Fatigue estimation device, fatigue estimation method, and program

Abstract

This fatigue estimation device (100) comprises: a fatigue estimation model (12) constructed in advance on the basis of data including a relationship between a physiological quantity outside an activity period of a person and the amount of human herpesvirus outside the activity period, the fatigue estimation model (12) being capable of outputting a fatigue degree estimated from the corresponding amount of human herpesvirus in response to an input of biometric information; a physiological quantity measurement unit (20) for measuring a physiological quantity of a subject (user (99)); and a fatigue estimation unit (10) for outputting the fatigue degree of the subject on the basis of the measured physiological quantity by using the fatigue estimation model (12).

Description

Fatigue Estimation Device, Fatigue Estimation Method, and Program

[0001] The present disclosure relates to a fatigue estimation device, a fatigue estimation method, and a program.

[0002] Conventionally, there has been a need to estimate a person's degree of fatigue. For example, if a user during work can know their own degree of fatigue, it is useful as information for dealing with the degree of fatigue, such as taking a break at an appropriate timing. On the other hand, as a concept of fatigue, there is fatigue (also referred to as subacute fatigue or circadian fatigue) that accumulates daily by continuously performing work or the like. Such fatigue is estimated based on how much fatigue accumulates daily and how sleep for recovering fatigue is taken. Patent Document 1 discloses a sleep index calculation device that calculates a sleep index from the sleep time, the degree of fatigue immediately before sleep, and the degree of fatigue immediately after sleep.

[0003] Japanese Patent Application Laid-Open No. 2020-121035 International Publication No. 2006 / 006634 Japanese Patent Application Laid-Open No. 2007-330263

[0004] By the way, in conventional fatigue estimation systems and the like, depending on the measurement timing of the physiological quantity used for fatigue estimation, an appropriate degree of fatigue may not be obtained and the degree of fatigue may not be presented appropriately. Therefore, the present disclosure provides a fatigue estimation device and the like capable of estimating a degree of fatigue for more appropriate presentation.

[0005] In order to solve the above problems, a fatigue estimation device according to one aspect of the present disclosure is a fatigue estimation model constructed in advance based on data including the relationship between the physiological quantity outside a person's activity period and the amount of human herpesvirus during the outside of the activity period, and can output the degree of fatigue estimated from the corresponding amount of human herpesvirus in response to the input of biological information. And a physiological quantity measurement unit that measures the physiological quantity of the subject, and a fatigue estimation unit that causes the fatigue estimation model to output the degree of fatigue of the subject using the measured physiological quantity.

[0006] Furthermore, a fatigue estimation method according to one aspect of the present disclosure is a fatigue estimation method performed by a computer, comprising the steps of: constructing a fatigue estimation model based on data including the relationship between physiological amounts outside of a person's activity period and the amount of human herpesvirus outside of said activity period, which can output a fatigue level estimated from the corresponding amount of human herpesvirus in response to the input of biological information; having a physiological amount measurement unit measure the physiological amounts of a subject; and using the constructed fatigue estimation model based on the measured physiological amounts to output the fatigue level of the subject.

[0007] Furthermore, one aspect of this disclosure can be implemented as a program for causing a computer to execute the fatigue estimation method described above. Alternatively, it can be implemented as a computer-readable recording medium storing the program.

[0008] According to this disclosure, the level of fatigue required to make a more appropriate presentation can be estimated.

[0009] Figure 1 is a diagram illustrating the challenges of fatigue estimation. Figure 2 is a block diagram showing the functional configuration of a fatigue estimation device according to an embodiment. Figure 3 is a flowchart showing a fatigue estimation method according to an embodiment. Figure 4 is a diagram showing a correction table according to a modified example of the embodiment. Figure 5 is a diagram showing an accuracy table according to a modified example of the embodiment.

[0010] (Knowledge leading to this disclosure) As shown in Patent Document 1, there has been a need to estimate the degree of fatigue in people, particularly the degree of fatigue that accumulates over a medium period, also known as subacute fatigue or circadian fatigue, and various technological developments have been underway. On the other hand, one method is to measure a person's physiological quantities and estimate their fatigue level from those quantities. It is believed that there is a certain relationship between physiological quantities and fatigue levels. In particular, if several indicators that have a strong relationship with fatigue levels are extracted from physiological quantities as biological information, and fatigue levels are estimated using this biological information, it will be possible to estimate a person's fatigue level with high accuracy.

[0011] On the other hand, since people's subjective perceptions of fatigue vary, a certain standard index is needed to quantify fatigue. Research has shown that, for example, in Patent Documents 2 and 3, the use of human herpesvirus as a standard index for fatigue is proposed. Specifically, the amount of human herpesvirus is low if fatigue has not accumulated, and increases as the level of fatigue rises. Here, Figure 1 is a diagram to explain the challenges of fatigue estimation. In Figure 1, the amount of human herpesvirus (herpes amount) is shown on the vertical axis, and the elapsed time is shown on the horizontal axis. In Figure 1, the period with dot hatching is shown to indicate when a person was active (exercising). If a person is not active, the herpes amount remains at a constant low value, as shown by the white circle, whereas if a person is active, the herpes amount transiently increases over a certain period from the start of the activity to after the end of the activity, in accordance with the duration of the activity, as shown by the hatched circle.

[0012] To use the amount of human herpesvirus (HPV) as a reference index, it is necessary to investigate the relationship between the amount of HPV outside of activity periods (excluding periods of human activity) and a person's physiological load (more precisely, the relationship with biological information) in order to eliminate influences such as transient increases in herpes levels. By using this relationship between physiological load outside of activity periods and the amount of HPV, it becomes possible to obtain a degree of fatigue relative to physiological load, that is, a degree of fatigue estimated from the amount of HPV corresponding to physiological load.

[0013] However, in this relationship, if physiological quantities outside the activity period are used to estimate fatigue, the fatigue level can be obtained with good accuracy, but if physiological quantities within the activity period are used to estimate fatigue, it is difficult to maintain accuracy. Therefore, the following embodiments describe a fatigue estimation device that uses physiological quantities outside the activity period to estimate fatigue, and a fatigue estimation device that takes into account the effects when physiological quantities within the activity period are used to estimate fatigue.

