JP7054820B2 - Drowsiness detection device, drowsiness detection method, and program recording medium - Google Patents

Description

The present disclosure relates to a drowsiness detection device for detecting drowsiness of a user, a drowsiness detection method, and a program recording medium.

Patent Document 1 discloses a drowsiness detection device.

International Publication No. 2016/1044498

The present disclosure provides a drowsiness detection device that performs accurate drowsiness detection by normalizing the value for drowsiness detection based on the drowsiness detection result.

The drowsiness detection device in the present disclosure includes a feature amount extraction unit, a detection rule storage unit, a normalization unit, and a drowsiness detection unit. The feature amount extraction unit extracts the feature amount related to the heartbeat based on the interval of the user's heartbeat. The detection rule storage unit holds a detection rule for detecting drowsiness. The normalization unit updates the normalization coefficient. The drowsiness detection unit detects the drowsiness of the user based on the feature amount, the detection rule, and the normalization coefficient. The normalization unit updates the normalization coefficient based on the feature amount when the drowsiness detection unit does not detect the user's drowsiness.

It should be noted that these general or specific embodiments may be realized in a recording medium such as a system, method, integrated circuit, computer program or computer readable CD-ROM, and the system, method, integrated circuit, computer program. And may be realized by any combination of recording media.

The drowsiness detection device in the present disclosure can perform accurate drowsiness detection by normalizing the value for drowsiness detection based on the drowsiness detection result. Furthermore, even in the same person, the feature amount related to the heartbeat changes over time. The present disclosure can accurately detect drowsiness even over time in the same person.

FIG. 1 is a block diagram of the drowsiness detection device 10 according to the first embodiment. FIG. 2 is a schematic diagram showing an example of an electrocardiographic waveform input to the RRI detection unit 11. FIG. 3 is a schematic diagram showing an example of a detection rule held by the detection rule storage unit 14. FIG. 4 is a flowchart showing the operation of the drowsiness detection device 10 according to the first embodiment. FIG. 5A is a diagram showing an example of drowsiness detection based on a driver's declared value in the first embodiment. FIG. 5B is a diagram showing an example of drowsiness detection when the feature amount is not normalized in the first embodiment. FIG. 5C is a diagram showing an example of drowsiness detection when the feature amount is normalized in the first embodiment. FIG. 6 is a configuration diagram of the drowsiness detection device 20 in the second modification.

Hereinafter, embodiments will be described in detail with reference to the drawings as appropriate. However, more detailed explanation than necessary may be omitted. For example, detailed explanations of already well-known matters and duplicate explanations for substantially the same configuration may be omitted. This is to avoid unnecessary redundancy of the following description and to facilitate the understanding of those skilled in the art.

It should be noted that the inventor intends to limit the subject matter described in the claims by those skilled in the art by providing the accompanying drawings and the following description in order to fully understand the present disclosure. not.

(Embodiment 1)
Hereinafter, the drowsiness detection device 10 of the first embodiment will be described with reference to FIGS. 1 to 5C.

[1. Constitution]
FIG. 1 is a configuration diagram showing a drowsiness detection device 10 according to the first embodiment. The drowsiness detection device 10 includes an RRI detection unit 11, a feature amount extraction unit 12, a feature amount normalization unit 13, a detection rule storage unit 14, a drowsiness detection unit 15, and an accuracy determination unit 16. The feature amount normalization unit 13 is an example of the normalization unit.

The RRI detection unit 11 detects the RR interval (RRI: RR Interval) based on the user's electrocardiographic waveform (electrocardiographic information) input from the outside. The RR interval is the time interval between two adjacent R waves in the electrocardiographic waveform. The feature amount extraction unit 12 extracts the feature amount related to the heart rate variability described later based on the RR interval detected by the RRI detection unit 11. The feature amount normalization unit 13 normalizes the feature amount extracted by the feature amount extraction unit 12 based on the normalization coefficient described later. The detection rule storage unit 14 holds a detection rule for detecting drowsiness from the normalized feature amount. The drowsiness detection unit 15 detects the presence or absence of drowsiness based on the feature amount normalized by the feature amount normalization unit 13 and the detection rule held in the detection rule storage unit 14. The accuracy determination unit 16 determines the accuracy of the presence or absence of drowsiness detected by the drowsiness detection unit 15.

