JP2012061222A - Driver condition estimating device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a driver condition estimating device which more accurately estimates a driver condition.SOLUTION: The feature quantity is set, which is biological information to be used in estimating a mental load on a driver for each of three classes classified in terms of aggression which is one of driving characteristics. The mental load on a driver is calculated by multiple regression analysis using the feature quantity as a variable, and a mental load level on a driver is estimated based on the calculation result. The accuracy of estimating the driver can be improved by discriminating the driving characteristics of the driver by aggression, and selectively using the feature quantity to be used in estimating the mental load for each aggression classes. The estimation uses the feature quantity suitable for the driver classified into the same class of more similar driving characteristics (aggression).

Description

  The present invention relates to a driver state estimating device that estimates a driver's state.

  2. Description of the Related Art Conventionally, as disclosed in Patent Document 1, for example, an apparatus that estimates a driver's arousal level, fatigue, or emotional state through measurement of biological information such as a heartbeat is known. The apparatus determines the driver's consciousness level based on the driver's heartbeat information obtained through the sensor, and issues a warning when it is determined that the consciousness level is low. Further, as shown in Patent Document 2, there is also known an apparatus that detects a driver's driving behavior and estimates the driver's state based on the driving behavior. In this device, the frequency of visually observing the side mirror or the like is detected, and the state of the driver is estimated through comparison between the detection result and a threshold value.

Japanese Patent Laid-Open No. 5-184558 Japanese Patent Laid-Open No. 2005-4414

  In the apparatus of Patent Document 1, the state of the driver is uniformly determined based on the determination condition related to specific biological information. However, the relationship between the biological information and the driver's state depends on the driver's personality. For this reason, when estimating a driver | operator's state uniformly by the same determination conditions, there existed a possibility that the determination result may shift | deviate from the actual driver | operator state. Further, since the relationship between the driving behavior and the driver's state is different for each driver, there is a similar concern in the apparatus of Patent Document 2.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a driver state estimation device that can perform more accurate driver state estimation.

  The invention according to claim 1 is characterized in that after setting a feature amount that is biometric information used for estimating a driver's mental load for each of a plurality of classes classified for aggression which is one of driving characteristics, The gist is to calculate the driver's mental load through statistical analysis using the feature quantity as a variable, and to estimate the driver's mental load level based on the calculation result.

  Even in the case of the same situation, the driver feels a different degree of burden. That is, the change in the feature amount (biological information) under the same situation is different for each driver. For this reason, when the same feature value is used for all the drivers, there is a concern that the estimation result based on the feature value may vary individually. In this respect, as in the present invention, the driver's mental load is identified by identifying the driver's driving characteristics by aggressiveness, and using the characteristic amount used for estimating the mental load for each of the aggressive classes. The estimation accuracy can be improved. This is because estimation is performed by using feature quantities suitable for drivers classified into the same class that has closer driving characteristics (here, aggressiveness).

  According to a second aspect of the present invention, in the driver state estimating device according to the first aspect, the vehicle state detecting device for detecting the state of the driving operation by the driver is provided, and the aggressiveness of the driver is determined based on the detection result. The gist is to determine and classify the aggression of the driver into a plurality of classes based on the determination result.

  According to the present invention, driver aggression is classified at any time during driving. When it is assumed that the driver's psychological state also changes each time, the driver's state can be estimated according to such a change.

  According to a third aspect of the present invention, in the driver state estimating device according to the second aspect, the aggressiveness of the driver is determined based on the initial information stored in advance until a certain period of time elapses after the vehicle starts driving. The gist is to classify into multiple classes.

  According to the present invention, information from the vehicle state detection device is accumulated until a certain period of time has elapsed since the start of driving of the vehicle. The driver's aggressiveness is suitably classified based on this accumulated information. Until a certain amount of information is accumulated, the aggressiveness of the driver is determined based on the initial information stored in advance. For this reason, the aggressive classification and the calculation and estimation processing of the driver's mental load are executed without interruption.

  According to a fourth aspect of the present invention, in the driver state detection device according to the third aspect, the initial information is created from the viewpoint of determining information related to a driver's past driving history or driver's aggression. The gist is that the information indicates a standard attack level set based on the result of the aggression questionnaire.

  As the initial information used in the invention of claim 3, according to the present invention, information on the driver's past driving history or the result of an aggression questionnaire created from the viewpoint of judging the driver's aggression It is preferable to use information indicating a standard attack level set based on the information. A decrease in estimation accuracy due to a difference in aggression level for each driver can be minimized.

