WO2014034779A1

WO2014034779A1 - Individual characteristic estimation unit and individual characteristic estimation method

Info

Publication number: WO2014034779A1
Application number: PCT/JP2013/073140
Authority: WO
Inventors: 充伸神沼
Original assignee: 日産自動車株式会社
Priority date: 2012-08-31
Filing date: 2013-08-29
Publication date: 2014-03-06

Abstract

Individual characteristic estimation devices (100, 101) comprise a signal detection unit (11), a scene estimation unit (13), an instantaneous characteristic estimation unit (14), an instantaneous characteristic model storage unit (15), and a trend estimation unit (16). The signal detection unit (11) acquires vehicle information. The scene estimation unit (13) estimates, on the basis of the vehicle information, scenes wherethrough a host vehicle travels. The instantaneous characteristic model storage unit (15) stores as an instantaneous characteristic model a characteristic model wherein a characteristic trend label whereby the scene is easily expressed is associated with each scene. On the basis of the scene which is estimated with the scene estimation unit (13), the instantaneous characteristic estimation unit (14) queries the instantaneous characteristic model and estimates an instantaneous characteristic trend of a driver. The trend estimation unit statistically analyses the estimated instantaneous characteristic trend, using HMM or other algorithm, and estimates a static characteristic trend of the driver.

Description

Personal characteristic estimation apparatus and personal characteristic estimation method

The present invention relates to a personal characteristic estimation device and a personal characteristic estimation method for estimating a driver's characteristics based on driver's operation information, vehicle behavior, and driving scene when driving the vehicle.

Japanese Patent Application Laid-Open No. 2011-34430 (Patent Document 1) describes an apparatus for estimating driver characteristics (a general term for personality, tendency, temperament, etc.) based on behavior when a driver drives a vehicle. Things are known. Patent Document 1 discloses that a driver recognizes a driver's kite based on a driving operation when driving the vehicle, and supports driving based on the driver's kite.

JP 2011-34430 A

However, the conventional example disclosed in Patent Document 1 recognizes a driver's habit and supports driving, and does not estimate the basic characteristics of the driver.

An object of the present invention is to provide a personal characteristic estimation device and a personal characteristic estimation method capable of estimating a basic characteristic of a driver.

A personal characteristic estimation device according to an aspect of the present invention includes a vehicle information acquisition unit that acquires vehicle information, a scene estimation unit that estimates a scene in which the vehicle travels based on the vehicle information, a plurality of scenes, and driving Instantaneous characteristic model storage means for setting a plurality of characteristic tendency labels indicating an instantaneous characteristic tendency of a person and storing a characteristic model associating a characteristic tendency label easy to be expressed in each scene as an instantaneous characteristic model Based on the scene estimated by the scene estimation means, referring to the instantaneous characteristic model, the instantaneous characteristic estimation means for estimating the instantaneous characteristic tendency of the driver, and the instantaneous characteristic estimated by the instantaneous characteristic estimation means Characteristic tendency estimating means for statistically analyzing the characteristic tendency and estimating a static characteristic tendency of the driver.

The personal characteristic estimation method according to an aspect of the present invention sets a plurality of scenes, sets a plurality of characteristic tendency labels indicating an instantaneous characteristic tendency of the driver, and displays each scene in the scene. The characteristic model associated with the easy characteristic tendency label is stored as the instantaneous characteristic model, the scene where the vehicle travels is estimated based on the vehicle information, and the instantaneous characteristic model of the driver is referred to based on the estimated scene. A static characteristic tendency of the driver is estimated by statistically analyzing the estimated instantaneous characteristic tendency and statistically analyzing the estimated instantaneous characteristic tendency.

FIG. 1 is a block diagram showing the configuration of the personal characteristic estimation apparatus according to the first embodiment of the present invention. FIG. 2 is an explanatory diagram schematically showing processing of the estimation unit of the personal characteristic estimation device according to the first embodiment of the present invention. FIG. 3 is a flowchart showing a processing procedure of the personal characteristic estimation apparatus according to the first embodiment of the present invention. FIG. 4 is a block diagram showing a configuration of the personal characteristic estimation apparatus according to the second embodiment of the present invention. FIG. 5 is an explanatory diagram schematically showing an instantaneous characteristic model of the personal characteristic estimation device according to the second embodiment of the present invention. FIG. 6 is an explanatory diagram schematically showing how a characteristic model is selected by the model switching unit of the personal characteristic estimation device according to the second embodiment of the present invention. FIG. 7 is a flowchart showing a processing procedure of the personal characteristic estimation apparatus according to the second embodiment of the present invention. FIG. 8A is an explanatory diagram illustrating an example of a likelihood threshold according to each embodiment of the present invention. FIG. 8B is an explanatory diagram illustrating an example of a calculated likelihood value according to each embodiment of the present invention. FIG. 9A is an explanatory diagram illustrating a relationship between a frequency threshold value and a frequency result according to each embodiment of the present invention. FIG. 9B is an explanatory diagram illustrating a relationship between a frequency threshold value and a frequency result according to each embodiment of the present invention. FIG. 10 is an explanatory diagram showing a relationship between a plurality of scenes and a characteristic tendency label that can be detected in each scene according to each embodiment of the present invention. FIG. 11A is an explanatory diagram showing the situation of the scene (1) according to each embodiment of the present invention. FIG. 11B is an explanatory diagram showing the situation of the scene (2) according to each embodiment of the present invention. FIG. 11C is an explanatory diagram showing a situation of the scene (3) according to each embodiment of the present invention. FIG. 11D is an explanatory diagram showing a situation of the scene (4) according to each embodiment of the present invention. FIG. 12A is an explanatory diagram showing the situation of the scene (5) according to each embodiment of the present invention. FIG. 12B is an explanatory diagram showing the situation of the scene (6) according to each embodiment of the present invention. FIG. 12C is an explanatory diagram showing the situation of the scene (7) according to each embodiment of the present invention. FIG. 12D is an explanatory diagram showing a situation of the scene (8) according to each embodiment of the present invention. FIG. 13 is an explanatory diagram showing the relationship between the driving signal and the driver's static characteristics according to each embodiment of the present invention. FIG. 14A is an explanatory diagram showing a procedure for designing an environmental corpus according to each embodiment of the present invention. FIG. 14B is an explanatory diagram showing a procedure for designing an environmental corpus according to each embodiment of the present invention. FIG. 15 is an explanatory diagram showing a procedure for designing a characteristic corpus according to each embodiment of the present invention. FIG. 16 is an explanatory diagram showing the structure of the environmental model and the characteristic model according to each embodiment of the present invention.