The drowsiness detection device 10 has a processor, a volatile storage area (main storage device), and a non-volatile storage area (auxiliary storage device). The volatile storage area is used as a work area to be used when executing the control program. The non-volatile storage area stores control programs, contents, and the like. The volatile storage area is, for example, a RAM (Random Access Memory). The non-volatile storage area is, for example, a ROM (Read Only Memory), a flash memory, an HDD (Hard Disk Drive), or the like. All or part of the functions of the drowsiness detection device 10 are realized, for example, by the processor executing a control program.

Next, each component will be described in detail.

[1-1. RRI detection unit 11]
The RRI detection unit 11 receives the user's electrocardiographic waveform from an external device. Generally, in the electrocardiographic waveform, P wave, Q wave, R wave, S wave, T wave, and U wave appear in synchronization with the movement of the heart for each beat as shown in FIG. In FIG. 2, the horizontal axis represents time t and the vertical axis represents the electrocardiographic potential V. In the electrocardiographic waveform, the R wave has a large amplitude and changes rapidly per hour. Therefore, the R wave is used for heartbeat detection. The specific method for detecting the R wave is not particularly limited. For example, the R wave can be detected by detecting the time corresponding to the maximum amplitude in the section where the amplitude becomes equal to or more than a predetermined threshold value as the R wave time. The interval between the detected R wave time and the previously detected R wave time is called an RR interval (RRI: RR Interval). The RR detection unit 11 detects this RR interval.

[1-2. Feature extraction unit 12]
The feature amount extraction unit 12 extracts the feature amount in order to perform drowsiness detection based on the RR interval input from the RRI detection unit 11. As the feature amount, a feature amount related to heart rate variability can be used. The characteristic amount of heart rate variability can be calculated using a time series of RR intervals included in a predetermined time interval. There is no particular limitation on the method of setting the time interval. The length of the time interval can be set to, for example, 60 seconds. If the length of the time interval is short, the time tracking of the heart rate variability can be accelerated. On the other hand, if the length of the time interval is short, the RR interval may react sensitively to the error when an error occurs in the RR interval due to a cause such as body movement. Therefore, it is desirable to set the length of the time interval appropriately.

The feature amount related to the heart rate variability can be calculated by using the time series of the RR interval included in the predetermined time interval. The feature amounts related to such heart rate variability include the feature amount in the time domain and the feature amount in the frequency domain.

The following features can be used as the features related to heart rate variability in the time domain.

・ Mean RR interval: Mean value of RR interval in time interval ・ Mean heart rate: Mean value of heart rate in time interval ・ SDNN: Standard deviation of RR interval in time interval ・ RMSSD: Continuous in time interval Square root of the mean squared of the difference between adjacent RR intervals ・ NN50: Total number of consecutive adjacent RR intervals with a difference of more than 50 ms in a time interval ・ pNN50: Difference between consecutive adjacent RR intervals within a time interval Percentage over 50 ms

On the other hand, as the feature amount related to the heart rate variability in the frequency domain, the following feature amount can be used by performing spectral analysis on the RR interval in the time interval.

・ LF (low frequency): spectral power in the frequency band of 0.04 to 0.15 Hz ・ HF (high frequency): spectral power in the frequency band of 0.15 to 0.4 Hz ・ LF / HF: LF (low frequency) HF (radio frequency) spectral power ratio

The feature amount extracted by the feature amount extraction unit 12 is not limited to the above. In addition to the above features, another feature obtained from the electrocardiographic waveform may be added. Alternatively, another feature amount obtained from the electrocardiographic waveform may be replaced with the above feature amount. When the drowsiness detection device 10 uses a decision tree as shown in FIG. 3, the feature amount extraction unit 12 may extract a plurality of feature amounts.

The feature amount F extracted as described above is expressed by Equation 1.

Here, F _k represents a feature amount in the kth time interval. The time interval is 1 to N.

[1-3. Feature normalization unit 13]
The feature amount normalization unit 13 normalizes the feature amount F _k of the current time interval k extracted by the feature amount extraction unit 12. Here, the feature amount normalization unit 13 calculates the normalization coefficient C _k based on the feature amount in the time interval detected by the accuracy determination unit 16 described later as “no drowsiness” higher than the predetermined accuracy. Then, the feature amount normalization unit 13 normalizes the feature amount F _k using the normalization coefficient C _k .

