Analysis device and analysis method

Abstract

To present a driving route to a driving tendency for each user.SOLUTION: An analysis device 1 includes: a setting information acquisition section 10 for acquiring a departing spot and a destination spot of a vehicle 2; an output symbol acquisition section 12 for acquiring output symbol series including a history of operation of a travel direction in each point (p) that is passed until the vehicle 2 reaches a destination spot; a parameter storage section 16 for storing a learned Hidden Markov Model in which parameters of the Hidden Markov Model are estimated in advance, where each point (p) that the vehicle 2 passes from a departing spot to a destination spot is set as a finite set of a hidden state, and operation of a travel direction by the vehicle 2 in each point (p) is set as a finite set of observable output symbols; and a calculation section 14 for allowing the vehicle 2 to obtain a probability that each operation in a travel direction acquired by the output symbol acquisition section 12 is performed in each point (p) by using a learned Hidden Markov Model.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an analysis apparatus and an analysis method, and more particularly to a technique for analyzing a driving route of a vehicle.

Background Art

[0002] Conventionally, a car navigation system mounted on a vehicle or the like has been known. For example, Patent Document 1 discloses a technique for setting a driving route that minimizes the distance from the current position of a vehicle to a destination or a driving route that reaches the destination in the shortest time, and guiding the vehicle along the route. Further, in Patent Document 1, a route is set such that the vehicle can reach the destination more efficiently, taking into account the presence or absence of traffic lights, intersections, and the ease of access of the vehicle to facilities.

[0003] Here, a user who is a driver of a vehicle may frequently drive using a shortcut or the like that is different from the road along the guidance, with respect to the guidance route uniformly set based on the shortest time or the shortest distance. However, in the route setting by the car navigation system disclosed in Patent Document 1, it was not possible to present an individualized driving route according to the driving tendency of each user.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] As described above, according to the conventional technology, it was not possible to present a driving route that matches the driving tendency of each user.

[0006] The present invention has been made to solve the above-described problems, and an object thereof is to present a driving route that matches the driving tendency of each user.

Means for Solving the Problem

[0007] In order to solve the above problems, the analysis apparatus according to the present invention includes a setting information acquisition unit configured to acquire a departure point and a destination point of a vehicle, and an output symbol acquisition unit configured to acquire an output symbol sequence including a history of operations of the vehicle in the traveling direction at each point passed by the vehicle until the vehicle reaches the destination point. The analysis apparatus further includes a storage unit configured to store a learned hidden Markov model in which each point passed by the vehicle from the departure point to the destination point is a finite set in a hidden state, and an output symbol that can observe the operation of the vehicle in the traveling direction at each point is a finite set, and the parameters of the hidden Markov model have been estimated in advance. The analysis apparatus also includes an arithmetic unit configured to use the learned hidden Markov model to obtain the probability that each of the operations in the traveling direction acquired by the output symbol acquisition unit is performed at each point of the vehicle, and a determination unit configured to determine a traveling route to the destination point according to the operation in the traveling direction with the highest probability of being performed by the vehicle among the operations in the traveling direction based on the probability that each of the operations in the traveling direction obtained by the arithmetic unit is performed. The analysis apparatus further includes a presentation unit configured to present the determined traveling route.

[0008] Further, in the analysis apparatus according to the present invention, the setting information acquisition unit may acquire a plurality of different traveling routes by which the vehicle can reach the destination point from the departure point, and each of the plurality of different traveling routes may be specified by a plurality of points.

[0009] Also, in the analyzer according to the present invention, the output symbol acquisition unit acquires, as learning data, a series of output symbols indicating the operation in the traveling direction at each point by the vehicle, and further, a state transition probability distribution that is the probability of the vehicle moving from a predetermined point to another point and that maximizes the likelihood with respect to the learning data, and a symbol output probability distribution that is the probability of the vehicle performing a predetermined operation in the traveling direction at each point. The learning unit is configured to estimate the parameters of the hidden Markov model including the above, and the learned hidden Markov model may include the parameters estimated by the learning unit.

[0010] Also, in the analyzer according to the present invention, the operation in the traveling direction may include a plurality of preset steering operations in the vehicle.

