ≦ r ₆ ) ∧ ((y _{7, k−1} ≧ r ₇ ) ∧ (y _{7, k-2} ≧ r ₇ ) ∧ (y _{7, k-3} ≧ r ₇ ))
It can be set to 1 if the above estimation function (conditional expression) is TRUE, and 0 if it is FALSE.

As described above, the index value calculation unit 23 (FIG. 1) calculates the activity index for unit time using the RRI data for each first time, collects the activity index for unit time for each type of activity index, and vectorizes it. Then, the vectorized index values are arranged in time series to create an index value vector time series, and the sleepiness estimation means 24 performs sleepiness estimation using a threshold vector and a function that is evaluated using the index value vector time series.

Next, the output means 25 that performs processing following the sleepiness estimation means 24 will be described. The output unit 25 performs a process of returning the estimated value f (DD, RR) calculated by the sleepiness estimation unit 24 as a return value to the caller / upper side of the sleepiness estimation process.

Next to the processing of the output means 25, the loop end determination unit 203 performs processing. In response to an end instruction from the caller / upper side of the drowsiness estimation process, the loop end determination unit 203 branches to YES, terminates and stops data acquisition by the RRI sensor 10, and controls the caller / upper function of the drowsiness estimation process. Will pass. When the loop end determination unit 203 branches to NO, the loop start unit 202 returns to the processing of the RRI acquisition unit 21, and the loop processing after the next step is performed.

After the end determination (branch to YES) is made by the loop end determination unit 203, the loop end unit 204 includes at least the RRI2 (before shaping) 1012 acquired from the RRI sensor 10 and the index value, as shown in FIG. The index value vector time series 1051 that is a part or all of the vector time series data DD is transferred to the cloud storage 30 together with a flag that preferably indicates whether or not the drowsiness occurs. As a result, the cloud storage 30 stores the acquired RRI 1012X and the index value vector time series 1051X for a predetermined period.

The data transferred to the cloud storage 30 may be used for recalculation of the sleepiness estimation rule 1102 by a server (not shown) that can access the cloud storage 30. The drowsiness estimation rule 1102 may be updated every time the drowsiness estimation process is executed, and a configuration that improves the estimation accuracy may be adopted.

The second embodiment of the sleepiness estimation apparatus is configured as shown in FIG. In the second embodiment, the terminal 20 holds the estimation rule 1102. Therefore, the start processing unit 201 reads the estimation rule 1102 from the storage device of the terminal 20 and uses it. The other configuration is the same as that of the first embodiment. In the second embodiment, the process is executed according to the flowchart shown in FIG. 26 having a process of reading the estimation rule 1102 from the storage device of the terminal 20.

The third embodiment of the sleepiness estimation apparatus is configured as shown in FIG. That is, the RRI sensor 10 and the server 60 are connected to the terminal 20, and the cloud storage 30 is connected to the server 60. The terminal 20 is provided with RRI acquisition means 21 and output means 25. Further, the server 60 is provided with data shaping means 22, index value calculation means 23, and sleepiness estimation means 24. An estimated value 1005 is obtained in the server 60. This estimated value 1005 is transmitted to the calling side / upper side of the sleepiness estimation process provided in the server 60 and / or the terminal 20, and used there.

The cloud storage 30 includes sensor characteristic information 1101 and sleepiness estimation rules 1102 in advance. The cloud storage 30 can store the acquired RRI data acquired in the terminal 20 as the acquired RRI 1012X of history data.

FIG. 28 shows the configuration of the terminal 20, the server 60, and the cloud storage 30 in the third embodiment. The configuration of the cloud storage 30 is the same as that of the first embodiment, and the configuration of the terminal 20 includes a data processing program for performing data acquisition processing, and performs data processing such as data shaping and sleepiness estimation processing. This is different from that of the first embodiment in that a sleepiness estimation program is not provided.

The server 60 performs processing under the control of the CPU, and is composed of a memory that temporarily stores data being processed and the like, a non-volatile memory that stores various types of processing data, and the like. And a communication unit for performing communication via the network. Communication with the terminal 20 or the cloud storage 30 can be performed by the communication unit. The server 60 does not include a data processing program for performing data processing for data acquisition, but includes a sleepiness estimation program for performing data processing such as data shaping and sleepiness estimation processing.