Specifically, the time interval determined to be "no drowsiness" with high accuracy between the time before a predetermined time and the current time interval k is defined as p1 ... pL. Here, L is the number of sections determined to be "no drowsiness" with high accuracy between the time before a predetermined time and the current time section k. The feature amount normalization unit 13 can calculate the normalization coefficient by the mathematical formula 2.

That is, the feature amount normalization unit 13 uses the average value of the feature amounts in the time section determined to be “no drowsiness” with high accuracy between the time before the predetermined time and the current time interval k as the normalization coefficient C _k . Calculated as. The feature amount normalization unit 13 calculates the normalized feature amount FN _k by the mathematical formula 3 using the feature amount F _k and the calculated normalization coefficient C _k .

The formula for calculating the normalization coefficient C _k is not limited to the formula 2. The normalization coefficient C _k may be determined based on the feature amount of the section determined to be “no drowsiness” with high accuracy. For example, the normalization coefficient C _k is calculated by the weighted sum of the previously calculated normalization coefficient C _k-1 and the feature amount F _pL of the section determined to be “no drowsiness” with a high degree of probability as shown in Equation 4. It may be one.

Here, the coefficient r is a coefficient for calculating the weighted sum. The coefficient r is a value of 0 to 1, and indicates how much priority is given to the feature amount F _pL .

The method for setting the coefficient r is not particularly limited. A fixed value such as 0.8 may be set as the coefficient r. Alternatively, if the accuracy determined by the accuracy determination unit 16 described later is high, the coefficient r may be set large, and if the accuracy is low, the coefficient r may be set small. In this way, the coefficient r may be set so as to give priority to the latest feature amount with high accuracy.

In Equation 3, the normalized feature amount FN _k was calculated by dividing by the normalization coefficient C _k . However, as shown in Equation 5, the normalized feature amount FN _k may be calculated by subtracting with the normalization coefficient C _k .

That is, drowsiness may be detected by the ratio with the feature amount at the time of "no drowsiness", or drowsiness may be detected by the difference with the feature amount at the time of "no drowsiness".

[1-4. Detection rule storage unit 14]
The detection rule storage unit 14 stores the detection rule for determining "sleepy" or "not sleepy" by using the normalized feature amount FN _k calculated by the feature amount normalization unit 13. For example, when determining drowsiness by a decision tree, the detection rule storage unit 14 stores the detection rule by the decision tree as shown in FIG.

The decision tree shown in FIG. 3 is generated based on, for example, a learning algorithm such as ID3 (Iterative Dichotomiser 3). Each of the plurality of nodes N1 to N7 of the decision tree corresponds to the plurality of feature quantities F1 to F7 related to the user's heartbeat described above. Further, this decision tree has a plurality of threshold values T1 to T7 corresponding to a plurality of nodes N1 to N7. Specifically, the feature amount F1 and the threshold value T1 correspond to the node N1, and the feature amount F2 and the threshold value T2 correspond to the node N2. Features and thresholds correspond to other nodes as well.

For example, in the node N1, when the value of the feature amount F1 is smaller than the threshold value T1, the transition to the node N2 occurs. When the feature amount F2 is smaller than the threshold value T2 in the node N2, the transition to the node N4 occurs. If the feature amount F4 is smaller than the threshold value T4 in the node N4, the leaf node L1 is reached. Then, the drowsiness detection unit 15 determines that there is "sleepiness". On the other hand, in the node N4, if the feature amount F4 is equal to or higher than the threshold value T4, the leaf node L2 is reached. Then, the drowsiness detection unit 15 determines that there is no drowsiness. In this way, in each node, the feature amount corresponding to the node and the threshold value are compared. Each leaf node stores the accuracy of "no drowsiness" and "drowsiness". For example, in the leaf node L1, the accuracy of "with drowsiness" is 0.8, and in the leaf node L2, the accuracy of "without drowsiness" is 0.8. The detection rule storage unit 14 stores the above detection rules. The accuracy of "no drowsiness" and "with drowsiness" is determined by the accuracy determination unit 16 described later.