[0011] In order to solve the above-described problems, an analysis method according to the present invention includes a setting information acquisition step of acquiring a departure point and a destination point of a vehicle, an output symbol acquisition step of acquiring a series of output symbols including a history of an operation in the traveling direction by the vehicle at each point passed until the vehicle reaches the destination point, a storage step of storing, in a storage unit, a learned hidden Markov model in which each point passed by the vehicle from the departure point to the destination point is a finite set of hidden states and each operation in the traveling direction by the vehicle at each point is a finite set of observable output symbols, and the parameters of the hidden Markov model have been estimated in advance, an arithmetic step of obtaining, using the learned hidden Markov model, the probability that each of the operations in the traveling direction acquired in the output symbol acquisition step is performed at each point by the vehicle, and a determination step of determining a travel route to the destination point according to the operation in the traveling direction having the highest probability of being performed by the vehicle among the operations in the traveling direction, based on the probability that each of the operations in the traveling direction obtained in the arithmetic step is performed, and a presentation step of presenting the determined travel route.

[0012] Further, in the analysis method according to the present invention, in the setting information acquisition step, a plurality of different driving routes by which the vehicle can reach the destination from the departure point are acquired, and each of the plurality of different driving routes may be specified by a plurality of points.

[0013] Further, in the analysis method according to the present invention, in the output symbol acquisition step, a series of output symbols indicating the operation of the traveling direction at each point by the vehicle is acquired as learning data, and further, the state transition probability distribution, which is the probability that the vehicle moves from a predetermined point to another point and maximizes the likelihood for the learning data, and the symbol output probability distribution, which is the probability that the vehicle performs a predetermined operation of the traveling direction at each point, are included, and a learning step of estimating the parameters of the hidden Markov model is provided, and the learned hidden Markov model may include the parameters estimated in the learning step.

Advantages of the Invention

[0014] According to the present invention, using a learned hidden Markov model, the probability that the vehicle performs each of the operations of the traveling direction at each point is obtained, and based on the obtained probability, the driving route to the destination point corresponding to the operation of the traveling direction with the highest probability of the operation by the vehicle at each point is determined. Therefore, it is possible to present a driving route that matches the driving tendency of each user.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

[0016] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to FIGS. 1 to 6.

[0017] [Configuration of Analysis System] FIG. 1 is a block diagram showing the configuration of an analysis system including an analyzer 1 according to an embodiment of the present invention. The analysis system according to the present embodiment analyzes the tendency of a driving route by the vehicle 2, determines an optimal driving route based on the analysis result, and presents it.

[0018] The analysis system includes an analyzer 1 and a vehicle 2. The analyzer 1 shown in FIG. 1 is configured as an in-vehicle device provided in the vehicle 2, but may be configured as a cloud system or the like provided on the network side.

[0019] The vehicle 2 includes an automobile, a motor vehicle, a motorcycle, etc. The vehicle 2 includes a steering wheel 2a, a sensor 20, a GPS module 21, an ECU (Electronic Control Unit) 22, a display device 23, a communication interface 24, and a map database 25. The steering wheel 2a constitutes a steering mechanism for operating the traveling direction of the vehicle 2. Hereinafter, the case where the vehicle 2 is manually driven by a driver (user) will be illustrated and described.

[0020] When the steering wheel 2a is rotated by the driver of the vehicle 2, the rotational motion is converted into a reciprocating motion in the left-right direction of the vehicle 2, and the angle of the front wheels of the vehicle 2 is changed. The traveling direction of the vehicle 2 is operated by changing the angle of the front wheels of the vehicle 2 according to the amount of rotation of the steering wheel 2a.

[0021] In this embodiment, as examples of operations in the traveling direction of the vehicle 2, three types of operations are used: an operation (right rotation) of the steering wheel 2a for turning the vehicle 2 to the right, an operation (left rotation) of the steering wheel 2a for turning the vehicle 2 to the left, and a straight-ahead operation (no rotation) in which the traveling direction does not change. The rotation angles of the steering wheel 2a for right rotation and left rotation and the rotation angles of the front wheels with respect to the rotation angles are defined in advance.

[0022] The sensor 20 includes various sensors such as a geomagnetic sensor, a gyro sensor, an acceleration sensor, a vehicle speed sensor, a camera, and LiDAR. The direction, tilt angle, moving distance, etc. of the vehicle 2 are detected by the sensor 20 and used to specify the position of the vehicle 2. In addition, the vehicle 2 includes a GPS module 21 having a GPS function. Based on the absolute position and time of the GPS satellite received by the GPS module 21, the current position of the vehicle 2 is specified.

[0023] The ECU 22 executes steering control, drive control, brake control, car navigation, etc. of the vehicle 2 by software incorporated in the ECU 22. In this embodiment, the ECU 22 cooperates with the analysis device 1 via the in-vehicle network NW. The ECU 22 is composed of various ECUs such as a steering ECU, a drive ECU, a brake ECU, and a car navigation ECU, and each ECU cooperates via the in-vehicle network NW. Also, in this embodiment, the history of the steering operation of the steering wheel 2a by the driver is stored in a memory (not shown) of the vehicle 2.