Hereinafter, first and second embodiments of the present invention will be described with reference to FIGS.

(First embodiment)
FIG. 1 is a block diagram showing the configuration of the personal characteristic estimation apparatus according to this embodiment. As shown in FIG. 1, the personal characteristic estimation apparatus 100 includes a signal detection unit (vehicle information acquisition unit) 11, an estimation unit 12, an instantaneous characteristic model storage unit (instantaneous characteristic model storage unit) 15, and a tendency estimation unit ( Characteristic tendency estimation means) 16. The estimation unit 12 includes a scene estimation unit (scene estimation unit) 13 and an instantaneous characteristic estimation unit (instantaneous characteristic estimation unit) 14.

The signal detection unit 11 acquires information related to the operation of the vehicle and information related to the behavior of the vehicle from a CAN (Controller Area Network) mounted on the vehicle. Examples of information relating to the operation of the vehicle include information such as a brake operation, a steering angle of the steering, and an accelerator opening. Examples of the information regarding the behavior of the vehicle include information such as the traveling speed of the vehicle, the yaw rate, and the discharge level of the battery.

Furthermore, the signal detection unit 11 acquires information on the outside of the vehicle. For example, the signal detection unit 11 acquires information related to the external environment based on a surrounding image captured by an in-vehicle camera (not shown) or a traveling path of the host vehicle detected by a GPS device or the like. In addition, the signal detection unit 11 acquires information related to operation history of the navigation device, the air conditioner, and the audio mounted on the vehicle. Hereinafter, the information related to the operation of the vehicle, the information related to the behavior of the vehicle, and the information related to the operation history will be collectively referred to as a “vehicle signal”. Moreover, the concept which shows at least one of the information regarding the operation of the vehicle, the information regarding the behavior of the vehicle, the information regarding the operation history, and the information on the outside of the vehicle is referred to as vehicle information.

The instantaneous characteristic model storage unit 15 stores a plurality of preset scenes (scenes (1) to (8) shown in FIGS. 11A to 12D described later). The instantaneous characteristic model storage unit 15 sets a plurality of characteristic tendency labels indicating the characteristics of the driver, an environmental model used for estimating the scene, and a characteristic tendency that can be easily expressed in each scene. A characteristic model associated with a label (hereinafter abbreviated as “label”) is stored as an instantaneous characteristic model. Specifically, the instantaneous characteristic model storage unit 15 estimates the scene as shown in the correspondence relationship in the correspondence table shown in FIG. 10 in which a plurality of scenes related to vehicle travel are taken in the row direction and a plurality of labels are taken in the column direction. In addition, an environment model that is used for the purpose and a characteristic model that associates each scene with a label (instantaneous characteristic of the driver) that is prominently displayed in the scene are stored as an instantaneous characteristic model.

In the correspondence table shown in FIG. 10, scenes (1) to (1) to (3) are classified according to conditions such as road shape in the row direction, time zone, traffic jam information, external state, presence of pedestrians, presence of vehicles, presence of oncoming vehicles, and the like. 8) is set. In the column direction, various characteristic tendency labels (labels (1) to (9)) indicating the characteristics of the driver are set. In the correspondence table shown in FIG. 10, one or more labels that are easy to be displayed for each scene are indicated by circles. For example, in a scene (1) (FIG. 11A) in which a vehicle travels on a “crosswalk without a signal on a narrow street” (FIG. 11A), “other-person sympathy” in label (1) and “non-existence” in label (4) Since a behavior that makes it easy to determine the driver's characteristics such as “sympathy” appears, in the correspondence table shown in FIG. 10, the labels (1) and (4) are circled. In other words, by detecting the driver's behavior when the vehicle travels on a pedestrian crossing without a signal on a narrow road, the driver's "other-sympathetic empathy" and "non-sympathy" are determined. can do. Therefore, the instantaneous characteristic model storage unit 15 stores two characteristic models in which the labels (1) and (4) are associated with the scene (1). As shown in FIG. 2, since the estimation unit 12 uses a characteristic model that combines a scene and a characteristic (label) when estimating the instantaneous characteristic of the driver, the instantaneous characteristic model storage unit 15 stores the characteristic model in FIG. Sixteen characteristic models corresponding to the number of circles shown are stored.

Note that the correspondence table shown in FIG. 10 is merely an example. For example, with respect to the “crosswalk without a signal on a narrow road” in the scene (1) shown in FIG. 10, the “determinism” column in the label (9) is not circled. This does not mean that the driver's characteristics such as determinism are difficult to express in the scene (1), but data indicating the relationship between the scene (1) and the label (9) is not obtained. It is shown that. Therefore, the correspondence table shown in FIG. 10 may be changed by future research.

In addition, compared with a static characteristic detected by a general and static characteristic tendency measuring method such as AQ test (Autism-Spectrum Quotient), it is greater than a predetermined value (for example, 0 .2 or higher) can be set as a label. Furthermore, it is also possible to assume the driver's introspection from a moving image or sound and assign the assumed driver's introspection as a label.

