JP2016129629A - Biological state estimation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device estimating a biological state of a subject by using a normalized biological measurement feature amount, which suppresses an influence that a state of the subject and the biological measurement feature amount are easily changed on a statistical value for normalization.SOLUTION: The device normalizes a biological measurement feature amount of the subject and determines a biological state of a subject estimated on the basis of the normalized biological measurement feature amount. In setting of a width of a time range of a group of biological measurement feature amounts or a group of measurement data in the past used in calculating a statistical value of the biological measurement feature amount used in normalization, a width of the time range in a time zone where a change of the biological measurement feature amount in the past is large is set to be shorter than a width of the time range in a time zone where a change of the biological measurement feature amount in the past is small.SELECTED DRAWING: Figure 3

Description

  The present invention relates to an apparatus for estimating a psychological or physiological state of a person (hereinafter referred to as “biological state” in the present specification), and more specifically, the biological state as described above. The present invention relates to an apparatus that estimates by biological measurement.

  Psychological or physiological states (biological states) related to various emotions and consciousness caused by human brain and nerve activities, such as concentration, stress, tension, rest, awakening, sleepiness, sleep, etc. It is also reflected in the physical state of the brain, for example, heart rate, respiration, body temperature, skin potential, blink, blood pressure, blood flow, blood oxygen concentration (light transmittance), electroencephalogram and the like. Therefore, in recent years, the measurement of the physical state as described above obtained by physical measurement such as pressure, potential, current or optical measurement in the field of the detection or estimation technique of the human biological state as described above. Attempts have been made to estimate the biological state as described above using a value or its characteristics (frequency characteristics, waveform characteristics, etc. of measured values). In the case of estimation of a biological state based on such a physical measurement value of the living body or an index value of the characteristic thereof (hereinafter, generally referred to as “biological measurement feature amount”), typically, the state in which the biological state is known first. A large amount of data of biometric features obtained in step 1 is collected, and a statistical analysis method is used to search for biometric features that are highly correlated with the biological state, and biometric features and biological states that are highly correlated And the correlation or correspondence is determined. Then, in the estimation of the actual biological state, the above-mentioned pre-obtained value is obtained by referring to the biological measurement feature amount (highly correlated with the biological state) obtained from the measurement value in the subject as a variable parameter. It is estimated that the biological state corresponding to the biological measurement feature at that time is the biological state at that time using the correlation between the measured biological characteristic and the biological state (hereinafter, unless otherwise specified) , "Biometric measurement feature value" refers to a biometric feature value having a high correlation with the biological state.)

  In the estimation of the biological state based on the above-described biological measurement features, the particular problem is that the correlation between the biological measurement features and the biological state depends on individual differences or individual differences (daily fluctuations, seasonal fluctuations, etc.). The point is that it often changes. That is, the correlation between the biometric feature quantity and the biometric state determined using the biometric feature quantity data collected in a certain situation is different in other situations (different individuals, Therefore, it is not always possible to use the correlation in one situation as it is to estimate the biological state in another situation. It is difficult (if it is used as it is, it may result in a decrease in estimation accuracy or erroneous estimation) Therefore, from the past, taking into account changes in the situation, i.e., individual differences and individual differences, a method for determining the correlation between biometric features and biological states and estimating biological states based on actual biometric features Various proposals have been made.

  For example, in Patent Document 1, the degree of sleepiness of a subject is estimated from fluctuation in a certain time width (reference time width) of a feature amount (heart rate feature amount) obtained from the heartbeat data of the subject. In this apparatus, a method for determining a reference time width for fluctuation calculation for each subject using a frequency spectrum of a heartbeat feature amount and suppressing an estimation error due to individual differences has been proposed. In Patent Document 2, the driver's arousal level (degree of arousal) is determined from an index value determined based on the length of eye blink time detected in a photographed image of the driver's face of the vehicle. ), The index corrected based on the measured heart rate variability corresponding to the circadian rhythm so that the influence of the fluctuation of the blink time due to the circadian rhythm of the driver is reduced. It has been proposed to use values for arousal level estimation. Furthermore, in Patent Document 3, an apparatus for determining whether or not a subject is healthy based on biological measurement features such as weight and blood pressure related to the health condition of the subject, the past biological measurement features When judging whether or not it is healthy from the magnitude of the occurrence probability of the current biometric feature calculated in the frequency distribution of the quantity data, the day is divided into a plurality of time zones, The occurrence probability of the current biometric feature amount is determined using the occurrence frequency distribution of the past biometric feature amount data generated for each divided time zone, and thereby the biometric feature in the determination An attempt is made to reduce the influence of time-dependent fluctuations in a day on the correlation between the quantity and the health condition of the subject. In Patent Document 4, in a system for performing a threshold determination as to whether or not a user's state is normal based on statistics of sensor data worn by the user, a threshold value is stored as a pre-stored data group. When dividing into sections and determining by statistical processing of the data group for each section, when dividing the time section, the time for the division so that the dispersion of the average value and variance of the data group for each section is minimized It has been proposed to determine the width and the starting point of the section (corresponding to performing the time section division of the data group so that the reproducibility of the statistical value of the sensor data is maximized).