[0014] The embodiments of this disclosure will be described below with reference to the drawings. The embodiments described below are all general or specific examples of this disclosure. Therefore, the numerical values, components, their arrangement and connection configurations, as well as the steps and their order, shown in the following embodiments are examples and are not intended to limit this disclosure. Accordingly, any components in the following embodiments that are not described in the independent claims of this disclosure will be described as optional components.

[0015] Furthermore, each figure is a schematic diagram and not necessarily a strictly accurate representation. Therefore, the scale and other aspects may not necessarily be consistent across all figures. In each figure, substantially identical components are given the same reference numerals, and redundant explanations are omitted or simplified.

[0016] (Embodiment) [Configuration of Fatigue Estimation Device] First, the functional configuration of the fatigue estimation device 100 in this embodiment will be explained in detail using Figure 2. Figure 2 is a block diagram showing the functional configuration of the fatigue estimation device according to this embodiment. The functional configuration of the fatigue estimation device 100 described below is a compilation of configurations used in several embodiments. Depending on the operation of the fatigue estimation device 100, some functional configurations may be included that are not necessary. In other words, among the components shown in Figure 2, the components other than the physiological quantity measurement unit 20, the fatigue estimation unit 10, and the fatigue estimation model 12 are not essential.

[0017] As shown in Figure 2, the fatigue estimation device 100 in this embodiment comprises a fatigue estimation unit 10, a physiological quantity measurement unit 20, an indicator extraction unit 30, an activity level extraction unit 40, an accuracy calculation unit 50, and a presentation unit 60. Each of these components constituting the fatigue estimation device 100 may be housed in a single enclosure or the like and integrated, or they may be realized as multiple individual devices connected to each other via a communication line. The fatigue estimation device 100 is a computer including a presentation device such as a smartphone, PC, or tablet terminal, and is realized by a processor, memory, and a program executed using these.

[0018] The fatigue estimation unit 10 includes a wake-up detection unit 11, a fatigue estimation model 12, a correction unit 13, and an output unit 14. The fatigue estimation unit 10 is, for example, installed as one of the functions of a fatigue estimation device 100.

[0019] The physiological volume measurement unit 20 is a measuring device such as a wearable sensor, which is worn by the user 99, the subject of fatigue estimation. The physiological volume measurement unit 20 is, for example, a wearable electrocardiograph. A wearable electrocardiograph can be attached directly to the user 99's body to measure the user 99's electrocardiogram information. For example, wristwatch-type wearable electrocardiographs, wearable electrocardiographs that are attached directly to the user 99's solar plexus area with a belt or gel pad, and ring-shaped wearable electrocardiographs are known, and the form of the wearable electrocardiograph to which this application applies is not particularly limited.

[0020] The index extraction unit 30 is a processing unit that extracts biological information for indicators used to estimate fatigue levels from physiological amounts measured by the physiological amount measurement unit 20. For example, the index extraction unit 30 extracts LF / HF as one of the biological information components from the electrocardiogram information measured by the physiological amount measurement unit 20. LF / HF is the ratio of the LF component (low frequency component: generally 0.05 to 0.15 Hz) and the HF component (high frequency component: generally 0.15 to 0.40 Hz) in the power spectrum obtained by frequency conversion of a waveform obtained by arranging the heart rate intervals (R-R intervals) obtained from the electrocardiogram in a time series, and is obtained as time-series data.

[0021] The fatigue estimation unit 10 has a processor and a storage medium such as an SSD as memory for saving data, and has the function of estimating and outputting the fatigue level of the user 99 by executing a program etc. stored in the memory by the processor.

[0022] The memory contains a pre-built fatigue estimation model 12 (or fatigue estimation algorithm). The fatigue estimation model 12 is pre-built through pre-training using a dataset of "true fatigue values" and "explanatory variables" which are biological information.

[0023] Here, the "true value of fatigue" in the dataset is the amount of human herpesvirus. The "explanatory variable" in the dataset is LF / HF, extracted by the indicator extraction unit 30. We will now explain the pre-training of the fatigue estimation model 12 using these variables.

[0024] First, a dataset is a collection of multiple pairs of data, each pairing a "true value of fatigue" with an "explanatory variable." For example, a sample of saliva from subject A is collected every morning, and the amount of human herpesvirus contained in a unit volume of collected saliva is quantified using PCR or a similar method. Meanwhile, heart rate data from subject A is acquired before and after the time the saliva is collected. Longer acquisition times result in better accuracy, but if the time is too long, the computational load increases, so the acquisition time should be determined according to the processing resources available for the computation. Here, for example, we will use 3 minutes of heart rate data. The R-R interval is calculated from the heart rate waveform measured over 3 minutes, and by frequency-converting the time-series data of the R-R interval, one LF / HF data can be obtained.

[0025] By acquiring this data daily, a dataset for the number of days the data was collected can be obtained. Furthermore, data from multiple subjects other than subject A may also be acquired. This results in a dataset containing pairs of "true fatigue values," namely "human herpesvirus amounts," and "explanatory variables," namely "LF / HF." Using this dataset, a machine learning model is trained to construct a fatigue estimation model 12. Any machine learning model can be used; for example, regression analysis and SVR can be applied. A new user 99 has their electrocardiogram information measured for 3 minutes by the physiological volume measurement unit 20, and the LF / HF extracted from the 3-minute electrocardiogram information by the index extraction unit 30 is input to the constructed fatigue estimation model 12 to estimate the fatigue level of the user 99. The estimated fatigue level is output to the presentation unit 60 by the output unit 14, for example.

[0026] The output unit 14 is a functional unit that outputs the estimated fatigue level. The display unit 60 is a functional unit that displays the accumulated fatigue level as an estimated result to a mobile terminal (e.g., a PC or smartphone) owned by the user 99.

[0027] By doing so, it becomes possible to quantify the fatigue level of a new user 99 from 3 minutes of electrocardiogram data. As already mentioned, in order to directly quantify the amount of human herpesvirus, it is necessary to collect saliva and analyze the collected saliva using the PCR method or the like. Collecting saliva requires oral rinsing before collection, collecting saliva by holding cotton in the mouth, and extracting saliva from the cotton (centrifugation, etc.), and these operations alone take more than 3 minutes.