  The invention according to claim 5 is the driver state estimation device according to any one of claims 2 to 4, wherein the aggression of the driver is re-determined at regular intervals, and the determination result is The gist is that the class based on the driver's aggression is reclassified.

  The situation of the driver or vehicle changes constantly. For this reason, a driver | operator's mental load state also changes with time. In this regard, according to the present invention, the driver's aggression is re-determined at regular intervals, and the class based on the driver's aggression is also reclassified based on the determination result. For this reason, the mental load of the driver at that time can be accurately estimated.

  According to a sixth aspect of the present invention, in the driver state estimating device according to the first aspect, the driver aggression is determined based on the result of the aggression questionnaire created from the viewpoint of determining the driver aggression. The gist is to classify the driver's aggression into a plurality of classes based on the determination result.

  Even in the case of the present invention, estimation is performed by using feature quantities suitable for drivers classified into the same class having closer driving characteristics (here, aggressiveness). For this reason, the estimation accuracy of the driver's mental load is improved.

  According to the present invention, it is possible to estimate the driver's state with higher accuracy.

The block diagram which shows the outline of the driving assistance system in this Embodiment. (A) is the list figure which similarly shows the contents of a questionnaire, (b) is a bar graph which shows the result of a questionnaire. Similarly, a scatter diagram showing the results of AWWL (with class classification). Similarly, a scatter diagram showing the AWWL estimation results (no class classification). The graph for demonstrating the detection method of a pulsation interval (RRI). The flowchart which similarly shows the procedure of a driver state estimation process.

Hereinafter, an embodiment in which the present invention is embodied in a vehicle driving support system will be described with reference to FIGS.
As shown in FIG. 1, the driving support system includes a driver state estimation device 10 and a notification device 11. The driver state estimation device 10 determines whether or not the driver of the vehicle is in a state of reduced attention, such as concentration on acts other than driving such as conversation, sleepiness, fatigue, or ill-mannered state. It is estimated through the mental load state. Based on the estimation result of the driver state estimation device 10, the driver's recognition, determination, or operation of the vehicle when the vehicle is traveling is supported. For example, when it is estimated that the state of attention is reduced, this is indicated to the driver through the notification device 11.

  The driver state estimation device 10 includes an ECU (electronic control device) 21. A driver observation device 22, a vehicle state detection device 23, and a surrounding state detection device 24 are connected to the ECU 21. The ECU 21 executes a driver's mental load state estimation process based on the detection results from these devices.

  The driver observation device 22 detects a feature amount that is biological information of the driver, and generates an electrical signal corresponding to the detection result. Examples of the feature amount include an R wave interval (RRI) based on a driver's electrocardiogram, a heart rate, a blink rate, a pupil diameter, and a head movement amount. The driver observation device 22 includes a camera that captures a face, an image processing device that extracts physiological features from an image captured by the camera, an electrocardiograph that measures electrocardiograms, and the like. The driver observation device 22 detects the above-described feature amount through these devices.

  The vehicle state detection device 23 detects the state of the driving operation by the driver and generates an electrical signal corresponding to the detection result. The information indicating the state of the driving operation includes, for example, a steering angle, an accelerator opening degree, presence / absence of a brake operation, a frequency of a whistle blowing operation, a number of passing times, and the like. The vehicle state detection device 23 includes an accelerator sensor, a brake sensor, a steering angle sensor, and the like. The vehicle state detection device 23 acquires various types of information indicating the state of the driving operation described above through these sensors. Further, the vehicle state detection device 23 detects the presence or absence of passing through a lever combination switch.

  The surrounding state detection device 24 detects a state outside the vehicle and generates an electric signal corresponding to the detection result. The state outside the vehicle includes, for example, a relative distance from a preceding vehicle, a position of a traveling lane, a lane departure amount, and the like. The peripheral state detection device 24 includes a radar using a laser or the like provided at the front and rear of the vehicle, a camera, and the like. The peripheral state detection device 24 detects the information outside the vehicle through these devices.

  The ECU 21 collects the driver's biological information based on the electrical signal generated by the driver observation device 22 and detects a feature amount related to the driver's biological at that time. The ECU 21 collects information related to the driving operation based on the electric signal generated in the vehicle state detection device 23, and detects information indicating the driving operation of the driver at that time. The ECU 21 collects information related to the state outside the vehicle based on the electrical signal from the surrounding state detection device 24 and detects the state around the vehicle at that time.