[0024] The sensor 20, GPS module 21, ECU 22, display device 23, communication interface 24, and map database 25 provided in the vehicle 2 constitute the car navigation system 2A. The analysis device 1 cooperates with the ECU 22 (car navigation ECU) that controls the car navigation system 2A via the in-vehicle network NW. The car navigation system 2A includes a display and a touch panel as the display device 23. The car navigation system 2A further includes a speaker, a connector, etc. (not shown).

[0025] In the present embodiment, the car navigation system 2A provides the analysis device 1 with the vehicle position, destination point, and map data of the vehicle 2. Further, the car navigation system 2A can perform a map matching process of reflecting the driving route and the vehicle position determined by the analysis device 1 on the map data in the map database 25. The analysis device 1 can cause the ECU 22 that controls the car navigation system 2A, that is, the car navigation ECU, to display the determined driving route on the map data of the display device 23.

[0026] The map database 25 stores, for example, road map data of the whole of Japan and facility data such as various facilities and stores associated therewith. For example, the map data can be composed of a road network representing roads on the map as lines, and can be provided as section data obtained by dividing the roads into a plurality of sections based on intersections, branch points, etc. Each section includes position information consisting of the longitude and latitude of the start point and the end point, length, data such as the width and type of the road. Note that the map database 25 may be a configuration provided in the analysis device 1.

[0027] As shown in Fig. 1, vehicle 2 travels on road R registered in map database 25, and travels from the starting point "START", which is the current position, to the destination point "GOAL" while passing through point p arranged along road R. As shown in the example of Fig. 1, point p can be arranged at fixed positions such as for each intersection or branch point, or for one or more sections obtained by dividing the road into a plurality of sections. Alternatively, each position on road R reached by vehicle 2 at set time intervals (for example, 1 to several seconds) can be set as point p. Also, when setting by time intervals, the time interval can be arbitrarily set according to the speed of vehicle 2 and the like. In any case, the position of point p on the map including longitude and latitude is specified.

[0028] The current position of vehicle 2 is assumed to be due to any one of the steering operations of right turn, left turn, or no turn in vehicle 2 at the previous point p. In the following, for simplicity of explanation, it is assumed that when the vehicle makes a right turn operation at the previous point p, it always reaches a certain specific point p. In the example of Fig. 1, when a right turn steering operation is performed at point p "1" which is the next point reached from the starting point "START", vehicle 2 reaches point p "2", and a left turn steering operation is further performed at point p "2". Thereafter, vehicle 2 reaches point p "3". In this way, the steering operations at each point p by vehicle 2 are sequentially performed, and finally the destination point is reached.

[0029] In the analysis system according to this embodiment, using a hidden Markov model as a machine learning model, at each point p included in all the driving routes by which vehicle 2 can reach from the starting point to the destination point, the steering operation with the highest probability performed by vehicle 2 is obtained. Further, based on the probabilities of the right turn, no turn, and left turn steering operations at each point p, the point p passed through when going from the starting point to the destination point is specified, and a recommended driving route is determined and presented.

[0030] [Function Blocks of Analysis Device] The analysis device 1 includes a setting information acquisition unit 10, a setting information storage unit 11, an output symbol acquisition unit 12, a learning unit 13, an arithmetic unit 14, a determination unit 15, a parameter storage unit (storage unit) 16, and a presentation unit 17.

[0031] The setting information acquisition unit 10 acquires the departure point and the destination point of the vehicle 2. Further, the setting information acquisition unit 10 acquires a plurality of different driving routes by which the vehicle 2 can reach the destination point from the departure point. More specifically, the setting information acquisition unit 10 acquires the information of the destination point input in the car navigation system 2A of the vehicle 2 via the network NW. Further, the setting information acquisition unit 10 can acquire, via the in-vehicle network NW, the current position of the own vehicle specified by the data of the sensor 20 and the GPS module 21 of the vehicle 2 as the departure point.

[0032] The setting information acquisition unit 10 acquires all the driving routes from the position of the departure point to the position of the destination point based on the map data of the map database 25. For example, all the driving routes can be acquired from the car navigation system 2A via the in-vehicle network NW. Further, the setting information acquisition unit 10 specifies each of all the driving routes as a plurality of points p through which the vehicle 2 passes. As described above, the point p can be, for each divided section of a predetermined road R, for each intersection or branch point, or the position where the vehicle 2 arrives at a predetermined time interval (for example, every 1 to several seconds).