[0028] Furthermore, collecting saliva would impose a time and mental burden on the user 99, as they would have to undergo an operation. However, with a wearable electrocardiograph, the user 99 can wear it on their body and perform other tasks as usual. Thus, the measurement of physiological volume by the physiological volume measurement unit 20 in this embodiment is a passive measurement, eliminating the need to collect saliva and other procedures. Since the measurement is performed automatically, it reduces the time and mental burden and increases the likelihood of continuity, making it effective.

[0029] Furthermore, the PCR method typically takes several hours, thus lacking the immediacy of fatigue estimation. On the other hand, fatigue estimation using the fatigue estimation model 12 allows for near real-time assessment of fatigue levels, provided physiological data can be measured. Therefore, for example, a worker can adjust their daily work schedule according to their fatigue level, making the estimated fatigue level useful in practical applications.

[0030] The physiological quantity measurement unit 20 uses a wearable electrocardiograph as an example, but it does not have to be a wearable electrocardiograph. For example, it could be an electrocardiograph using a non-contact sensor such as millimeter waves, or a camera that measures the R-R interval from minute color changes caused by heartbeat in a facial image. Also, the indicators extracted from the wearable electrocardiograph do not have to be LF / HF; for example, RMSSD or simply heart rate could be used. In other words, there are no particular limitations on the biological information extracted by the indicator extraction unit 30.

[0031] Furthermore, the physiological quantity measurement unit 20 does not have to be an electrocardiograph, and the type of physiological quantity measured is not particularly limited as long as it can be an explanatory variable for the fatigue estimation model 12. For example, the physiological quantity measurement unit 20 may be a camera that takes facial images. From the facial image taken by the camera acting as the physiological quantity measurement unit 20, the index extraction unit 30 can identify the position of facial feature points, and for example, the number of blinks in 3 minutes may be extracted as an index. In this case, the "explanatory variable" of the dataset pre-trained in the construction of the fatigue estimation model 12 is changed to "number of blinks". That is, without changing the fact that subject A's saliva is collected every morning and the amount of human herpesvirus contained in the collected saliva per unit volume is quantified by PCR or the like, facial images are acquired before and after the time of saliva collection instead of heart rate data. As with the electrocardiograph, a longer acquisition time is better from the viewpoint of accuracy, but if it is too long the computational load increases, so the length of the acquisition time should be determined according to the processing resources that can be used for computation. Here, for example, 3 minutes of heart rate data will be used. By calculating the number of blinks from facial images measured over a three-minute period, one data point for "number of blinks" can be obtained.

[0032] By collecting this data daily, a dataset can be obtained for each day the experiment was conducted. Data from multiple subjects other than subject A may also be collected. This results in a dataset containing pairs of "true fatigue" ("amount of human herpesvirus") and "explanatory variable" ("number of blinks"). Subsequent processing is as previously described.

[0033] The fatigue level of a new user 99 is estimated by inputting the number of blinks extracted from the three minutes of facial images by the physiological measurement unit 20 into the fatigue estimation model 12 of the fatigue estimation unit 10.

[0034] Furthermore, the indicator extracted from the facial image does not have to be "number of blinks"; for example, it could be the angle of the corners of the mouth, the opening of the mouth, skin color, the angle of the eyebrows, etc., and the indicator is not limited here. Also, the physiological quantity measurement unit 20 does not have to be a camera; for example, it could be a skin potential meter. Skin potential is known to decrease due to sweating caused by the effects of tension, etc. Similar to the case of an electrocardiograph or camera, the "explanatory variable" skin potential is obtained by averaging the skin potential values ​​for, for example, 3 minutes before and after saliva collection, and a dataset is created that includes pairs of "human herpesvirus amount," which is the "true value of fatigue," and "skin potential," which is the "explanatory variable." The number of explanatory variables described here is one, but multiple explanatory variables (for example, "LF / HF" and "number of blinks") may be used, and the number of explanatory variables is not limited.

[0035] Several components not mentioned in the above explanation will be explained in detail in the modified examples described later.

[0036] [Operation of the Fatigue Estimation Device] The fatigue estimation device 100 described above operates as shown in Figure 3 below. Figure 3 is a flowchart of the fatigue estimation method according to the embodiment. As shown in Figure 3, the fatigue estimation device 100 first constructs a fatigue estimation model 12 by training a machine learning model using a pre-prepared dataset (the dataset already described) (S101). The construction of the fatigue estimation model 12 may be performed only once, or it may be reconstructed by retraining each time a new dataset is obtained.

[0037] Next, the physiological volume measurement unit 20 measures the physiological volume of the user 99 (S102). Then, the index extraction unit 30 extracts biological information indices from the measured physiological volume to be used for estimating the fatigue level. The fatigue estimation unit 10 uses the biological information extracted from this physiological volume to estimate the fatigue level of the user 99 using the constructed fatigue estimation model 12 (S103). After that, the output unit 14 outputs the estimated fatigue level to the presentation unit 60, and the fatigue level is presented (S104).

[0038] [Modification 1] In the modification described below, an example is described in which a wake-up detection unit 11 is used in addition to the above configuration. Referring again to Figure 2, the wake-up detection unit 11 detects the wake-up timing, which is the timing when the sleeping user 99 wakes up, based on the indicators extracted by the indicator extraction unit 30. In order to detect the wake-up timing, the physiological volume measurement unit 20 is configured to include a wearable accelerometer in addition to a wearable electrocardiograph. For example, a wristwatch-type wearable electrocardiograph, a wearable electrocardiograph that is attached directly to the user 99's solar plexus with a belt or gel pad, or an accelerometer attached to a ring-shaped wearable electrocardiograph may be used, or an accelerometer independent of the electrocardiograph may be used.

[0039] For example, in a wristwatch-type wearable accelerometer, acceleration in each of the three axes corresponding to each direction in three dimensions is measured, mainly when the arm moves. The time-series signals of the measured acceleration in each of the three axes are input to the index extraction unit 30, and for example, body movement quantity is extracted as an index. Body movement quantity is calculated by squaring the values ​​of each of the three axes of acceleration and taking the square root, and it represents the amount of body movement (scalar quantity) that is independent of direction. If the wearable accelerometer is attached to the arm, for example, during sleep, no body movement occurs during sleep except for short movements such as turning over in bed. However, when the person wakes up and moves from the bedding, body movement of a predetermined duration or longer will be detected, and the time when body movement of a predetermined duration or longer begins to be detected can be detected as the wake-up timing. In this way, the body movement data obtained by the index extraction unit 30 is input to the wake-up detection unit 11. The wake-up detection unit 11 can detect the time when body movement of a predetermined duration or longer than a predetermined amount is detected as the wake-up timing.