  The storage device 21a of the ECU 21 stores various control programs executed by the ECU 21 and various information. As the control program, there are an aggression determination program for determining a driver's aggression, an estimation program for estimating a driver's mental load state, and the like. The information stored in the storage device 21a includes various map data used for the driver state estimation process.

  The ECU 21 determines the aggression in the driving of the driver based on various information from the vehicle state detection device 23 and the surrounding state detection device 24, and performs an estimation process of the driver's mental load state according to the determination result. Do. When it is determined that the estimated mental load level exceeds a predetermined level, the ECU 21 notifies the driver through the notification device 11. The notification device 11 issues an alarm or displays it on a meter panel or the like.

  By the way, driving aggression is aggressive to other vehicles, such as showing anger when overtaken, feeling pedestrians in the way, or blowing a horn on other vehicles you do not like. Driving consciousness that shows a tendency. In addition, although driving frequency, driving aggression, driving habituation, and physical condition can be considered as factors that affect the subjective evaluation of the driver's mental load, in this example, driving aggression is a major factor. The estimation method based on this is adopted. The driving aggressiveness determination process will be described in detail later.

<Classification by aggression>
Next, determination of driver aggression will be described. In this example, drivers are classified into three classes based on driving aggression, and feature quantities (physiological quantities) used for estimating mental loads (variables of multiple regression analysis described later) are set for each class. Yes. The storage device 21a of the ECU 21 stores map data indicating the relationship between the three classes of aggressiveness and the feature values used for estimating the mental load. This map data is shown in Table 1. As shown in the table, the level of aggression is classified into three, high, medium and low. The head rotation angle, blink change rate, pupil diameter change, and heart rate are set as the feature quantities used for estimating the driver's mental load state with high aggressiveness. As feature quantities used for estimating the mental load state of a driver who is moderately aggressive, a head rotation angle and a heart rate are set. As feature amounts used for estimating a mental load state of a driver with low aggression, a head rotation angle and a pupil diameter change are set.

攻撃性は、表２に示される項目に基づき判定される。これらの項目は、運転者の攻撃性を判定する観点から設定される。各項目の検出結果の値が大きいほど攻撃性レベルが大きいといえる。例えばパッシング回数が多いほど他車両への威嚇の度合いが大きく攻撃性が高い傾向にあると想定される。

Aggressiveness is determined based on the items shown in Table 2. These items are set from the viewpoint of determining the driver's aggression. It can be said that the aggression level is higher as the value of the detection result of each item is larger. For example, it is assumed that the greater the number of passing times, the greater the degree of threat to other vehicles and the higher the aggressiveness.

運転者の攻撃性レベルの判定基準はつぎのようにして設定される。すなわちまず、ドライビングシミュレータを使用した実験等を通じて表２に示される各項目の値を取得する。そして各項目の検出結果あるいはその平均値としきい値との比較を通じて各項目の検出結果を、攻撃性レベルの観点から３段階（「大、中、小」あるいは「多、中、少」など）に分類する。次いで攻撃性について、「高レベル＝２」、「中レベル＝１」、「低レベル＝０」の得点を与え、その算出されるスコアに基づき攻撃性レベルの判定基準として次式（１）を設定した。当該式（１）により、運転者（被験者）の運転攻撃性は３つのクラス（高、中、低）に分類される。

The criterion for determining the driver's aggression level is set as follows. That is, first, the value of each item shown in Table 2 is acquired through an experiment using a driving simulator. Then, the detection result of each item or the average value and comparison with the threshold value are compared with the detection result of each item in three levels (“large, medium, small” or “many, medium, small”, etc.) Classify into: Next, for aggression, scores of “high level = 2”, “medium level = 1”, and “low level = 0” are given, and the following formula (1) is used as a criterion for aggression level based on the calculated score: Set. According to the formula (1), the driving aggression of the driver (subject) is classified into three classes (high, medium, and low).

  The ECU 21 detects the value of each item shown in Table 2 based on various information from the vehicle state detection device 23 and the surrounding state detection device 24, and applies the following expression (1) to these detection results, thereby driving the vehicle. It is determined which of the three classes the aggression level of the middle driver belongs.

ただし、「Ａｖｅ」は算出されたスコアの平均値、「ＳＤ」は標準偏差である。

However, “Ave” is an average value of the calculated scores, and “SD” is a standard deviation.