In the third embodiment, the processing is executed according to the flowchart shown in FIG. Since the processes with the same reference numerals as in FIG. 4 are basically the same processes, the different parts will be described. In the terminal 20, the start processing unit 201 does not capture the sensor characteristic information or the estimation rule, but the reprocessing time interval DT [seconds] by loop processing (between the loop start unit 202 and the loop end determination unit 203), 1 Parameters such as the RRI data length LT [seconds] handled in the loop processing of the number of times are fetched. The start processing unit 601 of the server 60 performs the same processing as the start processing unit 201 in FIG.

In the third embodiment, in order for the terminal 20 and the server 60 to operate synchronously, the terminal 20 is provided with a server synchronization start unit 206 and a server synchronization end unit 207, and the server 60 has a terminal synchronization A start unit 606 and a terminal synchronization end unit 607 are provided. From the synchronization start by communication between the server synchronization start unit 206 and the terminal synchronization start unit 606 to the synchronization end by communication between the server synchronization end unit 207 and the terminal synchronization end unit 607, RRI data collected by the terminal 20 is sent to the server 60, The server 60 performs data processing and sleepiness estimation processing using the sent RRI data.

During the period from the server synchronization start unit 206 to the server synchronization end unit 207, the terminal 20 acquires RRI data by the RRI acquisition unit 21 and sends the acquired RRI data to the server 60. The server 60 includes an RRI acquisition unit 21A that fetches data in which the time 1002 sent from the terminal 20 and the RRI (before shaping) 1001 are associated with each other, and the data shaping provided in the terminal 20 in the first embodiment of FIG. Means 22, index value calculation means 23, and sleepiness estimation means 24 are provided. In the server 60, data in which the sent time 1002 and RRI (before shaping) 1001 are associated with each other is fetched, and processing by the data shaping means 22, the index value calculating means 23, and the drowsiness estimating means 24 is executed, and an estimated value 1005 is obtained. Is obtained and sent to the terminal 20.

In the terminal 20, the estimated value 1005 is received, and the output means 25 performs a process of returning it as a return value to the caller / upper side of the sleepiness estimation process. The loop end determination unit 203 of the terminal 20 performs processing for continuing the loop processing by the RRI acquisition unit 21 and the output unit 25 until the predetermined condition as described above is satisfied, and the loop end determination unit 603 of the server 60 performs predetermined processing. Until the above condition is satisfied, a process of continuing the loop process by the data shaping means 22, the index value calculating means 23, and the sleepiness estimating means 24 is performed.

In the terminal 20, the end determination by the loop end determination unit 203 and the end process by the loop end unit 204 are performed, and after the end notification by the server synchronization end unit 207, the end processing unit 205 ends the sleepiness estimation process. In the server 60, the end determination by the loop end determination unit 603 and the end process by the loop end unit 604 are performed. In the end process by the loop end unit 604, the RRI 2 (before shaping) 1012 and the index value vector time series 1051 used by the server 60 for the process are transferred to the cloud storage 30, and the acquired RRI 1012X and the index value vector time series for a predetermined period are transferred. Accumulate as 1051X. Furthermore, the end notification is performed by the terminal synchronization end unit 607, and then the end processing unit 605 ends the sleepiness estimation process.

FIG. 30 shows a connection configuration of the terminal 20, the server 60, and the cloud storage 30 in the fourth embodiment. In this embodiment, the cloud storage 30 and the server 60 are connected to the terminal 20, and the cloud storage 30 and the server 60 are connected to each other. Since the configuration of each part is the same as in FIG. 27, the description thereof is omitted.

FIG. 31 shows the configuration of the terminal 20, the server 60, and the cloud storage 30 in the fourth embodiment. Since the configuration of each part is the same as in FIG. 28, the description thereof is omitted. Furthermore, the operations of the terminal 20, the server 60, and the cloud storage 30 in the fourth embodiment are as shown in the flowchart of FIG. 29 described above. As in the third embodiment, the terminal 20 acquires the RRI data. Processing is performed, and the server 60 performs data processing such as data shaping and sleepiness estimation processing using the RRI data acquired by the terminal 20.

