WO2021199175A1 - Data generation device, data generation method, method for manufacturing data generation device, and program - Google Patents

Data generation device, data generation method, method for manufacturing data generation device, and program Download PDF

Info

Publication number
WO2021199175A1
WO2021199175A1 PCT/JP2020/014606 JP2020014606W WO2021199175A1 WO 2021199175 A1 WO2021199175 A1 WO 2021199175A1 JP 2020014606 W JP2020014606 W JP 2020014606W WO 2021199175 A1 WO2021199175 A1 WO 2021199175A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
pseudo
gnss
data generation
vehicle
Prior art date
Application number
PCT/JP2020/014606
Other languages
French (fr)
Japanese (ja)
Inventor
剛志 是永
健司 ▲高▼尾
山田　昌弘
陽平 知識
Original Assignee
三菱重工機械システム株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 三菱重工機械システム株式会社 filed Critical 三菱重工機械システム株式会社
Priority to PCT/JP2020/014606 priority Critical patent/WO2021199175A1/en
Priority to JP2022512915A priority patent/JP7381719B2/en
Publication of WO2021199175A1 publication Critical patent/WO2021199175A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
    • G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
    • G01S19/03—Cooperating elements; Interaction or communication between different cooperating elements or between cooperating elements and receivers
    • G01S19/10—Cooperating elements; Interaction or communication between different cooperating elements or between cooperating elements and receivers providing dedicated supplementary positioning signals
    • G01S19/11—Cooperating elements; Interaction or communication between different cooperating elements or between cooperating elements and receivers providing dedicated supplementary positioning signals wherein the cooperating elements are pseudolites or satellite radio beacon positioning system signal repeaters
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Definitions

  • the present invention relates to a data generation device, a data generation method, a method for manufacturing a data generation device, and a program.
  • GNSS data positioning data by a GNSS receiver
  • server a server
  • map matching processing a large amount of GNSS data is required for verification of map matching processing and service processing using map matching processing (for example, toll road billing processing).
  • the data generator creates a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
  • the learning unit to be learned, simulated travel path data representing a time series of virtual positions where the vehicle virtually moves, and 3D map data including the virtual position are input to the pseudo data generation model, and the virtual position is input. It is provided with a pseudo data generation unit that generates pseudo GNSS data in which the GNSS data obtained in each of the above is estimated.
  • the data generation method creates a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
  • the step to be learned, the simulated travel path data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model to obtain the virtual position.
  • Each has a step of generating pseudo GNSS data inferring the GNSS data obtained in each.
  • the manufacturing method of the data generator (1) is based on the measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
  • the step of learning the pseudo data generation model and the simulated travel route data representing the time series of the virtual positions where the vehicle virtually moves and the 3D map data are input to the learned pseudo data generation model. It includes a step of outputting pseudo GNSS data, a step of determining the authenticity of the input GNSS data, and a step of further learning the pseudo data generation model based on the authenticity determination result.
  • the program learns a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
  • the step, the simulated travel route data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model, and at each of the virtual positions.
  • the computer of the data generator is made to execute the step of generating the pseudo GNSS data inferring the obtained GNSS data.
  • the data generation device the data generation method, the manufacturing method of the data generation device, and the program according to the present disclosure, it is possible to generate pseudo GNSS data that closely resembles the actually measured GNSS data.
  • FIG. 1 is a diagram showing a functional configuration of a data generation device according to an embodiment of the present disclosure. As shown in FIG. 1, the data generation device 1 includes a CPU 10 and a storage medium 20.
  • the CPU 10 is a processor that controls the operation of the entire data generation device 1, and by operating according to a predetermined program, it exhibits functions as a learning unit 11 and a pseudo data generation unit 12.
  • the learning unit 11 learns the pseudo data generation model based on the actually measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the actually measured GNSS data.
  • the measured GNSS data includes information such as latitude, longitude, and altitude that represent the position of the vehicle, for example. Further, the measured GNSS data may include the moving speed, the moving direction, the number of captured satellites, the satellite number capable of identifying the captured satellite, the signal strength, and the like.
  • the 3D map data is map data capable of specifying a three-dimensional shape of a terrain, a building, or the like.
  • the pseudo data generation unit 12 inputs simulated travel route data representing a time series of virtual positions where the vehicle virtually moves and 3D map data including the virtual positions into the pseudo data generation model, and at each of the virtual positions. Pseudo GNSS data is generated by estimating the obtained GNSS data. That is, the pseudo data generation unit 12 pseudo-generates GNSS data that is presumed to be obtained in the simulated travel route without actually driving the vehicle.