<Verification of the effectiveness of classification>
In the following, the effectiveness of classification by aggression is verified. Here, the criteria for determining the aggression level were set based on the questionnaire (16 items in total) shown in FIG. The questionnaire items are created from the viewpoint of determining the driver's subjective aggression. Then, in order to classify the driver's (subject) 's irritation or hostility, the classification was performed by giving a score of "Yes = 2", "None = 1", and "No = 0". From the result of the questionnaire shown in FIG. 2B (more precisely, the calculated score), the subjects were classified into three classes (high, medium, and low) according to the previous equation (1). The results are shown in Table 3. In this example, the number of subjects is 16 from A to T.

ここで、表１に示されるクラス毎の特徴量は、つぎのように設定される。すなわち、クラス毎に使用する特徴量の組合せを変えて重回帰分析（ステップワイズ変数選択法）を行い、その分析結果に基づき運転者の状態推定に使用される特徴量が決定される。回帰分析においては、相関係数が高く、誤差標準偏差が小さい場合に、そのモデルは高精度に基準変数を推定できているといえる。

Here, the feature quantity for each class shown in Table 1 is set as follows. That is, multiple regression analysis (stepwise variable selection method) is performed by changing the combination of feature values used for each class, and the feature value used for estimating the driver's state is determined based on the analysis result. In regression analysis, when the correlation coefficient is high and the error standard deviation is small, it can be said that the model can estimate the reference variable with high accuracy.

  As a result of performing multiple regression analysis by changing the combination of feature quantities used for each class, it was confirmed that the estimation accuracy was most secured when the feature quantities shown in Table 1 were used. Therefore, in this example, the head angle, blink rate, pupil diameter, and heart rate are used for drivers with high aggressiveness, and the head angle and heart rate are used for drivers with moderate aggressiveness. For a driver of a class with low aggression, the head angle and the pupil diameter are set as feature quantities used for state estimation.

<Estimation method>
In this example, a multiple regression analysis method, which is one of multivariate analysis methods, is used for estimating the driver's state. The multiple regression equation with p variables as explanatory variables is expressed by the following equation (2).

ただし、「Ｙ」は基準変数、「Ｘ」は説明変数である。本例では、基準変数「Ｙ」は心的負荷推定値〔ＡＷＷＬ（Ａｄａｐｔｉｖｅ Ｗｅｉｇｈｔ Ｗｏｒｋ−Ｌｏｒｄ）の推定値〕、説明変数「Ｘ」は特徴量である。「β」，「ε」は、係数である。

However, “Y” is a reference variable, and “X” is an explanatory variable. In this example, the reference variable “Y” is an estimated mental load value (estimated value of AWEWL (Adaptive Weight Work-Lord)), and the explanatory variable “X” is a feature quantity. “Β” and “ε” are coefficients.

  Further, in this example, in order to evaluate the estimation accuracy of the AWWL calculated by the multiple regression equation represented by the equation (2), the correlation coefficient R with the AWWL calculated from the subjective evaluation (NASA-TLX), and the error standard Deviation SD is used as an evaluation index.

  NASA-TLX used for subjective evaluation is an evaluation method for determining the mental burden based on a questionnaire consisting of six scale items consisting of mental demands, physical demands, time pressure, work achievement, effort, and dissatisfaction. . These six items are evaluated in 20 steps. Thereafter, all 15 combinations are compared, and a weighting coefficient is calculated. By obtaining a linear sum of the evaluation value of each item and the weighting coefficient, the mental load state is calculated with a maximum score of 100 points.

  The correlation coefficient R is expressed by the following equation (3), and the error standard deviation SD is expressed by the following equation (4).

ただし、「ｘ」は推定ＡＷＷＬ、「ｙ」はＡＷＷＬ、「ｎ」はデータ数である。

However, “x” is the estimated AWWL, “y” is the AWWL, and “n” is the number of data.

<Estimation result>
Next, the results of multiple regression analysis when classification by aggression is performed are shown in the scatter diagram of FIG. 3, and the results of multiple regression analysis when the classification is not performed as a comparative example are shown in the scatter diagram of FIG. 3 and 4, the vertical axis represents AWWL, and the horizontal axis represents estimated AWWL. Further, Table 4 shows the relationship between the class classification method (the presence or absence of class classification) and the evaluation index (correlation coefficient R and error standard deviation SD).

  The analysis results shown in FIG. 3 are obtained when different feature amounts are used for each class classified by driving aggression. In other words, the head rotation angle, blink change rate, pupil diameter change and heart rate (heart rate interval) for drivers with high aggressiveness, and head rotation for drivers with moderate aggressiveness The angle and heart rate are used as feature quantities for a driver of a class with low aggression, such as head rotation angle and pupil diameter change. The analysis results shown in FIG. 4 are obtained when the head rotation angle, heart rate, and CVRR (electrocardiogram heartbeat interval variation coefficient) are commonly used as feature quantities regardless of the driver.