JP2010-131061 JP2008-186263 JP 2010-152658 A JP 2010-238105 A

  By the way, one of the causes of the change in the correlation between the biological measurement feature amount and the biological state as described above is that the biological measurement feature amount is different from the change of the biological state to be estimated or determined. It changes depending on the situation (daily fluctuations, seasonal fluctuations, etc.). For example, it is known that the heartbeat, respiration, and other physically measurable states listed above vary depending on the time zone of the day, regardless of changes in the biological state to be observed. Therefore, in the determination of the correlation between the biometric feature quantity and the biological state and the actual estimation, the biometric feature is excluded so that the change in the biometric feature quantity due to the change in the condition of the subject is excluded. The quantity is statistically normalized, and the normalized biometric feature quantity (normalized biometric feature quantity) is adopted as a variable parameter. In short, normalization of such biometric feature amounts means that a plurality of data groups generated in different frequency distributions (that is, data groups obtained in different situations) have a high occurrence frequency distribution. In order to calculate the normalized biometric feature quantity, a calculation using a statistical value such as an average value or standard deviation of past data of the biometric feature quantity is performed. A certain normalized biological measurement feature value thus calculated is associated with a certain biological condition regardless of the condition of the subject. (As in the example of the embodiment described later, typically, the normalized biometric feature quantity is given as a vector having a plurality of elements. In that case, for one normalized biometric feature quantity vector, A certain biological state is associated with it.)

  Regarding the normalization of the above-mentioned biometric feature quantity, the past data group of the biometric feature quantity used for calculating the statistical value for the normalization calculation is ideally a data group that is all detected or measured in the same situation And the number of data needs to be a statistically significant number of data. However, in practice, for example, it is very difficult to collect a statistically significant number of data under exactly the same circumstances, for example at a certain time, so in practice The statistical value of data for a specific situation is calculated using a data group obtained in a situation close to the specific situation. Specifically, a statistical value for normalization of data at a specific time is calculated using a data group obtained with an appropriate time width including the time. For example, the statistical value for 11:00 am is calculated using data in the range of 10:00 am to 12:00 am over the past several days.

  In this regard, setting the width of the time range of the data group for calculating the statistical value for normalization of the biometric feature quantity at a specific time increases the number of data used for calculating the statistical value. Therefore, it is preferable in terms of improving the significance of the statistical value, and if the time range is too wide, the data group included in the calculation will include a lot of data that is different from the situation at a specific time, There is a possibility that a decrease in accuracy is caused. Moreover, in the measurement in the actual subject, the ease of change of the subject's situation varies depending on the time or the time zone, so in the data group included in a certain time range. The degree of change in the biometric feature amount also varies depending on the time or time zone. For example, in general, changes in the subject's situation and biometric features are small at night when the subject is often asleep, whereas the subject is active during the day. It becomes large in the time zone that is often. That is, in this case, there is a high possibility that the data group in the daytime time zone includes data in a situation different from the situation at the specific time as compared with the data group in the nighttime time zone. Therefore, it is preferable that the time width of the data group used in the calculation of the statistical value for normalization is set according to the condition of the subject and the ease of change of the biometric feature quantity. The above findings are utilized in the present invention.

  Thus, one object of the present invention is to calculate statistical values for normalizing biometric feature quantities in the technique for estimating the biological state of a subject using normalized biometric feature quantities. The time range width of the data group used in the process is set according to the condition of the subject and the ease of change of the biometric feature, and the condition of the subject in the statistical value for normalization This is to suppress the influence of the ease of changing the biometric feature quantity.

  According to the present invention, the above-described problem is an apparatus for estimating a biological state of a subject, a biological measurement feature acquisition unit that acquires the biological measurement feature of the subject, and normalizing the biological measurement feature A biometric feature normalization unit that calculates the normalized biometric feature value and an estimated biological state determination unit that determines the biological state of the subject estimated based on the normalized biometric feature value A biometric feature normalization unit that stores a group of past biometric features of the subject or a group of measurement data used for determination of the biometric feature, and within a set time range A normalization statistical value calculation unit that calculates a statistical value used for normalization of the biometric feature value using the stored past biometric feature value group or measurement data group, and using the statistical value To calculate normalized biometric features from biometric features A normalized biometric feature value calculation unit, and a normalization statistical value calculation unit that stores a group of past biometric feature values or measurement data groups used to calculate the statistical value. In the setting of the range width, the width of the time range in the time zone in which the change in the past biometric feature amount is large is the time in the time zone in which the change in the past biometric feature amount is small. This is achieved by a device that is set shorter than the width of the range.

  In the above configuration, the “subject's biological state” means various emotions caused by human brain and nerve activities such as concentration, stress, tension, rest, awakening, sleepiness, and sleep, as already mentioned. Or a psychological or physiological state related to consciousness (which may be any one of them), and “biometric feature” refers to heart rate, respiration, and body temperature in a subject. , Skin potential, blink, blood pressure, blood flow, blood oxygen concentration (light transmittance), measurement data such as electroencephalogram, etc., or index values representing characteristics such as frequency characteristics, waveform characteristics, etc., and correlation with a biological state The measurement data is a measurement value measured using a physical sensor such as pressure, potential, current, or optical measurement. In addition, in the apparatus of the present invention, “normalization” of the biometric feature amount is a biometric feature in a different time zone in a group of biometric feature amounts obtained from measurement data in different time zones. The biometric feature values are corrected so that the occurrence frequency distribution of the values among the groups of quantities is substantially similar to each other, so that fluctuations in the biometric feature values due to the measurement time zone are eliminated as much as possible (biological This is a process of calculating a normalized biometric feature value that is a variable parameter (used for determining the state).

