JP6358051B2 - Transition prediction data generation device and transition prediction device - Google Patents

Description

  The present invention relates to a transition prediction data generation device and a transition prediction device for predicting a transition of a driving scene.

  As a technique for predicting the transition of the above driving scene, the driving scene such as the positional relationship with the surrounding vehicles and the signal state is expressed as a symbol, and the temporal transition of this symbol is modeled. There is known one that predicts a driving scene corresponding to a symbol to be performed (see, for example, Patent Document 1).

Japanese Patent No. 5278419

  However, in the above technique, for example, in a frequent driving scene such as “stop”, there is a tendency that the estimation accuracy of the next driving scene is lowered. In other words, the driving scene of “stop” transitions from various driving scenes while traveling, and from this driving scene, transitions to various driving scenes such as “loose start”, “rapid acceleration”, “reverse”, etc. , Difficult to predict.

  Therefore, in view of such problems, it is an object of the present invention to be able to predict the transition of the driving scene with higher accuracy in the technology for predicting the transition of the driving scene.

  In the transition prediction data generation device of the present invention, the vehicle state acquisition means repeatedly acquires a vehicle state indicating at least one of the traveling environment of the vehicle and the traveling state of the vehicle. Further, the associated data generation means sequentially identifies one of a plurality of preset driving scenes according to the vehicle state for each preset driving interval, and sets the appearance of the driving scene as the driving interval. Corresponding association data is generated.

  The classification data generation means generates classification data obtained by classifying (clustering) the associated data according to the driving scene included in each travel interval. The storage unit stores the association data and the classification data in the recording unit as transition prediction data.

  According to such a transition prediction data generation device, classification data is generated by classifying a travel interval according to a driving scene that appears for each driving interval. Therefore, when predicting a driving scene transition, any vehicle state is What is necessary is just to determine whether it belongs to the classification | category of an interval. That is, it is possible to predict a driving scene that frequently appears according to the classification of the travel interval using the transition prediction data (corresponding data and classification data). Therefore, it is possible to accurately predict a driving scene that will appear in the future as compared with a configuration that predicts a driving scene without classifying the driving interval.

  In addition, description of each claim can be arbitrarily combined as much as possible. At this time, a part of the configuration may be excluded.

1 is a block diagram showing a schematic configuration of a transition prediction system 1 to which the present invention is applied. It is a flowchart which shows the prediction data generation process which the server 40 performs. It is explanatory drawing which shows an example of the relationship between the symbol contained in each trip, and a trip. It is explanatory drawing which shows an example of a clustering result. It is a state transition diagram which shows an example of the effect by clustering. It is a flowchart which shows the scene transition prediction process which the control part 10 of the transition prediction apparatus 5 performs.

Embodiments according to the present invention will be described below with reference to the drawings.
[Configuration of this embodiment]
The transition prediction system 1 to which the present invention is applied has a function of generating transition prediction data for predicting a transition of a driving scene, and a function of predicting a transition of a driving scene using the transition prediction data. As illustrated in FIG. 1, the transition prediction system 1 includes a transition prediction device 5 and a server 40.

  The transition prediction device 5 is mounted on a vehicle such as a passenger car, for example, estimates a driving scene of the vehicle, and predicts what driving scene it will transition to in the future. As illustrated in FIG. 1, the transition prediction device 5 includes a control unit 10, various sensors 21, a database 25, and a driving support unit 31.

  Here, the driving scene indicates at least a part of a vehicle state expressed by a combination of a vehicle environment, a running state, and the like. As the driving scene, for example, even “acceleration” can be classified into different driving scenes depending on the situation such as gradual acceleration and sudden acceleration. Further, for example, even “stop” can be classified into different driving scenes depending on the situation such as idling stop or idling stop.

  The various sensors 21 include sensors for repeatedly detecting the vehicle state. Examples of the sensor that detects the vehicle state include a sensor that detects an operation state such as an accelerator, a brake, and a steering wheel, a sensor that detects a traveling state such as a traveling speed, acceleration, and yaw rate of the vehicle, a road surface state, an air temperature, a brightness, A sensor for detecting a traveling environment such as the presence or absence of traffic congestion is provided. Various sensors 21 send the detection result of the vehicle state to the control unit 10.

  The database 25 stores data for predicting driving scene transitions. As the database 25, an operation symbol DB (database) 26, a cluster DB 27, and a symbol transition model DB 28 are provided. Details of the data stored in these databases 25 will be described later.