図３及び図４の散布図に示される推定結果から、攻撃性によってクラス分類を行うことにより大幅に精度が向上していることが確認できる。

From the estimation results shown in the scatter diagrams of FIG. 3 and FIG. 4, it can be confirmed that the accuracy is greatly improved by performing the classification according to the aggressiveness.

  Further, as shown in Table 4, from the correlation coefficient R and the error standard deviation SD, the method of the present example in which the class classification is performed is more accurately estimated than the case where the class classification is not performed. It can be confirmed that it is made. That is, the correlation coefficient between AWWL and estimated AWWL is improved from 0.79 to 0.89, and the error standard deviation SD is also decreased from 12.6 to 9.6, compared to the case without class classification. . This indicates that the estimation accuracy of the mental load state is improved.

<About the number of classifications>
As a result of verifying the optimum number of class classifications, as shown in Table 5, it was confirmed that the 3-class classification was the most accurate among the 2-class classification, the 3-class classification, and the 5-class classification. It was. As a result of calculating the estimation model in subjects with similar driving characteristics in consideration of aggression, it is considered that the estimation accuracy has improved.

表１に示されるように、採用された特徴量の中では、全てのクラスにおいて頭部回転角度が採用されていることから、頭部回転角度と心的負荷状態との相関の高さが分かる。また単回帰分析の結果から、その他の特徴量も心的負荷状態との相関があり、その中でもより強く傾向が現れた特徴量が採用されたことも、精度向上に貢献したと考えられる。ちなみに、相関係数から相関の強弱を判断する目安はつぎの通りである。

As shown in Table 1, since the head rotation angle is adopted in all the classes among the adopted feature amounts, the correlation between the head rotation angle and the mental load state can be understood. . From the results of simple regression analysis, the other feature quantities have a correlation with the mental load state, and among them, the feature quantities that showed a stronger tendency were adopted. Incidentally, the standard for judging the strength of correlation from the correlation coefficient is as follows.

“0.4 <R ≦ 0.6”: weak correlation “0.6 <R ≦ 0.8”: correlation “0.8 <R ≦ 1.0”: strong correlation From the above results, It can be seen that more accurate estimation is possible by monitoring the feature quantity of the vehicle during driving, classifying it by aggressiveness, and applying it to the optimal estimation model.

  In the verification of the effectiveness of the classification, the feature amount used for the multiple regression analysis is calculated from data acquired through an experiment using a driving simulator. In this example, experiments were conducted on three types of courses: straight roads, highways, and suburban roads. The characteristic amount (physiological amount) by these experiments was used as the explanatory variable X of the multiple regression equation represented by the equation (2). Here, for example, the heart rate, which is one of the feature quantities, is detected as follows. That is, the ECU 21 accumulates a driver's heartbeat signal detected through the driver observation device 22 (more precisely, an electrocardiograph) for a predetermined time (for example, 30 seconds or more) and samples it at a predetermined frequency. Further, the ECU 21 performs a filtering process on the sampled heartbeat signal. As shown in FIG. 5, a wave exceeding a predetermined threshold is set as an R wave from the heartbeat signal thus obtained, and RRI data is detected with the time interval as a pulsation interval RRI.

<Driver state estimation process>
Next, the procedure of the driver state estimation process executed by the ECU 21 will be described according to the flowchart of FIG. This flowchart is executed according to an estimation program stored in the storage device of the ECU 21. The program is executed when the vehicle is turned on (ignition on).

  First, the ECU 21 sends various information such as biological information of the driver, information on the state of driving operation by the driver, and information on the state outside the vehicle through the driver observation device 22, the vehicle state detection device 23, and the surrounding state detection device 24. Obtain (step S101).

Next, the ECU determines whether or not a predetermined time has elapsed (step S102). This is to accumulate a certain amount of data.
If it is determined that a certain time has elapsed (YES in step S102), the driver's aggression is determined based on information from the vehicle state detection device 23 and the surrounding state detection device 24 (step S103).

  Next, the ECU 21 classifies the driver aggression determined in the previous step S103 (step S104). That is, the class to which the determined aggression belongs is determined based on the above formula (1).