  In the configuration of the present invention, when normalizing the biometric feature amount, a statistical value calculated using data within a set time range in the past of the biometric feature amount, More specifically, when an average value, standard deviation value, variance value, or the like is used, the width of the time range to which the data used for calculation of the statistical value belongs changes as described above. It will be set according to the magnitude of the degree. According to such a configuration, in the calculation of statistical values for normalization, in the time zone in which the change in the past data of the biometric feature amount is large, that is, in the time zone in which the change in the condition of the subject is large. The time range of past data used is set to be narrow, thereby reducing fluctuations caused by changes in the subject's situation included in the group of past data, and the subject's situation according to the measurement time zone It is expected to suppress the decrease in accuracy in the statistical value due to the change in the value. On the other hand, for time zones where changes in biometric feature values in the past data are small, that is, time zones where changes in the condition of the subject are small, the width of the time range of past data used in the calculation of statistical values Therefore, the frequency of using past data at a point in time away from the measurement point that is the object of determination is reduced, so that the change in the condition of the subject is small More data can be used, and statistical values with higher statistical accuracy can be used.

  In the embodiment, the past data used in the calculation of the statistical value for normalization is the same time before the previous day of the time range including the measurement time of the measurement data of the biometric feature value to be normalized. It may be data of a time range including. In order to satisfy the statistically significant number of the past data used for calculating the statistical value, the number of days going back is increased according to the width of the time range in which the past data is collected. In that case, the possibility that the condition of the subject has changed increases as the measurement time of the past data becomes farther from the measurement time of the measurement data of the biometric feature quantity to be normalized. Therefore, in order to reduce the contribution of the data in which the condition of the subject is changing in the statistical value calculation, the calculation of the statistical value is performed by the weighted sum calculation of the past data. At this time, the weight of the past data may be reduced as the measurement time of the past data is further away from the measurement time of the measurement data of the biometric feature quantity to be normalized. Further, the setting of the time range width of past data used in the calculation of statistical values may be appropriately determined in advance by referring to the change in the biometric feature amount in the past data. In one aspect, the specific time range may be set by determining the starting point and the ending point so as to have a width corresponding to the magnitude of the change in the biometric feature quantity. In another aspect, the weight function assigned to the past data in the statistical value calculation is typically a small bell that becomes smaller as the measurement time of the measurement data of the biometric feature quantity to be normalized becomes farther away. Thus, the width of the time range may be adjusted by increasing or decreasing the bell-shaped width of the weight function within the time range of the arbitrarily set width. That is, in this case, the width in the time direction of the weight function assigned to the past data is set to a smaller value as the width to be set becomes narrower.

  Thus, according to the present invention described above, in the calculation of the statistical value of the past data of the biometric feature used for normalizing the biometric feature for estimating the biological state of the subject, the calculation The width of the time range to which the past data used in the field belongs is appropriately set according to the degree of change in the condition of the subject. According to such a configuration, in a group of past data that gives a statistical value at a certain time (particularly, data in a time zone in which the situation changes rapidly), data measured under different situations can be included. Therefore, the accuracy or reliability of statistical values is expected to be improved.

  Other objects and advantages of the present invention will become apparent from the following description of preferred embodiments of the present invention.

FIG. 1A shows an example of an apparatus (biological condition estimation apparatus) that estimates the biological state of a subject to which the present invention is applied, whether the subject is in a concentrated state or a resting state based on heartbeat information. It is the figure which represented the structure of the apparatus (concentration state estimation apparatus) which estimates which it is in the format of a block diagram. FIG. 1B is a diagram showing, in a tabular form, data of the measurement time and the heart rate feature amount obtained from the measured heart rate information, which are sequentially stored in the storage unit. FIG. 1C is a diagram showing an outline of the operation of the concentration-rest state estimation device according to the present invention in the form of a flowchart. FIG. 2 is a diagram schematically showing the fluctuation of the heart rate feature amount according to the time of the day. In the figure, each plot is the value of the heart rate feature value at each time, the thick solid line is the daily fluctuation of the boundary between the concentrated state and the rest state, and the thin dotted line is the frequency distribution of the heart rate feature value at each time Represents. The time window indicates a range of acquisition time of past data that is referred to when calculating a normalization parameter for normalizing the heartbeat feature amount. FIG. 3 shows an example of the actual measurement data of the heartbeat feature amount. In the figure, X is a point where it is estimated that the heartbeat information has temporarily changed due to exercise such as walking of the subject. Each of the time windows a to g is a time width of past data used when calculating the normalization parameter at the time indicated by the arrow. FIG. 4A is a diagram showing, in the form of a flowchart, a process from a search process of past data used for a normalization parameter for normalizing a heartbeat feature value to a calculation process of a normalization parameter. FIG. 4B shows a window width setting process in the process of FIG. 4A in the form of a flowchart, and FIG. 4C shows a table referred to at that time. FIG. 5 shows the outlier removal processing in the processing of FIG. 4A in the form of a flowchart. FIG. 6 shows an example of the storage status of past data referred to in the normalization parameter calculation processing in the processing of FIG. 4A, and the weight given to the past data when normalization parameters are calculated. The change in length is shown. (A) shows an example of using past data for the past several days for normalization parameter calculation processing, and (B) shows an example of using past data for the same day and the same time before the previous year for normalization parameter calculation processing. Is shown. In FIG. 6A, the hatched area indicates an area where data exists. FIG. 7 shows an example of the result of the concentration state estimated from the heartbeat feature amount. For comparison, the determination result based on the θ wave content rate of the electroencephalogram and the electroencephalogram at the time of the estimation is also shown.