As described above, in the present embodiment, RRI (heart rate interval) data is obtained from the RRI sensor 10, and an index (a plurality of indexes reflecting the activity state of the sympathetic nerve / parasympathetic nerve is used against the background of the autonomic nervous function evaluation method based on heart rate variability. ) Is calculated and time-series. Next, in this embodiment, the sleepiness estimation level is calculated based on the time series data of the index and the sleepiness estimation rule. The sleepiness estimation level can be, for example, a discretized integer of (4. Large, 3. Caution, 2. Low, 1. None), and the possibility of a state of high sleepiness is estimated. Note that the sleepiness estimation level handled in the present embodiment is a word indicating the possibility of sleepiness and does not mean the depth of sleepiness.

In the present embodiment, it is possible to perform sleepiness estimation with high real-time characteristics by repeating the processing from RRI measurement to sleepiness estimation level calculation at predetermined time intervals (for example, every 10 seconds). In this embodiment, RRI (heart rate interval data) output by a heart rate sensor (a heart rate measuring device including an electrocardiograph) is captured. In the present embodiment, the sleepiness estimation level can be calculated and output based on the calculation and determination inside the apparatus.

This embodiment has the following features and characteristics. The application / system implemented by the apparatus of this embodiment obtains and outputs the sleepiness estimation level. For example, it is possible to issue warnings to car drivers and machine operators, prompt them to take a break, and provide appropriate advice, services, and processing according to the sleepiness estimation level, such as safely stopping equipment. is there. The application / system implemented by the apparatus according to the present embodiment can be useful for workers who use the apparatus according to the present embodiment to know their sleepiness objectively. It is possible to take The application / system implemented by the apparatus according to the present embodiment can be used so that an administrator who manages the worker who uses the apparatus according to the present embodiment gives an appropriate instruction when estimating sleepiness of the worker. .

The apparatus according to this embodiment can be applied to a commercially available heart rate sensor, smartphone, mobile terminal, or the like. Therefore, it can be realized at a lower cost than a special device or a dedicated terminal in which the RRI sensor 10 and the computing device (hardware) and the analysis processing (software) system are integrated, and it is highly convenient. . Regarding the use of commercially available sensors, there is a concern that different determination results may be derived due to characteristic differences between different sensor models. The apparatus according to the present embodiment has a configuration in which correction according to sensor characteristics is performed on heartbeat interval data (RRI) acquired from a sensor, and information relating to the correction can be provided via the cloud. There is.

Furthermore, since individual differences, age differences, etc. appear in vital data, there is a problem that sleepiness estimation rules and thresholds need to be customized for each individual. Correspondingly, the apparatus according to the present embodiment has two features. First, after executing sleepiness estimation by sleepiness estimation processing, together with index calculation based on heartbeat data, sleepiness data is collected by sleepiness collection means other than heartbeat data, and both are collated and analyzed. It has a mechanism for determining different sleepiness estimation rules and thresholds. The second feature is that the terminal specializes in data collection for the purpose of reducing the calculation burden on the terminal side, and adopts a division process such as determining the data analysis and sleepiness estimation rules and thresholds at the server. It is required to be able to

10 RRI sensor 20 terminal 21 RRI acquisition unit 22, 221A, 221B, 221C data shaping unit 23 index value calculation unit 24 sleepiness estimation unit 25 output unit 30 cloud storage 60 server 201 start processing unit 222 trend removal unit 223 abnormal value removal unit 224 Data interpolation unit 225 Filter processing unit 233 Vectorization unit

Claims (12)