[0040] The fatigue estimation model 12 may, in addition to acceleration, include a wearable electrocardiograph as described above as the physiological quantity measurement unit 20, extract LF / HF in the index extraction unit 30, estimate the fatigue level using the pre-trained fatigue estimation model 12, and present it on the presentation unit 60, or, as described above, the physiological quantity measurement unit 20 may be equipped as a wearable accelerometer and estimate from the body movement volume extracted by the index extraction unit 30. In that case, the fatigue estimation model 12 needs to be pre-trained using the amount of human herpesvirus as the "true value of fatigue" in the dataset and body movement volume as the "explanatory variable" in the dataset.

[0041] For example, a sample of subject A's saliva is collected every morning, and the amount of human herpesvirus contained in each unit volume of the collected saliva is quantified using a method such as PCR. Meanwhile, body movement data is acquired before and after the time the saliva is collected. For example, one data point of total body movement over eight hours can be obtained from the acceleration measured in the eight hours prior to saliva collection.

[0042] By doing this daily, you can obtain a dataset for each day the experiment was conducted. You can also obtain data from multiple subjects. This will result in a dataset containing pairs of "true fatigue values," which are "human herpesvirus levels," and "explanatory variables," such as "physical movement."

[0043] This dataset is used to train a machine learning model to construct a fatigue estimation model 12. Then, the physiological measurement unit 20 measures acceleration information for a new user 99 over an 8-hour period, and the body movement amount extracted from the 8-hour acceleration information by the index extraction unit 30 is input into the fatigue estimation model 12 of the fatigue estimation unit 10, thereby estimating the fatigue level of the user 99. The subsequent steps are the same as in the embodiment described above.

[0044] The wake-up detection unit 11 obtains the wake-up timing of the user 99, and further estimates the fatigue level from the indices (such as body movement amount or LF / HF) obtained by the physiological quantity measurement unit 20 and extracted by the index extraction unit 30 before waking up (during sleep). By doing so, it is possible to estimate the fatigue level of the user 99 using the physiological quantity in the time period before the wake-up timing, which is a timing when the user 99 is surely not exercising (using biological information), and it is possible to reduce the influence of the increase in the amount of human herpesvirus immediately after exercise. Therefore, it becomes possible to estimate the fatigue level of the user 99 more accurately.

[0045] In addition, in order to eliminate the influence of the increase in the amount of human herpesvirus due to the influence of exercise, since a dataset at a timing not affected by exercise is used for learning during learning, it is easy to suppress the use of physiological quantities at timings such as after exercise that have not been learned beforehand during fatigue estimation.

[0046] As an example of the operation of the wake-up detection unit 11, an example was described in which the body movement amount extracted by the index extraction unit 30 using the physiological quantity obtained by the wearable acceleration sensor of the physiological quantity measurement unit 20 was used as an index, but the method is not limited, and other methods may be used. For example, as the wake-up detection unit 11, data from a pyroelectric sensor attached to the ceiling of the bedroom may be used. The pyroelectric sensor reacts to the movement of an isothermal object such as a human, and the output of the pyroelectric sensor may be input to the index extraction unit 30 using the pyroelectric sensor as the physiological quantity measurement unit 20, and the output of the pyroelectric sensor may be input to the wake-up detection unit 11 as an index.

[0047] The wake-up timing can be detected when an output above a predetermined value is output as an indicator, and it can be distinguished from movements such as turning over in sleep based on the signal strength. As another example, the wake-up detection unit 11 may use data from a radio wave sensor installed in the bedroom. The radio wave sensor can obtain the reflected signal of the radio waves it emits and analyze the movement of the reflected object. In other words, by frequency-separating the reflected signal of the sleeping user 99, both the user 99's heart rate signal and presence or absence can be detected. The reflected signal input from the radio wave sensor as the physiological quantity measurement unit 20 to the indicator extraction unit 30 is passed through a bandpass filter of about 0.5 to 2.0 Hz corresponding to the heart rate to obtain a heart rate signal, and further frequency analysis allows the LF / HF value to be extracted as an indicator. Furthermore, low-frequency components below 0.5 Hz can be extracted, and the presence or absence of the user 99 in the bedroom can be detected using these low-frequency components as an indicator. When the user is absent, signals such as breathing, which are low-frequency components, will be lost, so the proportion of low-frequency components will decrease, allowing it to be determined that the user is absent. If a radio wave sensor is used as the physiological quantity measurement unit 20, it is possible to realize a configuration that includes a wake-up detection unit 11 with only a single sensor. The data to be pre-trained in the fatigue estimation model 12 is the "true value of fatigue," which is the "amount of human herpesvirus," and the "explanatory variable," which is "LF / HF." As mentioned above, the physiological quantity measurement unit 20 does not need to be a single sensor; the biological information provided to the wake-up detection unit 11 and the biological information input to the fatigue estimation model 12 may originate from the measurement results of separate physiological quantity measurement units 20. When the index input to the fatigue estimation model 12 is "LF / HF," any biological information from before the wake-up time is acceptable, but it is particularly preferable that it is closer to the wake-up time.

[0048] [Modification Example 2] Hereinafter, an operation example in the case where the user 99 removes the wearable sensor at bedtime will be described. For example, assume that the physiological quantity measurement unit 20 is a wristwatch-type smartwatch, and the pulse wave measured by the smartwatch is input to the index extraction unit 30. The pulse wave is a vascular conduction wave caused by the change in the internal pressure of the artery generated when the heart pumps out blood, and is mainly measured using a photodetector that irradiates light from a green LED attached to a smartwatch or the like and detects the reflected light as a time waveform. That is, the fluctuation of the blood flow volume in the blood vessel is detected as the fluctuation of the reflection intensity of the green light emitted from the green LED. The peak-to-peak interval (P-P interval) of this pulse wave can be used as an index similar to the R-R interval of an electrocardiograph.