In the above explanation, the rule storage method using the decision tree has been described, but the present invention is not limited to this. For example, the detection rule may be realized by a method using a plurality of decision trees called a random forest. The number of multiple decision trees in a random forest is, for example, 100. When a random forest is used, the one with the larger sum of the numbers judged as "no drowsiness" or "drowsiness" in each decision tree is used as the judgment result of drowsiness.

[1-5. Drowsiness detection unit 15]
The drowsiness detection unit 15 determines "no drowsiness" or "drowsiness" based on the feature amount FN _k normalized by the feature amount normalization unit 13 and the detection rule stored in the detection rule storage unit 14. .. The drowsiness detection unit 15 detects the drowsiness of the user according to the detection rule stored in the detection rule storage unit 14.

[1-6. Accuracy determination unit 16]
When the drowsiness detection unit 15 detects "no drowsiness" or "drowsiness", the accuracy determination unit 16 determines with what degree of accuracy CM the drowsiness is detected.

An example in the case of using the decision tree stored in the detection rule storage unit 14 will be specifically described. When the decision tree is created by the ID3 algorithm or the like, the one with the most determination results (in this case, "no drowsiness" or "drowsiness") in the leaf node is used as the determination result. At this time, the accuracy CM can be calculated from the ratio of "no drowsiness" and "with drowsiness" in the leaf node. For example, the ratio of “without drowsiness” to “with drowsiness” in the leaf node L1 in FIG. 3 is 2: 8. In this case, the accuracy CM of "sleepy" in the leaf node L1 is set to 0.8. As described above, the accuracy determination unit 16 can determine the accuracy of the detection of "no drowsiness" or "with drowsiness" detected by the drowsiness detection unit 15. The accuracy as described above can be determined from the trained model generated in advance.

In the above description, the case where the drowsiness detection unit 15 detects drowsiness by the decision tree has been described, but the present invention is not limited to this. For example, when the drowsiness detection unit 15 detects drowsiness by a random forest, the accuracy determination unit 16 determines that the determination result is "no drowsiness" or "drowsiness" among the determination results of a plurality of decision trees (M pieces). The accuracy CM can be calculated using the number of determined decision trees (W). That is, the accuracy determination unit 16 can calculate the accuracy CM by the following formula.

CM = W / M

[2. motion]
Hereinafter, a more detailed operation of the first embodiment will be described with reference to the flowchart of FIG.

The RR detection unit 11 detects the RR interval based on the user's electrocardiographic waveform input from the outside (S41). The specific method for detecting the R wave is not particularly limited. For example, the R wave can be detected by detecting the time corresponding to the maximum amplitude in the section where the amplitude becomes equal to or more than a predetermined threshold value as the R wave time. The interval between the detected R wave time and the previously detected R wave time is called the RR interval. The RR detection unit 11 detects this RR interval.

The feature amount extraction unit 12 extracts the feature amount in order to perform drowsiness detection based on the RR interval input from the RRI detection unit 11 (S42). The feature amount extraction unit 12 extracts, as the feature amount, the average RR interval, the average heart rate, SDNN, RMSSD, NN50, pNN50, LF, HF, LF / HF, etc., which are the feature amounts related to heart rate variability. The feature amount is not limited to the above feature amount as long as it is calculated based on the heartbeat.

The feature amount normalization unit 13 normalizes the feature amount F _k extracted by the feature amount extraction unit 12 (S43). The feature amount normalization unit 13 updates the normalization coefficient _Ck based on the feature amount in the time interval detected by the accuracy determination unit 16 as “no drowsiness” higher than the predetermined accuracy. Then, the feature amount normalization unit 13 normalizes the feature amount F _k of the current time interval k based on the normalization coefficient C _k . Since the accuracy determination by the accuracy determination unit 16 has not been performed at the start of this operation, the feature amount normalization unit 13 should normalize the feature amount F _k with the normalization coefficient C _k as 1. Just do it. Alternatively, since it can be assumed that there is no drowsiness at the start, the normalization coefficient at the start may be set to F ₁ (feature amount at the start). Alternatively, a predetermined normalization coefficient may be used.

The drowsiness detection unit 15 is drowsy, that is, "no drowsiness" or "drowsiness" based on the feature amount FN _k normalized by the feature amount normalization unit 13 and the detection rule stored in the detection rule storage unit 14. Is determined (S44). When the determination result is "sleepy", the user can be notified of the determination result.