DESCRIPTION OF SYMBOLS 10 ... Mobile communication terminal 12 ... Heart rate sensor 20 ... Network 30 ... Server

  The configuration of the apparatus for estimating the biological state of the subject according to the present invention described above is, for example, a state in which the subject (human) is concentrated on one as an example of the biological state (hereinafter referred to as “concentrated state”). .) And an unconcentrated state (hereinafter referred to as “rest state”) from the heart rate information of the subject (hereinafter referred to as “concentrated state estimation device”). Is applicable. Therefore, hereinafter, as an example of the embodiment of the present invention, a subject concentration state estimation apparatus based on heartbeat information will be described.

Regarding determination of human concentration state Among the human brain waves, the band of 4 to 8 Hz is generally called a θ wave and is known to occur when drowsiness occurs. However, for example, when a person concentrates on something such as computer operation or computation, a characteristic component called Fmθ wave is also generated in the brain waves measured from the human frontal region. It is known to do. Therefore, whether or not a person is in a concentrated state (whether they are in a concentrated state or in a state of being relaxed or resting without concentrating) depends on the θ wave power or content (hereinafter, It can be determined by simply referring to “the amount of θ wave”. Further, whether or not a person is in a concentrated state is reflected on heartbeat, respiration, skin potential, or other physiological indicators that can be measured physically and easily, in addition to brain waves. Therefore, it has been attempted to estimate whether or not a person is in a concentrated state using a physiological index that can be easily measured as described above, instead of determination based on an electroencephalogram. For example, in estimating whether or not a person is in a concentrated state using heart rate information (the same applies to other physiological indicators that can be easily measured), first, focus on a plurality of subjects. The subject's heart rate and electroencephalogram are measured while the task to perform and the task to rest are executed under controlled conditions. Then, for example, using a method of statistical analysis such as secondary discriminant analysis, the state of the subject determined based on the amount of θ wave in the electroencephalogram (in this case, whether it is in a concentrated state or in a resting state) Any one) is referred to as a correct answer value, and n heartbeat feature values (biometric measurement feature values) having a high correlation with the state of the subject in the heartbeat measurement values are searched. Is done. Along with this, in an n-dimensional space with n highly correlated heart rate features as variables, a point group of heart rate features associated with the state of the subject determined by the θ wave amount is obtained. A point cloud region associated with the concentration state of the subject and a point cloud region associated with the resting state in the space of the heart rate feature amount is defined. It can be said that the distribution of the concentration state region and the resting state region in the space of the highly correlated heart rate feature amount represents the correlation between the heart rate information and the state of the subject. When the person's heart rate feature amount is obtained, by determining whether the point of the obtained heart rate feature amount exists in the concentration state region or the rest state region in the heart rate feature amount space, The state of the subject will be estimated.

  Regarding the correlation between the heart rate feature amount and the condition of the subject as described above, the distribution of the frequency of occurrence of the heart rate feature amount varies among individuals. (See FIG. 2), depending on whether the subject is in a concentrated state or not, depending on the measurement conditions, there is a correlation (that is, the distribution of the concentrated state region and the resting region in the heart rate feature space). (Position and size) can vary. Therefore, when determining the correlation between the heart rate feature quantity and the condition of the subject, normalization of the heart rate feature quantity is performed so that the distribution of the occurrence frequency of the heart rate feature quantity is uniform regardless of the measurement conditions. This is executed, and the correlation between the normalized heartbeat feature and the state of the subject is determined. FIG. 7 shows an example in which it is determined whether or not the subject is in a concentrated state using the normalized heart rate feature amount described above, together with the measurement of the θ wave content rate. Referring to the figure, when the θ wave content rate exceeds the boundary value, it is determined that the subject is in a concentrated state, and the estimation is performed using the heartbeat feature amount almost correspondingly. It will be understood that Details of normalization of the heartbeat feature amount will be described later.

Referring to block diagram of apparatus 1 (A), it is at the specific configuration of the system of the present embodiment, Stated in short, resulting in heartbeat sensor 12 attached to the subject (user) The obtained heart rate information is transmitted to the mobile communication terminal 10 such as a smartphone carried by the subject by wireless communication, and the mobile communication terminal 10 confirms that the concentration state of the subject is correlated from the heart rate information. The detected heart rate feature amount is extracted, and the estimated state of the subject at the time of measurement of the current or heart rate feature amount (hereinafter referred to as “estimation”) using the predetermined correlation with reference to the heart rate feature amount. Is referred to as a “state”). The mobile communication terminal 10 is in a state in which it can communicate with the server 30 on the network 20 through a public communication line or any other communication line, and the measured heart rate information, heart rate feature value and / or estimated state are While being stored in the storage unit 32 in the server 30 from the mobile communication terminal 10, the server 30 uses the stored information to perform “normalization of heart rate features” that is executed when the estimated state is determined. Necessary normalization parameters are calculated, and the calculated normalization parameters are received by the mobile communication terminal 10 via the network 20 and used for normalization of the heartbeat feature amount.