[0049] In the index extraction unit 30, LF / HF is similarly extracted from the time-series waveform of the P-P interval extracted from the time waveform of the reflection intensity. The extracted LF / HF is input to the fatigue estimation model 12 of the fatigue estimation unit 10, and the estimated fatigue level is output. Here, if the physiological amount measurement unit 20 is a smartwatch, many users 99 remove the smartwatch to charge it while sleeping. Alternatively, even if they do not charge it, some users remove the smartwatch before going to sleep because they dislike the feeling of being restricted while sleeping. In these cases, when the wake-up detection unit 11 detects the wake-up timing, the LF / HF during sleep that is suitable for estimating the user's fatigue level has not been extracted before the wake-up timing, so the fatigue level cannot be estimated. However, immediately after the wake-up timing, it means that no exercise has been performed immediately before, so if LF / HF is extracted from the pulse wave measured by the physiological amount measurement unit 20 as soon as possible after waking up and the fatigue level is estimated by the fatigue estimation model 12, it can be said that the effect of exercise is less likely to occur. Therefore, the wake-up detection unit 11 detects the timing of waking up, and instead of using data from before the waking time, it is sufficient to use data immediately after waking up as data for estimating the fatigue level (for input to the fatigue estimation model 12). This makes it possible to estimate the fatigue level with high accuracy even when the physiological volume measurement unit 20 is a removable wearable sensor and the user 99 is not wearing it while sleeping. However, since it is necessary to use physiological volume measured after such a removable wearable sensor is attached, it is desirable to use physiological volume measured as soon as possible after the waking time and immediately after the wearable sensor detects that the sensor is attached.

[0050] Here, a smartwatch is shown as an example of the physiological quantity measurement unit 20, but it is not limited to any sensor that the user 99 may remove during sleep. For example, smart glasses, a smart ring, or even a sensor attached to the body like a pedometer (registered trademark) could serve as the physiological quantity measurement unit 20.

[0051] [Modification 3] Modifications 1 and 2 described above describe a fatigue estimation device 100 in which physiological quantities outside the activity period are used to estimate the degree of fatigue. In this modification 3, a fatigue estimation device is described that takes into account the effect of using physiological quantities during the activity period to estimate the degree of fatigue.

[0052] As shown in Figure 2, the fatigue estimation device 100 includes an activity level extraction unit 40 and a correction unit 13 of the fatigue estimation unit 10. The activity level extraction unit 40 is a functional unit that extracts the activity level of the user 99 from the measurement results from the physiological amount measurement unit 20. The correction unit 13 corrects the fatigue level output from the fatigue estimation model 12 based on the activity level extracted by the activity level extraction unit 40. As explained using Figure 1, there is a difference in the estimated fatigue level (i.e., the amount of herpes virus) depending on the magnitude of the activity level and the duration of the activity, depending on whether the person is exercising or not. Conversely, if the magnitude of the activity level and the duration of the activity are known, it is possible to calculate the fatigue level when not exercising from the fatigue level when exercising.

[0053] Therefore, in this modified example, the physiological quantity measurement unit 20 includes a sensor capable of measuring activity levels. Examples of sensors capable of measuring activity levels include smartwatches equipped with accelerometers, wearable sensors such as pedometers, and cameras. Here, as an example, we will describe an example in which a smartwatch equipped with an accelerometer is provided as the physiological quantity measurement unit 20. The physiological quantities measured by the accelerometer of the smartwatch included in the physiological quantity measurement unit 20 are input to the activity level extraction unit 40 and extracted as time-series data of activity levels. This makes it possible to know how intensely and for how long the user 99 was active before the fatigue level is estimated by the fatigue estimation model 12.

[0054] On the other hand, the fatigue level estimated by the fatigue estimation model 12 is calculated from the LF / HF ratio extracted by the index extraction unit 30 from the electrocardiogram information measured by the smartwatch. In constructing the fatigue estimation model 12, the data used for training does not include data from periods when the amount of herpes virus is transiently increased, such as during and immediately after exercise. Therefore, the fatigue level estimated using the fatigue estimation device 100 during or immediately after exercise is considered to be incorrect. For example, as shown in Figure 1, the amount of herpes virus increases after the start of exercise, so the LF / HF value of the fatigue level estimated by the fatigue estimation model 12 becomes high, resulting in a higher fatigue level output than the actual value. In other words, the correction unit 13 reduces and corrects the deviation due to the increased fatigue level.

[0055] The correction unit 13 has previously measured the amount of discrepancy in fatigue level according to the activity level and duration of activity, and has stored a correction table containing the direction and amount of the discrepancy to cancel out the discrepancy. Figure 4 is a diagram showing a correction table according to a modified embodiment.

[0056] For example, the correction unit 13 holds the correction table shown in Figure 4 and corrects the fatigue level estimated by the fatigue estimation model 12 based on the activity level before the fatigue level is estimated. For example, if light exercise such as walking is performed for 15 minutes, the activity level is small (about 1.5 METs), and if the fatigue level is estimated 15 minutes after the end of the exercise, the fatigue level estimated by the fatigue estimation model 12 should be reduced by 6%.

[0057] Furthermore, if strenuous exercise of 5 METs or more (such as running or swimming) is performed for 20 minutes or more, and the fatigue level is estimated 35 minutes later, it is sufficient to reduce the fatigue level estimated by fatigue estimation model 12 by 15%.

[0058] This correction table can be obtained, for example, by measuring the fatigue level estimated by the fatigue estimation model 12 immediately before exercise in a state of not exercising for a long period of time, and then estimating the fatigue level in the same way using the fatigue estimation model 12 during and after exercise. In this embodiment, the circadian fatigue level that we aim to estimate is not affected by short-term exercise, so such a correction amount can be calculated from the difference between the fatigue level estimated during and after exercise and the fatigue level before exercise, and a correction table can be constructed.

[0059] Thus, in this modified example 3, the degree of fatigue can be accurately estimated even after exercise by applying a correction.