1. RRI acquisition means for acquiring RRI data, which is RR interval data, from a signal obtained by an RRI sensor that detects a signal corresponding to an R wave of an electrocardiogram signal;
   Index value calculation means for calculating an index value for a plurality of types of activity indexes related to autonomic nerve activity based on a result of statistical processing of the RRI data and a result of spectrum analysis of the RRI data;
   Drowsiness estimation means for evaluating drowsiness estimation by evaluating the index value calculated by the index value calculation means based on a drowsiness estimation rule composed of an estimation function that is evaluated based on a threshold value and / or a variation state for each activity index. The sleepiness estimation apparatus characterized by the above-mentioned.
2. The activity index as a result of statistical processing of the RRI data includes:
   SDRR: (standard deviation of RRI)
   RMSSD: (the square root of the root mean square of the difference between adjacent RRIs)
   SDSD: (standard deviation of the difference between adjacent RRIs)
   pRR50: (Percentage where the difference between adjacent RRIs exceeds 50 (milliseconds))
   The sleepiness estimation apparatus according to claim 1, wherein at least one of the following is included.
3. In the activity index as a result of spectrum analysis of the RRI data,
   LF: (PSD (power spectral density function) 0.04 to 0.15 [Hz] power)
   HF: (PSD 0.15-0.40 [Hz] power)
   HF / (LF + HF)
   p _i (i = 0, 1, 2,..., 9): (PSD 0.15 + i × 0.025 to 0.15+ (i + 1) × 0.025 [Hz] power)
   The sleepiness estimation apparatus according to claim 1, wherein at least one of the following is included.
4. The index value calculation means calculates an index value for unit time using the RRI data for each first time, collects the index values for unit time for each type of activity index, and vectorizes the index value. Create index value vector time series in time series,
   The sleepiness estimation apparatus according to claim 1, wherein the sleepiness estimation means performs sleepiness estimation using an estimation function evaluated using a threshold vector and the index value vector time series.
5. The estimation function of the drowsiness estimation rule includes a comparison condition between each activity index and the corresponding threshold value, and a condition for the presence or absence of a predetermined fluctuation state, and uses either AND or OR, or AND and OR. The sleepiness estimation apparatus according to claim 1, wherein two or more are used and the conditions are combined.
6. The drowsiness estimation apparatus according to claim 1, comprising data shaping means for performing at least one of trend removal, abnormal value removal, data interpolation, and filter processing, which are removal of a predetermined frequency component.
7. Computer
   RRI acquisition means for acquiring RRI data, which is RR interval data, from a signal obtained by an RRI sensor that detects a signal corresponding to an R wave of an electrocardiogram signal;
   Index value calculation means for calculating index values for a plurality of types of activity indices related to autonomic nerve activity based on a result of statistical processing of the RRI data and a result of spectrum analysis of the RRI data;
   Based on a drowsiness estimation rule constituted by a threshold value and / or a fluctuation state for each activity index, the index value calculated by the index value calculation means is evaluated to function as sleepiness estimation means for estimating sleepiness Drowsiness estimation program.
8. The activity index as a result of statistical processing of the RRI data includes:
   SDRR: (standard deviation of RRI)
   RMSSD: (the square root of the root mean square of the difference between adjacent RRIs)
   SDSD: (standard deviation of the difference between adjacent RRIs)
   pRR50: (Percentage where the difference between adjacent RRIs exceeds 50 (milliseconds))
   The sleepiness estimation program according to claim 7, wherein at least one of the following is included.
9. In the activity index as a result of spectrum analysis of the RRI data,
   LF: (PSD (power spectral density function) 0.04 to 0.15 [Hz] power)
   HF: (PSD 0.15-0.40 [Hz] power)
   HF / (LF + HF)
   p _i (i = 0, 1, 2,..., 9): (PSD 0.15 + i × 0.025 to 0.15+ (i + 1) × 0.025 [Hz] power)
   The sleepiness estimation program according to claim 7 or 8, wherein at least one of the following is included.
10. The index value calculation means calculates an index value for unit time using the RRI data for each first time, collects the index values for unit time for each type of activity index, and vectorizes the index value. Create index value vector time series in time series,
    The sleepiness estimation program according to claim 7, wherein the sleepiness estimation means performs sleepiness estimation using an estimation function evaluated using a threshold vector and the index value vector time series.
11. The estimation function of the drowsiness estimation rule includes a comparison condition between each activity index and the corresponding threshold value, and a condition for the presence or absence of a predetermined fluctuation state, and uses either AND or OR, or AND and OR. The sleepiness estimation program according to claim 7, wherein the sleepiness estimation program is formed by using two or more and combining the conditions.
12. The computer,
    8. The sleepiness estimation program according to claim 7, wherein the sleepiness estimation program functions as data shaping means for performing at least one of trend removal, abnormal value removal, data interpolation, and filter processing, which are removal of a predetermined frequency component.
    
    

Images