[0033] The setting information storage unit 11 stores the setting information of the point p, the departure point, and the destination point. Further, the setting information storage unit 11 stores the map data of the map database 25 in association with the identification information and the position information of each point p through which the vehicle 2 passes.

[0034] The output symbol acquisition unit 12 acquires an output symbol sequence including the history of the steering operation (operation of the traveling direction) by the vehicle 2 at each point p passed by the vehicle 2 until it reaches the destination. Further, the output symbol acquisition unit 12 acquires, as learning data, an output symbol sequence indicating the steering operation by the vehicle 2 at each point p. Specifically, the output symbol acquisition unit 12 acquires the history of the steering operation recorded in the vehicle 2 via the in-vehicle network NW.

[0035] Based on the history of the steering operation acquired by the output symbol acquisition unit 12, in the driving process including each point p passed by the vehicle 2, it is possible to grasp, as observation data, which steering operation among right rotation, no rotation, and left rotation was performed.

[0036] The learning unit 13 estimates the parameters of a hidden Markov model including a state transition probability distribution A, which is the probability that the vehicle 2 moves from a predetermined point to another point and maximizes the likelihood with respect to the learning data acquired by the output symbol acquisition unit 12, and a symbol output probability distribution B, which is the probability that the vehicle 2 performs a predetermined steering operation at each point p.

[0037] As described above, the analysis device 1 according to the present embodiment employs a hidden Markov model (Hidden Markov Model: HMM) as a machine learning model. The hidden Markov model is a model that probabilistically captures an output symbol, which is an observation value depending on a state variable that changes according to a Markov process. That is, it is a model in which a probability variable that moves between a plurality of states can estimate which state it is in at each time point.

[0038] More specifically, the learning unit 13 uses a hidden Markov model in which each point p passed at each predetermined time step due to a change in the traveling direction of the vehicle 2 is a finite set of hidden states, and the steering operation performed by the vehicle 2 at each point p is a finite set of observable output symbols.

[0039] (a) of FIG. 2 is a diagram for explaining the hidden Markov model adopted in the present embodiment. As shown in (a) of FIG. 2, in the present embodiment, a Left-to-Right HMM is used. The Left-to-Right HMM is a model based on the assumption that the state always transitions from left to right and has no ergodicity such that it cannot return to the previous state when transitioning to the next state. Each circle shown in (a) of FIG. 2 represents all the points p that the vehicle 2 can pass through from the starting point of the current position to the destination point. The point p passed by the vehicle 2 represents a hidden state that cannot be directly observed, and is represented by the set of states S = {S1, …, S2, …, S n}

[0040] Also, in the example of (a) of FIG. 2, the vehicle 2 moves by passing through each point p from the current position or the starting point, which is the initial state, to the destination point, which is the final state. The loop of each state indicates a self-loop, for example, indicating that the vehicle 2 is stopped. Further, in (a) of FIG. 2, at each point p where the vehicle 2 is in a hidden state, the probabilities of performing a right rotation, no rotation, and left rotation steering operation, that is, the output probabilities of the output symbol sequence that can be uniquely observed, are b i (1), b i (2), …, b i (k). Note that in the present embodiment, as described above, three types of K = {1 (right rotation), 2 (no rotation), 3 (left rotation)} are set. That is, the output probability of the output symbol sequence is the probability that the vehicle 2 performs each of the k = 3 types of preset steering operations at each point p where the state is S i .

[0041] In this way, it can be considered that a steering operation represented by the symbol O is output from the point p represented by the state S. Note that the state transition diagram in (a) of FIG. 2 is shown as a trellis diagram such that, corresponding to the three types of steering operations (k = 3), operations of k = 1 (right rotation), 2 (no rotation), and 3 (left rotation) can be performed as indicated by three arrows from each state.

[0042] The parameters of the hidden Markov model include, in addition to the set S of the above-mentioned hidden states and the types K = {1, 2, …, k} of output symbols, the set O = {O1, O2, …, O m}, the set π = {π i} of initial state probabilities, the set of final states, the set A = {a i,j} of state transition probabilities, and the set B = {b i (O t )}.

[0043] The sum of the initial state probabilities π i satisfies 1 as expressed by the following equation (1). This is because in a Left-to-Right HMM, it starts only from the initial state i = 0.

Equation

[0044] Regarding the final state, in a Left-to-Right HMM, there is only one final state.

[0045] The state transition probability a i,j is the probability of transitioning from state S i to state S j , and the sum of the state transition probabilities to all possible states S i from which a transition can occur to state S j satisfies 1 as shown by the following equation (2).