  In the above configuration, the heart rate sensor 12 may be a wristwatch type or a chest band type device, and specifically, an electrocardiogram for measuring the potential generated by the heart beat in such a form. Various sensors such as a sensor, a photoelectric pulse wave sensor that optically detects changes in subcutaneous blood flow, and an acceleration sensor that captures minute changes in the skin caused by the heartbeat may be used.

  The mobile communication terminal 10 includes a heart rate feature amount calculation unit 16 and an estimated concentration state determination unit 18, and further performs data transmission / reception through a public communication line or other arbitrary communication line according to an arbitrary mode. A communication unit 14 and a display unit 19 for displaying measurement results and / or determination results for a subject or a user are provided.

In the configuration of the mobile communication terminal 10, the heart rate feature amount calculation unit 16, more specifically, when a cardiac potential sensor is used as the heart rate sensor 12, the RR interval in the measured waveform of the cardiac potential. From the (peak interval) data, a plurality of heartbeat feature quantities that are known to have a high correlation with the condition of the subject, that is, a heartbeat feature quantity vector x is calculated. In the present embodiment, as the heartbeat feature vector x, for example, the following four quantities may be calculated from RR interval data in a waveform of a predetermined time interval, for example, 3 minutes.
(I) RRV ... dispersion value of RR interval (RRI) (ii) LF / HF ... RRI power spectrum of LF component (0.04-0.15 [Hz]) divided by HF component (0.15-0.4 [Hz]) (Iii) center of gravity frequency: frequency indicating the center of gravity of the RRI power spectrum (iv) peak frequency: frequency indicating the maximum value of the RRI power spectrum In addition to the above, the skewness and saliency of the RRI, VLF, Various heartbeat features such as features obtained from LF, HF, and Lorentz plots may be used.

  In the estimated concentration state determination unit 18, the normalization of the heart rate feature amount vector x calculated as described above is performed, and the subject's state is determined to be in a concentration state based on the normalized heart rate feature amount vector x. A determination is made as to whether there is a resting state.

Here, normalization of the heart rate feature vector x is performed by receiving normalization parameters calculated by a normalization parameter calculation unit of the server 30 described later via the network 20 and the communication unit 14, and normalizing them. Using parameters, for example, the following equation may be used.
X ′ = (X−Xa) / S (1)
Here, X is a heart rate characteristic amount (each element of the vector) to be determined or at the current time, and Xa and S are average values of past data of X calculated in a manner described later. It is a standard deviation, and X ′ is a heart rate feature value (normalized heart rate feature amount) to be determined or determined at the current time. The determination of the state of the subject typically includes a correlation obtained in advance, more specifically, n normalized heart rate features that are highly correlated with the state of the subject. Information indicating the position of the concentration state region and the resting state region in the space, for example, a map indicating the range of each of the concentration state region and the resting state region using the normalized heartbeat feature amount prepared in advance as a variable parameter With reference to this, it may be done by determining whether the point of the heart rate feature value to be determined or at the current time belongs to the concentrated state region or the rest state region.

  In the system of FIG. 1A described above, the server 30 includes a storage unit 32 that sequentially stores measured heartbeat information, heartbeat feature values, and / or estimated states, as already mentioned. Normalization parameter calculation that calculates average parameters and standard deviation parameters such as standard deviation of past data used for normalization of the elements of the heartbeat feature vector x described above, using the stored historical data of heartbeat features Part 34 is provided. More specifically, as shown in FIG. 1B, the storage unit 32 stores the values of each element of the heart rate feature amount vector obtained from the sequentially measured heart rate information. It is memorized with the date and time. In addition to these data, information such as actions and places such as walking and personal computer work performed at the same time may be stored. The normalization parameter calculation unit 34 extracts the past data of the heart rate feature vector accumulated in the storage unit 32 in a manner described in more detail later, and calculates an average value and a standard deviation (more strictly, by weighted calculation). (Average value and standard deviation) are calculated and transmitted to the mobile communication terminal 10 via the network 20. As for the calculation of the normalization parameters, the normalization parameters for each time period in one day are collectively calculated at a certain time, for example, at midnight on the day when the system is used or other time. It may be stored or transmitted to the mobile communication terminal 10, or the normalization parameter may be calculated when there is a request from the mobile communication terminal 10. .

  The configuration of each unit in the mobile communication terminal 10 and the server 30 described above may be realized by the operation of a computer according to a program. Each of the mobile communication terminal 10 and the server 30 is equipped with a CPU, a storage device, and an input / output device (I / O) that are interconnected by a bidirectional common bus (not shown) in a normal manner. The operation of each unit is achieved by executing a program in the CPU.

  As a configuration different from the configuration of the system illustrated in FIG. 1A, a storage unit and a normalization parameter calculation unit may be arranged in the mobile communication terminal 10, or a heart rate feature amount calculation unit, A concentration degree estimation unit may be arranged in the server 30.