The “instantaneous characteristic” is an instantaneous characteristic estimated based on the behavior of the driver. In the present invention, a “static characteristic” which is a personal characteristic of the driver (generic name such as personality, tendency, and temperament) is obtained based on a number of “instantaneous characteristics”.

The scene estimation unit 13 shown in FIG. 1 has instantaneous characteristics based on vehicle signals (information on vehicle operation, information on vehicle behavior, and information on operation history) detected by the signal detection unit 11 and information on the outside of the vehicle. The scene in which the host vehicle travels is estimated with reference to the environmental model stored in the model storage unit 15. For example, the scene estimation unit 13 estimates a scene by grasping a situation in which the host vehicle is currently traveling based on a surrounding image captured by an in-vehicle camera (not shown). At this time, the scene estimation unit 13 refers to the environment model and corresponds to which scene among the scenes (1) to (8) indicated in the row direction of the correspondence table shown in FIG. Estimate.

The instantaneous characteristic estimation unit 14 refers to the characteristic model stored in the instantaneous characteristic model storage unit 15 based on the vehicle signal detected by the signal detection unit 11 and the scene estimated by the scene estimation unit 13. The instantaneous characteristics of the person. Specifically, the instantaneous characteristic estimation unit 14 obtains the likelihood of each label using an HMM (Hidden Markov Model) algorithm (an algorithm for statistically analyzing characteristic trends) or the like based on the vehicle signal. Detailed procedures will be described later.

The personal characteristic estimation apparatus 100 can be configured as an integrated computer including a central processing unit (CPU), RAM, ROM, and storage means such as a hard disk.

Next, a method for estimating the driver's static characteristics will be described with reference to FIG. As a method for estimating the driver's static characteristics (character, tendency, temperament, etc.), a questionnaire method can be cited. The general questionnaire C1 asks the subject (driver in this embodiment) his / her behavior and introspection when a specific environment is given, and estimates the static characteristics of the subject from the answer result. Therefore, in order to estimate a static characteristic similar to the static characteristic estimated by the general questionnaire C1 from a driving operation signal or a vehicle operation signal (hereinafter referred to as a driving signal), based on the driving signal, It is only necessary to be able to estimate the specific environment given by the questionnaire C1 and the behavior of the driver when the specific environment is given.

Since driving behavior is determined instantaneously by the driver's characteristics, environment, surrounding conditions, traffic rules, and manners, the driver's instantaneous characteristics can be detected by grasping the driving behavior. However, since the driving signal does not include information representing the relationship between the driving behavior and the driver's static characteristics, the driving version questionnaire is used to represent the relationship between the driving behavior and the driver's static characteristics. C2 is required.

For example, as an example of a question for investigating the tendency regarding the presence or absence of empathy for others, a question such as “Do you think that pedestrians should give up in any case when they are trying to cross?” Is given to the subject. The subject answers the question while considering the actual driving scene, such as “Would you stop before the pedestrian crossing?” Or “Would you go first while observing the pedestrian?”

Thus, between the answer result of the driving version question sheet C2 having the question items assuming the actual driving scene and the answer result of the general question sheet C1 for examining the tendency regarding the presence or absence of general empathy for others. If there is a correlation, the static characteristics of the driver can be estimated by observing the driving behavior described in the driving questionnaire C2.

For example, in the case of “stop in front of a pedestrian crossing”, a signal indicating a one-time stop after stepping on the brake appears before the pedestrian crossing, and “going first while observing the pedestrian” Then, the action of opening and closing the accelerator is expressed. Based on these relationships, if the driving signal C3 when the driver takes an action in a specific environment is extracted and a label explaining the driving action (for example, “other-friendliness empathy”, etc.) can be given. The driving behavior can be estimated from only the driving signal, and the instantaneous characteristics of the driver observed instantaneously estimated from the driving behavior can be calculated. By measuring the frequency of occurrence of these instantaneous characteristics, as a result, static characteristics equivalent to the driver's static characteristics expected from the score (answer result) obtained from the psychological test (general questionnaire C1) are obtained. It can be estimated automatically.

The estimation unit 12 first divides the operation signal input from the signal detection unit 11 for each unit time and performs environment recognition. In the environment recognition, the estimation unit 12 uses an environmental model based on an HMM (Hidden Markov Model) and prepares in advance a road shape such as a right / left turn or a narrow road, a traffic jam situation, and the like based on the input driving signal. One pattern is recognized from among the patterns representing a plurality of environments. The estimation unit 12 selects a characteristic model based on the environment represented by the recognized pattern.

For example, when the “right turn” environment is recognized, the estimation unit 12 selects a characteristic model that has been learned in advance so that the instantaneous characteristics of the driver can be recognized in the “right turn” environment. Next, the estimation unit 12 recognizes the instantaneous characteristics of the driver observed instantaneously using the selected characteristic model. The recognized instantaneous characteristic is accumulated every time a recognition result is transmitted. When a certain amount of recognition results are accumulated, the tendency estimation unit 16 obtains the driver's static characteristics based on the occurrence frequency and distribution of a plurality of recognized instantaneous characteristics.

The environmental corpus and the characteristic corpus are designed in advance by the following method. An environmental model is obtained by pre-learning based on an environmental corpus, and a characteristic model is obtained by pre-learning based on a characteristic corpus.

(Design of environmental corpus)
14A and 14B are explanatory diagrams showing examples of collection and extraction of driving signals (vehicle speed in this example) in three environments by each of driver numbers D1 to Dn. Environment S1 shown to FIG. 14A has shown the situation of straight road congestion, S2 the narrow road crossing, and S3 the state of a T-shaped road right turn. FIG. 14B shows a state in which vehicle speed data for each environment (S1 to S3) is cut out and environmental symbols (straight road congestion, narrow road crossing, T-turn right turn) are added to the cut out vehicle speed data. By collecting driving signals from as many drivers as possible, it is possible to stabilize the form of distribution of driving signals (vehicle speed data) for each environmental symbol.