  The driving support unit 31 is an alarm device that issues a warning to the vehicle driver in response to a command from the control unit 10, or a brake control device or a steering control device that intervenes in a driving operation in response to a command from the control unit 10. It is configured.

  The control unit 10 is configured as a computer including a CPU 11 and a memory 12 such as a ROM or a RAM. The control unit 10 (CPU 11) performs various processes based on a program stored in the memory 12. The control unit 10 is configured to exchange data with the server 40 via a communication module (not shown).

  In addition, the control unit 10 instructs the driving support unit 31 to perform an alarm or intervention for driving operation when the driving scene is predicted to shift to a dangerous state for vehicles such as accidents in the near future. . Further, the control unit 10 transmits information obtained from the various sensors 21 to the server 40 as driving behavior data.

  The server 40 is configured as a well-known server that transmits and receives data in response to requests from one or more transition prediction devices 5. The server 40 includes a database 41. Similarly to the vehicle database 25, the database of the server 40 includes a driving symbol DB 42, a cluster DB 43, and a symbol transition model DB 44 corresponding to the driving symbol DB 26, the cluster DB 27, and the symbol transition model DB 28, respectively.

  However, the database 41 of the server 40 stores data on all vehicles (transition prediction device 5) to be managed, but the database 25 of the transition prediction device 5 is necessary for the vehicle on which the transition prediction device 5 is mounted. Only the data to be stored is stored. That is, the database 25 of the transition prediction device 5 stores driving scene data relating to the vehicle and data for predicting the transitioning driving scene.

[Process of this embodiment]
In the transition prediction system 1 configured as described above, the server 40 performs a prediction data generation process shown in FIG. The prediction data generation process is a process that is started when driving behavior data is transmitted from the transition prediction device 5, for example, and is then a process that is repeatedly performed at regular intervals.

  In the prediction data generation process, as shown in FIG. 2, first, driving behavior data is acquired from the transition prediction device 5 that has transmitted the driving behavior data (S10). Subsequently, a driving symbol is generated by estimating the driving symbol (S20). Here, the driving symbol is a symbolized combination of vehicle environment and vehicle state. Note that, as the driving symbol estimation method, for example, a method similar to the method described in Patent Document 1 can be adopted, and thus the details of this process are omitted.

  Subsequently, the generated symbol is stored in the database 41 as symbol data (association data) associated with a trip (running interval) (S40). In particular, it is stored in the operation symbol DB 42 of the database 41.

  Here, the trip is a unit for handling a group of symbols in the present process, and is set, for example, from the start of driving to the end of driving in each vehicle. Subsequently, it is determined whether or not the trip to be processed has ended (S50).

  If the trip to be processed has not ended (S50: NO), the process returns to S10. As described above, by repeating the processes of S10 to S50, the symbol data is associated with a large number of operation symbols for each trip.

  For example, as shown in FIG. 3, in the symbol data, the appearing operation symbol is stored in the database 41 for each trip to be processed. The database 41 also stores the order in which driving symbols appear in association with trips.

  Further, if the trip to be processed is completed in the process of S50 (S50: YES), clustering is performed on the appearance pattern of the driving symbol (S60). In this process, the symbol data is classified according to the operation symbols included in each trip.

  Clustering methods include, for example, an infinite relation model ([Kemp et al., AAAI2006] C. Kemp, JB Tenenbaum, TL Griffiths, T. Yamada, N. Ueda, “Learning Systems of Concepts with an Infinite Relational Model,” Nonparametric clustering methods such as National Conference on Artificial Intelligence, pp. 381-388, 2006) can be adopted. When the infinite relation model is used, driving symbols and trips can be naturally divided in consideration of the co-occurrence relationship of driving symbols.

  Further, since the number of divisions at the time of clustering can be determined from the data, it is not necessary to determine in advance. Note that the clustering method is not limited to the infinite relation model, and various methods such as k-means and the Latent Dirichlet Allocation method can be used.

  When such clustering is performed, for example, as shown in FIG. 4, a plurality of trips in which similar operation symbols appear are classified into a plurality of groups of clusters D1, D2, D3. In the example shown in FIG. 4, the vertical axis corresponds to trips and the horizontal axis corresponds to driving symbols, the trip classifications (clusters) obtained by clustering are D1, D2, D3..., And the symbol cluster is C1. , C2, C3... An operation symbol that appears for each cluster is represented by a single dot. That is, a symbol cluster having a larger number of dots tends to appear more easily.