Equation

[0046] The symbol output probability b i (O t ) indicates the probability of outputting the output symbol O i in state S t , and the sum of the symbol output probabilities of all symbols that can be output when transitioning from a certain state S i satisfies 1 as shown by the following equation (3). Note that the output symbol O t is the output symbol observed at time t. [Number]

[0047] The hidden Markov model defined as above is represented as λ = (A, B, π). The learning unit 13 estimates the parameters of the state transition probability distribution A and the symbol output probability distribution B of the hidden Markov model λ. In the hidden Markov model, since the hidden state sequence cannot be directly observed from the output symbol sequence, it is difficult to perform direct maximum likelihood estimation. Therefore, the learning unit 13 estimates these parameters by iterative operations based on the maximum likelihood method called the EM (Expectation-Maximization) algorithm. The learning unit 13 uses the Baum-Welch algorithm, which is well-known as one of the EM algorithms adapted for estimating the parameters of the hidden Markov model, to estimate the parameters of the state transition probability distribution A and the symbol output probability distribution B.

[0048] The learning unit 13 executes the learning steps of the Baum-Welch algorithm from the following Step 1 to Step 6.

[0049] [Step 1] First, the learning unit 13 sets the state transition diagram shown in Fig. 2(a) based on all the driving routes based on the current position of the vehicle or the starting point and the destination point obtained by the setting information acquisition unit 10, and the points p passed by the vehicle 2 on each driving route.

[0050] [Step 2] Next, the learning unit 13 sets the initial values of the state transition probability distribution A and the symbol output probability B. The learning unit 13 can use arbitrary values as the initial values.

[0051] [Step 3] Next, the output symbol acquisition unit 12 obtains the output symbol sequence O of the learning data t ={O1, O2, …, O mDetermine {}. The training data is a series of output symbols of the steering operations performed by the vehicle 2 at each point p, which is acquired by the output symbol acquisition unit 12. As shown in FIG. 3, the output symbol acquisition unit 12, during the process of the vehicle 2 traveling from the starting point to the destination point, takes the symbol output series O at time t as the operation history of the steering operations performed at each predetermined time step. t Determine. Note that the time step can be set as an arbitrary time interval. Also, the learning unit 13 may be configured to determine the symbol output series O t .

[0052] [Step 4] Next, the learning unit 13 defines the value of grid, which is a variable of the Forward algorithm, as α t (i) and calculates each α t according to the following formula (4). [Equation] In the above formula (4), j = 1, 2,..., n - 1.

[0053] [Step 5] Next, the learning unit 13 defines the value of grid, which is a variable of the Backward algorithm, as β t (t) and calculates each β t according to the following formula (5). [Equation] Here, j = 1, 2,..., n - 1.

[0054] [Step 6] Subsequently, the learning unit 13 calculates the transition probability Γ i from the state S at time t j to the state S t (i, j), and recalculates the state transition probability a i,j and the symbol output probability b j (O). [Equation] In the above formula (8), t ∈ 0 indicates that the symbol output O at time t is 0. t indicates that it is 0.

[0055] After that, the learning unit 13 repeats the above steps from the 4th step to the 6th step, takes the point where the parameters do not change or the likelihood does not change as the convergence point, and adopts the state transition probability distribution A and the symbol output probability distribution B at that time as the estimated values. In the Baum-Welch algorithm, in the 6th step, the transition probability Γ t (i, j) is the state transition probability a i,j and the symbol output probability b j (O) The calculation procedure corresponds to the expectation (E) step of the EM algorithm. Also, in the 6th step, the procedure of recalculating the symbol output probability b j (O) from the transition probability Γ t (i, j) corresponds to the maximization (M) step of the EM algorithm.

[0056] Returning to FIG. 1, the calculation unit 14 uses the learned hidden Markov model to obtain the probability that each of the steering operations acquired by the output symbol acquisition unit 12 is performed at each point p of the vehicle 2.

[0057] More specifically, as shown in FIG. 2(b), for example, the calculation unit 14 calculates the symbol output probability b i for each state S of each of the three types of output symbols, i.e., right rotation, no rotation, and left rotation steering operations. i (1), b i (2), b iObtain (3). For example, at point p corresponding to state S1, the probabilities that vehicle 2 performs a right - turning steering operation, a non - rotating steering operation, and a left - turning steering operation are obtained as b1(1) = 0.3, b1(2) = 0.6, and b1(3) = 0.1 respectively. In this way, the arithmetic unit 14 uses the learned hidden Markov model to obtain the probability of vehicle 2 performing each steering operation at each point p of all the driving routes that can reach the destination point.