Operation of the apparatus Referring to FIG. 1C , in the operation of the concentration state estimation apparatus for a subject according to one embodiment of the present invention illustrated in FIG. Then, in the mobile communication terminal 10, heart rate information such as an electrocardiogram is sequentially read from the heart rate sensor 12 attached to the subject (step 1), and the heart rate feature value (vector) is calculated from the waveform of the heart rate. Are sequentially calculated (step 2). Next, normalization of heart rate features (step 4) is executed using the normalization parameters (step 3) read from the server 30, and further normalization is performed with reference to the map prepared in advance as described above. The current (or time to be determined) normalized heart rate feature vector state (concentration or rest) in the heart rate feature space is the current (or time to be determined) test The estimated state of the person will be determined.

The normalization of the heart rate feature amount will be described in detail below in the apparatus according to the present invention as described above.

(A) Overview of normalization of heart rate feature amount As already described, the distribution of the occurrence frequency of the heart rate feature amount is psychological or related to consciousness in the brain such as concentration and stress that are targets of estimation of the apparatus of the present invention. In addition to changes in physiological state, it changes due to diurnal and seasonal variations such as sleep, dietary activity, and circadian rhythm. For example, as illustrated in FIG. 2, the value of a certain heart rate feature value from 6:00 to 18:00 indicates whether the subject is in an active state regardless of whether or not the subject is in a concentrated state. It has been found that the average magnitude and the degree of variation of the heart rate feature amount change as well as the state boundary (thick solid line) of the subject whether or not they are in a concentrated state. Therefore, for example, the value of the feature amount indicated by a in FIG. 2 is determined to be in a concentrated state in the morning, but in the afternoon, the boundary of the subject's state fluctuates, and the resting state Thus, the determination process of the subject's state becomes complicated (in the determination, it is necessary to simultaneously measure a state other than the state to be observed and consider it in the determination). ). Therefore, in order to reduce the complexity of such determination, first, in the apparatus of the present invention, normalization of the heart rate feature amount is executed as already mentioned. In the normalization of the heart rate feature value, in order to obtain a variable parameter having the same occurrence frequency distribution regardless of the time and / or time of the heart rate measurement, the above-described equation (1) is used to obtain the heart rate feature value. Quantity conversion is performed. That is, the normalization of the heart rate feature value is measured at different times and / or times as shown by the thin dotted lines in FIG. This corresponds to conversion into an amount in which the average value and spread of the distributions of the occurrence frequencies substantially coincide with each other. An amount obtained by converting a certain heartbeat feature amount, that is, a certain normalized heartbeat feature amount, does not depend on a measurement time and / or timing of the heartbeat, and is in a certain biological state. It will be associated. As described above, in the present embodiment, the heartbeat feature value is a vector composed of four elements, and thus the normalized heartbeat feature value is also a vector composed of four elements. Accordingly, in the present embodiment, one biological state is associated with one normalized heartbeat feature vector.

  With regard to the normalization of the heart rate feature value, as mentioned in the “Summary of Invention” section, in order to normalize the heart rate feature value at a certain time point, it is exemplified in the above formula (1). As shown in the figure, statistical values such as an average value and standard deviation of the heart rate feature amount data under the same situation as that time are required. For the calculation of statistical values such as the average value and standard deviation, ideally, a statistically significant number of groups of past data before the previous day obtained under the same situation as that time point is preferably used. However, in practice, it is often difficult to prepare the measurement data of the heart rate information before the day before a certain time point so as to reach a statistically significant number. Therefore, statistical values such as the average value and standard deviation of the heart rate feature value at a certain time point are data obtained in the past before the previous day within a time interval (time window) set to a certain width including that time point. Calculated using groups. For example, as shown in FIG. 2, the statistical values such as the average value and standard deviation of the data at 11:00 are 60 minutes before and after 11:00, that is, within the time window set from 10:00 to 12:00. (For example, obtained up to the previous day). Then, statistical values such as an average value and a standard deviation from time to time may be calculated sequentially using past data in the time window while sliding the time window as shown in the figure.

  In the apparatus of the present invention, in order to further improve the accuracy of statistical values, a new configuration described below is further adopted when extracting past data used for calculating statistical values.

(B) Setting of time window width As described above, statistical values such as average values and standard deviations of past data for normalization of heart rate feature values in measurement at a certain time point are set to a certain width. When calculating using a data group obtained in the past within a given time interval (time window), the width of the time window is such that the number of past data used for the calculation is a statistically significant number. It is preferable that the measured state of the data included in the time window is substantially the same as the state at the time of measurement of the normalized heartbeat feature value. If heart rate feature quantities measured in different situations are mixed, the accuracy of statistical values such as the calculated average value and standard deviation will decrease. In this regard, in the past, the time width was typically set to 60 minutes before and after each time point, regardless of the time zone. However, as illustrated in FIG. 3, when the transition of the heart rate feature amount of the subject in almost one day is observed, the time zone in which the change in the heart rate feature amount is relatively large and the time period in which the change is relatively small depending on the time zone. A band is found to exist. For example, in the example of FIG. 3, the change in the heartbeat feature amount is relatively small from about 9:00 to 12:00, but a large change in the heartbeat feature amount is observed from about 12:00 to 12:30. The In such a case, if a time zone with a relatively large change enters the time window at the time of a time zone with a relatively small change, the accuracy of the statistical values such as the calculated average value and standard deviation will be reduced. .