The accuracy determination unit 16 determines the accuracy when the drowsiness detection unit 15 determines “no drowsiness” or “drowsiness” (S45).

The feature amount normalization unit 13 confirms whether the drowsiness detection unit 15 detects “no drowsiness” in step S44 and whether the accuracy CM determined in step S45 is higher than the predetermined accuracy threshold value (S46).

If Yes (“no drowsiness” and the accuracy CM is higher than the predetermined accuracy threshold value) in step S46, the feature amount normalization unit 13 updates the normalization coefficient (S47). Then, the feature amount normalization unit 13 normalizes the feature amount based on the updated normalization coefficient.

The RRI detection unit 11 determines whether or not there is an input of a new electrocardiographic waveform, and if there is an input, the process returns to step S41. On the other hand, if there is no input, the process ends (S48).

If No (“sleepy” or the accuracy CM is equal to or less than a predetermined accuracy threshold value) in step S46, the process proceeds to step S48.

By the above operation, the drowsiness detection device 10 normalizes the feature amount for drowsiness detection based on the drowsiness detection result. As a result, it is possible to perform accurate drowsiness detection that is less dependent on individual differences.

FIGS. 5A to 5C show an example of drowsiness detection results when monotonous driving is continued for 30 minutes using a drive simulator. In FIGS. 5A to 5C, the horizontal axis indicates time, and the vertical axis indicates “no drowsiness” or “drowsiness”. FIG. 5A shows the presence or absence of drowsiness based on the driver's declaration. FIG. 5B is a detection result of the conventional method in which the feature amount is not normalized. That is, in FIG. 5B, drowsiness is detected by using the feature amount extracted by the feature amount extraction unit 12 as it is. FIG. 5C is a detection result of the method of the first embodiment in which the feature amount is normalized. That is, in FIG. 5C, drowsiness is detected by normalizing the feature amount based on the feature amount in the "no drowsiness" state.

In FIG. 5A, the driver declares that he is "sleepy" 14 minutes after the start of operation. On the other hand, in the conventional detection result of FIG. 5B, drowsiness could not be detected, and "drowsiness" was detected 26 minutes after the start of operation. On the other hand, in the detection result of the first embodiment of FIG. 5C, drowsiness can be detected based on the difference from "no drowsiness". Therefore, the drowsiness detection device 10 can detect "drowsiness" at the same timing as the driver's declaration 14 minutes after the start of operation.

[3. Effect, etc.]
As described above, the drowsiness detection device 10 according to the present embodiment includes a feature amount extraction unit 12, a detection rule storage unit 14, a feature amount normalization unit 13, and a drowsiness detection unit 15. The feature amount extraction unit 12 extracts the feature amount related to the heartbeat based on the interval of the user's heartbeat. The detection rule storage unit 14 holds a detection rule for detecting drowsiness. The feature amount normalization unit 13 updates the normalization coefficient. The drowsiness detection unit 15 detects the drowsiness of the user based on the feature amount, the detection rule, and the normalization coefficient. The feature amount normalization unit 13 updates the normalization coefficient based on the feature amount when the drowsiness detection unit 15 does not detect the user's drowsiness. Therefore, it is possible to perform accurate drowsiness detection that is less dependent on individual differences.

Further, the drowsiness detection device 10 further includes an accuracy determination unit 16 for determining the accuracy of the drowsiness detected by the drowsiness detection unit 15. When the accuracy when the drowsiness detection unit 15 determines that there is no drowsiness is higher than the predetermined accuracy threshold value, the feature amount normalization unit 13 updates the normalization coefficient. Therefore, drowsiness can be detected accurately even with time fluctuations in the same person.

Further, the drowsiness detection device 10 further includes an RRI detection unit 11 that receives an electrocardiographic waveform and detects an RR interval of the electrocardiographic waveform. The feature amount extraction unit 12 extracts the feature amount related to the heartbeat based on the RR interval. Therefore, drowsiness can be determined in real time by directly inputting the electrocardiographic waveform.

Further, the feature amount extraction unit 12 extracts a plurality of feature amounts F1 to F7. The plurality of feature quantities F1 to F7 correspond to the plurality of nodes N1 to N7 of the decision tree, respectively. With this configuration, drowsiness can be detected and the accuracy of drowsiness can be determined by using a plurality of feature quantities. Therefore, the drowsiness detection device 10 can detect drowsiness more accurately.