As in the present embodiment, by performing data-driven clustering, different transition models can be used even when the same driver passes different driving routes.
Subsequently, the cluster data (classification data) obtained by the clustering is stored in the database 41 (S70). In particular, it is stored in the cluster DB 43 of the database 41.

Subsequently, the appearance order of the driving symbols is acquired (S80). In this process, the appearance order of the driving symbols is read from the driving symbol DB 42.
Then, for each cluster, a symbol transition model in which an operation symbol that is likely to appear next to a certain operation symbol is extracted for each operation symbol based on the frequency distribution of the appearance order of operation symbols in trips included in the cluster ( (Transition data) is generated (S90).

  In this case, as shown in FIG. 5 (A), for example, the probability of transition to the symbol 6, 7 or 8 after the frequent driving symbol of the stop 10 is competing. In addition, as shown in FIG. 5B, it is possible to estimate a driving symbol that transitions after the stop 10 for each cluster. That is, in the example shown in FIG. 5B, a cluster that is likely to undergo the transition 1 → 2 → 10 → 6, a cluster that is likely to undergo the transition 3 → 4 → 10 → 7, and 5 → 10 → 8 → 9. Each is divided into clusters that are prone to transition.

  In this way, by reconstructing the transition model in each cluster, it is possible to construct a transition model that functions effectively even in a relatively small number of cases regardless of the existence of extremely general driving symbols (for example, the stop 10). it can. That is, even a driving symbol that appears next to a general driving symbol can be estimated with high accuracy.

Subsequently, the symbol transition model is stored in the symbol transition model DB 44 (S100), and the prediction data generation process is terminated.
Next, scene transition prediction processing for predicting driving scene transition in a vehicle will be described with reference to FIG. The scene transition prediction process is performed using the cluster data and the symbol transition model generated by the above-described prediction data generation process.

  For this reason, of these data stored in the database 41 of the server 40, in a state in which data suitable for the vehicle type and vehicle state of the vehicle on which the transition prediction device 5 is mounted is acquired by the transition prediction device 5. A prediction data generation process is performed. At this time, the data acquired from the database 41 of the server 40 is stored in the database 25 of the transition prediction device 5.

In addition, it can also be set as the structure which acquires data from the server 40 in the middle of the prediction data generation process being implemented.
In the predicted data generation processing, as shown in FIG. 6, first, processing similar to the processing of S10 to S40 described above is performed. The driving behavior data is acquired directly from the various sensors 21.

  Subsequently, the cluster corresponding to the driving pattern that appears on the current trip is estimated (S210). In this process, the appearance status of the driving symbol appearing in the cluster data stored in the cluster DB 27 is compared with the appearance status of the driving symbol appearing on the currently traveling trip, and the most similar cluster is selected.

  Subsequently, a symbol predicted to transition next to the latest driving symbol (the driving symbol recognized at the closest time from the current time) is extracted (S220). In this process, in the selected cluster, the driving scene indicated by the latest driving symbol is used as the current driving symbol, and in the symbol transition model DB 28, the driving symbol that is likely to appear next to the current driving symbol is extracted as the predicted symbol. To do.

  Subsequently, it is determined whether or not the predicted symbol is a dangerous symbol that is a preset driving symbol indicating danger (S230). Here, the danger symbol indicates a driving symbol indicating a vehicle state that exhibits a large acceleration such as an accident or a vehicle state that may cause an accident such as a sudden braking or a sudden turn.

  If the predicted symbol is a dangerous symbol (S230: YES), the driving support unit 31 is instructed to cause the driving support unit 31 to perform an operation corresponding to the predicted symbol (S240), and the scene transition prediction process is terminated. . If the prediction symbol is not a dangerous symbol (S230: NO), the scene transition prediction process is terminated.

[Effects of this embodiment]
In the transition prediction system 1 described in detail above, the server 40 repeatedly acquires a vehicle state indicating at least one of the traveling environment of the vehicle and the traveling state of the vehicle. Further, the server 40 sequentially identifies one or more of a plurality of preset driving scenes according to the vehicle state at each preset driving interval (trip), and travels the appearance status of the driving scene. Correspondence data associated with the interval is generated.