  • the virtual position included in the simulated travel route data is represented by, for example, a link ID that can identify the link (road) and information indicating the position on the link (distance from the link start point, etc.). Further, as the simulated travel route data, for example, the user of the data generation device 1 inputs in advance a travel route to be verified for the map matching process and / or the service process using the map matching process.
  • the storage medium 20 is a so-called auxiliary storage device, such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).
  • Various data acquired and generated by the data generation device 1 are stored in the storage medium 20.
  • the storage medium 20 stores actually measured GNSS data, 3D map data, a pseudo data generation model learned by the learning unit 11, pseudo GNSS data generated by the pseudo data generation unit 12, and the like.
  • the reception status of satellite signals by the GNSS receiver (number of satellites to be captured, strength of signals received from captured satellites, etc.) Change. Therefore, the measured GNSS data measured by the GNSS receiver includes a positioning error according to the reception state of the satellite signal.
  • GNSS data including the virtual position (latitude, longitude) of the vehicle is simply generated without considering this positioning error, there is a possibility that data that deviates significantly from the measured GNSS data will be generated.
  • the data generation device 1 learns a pseudo data generation model that reflects the positioning error according to the traveling route of the vehicle, and uses the learned pseudo data generation model to elaborate the measured GNSS data. Generate pseudo GNSS data that imitates.
  • the functional configurations of the learning unit 11 that learns the pseudo data generation model and the pseudo data generation unit 12 that generates pseudo GNSS data using the learned pseudo data generation model M1 will be described.
  • FIG. 2 is a diagram showing a functional configuration of a learning unit according to an embodiment of the present disclosure.
  • the learning unit 11 includes a generator 110, a discriminator 111, and a traveling route history creating unit 112.
  • the learning unit 11 learns the pseudo data generation model M1 that generates pseudo GNSS data by using the technology of the hostile generation network (GAN: Generative Adversarial Networks).
  • GAN Generative Adversarial Networks
  • the generator 110 learns the pseudo data generation model M1 based on the actually measured GNSS data and the 3D map data, and inputs the simulated travel route data and the 3D map data into the learned pseudo data generation model M1 to input the pseudo GNSS data. Output.
  • the discriminator 111 determines the authenticity of the input GNSS data. Specifically, the discriminator 111 is true using the discriminative model M2 learned based on the actually measured GNSS data, the pseudo GNSS data labeled to indicate that it is the pseudo GNSS data, and the 3D map data. Judge false.
  • the travel route history creation unit 112 creates a travel route history representing the time series of the vehicle's travel position based on the actually measured GNSS data.
  • FIG. 3 is a diagram showing a functional configuration of a pseudo data generation unit according to an embodiment of the present disclosure.
  • the pseudo data generation unit 12 inputs the simulated travel route data and the 3D map into the pseudo data generation model M1 learned by the learning unit 11 to generate pseudo GNSS data.
  • the pseudo data generation unit 12 may further input a random number into the pseudo data generation model M1.
  • the pseudo data generation unit 12 can generate a plurality of types of pseudo GNSS data changed by random numbers from the same simulated travel route data.
  • FIG. 4 is a first flowchart showing an example of processing of the learning unit according to the embodiment of the present disclosure.
  • the flow of the learning process of the generator 110 of the learning unit 11 will be described with reference to FIG.
  • the travel route history creation unit 112 creates a travel route history of the vehicle from the actually measured GNSS data obtained by actually traveling the vehicle (step S01). For example, the travel route history creation unit 112 performs map matching processing and creates a travel route history indicating a time series of the vehicle's travel position (position on the link) from the time history of the measured GNSS data.
  • the generator 110 learns the pseudo data generation model M1 based on the actually measured GNSS data and the 3D map around the traveling position of the vehicle specified from the actually measured GNSS data (step S02).
  • the pseudo data generation model M1 is, for example, a model using a recurrent neural network, and when 3D map data of the travel route and the vicinity of the travel route is input, the pseudo GNSS that estimates the GNSS data obtained when traveling on this travel route is estimated. Output data.
  • the generator 110 uses the 3D map data around the traveling position of the vehicle for learning, and uses the pseudo data generation model M1 to determine what kind of measured GNSS data can be acquired according to the influence of the terrain, buildings, etc. around the vehicle. Let them learn.