(Design of characteristic corpus)
Similarly to the environment corpus, the characteristic corpus collects driving signals in a plurality of environments by each of the drivers D1 to Dn over a long period of time, and extracts and classifies vehicle speed data for each environment symbol. Further, the plurality of vehicle speed data collected for each environmental symbol is classified for each driving action corresponding to the instantaneous characteristics of the driver, and a characteristic symbol is given to the vehicle speed data collected for each driving action. FIG. 15 gives characteristic symbols (emotional empathy behavior, non-sympathetic behavior) to the vehicle speed data collected for each driving behavior (a1, a2) when the environmental symbol is “narrow road crossing, with pedestrians” It shows how it was done.

When the environmental symbol is “narrow pedestrian crossing, with pedestrians”, the driving behavior of the driver is “pause after the pedestrian passes after driving temporarily” of driving behavior a1, and While watching the situation, "passing the pedestrian crossing without stopping" has been observed. “Emotional empathy behavior” is assigned as a characteristic symbol to the vehicle speed data obtained by observing the driving behavior a1. “Non-sympathetic behavior” is given as a characteristic symbol to the vehicle speed data obtained by observing the driving behavior a2. In the characteristic symbol, unlike the environmental symbol, the name of the driving action is not directly given.

FIG. 16 is an explanatory diagram showing the structure of the environment model and the characteristic model, and uses the Left_to_Right model used in speech recognition. Using an environmental model algorithm, an instantaneous characteristic model is generated by using each corpus to learn transition probabilities and output probabilities between states that are output when an input parameter series is observed for each symbol. .

Next, the operation of the personal characteristic estimation apparatus 100 will be described with reference to the flowchart shown in FIG.

First, in step S11, the signal detection unit 11 acquires various vehicle signals and information on the outside of the vehicle. Specifically, depending on the CAN mounted on the vehicle, vehicle signals such as accelerator opening, brake operation, steering angle, vehicle speed, yaw rate, etc., external video signals captured by an in-vehicle camera, or GPS current Get location information.

In step S <b> 12, the signal detection unit 11 determines the primary difference between the acquired signals at a certain time interval (this is referred to as a difference Δ) and the secondary difference between the primary differences of the acquired signal (this is referred to as a difference ΔΔ). Is calculated and parameterized. For example, when the acquired signal is the accelerator opening, when the accelerator opening is detected at each of the times t1, t2, t3, t4, and t5 having a certain time interval, the signal detector 11 Accelerator opening difference Δ1 obtained at t2, Accelerator opening difference Δ2 obtained at times t2 and t3, Accelerator opening difference Δ3 obtained at times t3 and t4, Accelerator opening difference obtained at times t4 and t5 The difference Δ4 is calculated. Further, the signal detector 11 obtains a difference ΔΔ1 between the differences Δ1 and Δ2, a difference ΔΔ2 between the differences Δ2 and Δ3, and a difference ΔΔ3 between the differences Δ3 and Δ4.

In step S13, the scene estimation unit 13 refers to the environmental model, and estimates the scene in which the host vehicle is traveling based on various vehicle signals detected by the signal detection unit 11 or information on the outside of the vehicle. . Furthermore, the instantaneous characteristic estimation unit 14 refers to the characteristic model and calculates the likelihood for the label (labeled with a circle in the correspondence table shown in FIG. 10) associated with the estimated scene. . In this process, the instantaneous characteristic estimation unit 14 calculates the likelihood of each label using an HMM (Hidden Markov Model) algorithm based on the difference Δ obtained in step S12 and the difference ΔΔ. Specifically, the instantaneous characteristic estimation unit 14 obtains the likelihood of each label by substituting the difference Δ obtained in step S12 and the data of the difference ΔΔ into the HMM algorithm. Note that the details of the HMM algorithm are well-known techniques and will not be described.

Specifically, in the instantaneous characteristic estimation unit 14, the vehicle is traveling on the “crosswalk without a signal on a narrow road” of the scene (1) in the correspondence table shown in FIG. 10 in the scene estimation unit 13. Is estimated, the likelihood of “other-person sympathy” of label (1) and the likelihood of “non-sympathy” of label (4) are calculated.

In step S14, the instantaneous characteristic estimation unit 14 determines whether or not the likelihood of the label having the maximum likelihood exceeds a preset threshold value. For example, in the case of the scene (1) in the correspondence table shown in FIG. 10, the likelihood threshold for “other-sympathetic empathy” in label (1) and the likelihood threshold for “non-sympathy” in label (4) are as follows: “3000” and “2000” shown in FIG. 8A are set.

The instantaneous characteristic estimation unit 14 sets a histogram for each label. When the likelihood of the label having the maximum likelihood exceeds the likelihood threshold, the trend estimation unit 16 increments the numerical value of the histogram of the label having the maximum likelihood in step S15.

As a specific example, as shown in FIG. 8B, the likelihood of “other-person sympathy” in label (1) and the likelihood of “non-sympathy” in label (4) are obtained by the processing by the HMM algorithm. Assume that “2500” and “2300” are recognized. When the label having the maximum likelihood is the “other-friendliness sympathy” of the label (1) at the detection time t1, the likelihood “2500” of the label (1) does not exceed the likelihood threshold “3000”. Therefore (NO in step S14), the tendency estimation unit 16 does not increment the numerical value of the histogram of the label (1).

On the other hand, when the label having the maximum likelihood is “non-sympathetic” of the label (4) at the detection time t2, the likelihood “2300” of the label (4) exceeds the likelihood threshold “2000”. Therefore (YES in step S14), in step S15, the tendency estimation unit 16 increments the numerical value of the histogram of the label (4). Here, the histogram is an accumulation of the number of times that the likelihood exceeds the likelihood threshold. For example, the histogram is stored in a memory or the like.