  Then, the server 40 generates classification data obtained by classifying (clustering) the association data according to the driving scenes included in the respective travel intervals. The server 40 stores the association data and the classification data in the recording means (database) as transition prediction data.

  According to such a transition prediction system 1, classification data is generated by classifying a travel interval according to a driving scene that appears at each traveling interval. Therefore, when predicting a transition of a driving scene, the vehicle state is any driving interval. It may be determined whether it belongs to the category. Then, the next transition symbol is predicted based on the corresponding symbol transition model. That is, it is possible to predict a driving scene that transitions next according to the classification of the travel interval using the transition prediction data (corresponding data and classification data). Therefore, it is possible to accurately predict a driving scene that will appear in the future as compared with a configuration that predicts a driving scene without classifying the driving interval.

In the transition prediction system 1, the server 40 generates correspondence data by replacing the driving scene specified according to the vehicle state with a preset symbol.
According to such a transition prediction system 1, since the driving scene is replaced with a symbol and held, the data configuration can be simplified.

  Further, in the transition prediction system 1, the server 40 obtains the transition pattern of the driving scene, and for each cluster obtained by classifying the driving intervals by clustering, the probability that each driving scene is applied next to each driving scene is set. The associated transition data is generated. The transition data is also stored in the recording means.

  According to such a transition prediction system 1, since transition data is stored, when a transition of a driving scene is predicted, if it is known whether it belongs to the classification of the driving interval, the driving scene to be transitioned next is easily predicted. can do.

Further, in the transition prediction system 1, the server 40 generates classification data by performing clustering using the infinite relation model on the association data.
According to such a transition prediction system 1, classification data can be easily generated using a known algorithm.

In the transition prediction system 1, the server 40 sets the travel interval from the start of driving of the vehicle to the end of driving.
According to such a transition prediction system 1, since the driver of the data of driving | running | working interval can be limited to one person, it can suppress that the transition pattern of several competing driving scenes mixes in one driving | running | working interval.

  In the transition prediction system 1, the server 40 acquires a vehicle state indicating at least one of the vehicle traveling environment and the vehicle traveling state, and sets one of a plurality of preset driving scenes to the vehicle state. In accordance with this, the appearance of the driving scene is recognized.

  In the transition prediction system 1, the control unit 10 of the transition prediction device 5 uses the association data in which the appearance state of the driving scene is associated with the travel interval, and is classified according to the driving scene included in each travel interval. It is determined to which of the classified classifications. Then, based on the transition model of the driving scene in the classification to which the appearance situation of the driving scene in the current traveling belongs, the driving scene that appears next is predicted.

  According to such a transition prediction system 1, it is possible to determine which class of classification data belongs according to the appearance state of a driving scene, and to predict a driving scene that appears frequently according to this classification. Therefore, it is possible to accurately predict a driving scene that will appear in the future as compared with a configuration that predicts a driving scene without classifying the driving interval.

[Other Embodiments]
The present invention is not construed as being limited by the above embodiment. Further, the reference numerals used in the description of the above embodiments are also used in the claims as appropriate, but they are used for the purpose of facilitating the understanding of the invention according to each claim, and the invention according to each claim. It is not intended to limit the technical scope of The functions of one component in the above embodiment may be distributed as a plurality of components, or the functions of a plurality of components may be integrated into one component. Further, at least a part of the configuration of the above embodiment may be replaced with a known configuration having the same function. Moreover, you may abbreviate | omit a part of structure of the said embodiment as long as a subject can be solved. In addition, at least a part of the configuration of the above embodiment may be added to or replaced with the configuration of the other embodiment. In addition, all the aspects included in the technical idea specified only by the wording described in the claim are embodiment of this invention.

  In addition to the transition prediction system 1 described above, a device (transition prediction device 5, server 40) that is a component of the transition prediction system 1, a program for causing a computer to function as the transition prediction system 1, and a medium on which the program is recorded The present invention can also be realized in various forms such as a transition prediction method.

  For example, in the above embodiment, the travel interval is set from the start of driving to the end of driving of the vehicle, but may be set at any other time. For example, it may be set from when the vehicle starts to when it stops. In this way, since a relatively short time is used as the travel interval, more data can be obtained in a short time. Therefore, transition prediction data can be generated in a short time.