Further, in the present embodiment, the feature amount normalization unit 13 is based on the feature amount when the accuracy determined by the accuracy determination unit 16 from a predetermined time before to the present is higher than the predetermined accuracy threshold value. Calculate the normalization coefficient. Therefore, drowsiness can be effectively detected.

Further, in the present embodiment, the feature amount normalization unit 13 is normalized by the previous normalization coefficient and the weighted sum of the feature amounts when the accuracy determined by the accuracy determination unit 16 is higher than the predetermined accuracy threshold value. Calculate the coefficient. Therefore, drowsiness can be effectively detected.

In the present embodiment, the drowsiness detection device 10 includes an RRI detection unit 11 that receives an electrocardiographic waveform and detects an RR interval. However, the drowsiness detection device 10 may include an RRI input unit that receives data of the RR interval separately detected by an external device instead of the RRI detection unit 11.

In the present embodiment, when the drowsiness detection unit 15 determines that there is no drowsiness and the accuracy detected by the accuracy determination unit 16 is equal to or greater than a predetermined accuracy threshold value, the feature amount normalization unit 13 normalizes. I decided to update the coefficient. However, the drowsiness detection device 10 does not have to include the accuracy determination unit 16. Then, the feature amount normalization unit 13 may update the normalization coefficient when the drowsiness detection unit 15 determines that there is no drowsiness. In that case, the configuration of the drowsiness detection device 10 can be simplified.

In this embodiment, a decision tree and a random forest are exemplified as a method for the drowsiness detection unit 15 to detect drowsiness. However, it is not limited to this. The method of detecting drowsiness may be any method of detecting drowsiness from the feature amount related to the fluctuation of the heartbeat. For example, drowsiness may be detected by a probabilistic model based on features related to heart rate variability. In that case, the probability value output by the probability model when "no drowsiness" is detected can be used as the accuracy to be determined by the accuracy determination unit 16.

[4-1. Modification 1]
The drowsiness detection device 10 of the first embodiment receives an electrocardiographic waveform as an input, detects an RR interval from the input electrocardiographic waveform, and detects drowsiness. However, the drowsiness detection device 10 may receive a pulse wave as an input instead of the electrocardiographic waveform. This makes it possible to detect drowsiness based not only on the electrocardiographic waveform but also on the pulse wave measured by the earlobe or the wrist.

[4-2. Modification 2]
The drowsiness detection device according to the second modification will be described with reference to FIG. FIG. 6 is a block diagram of the drowsiness detection device 20 according to this modification.

The drowsiness detection device 20 includes a normalization unit 23 in place of the feature amount normalization unit 13 (see FIG. 1) of the drowsiness detection device 10.

Similar to the feature amount normalization unit 13, the normalization unit 23 updates the normalization coefficient _Ck based on the feature amount when the drowsiness detection unit 15 does not detect the user's drowsiness. Then, the normalization unit 23 normalizes the rule threshold value of the detection rule held by the detection rule storage unit 14 by using the normalization coefficient C _k . Here, the normalization unit 23 may normalize each rule threshold value by multiplying each rule threshold value by the normalization coefficient C _k . In the drowsiness detection device 10 shown in FIG. 1, the normalized feature amount and the regular threshold value (threshold values T1 to T7) are compared at the node of the decision tree. However, in the drowsiness detection device 20, the non-normalized feature amount and the normalized rule threshold value are compared. Then, the drowsiness detection device 20 detects the drowsiness of the user based on the detection rule having the normalized rule threshold value and the feature amount. Even with this configuration, it is possible to perform accurate drowsiness detection as in the case of the drowsiness detection device 10.

Further, the normalization unit 23 may use the normalization coefficient to normalize both the feature amount and the rule threshold value of the detection rule. Then, the drowsiness detection device 20 may detect the drowsiness of the user based on the detection rule having the normalized rule threshold value and the normalized feature amount.

The drowsiness detection device and the like according to one or more aspects of the present disclosure have been described above based on the embodiments. However, the present disclosure is not limited to this embodiment. As long as it does not deviate from the gist of the present disclosure, one or more of the present embodiments may be modified by those skilled in the art, or may be constructed by combining components in different embodiments. It may be included within the scope of the embodiment.