  Next, the ECU 21 determines whether or not the determined aggressiveness belongs to the middle class (step S105). When the ECU 21 determines that it is a middle class (YES in step S105), the feature quantity corresponding to the class, that is, the head rotation angle and the heart rate is set with reference to the map shown in Table 1 ( Step S106).

  When ECU 21 determines that it is not a middle class (NO in step S105), it determines whether it is a high class (step S107). When it is determined that the ECU 21 is in the high class (YES in step S107), the feature amount corresponding to the class with reference to the map shown in Table 1, that is, the head rotation angle, blink change rate, pupil diameter A change and a heart rate are set (step S108).

  When ECU 21 determines that it is not a high class (NO in step S107), it determines that it is a low class (step S109). Then, the ECU 21 refers to the map shown in Table 1 and sets the feature amount corresponding to the class, that is, the head rotation angle and the pupil diameter change (step S110).

  Next, the ECU 21 uses the feature amount set in the previous step S106, step S108, or step S110 to execute a driver's mental load calculation process (step S111). That is, the ECU 21 obtains the driver's mental load (estimated AWWL) by applying the feature amount corresponding to each class to the explanatory variable “X” of the multiple regression equation of the previous equation (2). At this time, the mental load is calculated with a perfect score of 100 points.

  Next, the ECU 21 determines the state of the driver based on the calculated mental load (step S112). That is, the ECU 21 determines to which of the following determination conditions (A) to (J) the calculated mental load belongs.

(A) Level 0 when “mental load <28”
(B) When “28 ≦ mental load <36” Level 1
(C) When “36 ≦ mental load <44”, level 2
(D) When “44 ≦ mental load <52”, level 3
(E) When “52 ≦ mental load <60”, level 4
(F) When “60 ≦ mental load <68”, level 5
(G) When “68 ≦ mental load <76”, level 6
(H) When “76 ≦ mental load <84”, level 7
(I) When “84 ≦ Mental load <92” Level 8
(J) When “92 ≦ Mental load” Level 9
When the ECU 21 determines that the calculated mental load belongs to the range of (A) or (B), the ECU 21 determines that the load is lower than that of driving and that the same mental load continues or frequently occurs. In this case, the driver recognizes that he / she is drowsy. Similarly, when ECU 21 determines that it belongs to the range of (E) to (G), it is in a state where a load is applied rather than driving, and when it is determined that a similar mental load continues or frequently occurs. Recognizes that the driver is distracted. Similarly, when ECU 21 determines that it belongs to the range of (H) to (J), it is in a state where a load stronger than driving is applied, and when it is determined that the same mental load continues or frequently occurs. Recognizes that the driver is under stress. Then, the ECU 21 provides feedback to the driver in accordance with the recognized driver state. When it is determined that the calculated mental load belongs to the range (C) or (D), the ECU 21 determines that it is normal.

  By the way, the mental load level, to be precise, the score obtained on a scale of 100 is divided into 10 stages (A) to (J), depending on the strength of the mental load. This is to add a process such as changing the time until determination. For example, (A) and (B) are conditions under which it is possible to determine that the mental load is hardly felt. However, since (A) has a lower degree of load than (B), the driver corresponding to condition (A) feels sleepy in a shorter time (less times) than the driver corresponding to condition (B). It can be judged. In addition, although it is preferable to determine the driver's state with as fine a score as possible, since the calculation is originally performed based on the subjective evaluation value, the score is actually divided in increments of “8” as the score value.

  Next, the ECU 21 determines whether or not the vehicle power supply is shut off (ignition off) (step S113). When it is detected that the ignition has been turned off (YES in S113), the process ends. If it is not detected that the ignition has been turned off (NO in S113), it is determined whether or not a predetermined time has passed (step S114). If it is determined that the fixed time has not elapsed (NO in step S114), the process proceeds to step S111. If it is determined that the fixed time has elapsed (YES in step S114), the process proceeds to step S101. That is, class classification based on aggressiveness is performed again. Thereby, it is possible to suitably cope with the mental load state of the driver that changes every moment.

  In step S102, when it is determined that the ECU 21 has not elapsed for a certain period of time, that is, when it is determined that accumulation of a certain amount of data has not been completed (NO in step S102), the driver's The past driving history (aggression determination data) is read from the storage device (step S115), and the process proceeds to step S111. In step S111, the ECU 21 calculates a mental load (AWWL) based on the driving history, and then determines the driver's mental load state based on the calculated mental load (S112). Thereafter, the same processing as described above is executed. As described above, in the initial state immediately after the ignition is turned on, until the data is accumulated (until YES in step S102), the driver's state is determined based on the class classification based on the read past aggression data. Is estimated. Thereby, a driver | operator's state estimation process is continued without interruption.