[0060] Here, a smartwatch with a built-in accelerometer was used as an example of the physiological quantity measurement unit 20, but it does not have to be a smartwatch. For example, if a camera is used as the physiological quantity measurement unit 20, the activity level can be extracted from the amount of movement of a person in the frame. For example, an activity level can be extracted by using a camera that can capture the entire soccer field with a wide angle and tracking the position of a person within that field to quantify the amount they ran. Alternatively, the activity level can be extracted by capturing the body of a person during fitness activities and analyzing the amount of movement of their arms and legs. In addition, any sensor capable of measuring activity levels can be used in the physiological quantity measurement unit 20, as long as it can produce measurement results that allow for the extraction of the user's 99 activity level.

[0061] In this explanation, the fatigue estimation model 12 and the correction unit 13 within the fatigue estimation unit 10 are described as separate components; however, the functions of the correction unit 13 may be included in the fatigue estimation model 12. While the correction unit 13 is described as correcting the fatigue level from the fatigue estimation model using a correction table, other methods are also acceptable; for example, a continuous correction amount such as a correction function can be output. Furthermore, a pre-trained machine learning model constructed to output correction amounts may be used.

[0062] [Modification 4] In Modification 3 described above, a configuration was explained in which the correction amount is automatically determined and corrected. However, the fatigue level output from the fatigue estimation model 12 and the correction amount have different origins and calculation methods, so there is a need to know them separately. Therefore, in this Modification 4, an example is described in which the fatigue level output from the fatigue estimation model 12 is output as is, and separately an estimation accuracy that has the same meaning as the correction amount is calculated and presented.

[0063] The activity level output from the activity level extraction unit 40 is input to the accuracy calculation unit 50. The accuracy calculation unit 50 is a functional unit that calculates the estimation accuracy of the estimated fatigue level based on the activity level and activity duration. As described above, the fatigue level calculated by the fatigue estimation model 12 is output to the presentation unit 60 without correction.

[0064] In this example, we describe a case in which a smartwatch equipped with an acceleration sensor is used as the physiological volume measurement unit 20. The physiological volume measured by the smartwatch in the physiological volume measurement unit 20 is input to the activity level extraction unit, and the user's activity level is extracted as time-series data. In other words, it is possible to know how intensely and for how long the user 99 was active before the fatigue level is estimated by the fatigue estimation model 12.

[0065] On the other hand, the fatigue level output from the fatigue estimation model 12 is calculated from the LF / HF ratio extracted by the index extraction unit 30 from the electrocardiogram information measured by the smartwatch. The data used for training when constructing the fatigue estimation model 12 does not include data from periods when the amount of herpes virus is transiently increased, such as during and immediately after exercise. Therefore, the fatigue level estimated using the fatigue estimation device 100 during or immediately after exercise is considered to be inaccurate. The error is thought to be highest immediately after exercise.

[0066] This error is calculated by the accuracy calculation unit 50 from the time-series data of activity levels transmitted from the activity level extraction unit 40. The accuracy calculation unit 50, for example, measures the amount of the discrepancy in fatigue levels corresponding to the activity level and activity duration in advance, and then stores the data of the direction and amount of the discrepancy as an accuracy table. Figure 5 shows an accuracy table according to a modified example of the embodiment.

[0067] For example, the accuracy calculation unit 50 maintains the accuracy table shown in Figure 5, calculates the accuracy based on the activity level before estimating fatigue, outputs the accuracy to the presentation unit 60, and presents it together with the fatigue level estimated by the fatigue estimation model 12. For example, if light exercise such as walking is performed for 15 minutes, the activity level is small (about 1.5 METs), and in that case, if the fatigue level is estimated 15 minutes after the end of the exercise, it can be seen that there is an error of about 6% compared to the fatigue level estimated by the fatigue estimation model 12.

[0068] Furthermore, it was found that when strenuous exercise (such as running or swimming) of 5 METs or more is performed for 20 minutes or more, and the fatigue level is estimated 35 minutes later, there is an error of about 15% compared to the fatigue level estimated by fatigue estimation model 12.

[0069] This accuracy table can be obtained, for example, by measuring the fatigue level estimated by the fatigue estimation model 12 immediately before exercise, assuming a state of not exercising for a long period of time, and then estimating the fatigue level in the same way using the fatigue estimation model 12 during and after exercise. In this embodiment, the circadian fatigue level that we aim to estimate is not affected by short-term exercise, so the accuracy table can be constructed by taking the difference between the fatigue level estimated during and after exercise and the fatigue level before exercise.

[0070] Thus, in this modified example 4, by presenting the accuracy after exercise along with the estimated fatigue level, it becomes possible to inform the user 99 of the degree of estimation error included in their fatigue level, for example, by prompting them to remeasure at a time when the estimation error is small, or by warning them to be careful when comparing cases with low and high estimation errors.

[0071] Although the accuracy calculation unit 50 has been described as calculating the accuracy for fatigue level using an accuracy table, it is not limited to an accuracy table; for example, it can also output a continuous estimation accuracy such as an error function. Alternatively, a pre-trained machine learning model constructed to output estimation accuracy may be used.

[0072] [Modification 5] In Modification 4 described above, an example of calculating and presenting the estimation accuracy was explained, but other uses for the estimation accuracy are also possible. For example, in this Modification 5, the estimation accuracy is used for purposes other than presenting it. This Modification 5 differs from Modification 4 in that the activity amount and activity duration output from the activity amount extraction unit 40 are input to the accuracy calculation unit 50, and the output from the accuracy calculation unit 50 is input to the fatigue estimation unit 10.

[0073] In this modified example 5, we describe an example in which a smartwatch equipped with an acceleration sensor is used as the physiological quantity measurement unit 20. The physiological quantities of the user 99 measured by the smartwatch in the physiological quantity measurement unit 20 are input to the activity level extraction unit 40, and the user 99's activity level is extracted as time-series data. In other words, it is possible to know how intensely and for how long the user 99 was active before the fatigue level is estimated by the fatigue estimation model 12.

[0074] On the other hand, the fatigue level estimated by the fatigue estimation model 12 is calculated from the LF / HF ratio extracted by the index extraction unit 30 from the electrocardiogram information measured by the smartwatch. In constructing the fatigue estimation model 12, the data used for training does not include data from periods when the amount of herpes virus is transiently increased, such as during and immediately after exercise. Therefore, the fatigue level estimated using the fatigue estimation device 100 during or immediately after exercise is considered to be inaccurate. The error is thought to be highest immediately after exercise.