As described above, an embodiment has been described as an example of the technique in the present disclosure. To that end, an attached drawing and a detailed description are provided.

Therefore, among the components described in the attached drawings and the detailed description, not only the components essential for problem solving but also the components not essential for problem solving in order to illustrate the above-mentioned technology. Can also be included. Therefore, the fact that those non-essential components are described in the accompanying drawings or detailed description should not immediately determine that those non-essential components are essential.

Further, since the above-described embodiment is for exemplifying the technique in the present disclosure, various changes, replacements, additions, omissions, etc. can be made within the scope of claims or the equivalent thereof.

The present disclosure is applicable to a drowsiness detection device or the like that performs accurate drowsiness detection.

10, 20 Drowsiness detection device 11 RRI detection unit 12 Feature amount extraction unit 13 Feature amount normalization unit (normalization unit)
14 Detection rule storage unit 15 Drowsiness detection unit 16 Accuracy determination unit

Claims (14)

A feature amount extraction unit that extracts the feature amount related to the heartbeat based on the interval of the user's heartbeat, and
A detection rule storage unit that holds detection rules for detecting drowsiness,
The normalization part that updates the normalization coefficient, and
A drowsiness detection unit that detects drowsiness of the user based on the feature amount, the detection rule, and the normalization coefficient is provided.
The normalization unit updates the normalization coefficient based on the feature amount when the drowsiness detection unit does not detect the drowsiness of the user.
Drowsiness detector. The normalization unit normalizes the feature amount using the normalization coefficient.
The drowsiness detection unit detects the drowsiness of the user based on the normalized feature amount and the detection rule.
The drowsiness detection device according to claim 1. The normalization unit normalizes the rule threshold of the detection rule using the normalization coefficient.
The drowsiness detection unit detects drowsiness of the user based on the detection rule having the normalized rule threshold value and the feature amount.
The drowsiness detection device according to claim 1. Further provided with an RRI detection unit that receives an electrocardiographic waveform and detects the RR interval of the electrocardiographic waveform.
The feature amount extraction unit extracts the feature amount related to the heartbeat based on the RR interval.
The drowsiness detection device according to any one of claims 1 to 3. Further equipped with an RRI input unit that receives the RR interval of the electrocardiographic waveform,
The feature amount extraction unit extracts the feature amount related to the heartbeat based on the RR interval.
The drowsiness detection device according to any one of claims 1 to 3. Further, an accuracy determination unit for determining the accuracy of the user's drowsiness detected by the drowsiness detection unit is provided.
When the accuracy is higher than a predetermined accuracy threshold, the normalization unit updates the normalization coefficient.
The drowsiness detection device according to any one of claims 1 to 5. The detection rule is a rule based on a decision tree.
The accuracy determination unit determines the accuracy of the user's drowsiness based on the ratio of drowsiness in the leaf node of the decision tree.
The drowsiness detection device according to claim 6. The feature amount extraction unit extracts a plurality of the feature amounts,
Each of the plurality of features corresponds to a plurality of nodes of the decision tree.
The drowsiness detection device according to claim 7. The detection rule is a rule based on a random forest consisting of a plurality of decision trees.
The accuracy determination unit determines the accuracy of the user's drowsiness based on the ratio of drowsiness in the plurality of decision trees.
The drowsiness detection device according to claim 6. The normalization unit is a feature amount extracted from a predetermined time before to the present, and the normalization unit is based on the feature amount when the drowsiness detection unit does not detect the drowsiness of the user. Update the conversion factor,
The drowsiness detection device according to any one of claims 6 to 9. The normalization unit updates the normalization coefficient based on the weighted sum of the previous normalization coefficient and the feature amount.
The drowsiness detection device according to claim 10. The weight used for the weight sum is set larger as the accuracy is higher.
The drowsiness detection device according to claim 11. A feature amount extraction step for extracting the feature amount related to the heartbeat based on the interval of the user's heartbeat, and
Update steps to update the normalization factor, and
It has a drowsiness detection step for detecting drowsiness of the user based on the feature amount, a detection rule for detecting drowsiness, and the normalization coefficient.
The normalization coefficient is updated based on the feature amount when the user's drowsiness is not detected in the drowsiness detection step.
Drowsiness detection method. A program for causing a computer to execute the drowsiness detection method according to claim 13 is recorded.
Program recording medium.

Images