FIG. 9A is an explanatory diagram showing the relationship between the frequency threshold and the frequency result. FIG. 9B is an explanatory diagram showing a frequency threshold and a histogram of frequency results. As described above, in step S15, the trend estimation unit 16 increments the numerical value of the histogram of the label with the maximum likelihood when the likelihood of the label with the maximum likelihood exceeds the likelihood threshold. Specifically, as shown in FIG. 8B, since the likelihood of “non-sympathy” of the label (4) exceeds the likelihood threshold, the tendency estimation unit 16 displays the “frequency” shown in FIGS. 9A and 9B. Increment the value of “Result”.

In step S16, the tendency estimation unit 16 determines whether or not the maximum frequency value appearing in the histogram exceeds a preset frequency threshold value. In the example shown in FIG. 9B, the frequency result of “non-sympathy” of the label (4) which is the maximum frequency value exceeds the frequency threshold value (that is, the frequency threshold value is “4”). Therefore, the tendency estimation unit 16 determines that the maximum frequency value appearing in the histogram exceeds the frequency threshold (YES in step S16).

In step S17, the tendency estimation unit 16 outputs “non-sympathy” of the label (4) as a static characteristic of the driver. That is, the tendency estimation unit 16 determines that the static characteristic of the driver who drives the vehicle is “non-sympathetic”, and outputs the determination result to a subsequent device (not shown). Note that the frequency result shown in the histogram of FIG. 9B increases with time and eventually exceeds the frequency threshold. In order to avoid this phenomenon, it is also possible to accumulate the frequency result for a certain period of time or every predetermined number of detections, and use a numerical value obtained by normalizing the accumulated data as the frequency result.

In the above description, the scene in which the vehicle travels on the “crosswalk without a signal on a narrow road” in the scene (1) in the correspondence table shown in FIG. 10 is taken as an example. Scene (1) is the situation shown in FIG. 11A. Hereinafter, the situation of scenes (2) to (8) in the correspondence table shown in FIG. 10 will be described with reference to FIGS. 11B to 12D.

FIG. 11B shows a situation in which the host vehicle passes through the “crosswalk without a stop line on a narrow road” in the scene (2). FIG. 11C shows a situation in which the vehicle makes a “left turn from a narrow road to a wide road” in the scene (3). FIG. 11D shows a situation in which the host vehicle passes through “a pedestrian crossing on a one-lane road with good visibility” in scene (4). FIG. 12A shows a situation where the host vehicle makes a “left turn from a wide road to a narrow road” in the scene (5). FIG. 12B shows a situation in which the host vehicle performs the “straight line in a congested narrow road” in the scene (6). FIG. 12C shows a situation in which the host vehicle travels in “Merge with heavy traffic” in scene (7). FIG. 12D shows a situation in which the host vehicle performs the “lane changing action in a traffic jam” of the scene (8).

The instantaneous characteristic estimator 14 includes labels associated with the scenes estimated by the scene estimator 13 among the scenes (1) to (8) (labels that are circled in the correspondence table shown in FIG. 10). ) For the likelihood.

As described above, the personal characteristic estimation device 100 associates an environmental model used for estimating a scene with a label (a driver's instantaneous characteristic) that is prominently expressed in the scene. The model is stored as a characteristic model and an instantaneous characteristic model. The personal characteristic estimation device 100 estimates a scene in which the host vehicle travels based on the vehicle information with reference to the environment model. The personal characteristic estimation device 100 refers to the characteristic model and calculates the likelihood of each label associated with the estimated scene using an algorithm of the HMM or the like based on information operated by the driver. When the likelihood of the label having the maximum likelihood exceeds the likelihood threshold, the personal characteristic estimation device 100 increments the numerical value of the histogram of the label having the maximum likelihood. When the maximum frequency value appearing in the histogram exceeds a preset frequency threshold, the personal characteristic estimation device 100 determines that the label having the maximum frequency value (driver's instantaneous characteristic) is the static characteristic of the driver. To do.

Accordingly, the personal characteristic estimation device 100 estimates the driver's static characteristics (generic name, tendency, temperament, etc.) with high accuracy based on the behavior that the driver can take while driving the vehicle and the surrounding environment of the vehicle. can do. As a result, for example, when performing guidance by HMI (Human Machine Interface) for this driver based on the estimated static characteristics of the driver, an appropriate response according to the static characteristics of the driver is taken. be able to.

For example, the following three guidances can be considered when the navigation device detects a traffic jam and searches for a detour: (1) Explain to the driver the reason for detouring and automatically set the detour (2) Let the driver select whether to take a detour or the original route; (3) Automatically set a detour without explaining anything to the driver. If the driver's static characteristic estimated by the personal characteristic estimation device 100 is used, an appropriate guidance can be selected from the guidances (1) to (3) according to the driver's static characteristic. Become.

In the conventional method for estimating the driver's static characteristics, the driver's static characteristics are estimated by a psychological method such as having a subject perform an AQ test using a questionnaire. However, the conventional method has a problem that a true answer cannot always be obtained, and further, personal information leaks. On the other hand, the personal characteristic estimation device 100 of the present embodiment estimates the static characteristics of the driver based on the behavior of the driver in each scene when driving the vehicle. The static characteristics of the driver can be estimated accurately at a low cost.

As shown in FIG. 2, the instantaneous characteristic model storage unit 15 stores a characteristic model that combines a scene and a characteristic (label). In the present embodiment, since the number of combinations of scenes and labels represented by circles in the correspondence table shown in FIG. 10 is relatively small, the personal characteristic estimation device 100 can reduce the storage capacity.

The signal detection unit 11 acquires at least one signal among the accelerator opening of the vehicle, the brake operation, the steering angle, the yaw rate, the vehicle speed, and the battery discharge level. Since the estimation unit 12 estimates a scene using the scene estimation unit 13 based on the signal acquired by the signal detection unit 11 and detects the driver's behavior using the instantaneous characteristic estimation unit 14, The characteristic estimation device 100 can estimate the static characteristic of the driver with high accuracy based on the vehicle signal.