[0058] Based on the probabilities of vehicle 2 performing each of the steering operations at each point p obtained by the arithmetic unit 14, the determination unit 15 determines the driving route to the destination point according to the steering operation with the highest probability of being operated by vehicle 2 among the three types of steering operations. For example, as shown in the state transition diagram of (a) in FIG. 2 and the probability values in each state of (b), the determination unit 15 obtains the probabilities of the right - turning, non - rotating, and left - turning steering operations at point p when i = 0 as b0(1) = 0.7, b0(2) = 0.2, and b0(3) = 0.1.

[0059] In this case, since the probability b0(1) = 0.7 of the steering operation related to the right - turning is the highest, the determination unit 15 determines that for vehicle 2 in the next time step, among the probabilities of each steering operation at the point p (for example, S1) reached when moving rightward from the previous point p, the steering operation with the highest probability (b1(2) = 0.6), that is, the non - rotating (going straight) route, is determined.

[0060] The determination unit 15 can determine an individualized driving route according to the selection tendency of the driving route in vehicle 2 from the points p passed by vehicle 2 from the starting point to the destination point and the steering operations at each point p.

[0061] The parameter storage unit 16 stores the learned Markov model whose parameters are estimated by the learning unit 13.

[0062] The prompting unit 17 prompts the driving route of the vehicle 2 determined by the determination unit 15. For example, the prompting unit 17 can notify the driving route with the positions of each point p reflected in the map data to the car navigation system 2A of the vehicle 2 via the in-vehicle network NW. Further, the ECU 22 can cause the position and the driving route of the vehicle 2 to be displayed on the map data and output from the display device 23. Further, the car navigation system 2A of the vehicle 2 can display, as one of a plurality of selectable driving routes for the user, the driving route based on the tendency of the user's driving determined by the determination unit 15 together with the driving route calculated uniformly based on the shortest distance or the shortest time to the destination point.

[0063] [Hardware Configuration of the Analyzer] Next, an example of the hardware configuration for realizing the analyzer 1 having the functions described above will be described with reference to FIG. 4.

[0064] As shown in FIG. 4, the analyzer 1 is, for example, a computer including a processor 102, a main storage device 103, a communication interface 104, an auxiliary storage device 105, and an input / output I / O 106 connected via a bus 101, and can be realized by a program for controlling these hardware resources.

[0065] The main storage device 103 stores in advance a program for the processor 102 to perform various controls and calculations. The functions of the analyzer 1 such as the setting information acquisition unit 10, the output symbol acquisition unit 12, the learning unit 13, the calculation unit 14, and the determination unit 15 shown in FIG. 1 are realized by the processor 102 and the main storage device 103.

[0066] The communication interface 104 is an interface circuit for network-connecting the analyzer 1 and various external electronic devices.

[0067] The auxiliary storage device 105 is composed of a readable and writable storage medium and a driving device for reading and writing various information such as programs and data to and from the storage medium. As the storage medium of the auxiliary storage device 105, a semiconductor memory such as a hard disk or a flash memory can be used.

[0068] The auxiliary storage device 105 has a program storage area for storing the learning program executed by the analysis device 1. The setting information storage unit 11 and the parameter storage unit 16 described with reference to FIG. 1 are realized by the auxiliary storage device 105. Furthermore, for example, it may have a backup area for backing up the above-described data, programs, and the like.

[0069] The input / output I / O 106 is an input / output device that inputs signals from external devices and outputs signals to external devices.

[0070] [Operation of the analysis device] Next, the operation of the analysis device 1 having the above-described configuration will be described with reference to the flowcharts of FIGS. 5 and 6. FIG. 5 is a flowchart showing the learning process of the hidden Markov model by the analysis device 1. FIG. 6 is a flowchart showing the arithmetic process using the learned hidden Markov model whose parameters have been estimated by the analysis device 1.

[0071] First, with reference to FIG. 5, the learning process by the analysis device 1 will be described. First, the setting information acquisition unit 10 acquires setting information including the current position or the departure point and the destination point of the vehicle 2 (step S1). Specifically, the setting information acquisition unit 10 can acquire the destination point set in the car navigation system 2A and the current position of the specified vehicle via the in-vehicle network NW.

[0072] Next, the setting information acquisition unit 10 sets the entire driving route by referring to the map data of the setting information storage unit 11 and the map database 25 of the vehicle 2 based on the departure point and the destination point of the vehicle 2 (step S2). In step S2, the setting information acquisition unit 10 further specifies each of the entire driving routes to the destination point by a plurality of points p through which the vehicle 2 passes. For example, the positions reached by the vehicle 2 can be arranged as points p for each divided section of the map data, for each intersection or branch point, or at regular time intervals.