  As described above, according to the estimation method of this example, the mental load such as a state where the driver's consciousness is concentrated or a blurred state is statistically analyzed (in this example, multiple regression). When estimating using analysis, it is not necessary to set a characteristic amount (physiological amount) that is likely to cause a change in mental load for each driver. This is because the aggressiveness of the driver is automatically determined based on the operation state sequentially collected during driving and the surrounding situation of the vehicle. Since it is not necessary to change the mental load for each individual and specify a feature quantity suitable for use, adaptation to an unspecified number of drivers is easy. In addition, as described above, the driver's aggression is determined based on the surrounding conditions of the vehicle, etc., and this is caused by changes in the surrounding environment, unlike the case where a characteristic amount common to a large number of unspecified drivers is set. Individual differences can also be handled (by class). According to this example, an advantage in setting feature quantities for each driver (corresponding to individual differences) and an advantage in setting feature quantities common to an unspecified number of drivers (unspecified many drivers) Both).

<Effect of Embodiment>
Therefore, according to the present embodiment, the following effects can be obtained.
(1) A feature amount, which is biometric information used for estimating a driver's mental load, is set for each of a plurality of classes classified as aggressiveness, which is one of driving characteristics. Then, the driver's mental load is calculated through statistical analysis using the feature amount as a variable, and the driver's mental load level is estimated based on the calculation result.

  Even in the case of the same situation, the driver feels a different degree of burden. That is, the change in the feature amount (biological information) under the same situation is different for each driver. For this reason, when the same feature value is used for all the drivers, there is a concern that the estimation result based on the feature value may vary individually. In this regard, as in this example, the driver's mental load is identified by identifying the driver's driving characteristics by aggression and using the characteristic amount used for estimating the mental load for each class of aggression. The estimation accuracy can be improved. This is because the estimation is performed by using feature quantities suitable for drivers classified into the same class closer to driving characteristics (here, aggressiveness).

(2) The driver's aggression is determined based on the detection result of the vehicle state detection device 23, and the driver's aggression is classified into a plurality of classes based on the determination result.
For this reason, the driver's aggression is classified at the time of driving. When it is assumed that the driver's psychological state also changes each time, the driver's state can be estimated according to such a change. It corresponds to individual differences and sets the physiological amount to be used for an unspecified number of drivers, and accurately estimates the mental load.

(3) The driver's aggression is classified into a plurality of classes based on initial information stored in advance until a certain period of time elapses after the vehicle starts driving.
According to this example, information from the vehicle state detection device is accumulated until a certain period of time has elapsed since the start of driving of the vehicle. The driver's aggression is suitably classified based on this accumulated information. Until a certain amount of information is accumulated, the aggressiveness of the driver is determined based on the initial information stored in advance. For this reason, the aggressive classification and the calculation and estimation processing of the driver's mental load are executed without interruption.

  (4) Information regarding the past driving history of the driver is used as the initial information (see step S115). Thereby, the fall of the estimation precision by the difference in the aggression level for every driver | operator can be suppressed to the minimum.

(5) The driver's aggressiveness is re-determined at regular intervals, and the class based on the driver's aggressiveness is reclassified based on the determination result (see step S114).
The situation of the driver or vehicle changes constantly. For this reason, a driver | operator's mental load state also changes with time. In this regard, according to the present example, the aggression of the driver is re-determined every predetermined period, and the class based on the aggression of the driver is also reclassified based on the determination result. For this reason, the mental load of the driver at that time can be accurately estimated.

(6) Aggressiveness was classified into 3 classes. For this reason, higher estimation accuracy can be obtained.
<Other embodiments>
In addition, you may implement the said embodiment as follows.

  -In this example, when the mental load of the driver state is at a predetermined level, the notification device 11 is notified to that effect, but not only the notification, but also through the control of scent, lighting, meter display, etc. You may make it suppress a driver | operator's mental load. In this case, a scent generating device, a vibrating device, a cool air generating device, an oxygen concentration adjusting device, and the like are provided as appropriate as the notification device 11 or instead of this. These devices are operated according to the state of mental load of the driver.