Since the scene estimation unit 13 estimates the scene in which the host vehicle travels based on data such as road shape, time zone, traffic jam information, external conditions, presence / absence of pedestrians, presence / absence of vehicles, presence / absence of oncoming vehicles, etc. The personal characteristic estimation device 100 can estimate an appropriate scene according to the situation around the host vehicle, and can estimate the static characteristics of the driver with high accuracy.

The instantaneous characteristic estimator 14 uses a hidden Markov model (HMM) as a method for statistically analyzing the result estimated using the instantaneous characteristic model. The target characteristic can be estimated with high accuracy.

The personal characteristic estimation device 100 is a characteristic having a correlation of a predetermined value or more (for example, 0.2 or more) with respect to a characteristic detected by a conventional method of measuring a general and static characteristic tendency such as an AQ test. Is set as a characteristic tendency label, a label that can easily detect the static characteristics of the driver can be appropriately selected, and the static characteristics of the driver can be estimated with high accuracy.

The trend estimation unit 16 detects a label having the maximum likelihood every arbitrary time, accumulates the detected labels, forms a histogram, calculates a distribution for each label, and based on a label whose histogram value exceeds a threshold value Thus, since the static characteristic of the driver is estimated, the personal characteristic estimation device 100 can estimate the static characteristic of the driver with high accuracy.

(Second Embodiment)
FIG. 4 is a block diagram showing the configuration of the personal characteristic estimation apparatus according to this embodiment. As shown in FIG. 4, the personal characteristic estimation apparatus 101 includes a signal detection unit 11, a scene estimation unit 13, an instantaneous characteristic estimation unit 14, an instantaneous characteristic model storage unit 15, a trend estimation unit 16, and a model switching unit. (Model switching means) 17.

Since the signal detection unit 11 has the same configuration as the signal detection unit 11 shown in FIG.

The instantaneous characteristic model storage unit 15 stores a plurality of preset scenes (scenes (1) to (8) shown in FIGS. 11A to 12D). The instantaneous characteristic model storage unit 15 sets a plurality of characteristic tendency labels indicating the characteristics of the driver, an environmental model used for estimating the scene, and a characteristic tendency that can be easily expressed in each scene. A characteristic model associated with a label (hereinafter abbreviated as “label”) is stored as an instantaneous characteristic model. Specifically, the instantaneous characteristic model storage unit 15 estimates the scene as shown in the correspondence relationship in the correspondence table shown in FIG. 10 in which a plurality of scenes related to vehicle travel are taken in the row direction and a plurality of labels are taken in the column direction. And a characteristic model in which a label (a driver's instantaneous characteristic) prominently displayed in each scene is associated with each scene.

Furthermore, the instantaneous characteristic model storage unit 15 stores a plurality of characteristic models classified for each scene. Specifically, as shown in FIG. 5, label (1), label (4), and garbage (others) are stored as the characteristic model of scene (1). Label (2), label (6), and garbage are stored as a characteristic model of scene (2). Similarly, one or more labels associated with the scene are stored as a characteristic model of each scene after the scene 3.

As in the first embodiment, the scene estimation unit 13 is information about the external environment of the host vehicle detected by the signal detection unit 11 and vehicle signals (information about vehicle operation, information about vehicle behavior, and information about operation history). ), The scene in which the host vehicle travels is estimated with reference to the environmental model stored in the instantaneous characteristic model storage unit 15. For example, the scene estimation unit 13 grasps a situation in which the host vehicle is currently traveling based on a surrounding image captured by an in-vehicle camera (not shown), and estimates a scene. At this time, the scene estimation unit 13 refers to the environment model and corresponds to which scene among the scenes (1) to (8) indicated in the row direction of the correspondence table shown in FIG. Estimate.

The model switching unit 17 performs a process of selecting a desired characteristic model from a plurality of characteristic models stored in the instantaneous characteristic model storage unit 15 based on the scene estimated by the scene estimation unit 13. That is, as illustrated in FIG. 6, the model switching unit 17 performs a process of selecting a characteristic model associated with the scene estimated by the scene estimation unit 13 from a plurality of characteristic models. Specifically, when the current scene is “a crosswalk without a signal on a narrow road” in the scene (1) in the correspondence table shown in FIG. 10, the model switching unit 17 changes the scene (1) to the scene (1). The associated characteristic model including “other-person sympathy” of label (1), “non-sympathy” of label (4), and “garbage” is selected.

The instantaneous characteristic estimation unit 14 estimates the driver's instantaneous characteristic using the vehicle signal detected by the signal detection unit 11 and the characteristic model selected by the model switching unit 17. For example, as shown in FIG. 6, when the model switching unit 17 selects the scene 2 characteristic model, the label (2), label (6), and garbage stored as the scene 2 characteristic model The label with the maximum likelihood is extracted and output to the trend estimation unit 16. A method for extracting the label with the maximum likelihood will be described later.

Next, the operation of the personal characteristic estimation apparatus 101 will be described with reference to the flowchart shown in FIG.

First, in step S31, the signal detection unit 11 acquires various vehicle signals and vehicle external information. Specifically, depending on the CAN mounted on the vehicle, vehicle signals such as accelerator opening, brake operation, steering angle, vehicle speed, yaw rate, etc., external video signals captured by an in-vehicle camera, or GPS navigation Get information.

In step S <b> 32, the signal detection unit 11 determines the primary difference between the acquired signals at a certain time interval (this is referred to as a difference Δ) and the secondary difference between the primary differences of the acquired signal (this is referred to as a difference ΔΔ). Is calculated and parameterized.

In step S <b> 33, the scene estimation unit 13 refers to the environment model, and information on the outside of the vehicle detected by the signal detection unit 11 (specifically, measured by an outside image captured by the in-vehicle camera or a GPS device). Based on the position information of the own vehicle), the scene likelihood of the own vehicle is calculated, and the scene in which the own vehicle is traveling is estimated.