[0073] Next, the output symbol acquisition unit 12 acquires, as learning data, a series of output symbols indicating the steering operation at each point p by the vehicle 2 (step S3). Specifically, the output symbol acquisition unit 12 can acquire the history of the steering operation recorded in the vehicle 2 via the in-vehicle network NW.

[0074] Next, the learning unit 13 estimates the parameters of the hidden Markov model that maximize the likelihood with respect to the learning data acquired in step S3 (step S4). Specifically, the learning unit 13 executes the processing from the first step to the sixth step of the above-described Baum-Welch algorithm, and estimates the parameters including the state transition probability distribution A and the symbol output probability distribution B.

[0075] The parameters estimated in step S4 are stored in the parameter storage unit 16 as a learned hidden Markov model (step S5).

[0076] Next, with reference to FIG. 6, the arithmetic processing using the learned hidden Markov model by the analysis device 1 will be described.

[0077] First, the arithmetic unit 14 loads the learned hidden Markov model from the parameter storage unit 16 (step S10). Next, the output symbol acquisition unit 12 acquires a series of output symbols including the history of the steering operation by the vehicle 2 (step S11).

[0078] Next, the arithmetic unit 14 uses the learned hidden Markov model to determine the probability that each of the steering operations acquired in step S11 is performed by the vehicle 2 at each point p (step S12). For example, as shown in FIG. 2(b), the arithmetic unit 14 determines the probability value b i of the steering operation that the vehicle 2 performs at each point p corresponding to each state S i (1), b i (2), b i (3).

[0079] Next, based on the probability values obtained in step S12, the determination unit 15 determines the driving route to the destination point according to the steering operation with the highest probability of being operated by the vehicle 2 among the plurality of steering operations (step S13). For example, as shown in FIG. 1, the determination unit 15 determines the next point p "2" as the driving route according to the steering operation (right turn) at the first point p "1" from the starting point. Similarly, the determination unit 15 determines the next point p "3" as the route according to the steering operation at the point p "2", and can determine the driving route based on all the points p that the vehicle 2 passes through and the order in which the vehicle 2 passes through the points p until it reaches the destination point.

[0080] Next, the presentation unit 17 presents the driving route determined in step S13 to the vehicle 2 via the in-vehicle network NW (step S14). For example, the presentation unit 17 can notify the car navigation system 2A of the determined driving route.

[0081] As described above, according to the analyzer 1 according to the present embodiment, each point p that the vehicle 2 passes from the starting point to the destination point is set as a finite set in a hidden state, and the parameters of the hidden Markov model are set as a finite set of output symbols that can observe the operation of the traveling direction by the vehicle 2 at each point p. are estimated. Furthermore, the operation of the learned hidden Markov model is performed, and the probability that each of the traveling direction operations acquired by the output symbol acquisition unit 12 is performed at each point p is obtained. Therefore, it is possible to present a driving route according to the driving tendency of each user.

[0082] Note that in the above-described analyzer 1, the learning process for estimating the parameters of the hidden Markov model has been described for the case of using the Baum-Welch algorithm. However, the estimation of the parameters of the hidden Markov model is not limited to the Baum-Welch algorithm. For example, structured variational inference or the Viterbi algorithm may be used.

[0083] In addition, the analyzer 1 according to the above-described embodiment has been described as an example of a configuration mounted on the vehicle 2 and cooperating with the car navigation system 2A, but is not limited thereto. For example, the analyzer 1 can be configured as a part of the in-vehicle car navigation system 2A. Alternatively, the analyzer 1 can be configured as a cloud network that communicates with the vehicle 2 by V2N (Vehicle to Network) communication.

[0084] In addition, in the above-described embodiment, the vehicle 2 has been described on the premise of manual driving in which the steering operation is performed by the driver. However, the vehicle 2 may include a case corresponding to autonomous driving. In this case, the analyzer 1 analyzes the tendency of the user to change the route specified for the driving plan set in the vehicle 2 that performs autonomous driving. Furthermore, the analyzer 1 can be configured to execute the most probable steering operation obtained by the calculation unit 14 as the steering control at each point p in autonomous driving, together with the presentation of the driving route.

[0085] In the above-described embodiment, the steering operation has been described as an example of the operation of the traveling direction of the vehicle 2. However, the operation of the traveling direction of the vehicle 2 is not limited to the steering operation as long as it is operation information for specifying a change in the traveling direction of the vehicle 2.

[0086] In the above-described embodiment, three types of steering operations of the vehicle 2, i.e., right rotation, no rotation, and left rotation, have been exemplified. However, the steering operation or the operation of the traveling direction is not limited to this.