  Moreover, in the said embodiment, although the process of S10-S50 was demonstrated in the server 40 in the prediction data generation process, these processes may be implemented in the transition prediction apparatus 5 of a vehicle. In this case, the server 40 may acquire symbol data from the vehicle transition prediction device 5.

  In the above embodiment, the driving scene transition is predicted using the symbol data, the cluster data, and the symbol transition model. However, the driving scene transition is performed using the symbol data and the cluster data without using the symbol transition model. It may be predicted.

In this way, the transition of the driving scene can be predicted with a simpler configuration.
[Correspondence between Configuration of Embodiment and Means of Present Invention]
The server 40 in the above embodiment corresponds to the transition prediction data generation device referred to in the present invention, and the transition prediction device 5 in the above embodiment corresponds to the transition prediction device referred to in the present invention. Of the processes performed by the server 40 or the transition prediction apparatus 5 (control unit 10), the process of S20 corresponds to the association data generation means and the driving scene recognition means referred to in the present invention. The processes of S70 and S100 correspond to storage means in the present invention.

  Further, the process of S60 in the above embodiment corresponds to the classification data generation means referred to in the present invention, and the process of S80 in the above embodiment corresponds to the transition pattern acquisition means referred to in the present invention. In addition, the process of S90 in the above embodiment corresponds to the transition data generation unit referred to in the present invention, and the process of S210 in the above embodiment corresponds to the classification determination unit referred to in the present invention.

  Furthermore, the process of S220 in the above embodiment corresponds to the prediction means in the present invention.

  DESCRIPTION OF SYMBOLS 1 ... Transition prediction system, 5 ... Transition prediction apparatus, 6 ... Symbol, 10 ... Control part, 11 ... CPU, 12 ... Memory, 21 ... Various sensors, 25 ... Database, 26 ... Driving symbol DB, 27 ... Cluster DB, 28 ... symbol transition model DB, 31 ... driving support unit, 40 ... server, 41 ... database, 42 ... driving symbol DB, 43 ... cluster DB, 44 ... symbol transition model DB.

Claims (6)

A transition prediction data generation device (40) that generates transition prediction data used when predicting a transition of a driving scene,
Vehicle state acquisition means (S10) for repeatedly acquiring a vehicle state indicating at least one of a traveling environment of the vehicle and a traveling state of the vehicle;
For each travel interval set from the start of travel of the vehicle to the end of travel , one of a plurality of preset driving scenes is sequentially identified according to the vehicle state, and the appearance status of the driving scene is determined as the travel interval. Association data generation means (S20) for generating association data associated with
Classification data generation means (S60) for generating classification data obtained by classifying the association data according to the driving scene included in each driving interval;
Storage means (S40, S70, S100) for storing the association data and the classification data in the recording means as the transition prediction data;
A transition prediction data generation device characterized by comprising: In the transition prediction data generation device according to claim 1,
The transition-predicted data generating device, wherein the associated data generating means generates associated data by replacing a driving scene specified according to the vehicle state with a preset symbol. In the transition prediction data generation device according to claim 1 or 2,
Transition pattern acquisition means (S80) for acquiring the appearance order of driving scenes;
Transition model data generating means (S90) for generating transition model data in which the probability that each driving scene appears next to each driving scene for each classification of the driving interval;
With
The storage means stores the transition model data in the recording means. In the transition prediction data generation device according to any one of claims 1 to 3,
The classification data generation unit generates the classification data by performing clustering on the association data using an infinite relation model. A driving scene transition prediction device (5) for predicting driving scene transition,
Vehicle state acquisition means (S10) for acquiring a vehicle state indicating at least one of a traveling environment of the vehicle and a traveling state of the vehicle;
Driving scene recognition means (S20) for sequentially identifying one or more of a plurality of preset driving scenes according to the vehicle state and recognizing the appearance of the driving scene;
Any of the data classified according to the driving data included in each of the driving intervals included in the correspondence data that associates the appearance of the driving scene with the driving intervals set from the start of driving to the end of driving of the vehicle. Classification determination means (S210) for determining whether it belongs to the classification of
Prediction means (S220) for predicting a driving scene that appears next based on a distribution tendency of the driving scene in the classification to which the appearance situation of the driving scene belongs;
A transition prediction apparatus comprising: In the transition prediction device according to claim 5,
The association data and the classification data are respectively the association data and the classification data generated by the transition prediction data generation device according to any one of claims 1 to 4. Transition prediction device.