In step S34, the model switching unit 17 determines whether or not the scene estimated by the scene estimation unit 13 is garbage. If the estimated scene is garbage, the process returns to step S31. If the estimated scene is not garbage, the process proceeds to step S35.

In step S35, the model switching unit 17 selects a characteristic model associated with the estimated scene from among a plurality of characteristic models (see FIG. 5) respectively associated with the scenes (1) to (8). . For example, when the estimated scene is the scene (2) in the correspondence table shown in FIG. 10, the characteristic model associated with the scene (2) is selected as shown in FIG.

In step S36, the instantaneous characteristic estimation unit 14 calculates the likelihood (characteristic likelihood) of each label using an algorithm of HMM (Hidden Markov Model) using the difference Δ obtained in step S32 and the difference ΔΔ of the difference. calculate.

For example, when the acquired vehicle signal is the accelerator opening, when the accelerator opening is detected at each of the times t1, t2, t3, t4, and t5 having a certain time interval, the signal detection unit 11 determines that the time t1 , T2 of the accelerator opening obtained at time t2, the difference Δ2 of the accelerator opening obtained at time t2, t3 Δ2, the difference of accelerator opening obtained at time t3, t4, the accelerator opening obtained at time t4, t5 The difference Δ4 is calculated. Further, the signal detector 11 obtains a difference ΔΔ1 between the differences Δ1 and Δ2, a difference ΔΔ2 between the differences Δ2 and Δ3, and a difference ΔΔ3 between the differences Δ3 and Δ4. The instantaneous characteristic estimator 14 calculates the likelihood for each label by substituting these data into the HMM algorithm.

As a specific example, the instantaneous characteristic estimating unit 14 causes the scene estimating unit 13 to drive the host vehicle on the “crosswalk without a signal on a narrow road” of the scene (1) in the correspondence table shown in FIG. If it is estimated that there is, the likelihood of “other-person sympathy” in label (1) and the likelihood of “non-sympathy” in label (4) are calculated.

In step S37, the instantaneous characteristic estimation unit 14 determines whether or not the likelihood of the label having the maximum likelihood exceeds a preset threshold value. For example, in the case of scene (1) in the correspondence table shown in FIG. 10, the likelihood threshold for “other-sympathetic empathy” for label (1) and the likelihood threshold for “non-sympathy” for label (4) are , “3000” and “2000” shown in FIG. 8A, respectively.

The instantaneous characteristic estimation unit 14 sets a histogram for each label. When the likelihood of the label having the maximum likelihood exceeds the likelihood threshold (YES in step 37), in step S38, the tendency estimation unit 16 sets the numerical value of the histogram of the label having the maximum likelihood. Is incremented.

As a specific example, as shown in FIG. 8B, the likelihood of “other-person sympathy” in label (1) and the likelihood of “non-sympathy” in label (4) are obtained by the processing by the HMM algorithm. , “2500” and “2300”. When the label having the maximum likelihood is the “other-friendliness sympathy” of the label (1) at the detection time t1, the likelihood “2500” of the label (1) does not exceed the likelihood threshold “3000”. Therefore (NO in step S37), the tendency estimation unit 16 does not increment the numerical value of the histogram of the label (1).

On the other hand, when the label having the maximum likelihood is “non-sympathetic” of the label (4) at the detection time t2, the likelihood “2300” of the label (4) exceeds the likelihood threshold “2000”. Therefore (YES in step S37), in step S38, the tendency estimation unit 16 increments the numerical value of the histogram of the label (4). Here, the histogram is an accumulation of the number of times that the likelihood exceeds the likelihood threshold. For example, the histogram is stored in a memory or the like.

FIG. 9A is an explanatory diagram showing the relationship between the frequency threshold and the frequency result. FIG. 9B is an explanatory diagram showing a histogram of frequency results. As described above, in step S37, the trend estimation unit 16 increments the numerical value of the histogram of the label with the maximum likelihood when the likelihood of the label with the maximum likelihood exceeds the likelihood threshold. Specifically, as shown in FIG. 8B, since the likelihood of “non-sympathy” of the label (4) exceeds the likelihood threshold, the trend estimation unit 16 performs the “frequency” shown in FIGS. 9A and 9B. Increment the value of “Result”.

In step S39, the trend estimation unit 16 determines whether or not the normalized maximum frequency value appearing in the histogram exceeds a preset frequency threshold value. In the example shown in FIG. 9B, the frequency result of “non-sympathy” of the label (4) which is the normalized maximum frequency value exceeds the frequency threshold value (that is, the frequency threshold value is “4”). On the other hand, since the numerical value of the frequency result is “6”), the tendency estimation unit 16 determines that the normalized maximum frequency value appearing in the histogram exceeds the frequency threshold (YES in step S39). ).

In step S40, the tendency estimation unit 16 outputs “non-sympathy” of the label (4) as a static characteristic of the driver. That is, the tendency estimation unit 16 determines that the static characteristic of the driver who drives the vehicle is “non-sympathetic”, and outputs the determination result to a subsequent device (not shown).

As described above, the personal characteristic estimation device 101 associates an environmental model used for estimating a scene with a label (a driver's instantaneous characteristic) that is prominently expressed in the scene. The model is stored as a characteristic model and an instantaneous characteristic model. The personal characteristic estimation apparatus 101 estimates a scene in which the host vehicle travels based on the vehicle information with reference to the environment model. The personal characteristic estimation device 101 refers to the characteristic model, recognizes the label associated with the estimated scene, and uses the HMM algorithm or the like based on the information operated by the driver to estimate the likelihood of each label. Calculate the degree. When the likelihood of the label having the maximum likelihood exceeds the likelihood threshold, the personal characteristic estimation apparatus 101 increments the numerical value of the histogram of the label having the maximum likelihood. When the normalized maximum frequency value appearing in the histogram exceeds a preset frequency threshold, the personal characteristic estimation device 101 displays the label (the driver's instantaneous characteristic) having the normalized maximum frequency value as the driver's static value. Judged to be characteristic.