  In this example, the driver's aggressiveness is sequentially determined, but may be set as a fixed one. For example, when a driver's license is issued, the aggression questionnaire as shown in Fig. 2 (a) or a driving simulator is used to determine each person's aggression, and the aggression level is classified into classes based on the determination result. To do. Then, this class classification is registered in the vehicle (more precisely, the storage device of the ECU 21). During actual driving, feature quantities (head rotation angle, heart rate) corresponding to the class are set based on the class classification of the aggression level registered in the vehicle with reference to the map shown in Table 1. (Steps S106, S108, S110 in FIG. 6).

  Even in such a case, estimation is performed using feature quantities suitable for the drivers classified into the same class that has closer driving characteristics (here, aggressiveness). For this reason, the estimation accuracy of the driver's mental load is improved. In addition, from the start of driving until a certain amount of information is accumulated, information indicating the standard attack level set based on the results of the aggression questionnaire created from the viewpoint of judging the driver's aggression is used Thus, since the driver's aggression is determined, it is possible to minimize a decrease in estimation accuracy due to a difference in the aggression level for each driver. Furthermore, since the processing of step S101 to step S103 in the flowchart of FIG. 6 can be omitted, the calculation load on the ECU 21 is reduced. You may make it update the classification by aggression whenever a driver's license is updated.

  In this example, in the process of step S115 of FIG. 6, information related to the driver's past driving history (aggression determination data) is read from the storage device, and the mental load (AWWL) is based on the driving history information. However, it may be as follows. For example, instead of the driving history, data indicating a standard aggressiveness level may be read. The data is set based on, for example, the aggression questionnaire shown in FIG. 2A for an unspecified number of drivers and the questionnaire results. A driving simulator may be used to analyze the driving tendency of an unspecified number of drivers, and data indicating a standard attack level may be set based on the analysis result.

  In this example, when a certain period of time has elapsed since the estimation of the driver state in step S112 in FIG. 6 (YES in step S114), the process proceeds to step S101 and the aggression is determined again. However, the processing (step S114) may be omitted.

  In this example, various information from the peripheral state detection device 24 is used as information for determining the aggression level, but the aggression level may be determined without using these pieces of information.

  In this example, the number of class classifications based on aggression is three, but it may be two, five or more classifications. Even in this case, the estimation accuracy of the driver state can be improved as compared with the case where the classification is not performed.

The criteria for determining the aggression level or the criteria for determining the driver's mental load state may be appropriately changed and set.
In this example, multiple regression analysis is adopted as a method for estimating the driver state, but other methods may be adopted. For example, other statistical analysis (multivariate analysis) methods such as discriminant analysis and cluster analysis can be employed.

  In this example, the driver state estimation device 10 is applied to a vehicle driving support system, but may be applied to, for example, a simulation system. Note that automobiles, trains, and the like are assumed to be installed in these systems.

<Other technical ideas>
Next, a technical idea that can be grasped from the above embodiment will be added below.
(B) Classifying aggression, which is one of the driving characteristics, into multiple classes, setting feature quantities to be used for estimating mental load for each class, and driving through statistical analysis using the feature quantities as variables A driver state estimation method for calculating a driver's mental load and estimating a driver's mental load level based on the calculation result.

  DESCRIPTION OF SYMBOLS 10 ... Driver state estimation apparatus, 23 ... Vehicle state detection apparatus.

Claims (6)

  Statistical features that use feature values as variables after setting feature values, which are biometric information used to estimate the driver's mental load, for each of several classes classified for aggression, which is one of driving characteristics A driver state estimating device that calculates a driver's mental load through analysis and estimates a driver's mental load level based on the calculation result. The driver state estimation apparatus according to claim 1, wherein
Driving that includes a vehicle state detection device that detects the state of driving operation by the driver, determines the driver's aggression based on the detection result, and classifies the driver's aggression into a plurality of classes based on the determination result Person state estimation device. In the driver state detection device according to claim 2,
A driver state estimation device that classifies a driver's aggressiveness into a plurality of classes based on initial information stored in advance until a certain period of time has elapsed since the start of driving of the vehicle. In the driver state detection device according to claim 3,
The initial information is information indicating a standard attack level set based on information on a driver's past driving history or an aggression questionnaire created from the viewpoint of determining a driver's aggression. Person state estimation device. In the driver state estimating device according to any one of claims 2 to 4,
A driver state estimation device that re-determines a driver's aggression at regular intervals and reclassifies a class based on the result of the determination. The driver state estimation apparatus according to claim 1, wherein
A driver state that determines the driver's aggression based on the result of the aggression questionnaire created from the viewpoint of determining the driver's aggression, and classifies the driver's aggression into multiple classes based on the determination result Estimating device.

Images