[0075] This error is calculated by the accuracy calculation unit 50 from the time-series data of activity levels transmitted from the activity level extraction unit 40. The accuracy calculation unit 50, for example, measures the amount of discrepancy in fatigue levels corresponding to the activity level and activity duration in advance, and then stores the data of the direction and amount of the discrepancy as an accuracy table as shown in Figure 5.

[0076] The calculated estimation accuracy is input to the fatigue estimation unit 10. If the accuracy meets a predetermined standard, the fatigue level is calculated; if it does not meet the predetermined standard, the fatigue level is not calculated.

[0077] For example, if a criterion is set such that calculations are not performed if the error is 15% or more, then if the preceding exercise was moderate (around 3-5 METs) and continued for 20 minutes or more, fatigue levels would not be calculated from 10 minutes before the exercise and for 20 minutes after the exercise.

[0078] Furthermore, even when calculating fatigue levels, if the predetermined criteria are not met, the output unit 14 may be configured not to output the estimation result, thereby preventing the display unit 60 from displaying the fatigue level. In this case, for example, the display unit 60 may display a reason why the estimation accuracy is not shown, such as "The fatigue level cannot be calculated (displayed) due to a large estimation error."

[0079] This approach prevents discrepancies in fatigue levels caused by informing users of large errors in estimated fatigue levels, by not displaying the fatigue level if the error is significant. Furthermore, it eliminates the need for calculations required for fatigue estimation, resulting in faster response times and improved usability.

[0080] Although the error threshold for the specified standard was explained as 15% above, this is merely an example, and it may be permitted for user 99 to set it as appropriate.

[0081] [Effects, etc.] As described above, the fatigue estimation device 100 according to the first embodiment of this implementation comprises a fatigue estimation model 12 that is pre-constructed based on data including the relationship between a person's physiological amount outside of activity and the amount of human herpesvirus outside of said activity, and which can output a fatigue level estimated from the corresponding amount of human herpesvirus in response to the input of biological information; a physiological amount measurement unit 20 that measures the physiological amount of the subject (user 99); and a fatigue estimation unit 10 that outputs the fatigue level of the subject using the fatigue estimation model 12 based on the measured physiological amount.

[0082] According to this fatigue estimation device 100, a fatigue estimation model 12 based on data including the relationship between human herpesvirus load outside the activity period and physiological load can output the subject's fatigue level relative to the measured physiological load. Since the influence of transient increases in fatigue level during the activity period is suppressed, it becomes possible to more appropriately quantify fatigue level based on human herpesvirus load for medium- to long-term accumulated fatigue. Therefore, a fatigue level that can be presented more appropriately can be estimated.

[0083] Furthermore, the fatigue estimation device 100 according to the second embodiment of this implementation is the fatigue estimation device 100 described in the first embodiment, and further includes a biological information extraction unit (indicator extraction unit 30) that extracts biological information from measured physiological quantities, and the fatigue estimation unit 10 outputs the degree of fatigue of the subject using a fatigue estimation model 12 based on the biological information extracted from the measured physiological quantities.

[0084] According to this, the biological information extracted from the measured physiological quantities can be used to output the fatigue level of the subject by the fatigue estimation model 12.

[0085] Furthermore, the fatigue estimation device 100 according to the third embodiment of this implementation is the fatigue estimation device 100 described in the first or second embodiment, and further includes a display unit 60 that displays the fatigue level of the subject output by the fatigue estimation unit 10.

[0086] This allows us to display the fatigue level of the subject being analyzed.

[0087] Furthermore, the fatigue estimation device 100 according to the fourth embodiment of this implementation is the fatigue estimation device 100 described in any one of the first to third embodiments, wherein the measured physiological quantities include the subject's activity level and activity duration, and the fatigue estimation unit 10 estimates the subject's fatigue level by further correcting using the subject's activity level and activity duration.

[0088] This allows for the output of a more accurate assessment of the subject's fatigue level, corrected using the subject's activity level and duration.

[0089] Furthermore, the fatigue estimation device 100 according to the fifth embodiment of this embodiment is the fatigue estimation device 100 described in any one of the first to fourth embodiments, and further includes a wake-up detection unit 11 that detects the timing of the subject's waking up, the physiological amount measurement unit 20 measures the physiological amount of the subject before the detected waking up time, and the fatigue estimation unit 10 outputs the degree of fatigue of the subject using a fatigue estimation model 12 based on the physiological amount of the measured physiological amount before the waking up time.

[0090] According to this method, the degree of fatigue in a subject can be estimated by measuring physiological data at a time when the subject is likely to be outside of their active period, prior to when they wake up.

[0091] Furthermore, the fatigue estimation device 100 according to the sixth embodiment of this implementation is the fatigue estimation device 100 described in any one of the first to fifth embodiments, wherein the physiological volume measurement unit 20 is a sensor worn and used by the subject, and measures the physiological volume of the subject immediately after wearing it.

[0092] According to this method, the fatigue level of a subject can be estimated based on physiological data measured at appropriate times when sensors are attached.

[0093] Furthermore, the fatigue estimation device 100 according to the seventh embodiment of this implementation is the fatigue estimation device 100 described in any one of the first to sixth embodiments, wherein the measured physiological quantities include the subject's activity level and activity duration, and the fatigue estimation device 100 further includes an accuracy calculation unit 50 that uses the subject's activity level and activity duration to calculate the estimation accuracy for the subject's fatigue level estimated by the fatigue estimation unit 10.

[0094] According to this, the fatigue estimation unit 10 can calculate the estimation accuracy of the subject's fatigue level, using the subject's activity level and activity duration.

[0095] Furthermore, the fatigue estimation device 100 according to the eighth aspect of this embodiment is the fatigue estimation device 100 described in the seventh aspect, and further includes a presentation unit 60 that presents the fatigue level of the subject output by the fatigue estimation unit 10 and the estimation accuracy of the calculated fatigue level of the subject.

[0096] This allows us to present the calculated estimation accuracy.