Therefore, the personal characteristic estimation apparatus 101 estimates the driver's static characteristics (generic name, tendency, temperament, etc.) with high accuracy based on the behavior that the driver can take while driving the vehicle and the surrounding environment of the vehicle. can do. As a result, for example, when performing guidance based on the estimated driver static characteristics based on the HMI (Human_Machine_Interface) for the driver, an appropriate response corresponding to the driver static characteristics may be taken. it can.

The model switching unit 17 selects a characteristic model associated with the scene estimated by the scene estimation unit 13 from a plurality of characteristic models (see FIG. 5), and the instantaneous characteristic estimation unit 14 selects the selected characteristic model. Based on this, since the likelihood of each label is calculated, the personal characteristic estimation device 101 can reduce the calculation load when calculating the likelihood of each label.

The trend estimator 16 accumulates and normalizes the label detected by the instantaneous characteristic estimator 14 at any given time and having the maximum likelihood for a predetermined time or every predetermined number of detections. Create a histogram and calculate the distribution of each label. Therefore, the personal characteristic estimating apparatus 101 converts the numerical value of the histogram into an appropriate numerical value and estimates the driver's static characteristic, so that the driver's static characteristic can be estimated with high accuracy. .

As described above, the personal characteristic estimation apparatus and personal characteristic estimation method of the present invention are not limited to the first and second embodiments, and can be appropriately changed without departing from the gist of the present invention.

For example, in the first and second embodiments, the HMM (Hidden Markov Model) is used as a method for statistically analyzing the result of estimating the instantaneous characteristics of the driver, but the present invention is not limited to this. Instead, it is possible to use other techniques having equivalent functions.

The present invention can be used to estimate personal characteristics based on behavior during vehicle driving.

This application claims priority based on Japanese Patent Application No. 2012-190952 filed on August 31, 2012, the entire contents of which are incorporated herein by reference.

DESCRIPTION OF SYMBOLS 11 Signal detection part 12 Estimation part 13 Scene estimation part 14 Instantaneous characteristic estimation part 15 Instantaneous characteristic model memory | storage part 16 Trend estimation part 17 Model switching part 100,101 Personal characteristic estimation apparatus

Claims

In the personal characteristic estimation device that estimates the characteristics of the driver based on the driving operation of the vehicle,
Vehicle information acquisition means for acquiring, as vehicle information, at least one of information relating to operations by the driver, information relating to vehicle behavior, and information relating to the situation of the outside world;
Scene estimation means for estimating a scene in which the host vehicle travels based on the vehicle information;
Set up multiple scenes, set up multiple characteristic trend labels that indicate the instantaneous characteristic trends of the driver, and associate an instantaneous characteristic model with a characteristic model that associates characteristic trend labels that are easy to express in each scene. Instantaneous characteristic model storage means for storing as,
Based on the scene estimated by the scene estimation means, referring to the instantaneous characteristic model, the instantaneous characteristic estimation means for estimating the instantaneous characteristic tendency of the driver;
Statistical analysis of the instantaneous characteristic trend estimated by the instantaneous characteristic estimation means, characteristic tendency estimation means for estimating a static characteristic tendency of the driver,
An apparatus for estimating personal characteristics, comprising:
2. The vehicle information acquisition unit according to claim 1, wherein the vehicle information acquisition unit acquires at least one of a vehicle accelerator opening, a brake operation, a steering angle, a yaw rate, a vehicle speed, and a battery discharge level. Personal characteristic estimation device.
Model switching means for selecting a characteristic model classified into the scene estimated by the scene estimation means from the plurality of characteristic models classified for each scene in the instantaneous characteristic model storage means, The personal characteristic estimation apparatus according to claim 1, wherein the personal characteristic estimation apparatus is characterized.
The plurality of scenes are classified based on at least one of road shape, time zone, traffic jam information, external state, presence / absence of pedestrians, presence / absence of vehicles, and presence / absence of oncoming vehicles. The personal characteristic estimation device according to any one of claims 1 to 3.
5. The individual according to claim 1, wherein a hidden Markov model is used as a method of statistically analyzing the instantaneous characteristic tendency estimated by the instantaneous characteristic estimation unit. Characteristic estimation device.
The characteristic trend label is a characteristic having a correlation of a predetermined value or more as compared with a characteristic detected by means for measuring a general and static characteristic tendency. Item 6. The personal characteristic estimation device according to any one of Items 5 to 6.
2. The characteristic trend estimating means detects a label having the maximum likelihood at every arbitrary time, accumulates the detected label, forms a histogram, and calculates a distribution of each label. The personal characteristic estimation device according to any one of claims 6 to 6.
The characteristic tendency estimation means calculates a distribution of each label by histogramating a normalized value by accumulating a label having the maximum likelihood for every predetermined time or every predetermined number of detection times. The personal characteristic estimation device according to any one of claims 1 to 6, characterized in that:
In the personal characteristic estimation method for estimating the characteristics of the driver based on the driving operation of the vehicle,
Set multiple scenes,
Set multiple characteristic trend labels to indicate the driver's instantaneous characteristic trends,
For each scene, store a characteristic model associating characteristic tendency labels that are easy to express in the scene as an instantaneous characteristic model,
Obtaining at least one of information related to operations by the driver, information related to vehicle behavior, and information related to the situation of the outside world as vehicle information;
Estimating a scene in which the vehicle travels based on the vehicle information,
Based on the estimated scene, referring to the instantaneous characteristic model, an instantaneous characteristic tendency of the driver is estimated,
A personal characteristic estimation method characterized by statistically analyzing the estimated instantaneous characteristic tendency to estimate a static characteristic tendency of a driver.