  Therefore, in the apparatus of the present embodiment, the width of the time window of data used for calculating the statistical value is set in consideration of the magnitude of the change in the heartbeat feature value. More specifically, the time window width is set to be narrow in the time zone in which the change in the subject's situation is large, thereby causing the change in the subject's situation included in the group of past data. In a time zone where changes in the subject's situation are small, the time window width may be set wide and a larger number of data may be used. . In the case of the example of FIG. 3, the width of the time window is basically set to a width of 60 minutes before and after each time point (a, b, and g in the figure), but changes depending on the setting. In the time zone in which 12:00 to 12:30 having a large value is included, the width of the set time window is shortened (cf in the figure). According to such a configuration, it is possible to avoid as much as possible a mixture of a time zone with a large change in heart rate feature amount and a small time zone within one time window, and statistics such as average value and standard deviation at each time point. An improvement in the accuracy of the value is expected. Note that the specific width of the time window may be set in advance in consideration of past data trends, user activity patterns, and the like (see the description related to FIG. 4B described later).

(C) Outliers By the way, during the daytime, when the subject's activities are active, if the subject exercises such as walking, the state of the heartbeat temporarily fluctuates greatly. . In the example of FIG. 3, the spike-like displacement of the heartbeat feature amount at the time indicated by X in the figure is due to the subject's movement such as walking. If the data at the time of fluctuation of the temporary heartbeat state (outlier value) is mixed in the data for calculating the statistical values such as the average value and the standard deviation, the accuracy of the calculation result is lowered. Become. Therefore, in the apparatus of the present embodiment, it is preferable that the outliers as described above are removed prior to the extraction of past data for calculating statistical values such as average values and standard deviations. The removal of the outlier value may be performed by an arbitrary method. For example, it can be achieved by the processing of the flowchart illustrated in FIG. 5 described later.

(D) About past data weighting In the calculation of statistical values such as the average value and standard deviation of past data for normalization of heart rate feature values, the width of the time window is set as described above, and the outlier value is calculated. When the past data in the time window is extracted after the removal, the number of past data preferably reaches a statistically significant number. Therefore, as the data used for the calculation, preferably, a group of data in the time zone of the same time window that is traced back from the previous day to the previous day until the number of past data reaches a statistically significant number. Extracted. In that case, if it is too far from the measurement date of the current or determined heart rate feature value, there is a high possibility that the change in the state of the subject is excessive, so that the number of days going back may be appropriately limited. In addition, when the activity pattern of the subject is almost determined every day of the week, every day of the month, every day of the year, as past data, Data on the same day of the month and data on the same day of the past year may be used for calculation of statistical values such as an average value and a standard deviation for normalizing the heart rate feature amount.

  However, when calculating statistical values such as average values and standard deviations of past data for normalization of heart rate features by measurement at a certain time, the data used for calculating the statistical values is calculated from that time. The longer the measurement time is, the higher the possibility that the state of the subject is changing. Therefore, in order to reduce the influence of the change in the state of the subject, when calculating the statistical value, the measurement time is different from the measurement time of the heart rate feature quantity to be determined for each past data. As the statistical value is calculated, the past data may be weighted so that the contribution to the statistical value calculation is reduced as the value increases.

Specifically, when calculating the statistical value of past data for normalization of the heartbeat feature quantity to be determined at the measurement time t, the past data i measured at time ti d days before y years ago The weight w multiplied by is
w = w _TIME i · _DAY i · w _YEAR i… (2)
May be given by Where w _TIME i
w _TIME i = exp (− (ti−t) ² / h) (2a)
w _DAY i = 1.0− (d−1) / D (2b)
w _YEAR i = 1.0−y / Y (2c)
It is. Here, the equation (2a) is a weight for the time that is reduced as the time ti becomes farther from the measurement time t, as depicted at the right end of FIG. 6A (h is set in advance). (2b) is a weight for the number of days that is reduced as the date d moves away from the measurement date, as depicted at the top of FIG. 6 (A) (D is the number of days going back. When the right side becomes negative, it is assumed that w _DAY i = 0.) As shown in the right end of FIG. It is a weight for the number of years that decreases as the distance increases (Y is the maximum number of years that go back, and when the right side is negative, w _YEAR i = 0).

  In addition to the above formula, different weights may be added depending on the day of the week. For example, Monday to Friday are office work, and Saturday and Sunday are general office workers who are often at home. If the day on which the statistical value is calculated is Monday to Friday, , 1.0, Saturday and Sunday data may be multiplied by 0.5 (Similarly, if the day for which the statistical value is calculated is Saturday and Sunday, 0.5 to Saturday, Day data may be multiplied by 1.0.) In addition, if you use a smartphone acceleration sensor to estimate separately sitting, computer work, driving, and other behaviors and store them in the memory along with the heart rate features, past data corresponding to the current behavior It is possible to consider the difference due to the influence of the action by increasing the weight of.

The average value Xa and the standard deviation S of past data calculated by weighting the past data may be given by the following equation using the past data xi and its weight wi.
Xa = Σwixi / Σwi (3a)
S = (Σwi (xi-Xa) ² / Σwi) ^1/2 (3b)
Here, Σ is the total sum of all past data i to be extracted. When the heartbeat feature amount is a vector having a plurality of elements, the average value Xa and the standard deviation S are calculated for each element.