[0097] Furthermore, the fatigue estimation device 100 according to the ninth aspect of this embodiment is the fatigue estimation device 100 described in the seventh aspect, and further includes a presentation unit 60 that presents the fatigue level of a subject output by the fatigue estimation unit 10, wherein the presentation unit 60 does not present the fatigue level of the subject if the estimation accuracy of the calculated fatigue level of the subject does not meet a predetermined standard.

[0098] According to this, if the calculated estimation accuracy does not meet a predetermined standard, the fatigue level of the subject can be omitted.

[0099] Furthermore, the fatigue estimation device 100 according to the tenth embodiment of this implementation is the fatigue estimation device 100 described in the seventh embodiment, wherein the fatigue estimation unit 10 does not estimate the fatigue level of the subject if the estimation accuracy of the calculated fatigue level of the subject does not meet a predetermined standard.

[0100] According to this, if the calculated estimation accuracy does not meet a predetermined standard, the fatigue level of the subject can be avoided from being estimated.

[0101] Furthermore, the fatigue estimation method according to the 11th embodiment of this implementation is a fatigue estimation method performed by a computer, and includes the steps of: constructing a fatigue estimation model 12 based on data including the relationship between physiological amounts outside of a person's activity period and the amount of human herpesvirus outside of said activity period, which can output a fatigue level estimated from the corresponding amount of human herpesvirus in response to the input of biological information (S101); having a physiological amount measurement unit 20 measure the physiological amounts of a subject (S102); and using the constructed fatigue estimation model 12 based on the measured physiological amounts to output the fatigue level of the subject (S103).

[0102] According to this, the same effects as the fatigue estimation device 100 described above can be achieved.

[0103] Furthermore, the program relating to the twelfth aspect of this implementation is a program for causing a computer to execute the fatigue estimation method described in the eleventh aspect.

[0104] According to this, it is possible to achieve the same effect as the fatigue estimation device 100 described above using a computer.

[0105] (Other Embodiments) The fatigue estimation apparatus, fatigue estimation method, and program relating to this disclosure have been described above based on the embodiments described above, but this disclosure is not limited to the embodiments described above. For example, forms that can be obtained by applying various modifications to each embodiment that a person skilled in the art can conceive of, and forms that can be realized by arbitrarily combining the components and functions of each embodiment without departing from the spirit of this disclosure are also included in this disclosure.

[0106] Furthermore, for example, this disclosure can be implemented not only as a fatigue estimation device, but also as a program that includes the processing performed by each component of the fatigue estimation device as steps, and as a computer-readable recording medium on which the program is recorded. The program may be pre-recorded on the recording medium, or it may be supplied to the recording medium via a wide-area communication network, including the Internet.

[0107] In other words, the comprehensive or specific embodiments described above may be implemented in a system, device, integrated circuit, computer program, or computer-readable recording medium, or in any combination of a system, device, integrated circuit, computer program, and recording medium.

[0108] 10 Fatigue estimation unit 11 Wake-up detection unit 12 Fatigue estimation model 13 Correction unit 14 Output unit 20 Physiological volume measurement unit 30 Indicator extraction unit 40 Activity level extraction unit 50 Accuracy calculation unit 60 Presentation unit 99 User 100 Fatigue estimation device

Claims

1. A fatigue estimation device comprising: a fatigue estimation model pre-constructed based on data including the relationship between a person's physiological load outside of their activity period and the amount of human herpesvirus outside of that activity period, which can output a fatigue level estimated from the corresponding amount of human herpesvirus in response to the input of biological information; a physiological load measurement unit for measuring the physiological load of a subject; and a fatigue estimation unit that outputs the fatigue level of the subject using the fatigue estimation model based on the measured physiological load.

2. The fatigue estimation device according to claim 1, further comprising a biological information extraction unit that extracts biological information from measured physiological quantities, wherein the fatigue estimation unit outputs the degree of fatigue of the subject using the fatigue estimation model based on the biological information extracted from the measured physiological quantities.

3. The fatigue estimation device according to claim 1, further comprising a display unit that displays the fatigue level of the subject output by the fatigue estimation unit.

4. The fatigue estimation device according to claim 1, wherein the measured physiological quantities include the subject's activity level and activity duration, and the fatigue estimation unit estimates the subject's fatigue level by further correcting the subject's activity level and activity duration.

5. The fatigue estimation device according to claim 1, further comprising a wake-up detection unit for detecting the wake-up timing of the subject, the physiological amount measurement unit for measuring physiological amounts prior to the detected wake-up timing of the subject, and the fatigue estimation unit for outputting the subject's fatigue level using the fatigue estimation model based on the physiological amounts prior to the wake-up timing from the measured physiological amounts.

6. The fatigue estimation device according to claim 1, wherein the physiological volume measurement unit is a sensor worn and used by the subject, and measures the physiological volume of the subject immediately after wearing the sensor.

7. The fatigue estimation device according to any one of claims 1 to 6, wherein the measured physiological quantities include the activity level and duration of activity of the subject, and the fatigue estimation device further comprises an accuracy calculation unit that calculates the estimation accuracy for the degree of fatigue of the subject estimated by the fatigue estimation unit using the activity level and duration of activity of the subject.

8. The fatigue estimation device according to claim 7, further comprising a display unit that displays the fatigue level of the subject output by the fatigue estimation unit and the calculated estimation accuracy for the fatigue level of the subject.

9. The fatigue estimation device according to claim 7, further comprising a display unit that displays the fatigue level of the subject output by the fatigue estimation unit, wherein the display unit does not display the fatigue level of the subject if the estimation accuracy of the calculated fatigue level of the subject does not meet a predetermined standard.

10. The fatigue estimation device according to claim 7, wherein the fatigue estimation unit does not estimate the fatigue level of the subject if the estimation accuracy of the calculated fatigue level of the subject does not meet a predetermined standard.

11. A fatigue estimation method performed by a computer, comprising the steps of: constructing a fatigue estimation model based on data including the relationship between a person's physiological load outside of an active period and the amount of human herpesvirus outside of said active period, the model capable of outputting a fatigue level estimated from the corresponding amount of human herpesvirus in response to the input of biological information; having a physiological load measurement unit measure the physiological load of a subject; and using the constructed fatigue estimation model based on the measured physiological load, outputting the fatigue level of the subject.

12. A program for causing the computer to execute the fatigue estimation method described in claim 11.

Images