[0087] The embodiments of the analyzer and the analysis method of the present invention have been described above. However, the present invention is not limited to the described embodiments, and various modifications that can be assumed by those skilled in the art can be made within the scope of the invention described in the claims.

Description of Reference Numerals

[0088] 1... Analyzer, 10... Setting information acquisition unit, 11... Setting information storage unit, 12... Output symbol acquisition unit, 13... Learning unit, 14... Calculation unit, 15... Determination unit, 16... Parameter storage unit, 17... Presentation unit, 2... Vehicle, 2a... Steering wheel, 20... Sensor, 21... GPS module, 22... ECU, 23... Display device, 25... Map database, 101... Bus, 102... Processor, 103... Main storage device, 24, 104... Communication interface, 105... Auxiliary storage device, 106... Input / output I / O, NW... In-vehicle network.

Claims

1. A setting information acquisition unit configured to acquire a departure point and a destination point of a vehicle; An output symbol acquisition unit configured to acquire an output symbol sequence including a history of operations of the vehicle in the traveling direction at each point passed by the vehicle until the vehicle reaches the destination point; A storage unit configured to store a learned hidden Markov model in which each point passed by the vehicle from the departure point to the destination point is a finite set in a hidden state, and a finite set of output symbols observable for the operation of the vehicle in the traveling direction at each point, and parameters of the hidden Markov model have been estimated in advance; A calculation unit configured to use the learned hidden Markov model to obtain a probability that the vehicle performs each of the operations in the traveling direction acquired by the output symbol acquisition unit at each point; A determination unit configured to determine a driving route to the destination point according to the operation in the traveling direction having the highest probability of being an operation by the vehicle among the operations in the traveling direction based on the probability that each of the operations in the traveling direction obtained by the calculation unit is performed; A presentation unit configured to present the determined driving route An analysis device comprising the same.

2. The analysis device according to claim 1, wherein the setting information acquisition unit acquires a plurality of different driving routes by which the vehicle can reach the destination point from the departure point, and each of the plurality of different driving routes is specified by a plurality of points The analysis device is characterized by the above.

3. The analysis device according to claim 1, wherein the output symbol acquisition unit acquires, as learning data, an output symbol sequence indicating the operation of the vehicle in the traveling direction at each point by the vehicle; Furthermore, a learning unit configured to estimate parameters of the hidden Markov model including a state transition probability distribution which is the probability that the vehicle moves from a predetermined point to another point and maximizes the likelihood with respect to the learning data, and a symbol output probability distribution which is the probability that the vehicle performs an operation in a predetermined traveling direction at each of the points. The learned hidden Markov model includes the parameters estimated by the learning unit. An analyzer characterized by the above.

4. In the analyzer according to claim 1, The operation in the traveling direction includes a plurality of preset steering operations in the vehicle. An analyzer characterized by the above.

5. A setting information acquisition step of acquiring a departure point and a destination point of a vehicle; An output symbol acquisition step of acquiring an output symbol sequence including a history of operations in the traveling direction by the vehicle at each point passed by the vehicle until it reaches the destination point; A storage step of storing, in a storage unit, a learned hidden Markov model in which each point passed by the vehicle from the departure point to the destination point is a finite set of hidden states, and a finite set of observable output symbols of the operation in the traveling direction by the vehicle at each point, and parameters of the hidden Markov model have been estimated in advance; An arithmetic step of using the learned hidden Markov model to obtain the probability that each of the operations in the traveling direction acquired in the output symbol acquisition step is performed at each point by the vehicle; A determination step of determining a driving route to the destination point according to the operation in the traveling direction with the highest probability of being performed by the vehicle among the operations in the traveling direction based on the probabilities that each of the operations in the traveling direction obtained in the arithmetic step is performed; A presentation step of presenting the determined driving route And an analysis method comprising the above.

6. In the analysis method according to claim 5, the setting information acquisition step acquires a plurality of different driving routes by which the vehicle can reach the destination from the departure point, each of the plurality of different driving routes is specified by a plurality of points The analysis method is characterized in that.

7. In the analysis method according to claim 5, the output symbol acquisition step acquires, as learning data, a series of output symbols indicating the operation of the traveling direction at each point by the vehicle, Further, a learning step of estimating the parameters of the hidden Markov model including a state transition probability distribution that is the probability of the vehicle moving from a predetermined point to another point and maximizes the likelihood with respect to the learning data, and a symbol output probability distribution that is the probability of the vehicle performing a predetermined traveling direction operation at each point is provided, The learned hidden Markov model includes the parameters estimated in the learning step The analysis method is characterized in that.

Images