  Thus, the weighting of the past data described above makes it possible to reduce the contribution of past data away from the measurement time point of the heartbeat feature quantity to be determined. For example, as illustrated in FIG. 6A, when extracting past data in order to calculate a statistical value of a heartbeat feature value measured at a certain time, a relatively close past, for example, the previous day to 3 In some cases, the number of data acquired on a day is insufficient, and past data on a date that is traced back is used. In such a case, the past data from the previous day to the third day is not treated in the same way as the most recent data, but the contribution is reduced. It is expected that the effects of

(E) Normalization Parameter Calculation Flow Normalization parameters (average value Xa and standard deviation S of past data) are calculated in the case of the example of FIG. 34 may be executed. Specifically, referring to FIG. 4A, first, as described above, the width of the time window is set according to the measurement time of the heartbeat feature quantity to be determined (step 10). More specifically, as shown in FIG. 4B, first, the measurement time (determination time) (step 20) of the determined heart rate feature value is acquired from the mobile communication terminal 10, and then, FIG. As illustrated in (C), the window width corresponding to the determination time may be selected from a window width table prepared in advance (step 21).

  When the time window width is set, first, the past data in the time window including the time corresponding to the determination time of the previous day is searched and extracted (step 11). At this time, since there is a possibility that an outlier value exists in the past data in the corresponding time window, first, such “outlier value” removal processing may be executed. In the removal processing of “outlier value”, referring to the flowchart of FIG. 5, first, after selecting a feature amount to be processed (step 30), the average value of past data in the time window A standard deviation is calculated. Here, the average value and the standard deviation may be based on a simple calculation without weighting. Thereafter, if there is data outside the range of the average value ± 3 × standard deviation, the data is excluded (steps 32 and 33). This operation may be repeatedly executed until there is no data outside the range of the average value ± 3 × standard deviation (step 32). It should be understood that the outlier removal processing may be executed by any other method.

  Thus, when the outlier is removed from the past data within the corresponding time window, the number of data is counted (step 12), and the counted number of data is equal to a predetermined data number threshold, that is, It is determined whether a statistically significant number has been exceeded (step 13). Here, when the counted number of data does not reach the predetermined data number threshold, in order to extract more data, the past data in the corresponding time window one day before the previous day (the day before the previous day) is extracted. Reference is made (steps 14 and 15), and the same processing as described above is repeated (steps 11 to 13). That is, if the number of data does not exceed the data number threshold even after extraction from the data of the day before, the data of the previous day is further referred to. Then, when the above process is repeated and the number of data reaches the data number threshold, the average value Xa and the standard deviation of the past data are obtained by the equations (3a) and (3b) using the past data extracted so far. S is calculated (step 16). On the other hand, if the number of days going back is excessive while the processes of steps 11 to 15 are repeated, the possibility that the state of the subject is no longer the same as the state at the time of measurement to be determined is reduced. It is done. Therefore, when the number of days going back exceeds a predetermined number of days threshold (step 14), the above iterative process is terminated, and default values arbitrarily set as the average value Xa and the standard deviation S may be used. In addition, even when the number of days since the start of use of the apparatus is small and sufficient past data has not been acquired, default values arbitrarily set as the average value Xa and the standard deviation S may be used.

  Although the above description has been made in relation to the embodiment of the present invention, many modifications and changes can be easily made by those skilled in the art, and the present invention is limited to the embodiment exemplified above. It will be apparent that the invention is not limited and applies to various devices without departing from the inventive concept.

  It should be understood that the processing configuration in the normalization of the heart rate feature value in the present invention can be applied to other biometric feature values. In the above configuration, the setting of the width of the time window can also be achieved by adjusting the weight given to the past data. Specifically, for example, when the value of h in the expression (2a) is changed, the weight width that is a bell-shaped function is changed. In this case, by increasing h, the width of the weight is reduced, and the same effect as that of reducing the width of the time window can be obtained. It should be understood that such a configuration also belongs to the scope of the present invention. Furthermore, in the calculation of normalization of the heart rate feature value of the equation (1), when the difference in the occurrence frequency of the heart rate feature value due to the difference in measurement conditions is small or almost no, the division by the standard deviation is omitted. When the difference in the average value of the heart rate feature amount due to the difference in measurement conditions is small or almost no, the subtraction by the average value may be omitted.

Claims (1)

An apparatus for estimating a biological state of a subject,
A biometric feature acquisition unit for acquiring the biometric feature of the subject;
A biometric feature normalization unit that normalizes the biometric feature and calculates a normalized biometric feature;
An estimated biological state determination unit that determines the biological state of the subject estimated based on the normalized biological measurement feature amount;
A data storage unit that stores a group of past biometric feature quantities of the subject or a group of measurement data used to determine a biometric feature quantity, and a set time range; A statistical value calculation unit for normalization that calculates a statistical value used for normalization of the biometric feature quantity using the stored past biometric feature quantity group or measurement data group, and the statistics A normalized biometric feature quantity calculation unit that calculates a normalized biometric feature quantity from the biometric feature quantity using a value,
In the normalization statistical value calculation unit, in the setting of the time range width of the stored past biometric feature quantity group or measurement data group used for calculation of the statistical value, the past An apparatus in which the width of the time range in a time zone in which the change in the biometric feature amount is large is set shorter than the width of the time range in a time zone in which the change in the past biometric feature amount is small.

Images