  • the learned pseudo data generation model M1 is stored in the storage medium 20.
  • the generator 110 may make the pseudo data generation model M1 learn the measured GNSS data according to the travel route of the vehicle by using the travel route history created by the travel route history creation unit 112.
  • the generator 110 may further learn the pseudo data generation model M1 based on the vehicle type of the vehicle for which the actually measured GNSS data has been acquired.
  • the mounting position of the GNSS receiver may differ depending on the vehicle type. Then, the reception state of the satellite signal may change, and the tendency of the positioning error may also change. Therefore, the generator 110 learns the pseudo data generation model M1 by inputting the vehicle type of the vehicle that has acquired the actually measured GNSS data into the pseudo data generation model M1 in addition to the actually measured GNSS data.
  • the vehicle type of the vehicle for which the actually measured GNSS data has been acquired is input by, for example, the user of the data generation device 1.
  • the generator 110 inputs the simulated travel route data and the 3D map around the simulated travel route into the learned pseudo data generation model M1 and travels on the simulated travel route. Pseudo-GNSS data presumed to be obtained in this case is output (generated) (step S03).
  • the generator 110 may further input a random number into the pseudo data generation model M1. As a result, the generator 110 can generate different pseudo GNSS data even when the same simulated travel route data is input.
  • the generator 110 inputs the vehicle type of the vehicle and learns the pseudo data generation model M1
  • the generator 110 inputs an arbitrary vehicle type into the pseudo data generation model M1 and generates pseudo GNSS data according to the vehicle type specified in advance. You may.
  • the generator 110 outputs the generated pseudo GNSS data to the discriminator, and receives the authenticity determination result by the discriminator 111. It is assumed that the discriminator 111 has already learned the discriminative model M2. Then, the generator 110 further learns the pseudo data generation model M1 based on the determination result output from the discriminator 111 (step S04).
  • the generator 110 repeatedly executes steps S03 to S04 to learn the pseudo data generation model M1 so that the generated pseudo GNSS data is determined to be "true (actually measured GNSS data)" by the discriminator 111.
  • FIG. 5 is a second flowchart showing an example of processing of the learning unit according to the embodiment of the present disclosure.
  • the flow of the learning process of the discriminator 111 of the learning unit 11 will be described with reference to FIG.
  • the discriminator 111 learns the discriminative model M2 based on the GNSS data with the truth label and the 3D map around the traveling position of the vehicle identified from the GNSS data (step S11). .. Specifically, the discriminator 111 uses the identification model M2 with actually measured GNSS data with a label indicating that it is actually measured data (true) and pseudo data (false) with a label indicating that it is pseudo data (false). The GNSS data and the 3D map data are input to learn the authenticity identification of the GNSS data.
  • the learned discriminative model M2 is stored in the storage medium 20.
  • the discriminator 111 inputs, for example, the GNSS data from which the truth label has been removed and the 3D map data into the identification model M2 when the learning of a predetermined number of data is completed, and the input GNSS data is actually measured GNSS. It is determined whether the data is data (true) or pseudo GNSS data (false) (step S12). At this time, the discriminator 111 randomly inputs either the actually measured GNSS data or the pseudo GNSS data into the discriminative model M2.
  • the discriminator 111 further learns the discriminative model M2 based on its own determination result (step S13). For example, the discriminator 111 stores whether the GNSS data input to the identification model M2 is the measured GNSS data or the pseudo GNSS data, and compares it with the determination result of the identification model M2 to determine the success or failure of the determination.
  • the discriminative model M2 is trained according to the above.
  • the discriminator 111 repeatedly executes steps S12 to S13 to learn the discriminative model M2 so that the authenticity of the input GNSS data can be correctly determined.
  • the data generation device 1 uses the pseudo data generation model M1 to generate pseudo GSNN data that closely resembles the measured GNSS data by causing the generator 110 and the discriminator 111 of the learning unit 11 to alternately and repeatedly execute the above-mentioned learning. Can be generated.
  • FIG. 6 is a first flowchart showing an example of processing of the pseudo data generation unit according to the embodiment of the present disclosure. Hereinafter, the processing flow of the pseudo data generation unit 12 will be described with reference to FIG.
  • the pseudo data generation unit 12 first accepts the input of the simulated travel route data by the user of the data generation device (step S21).
  • the pseudo data generation unit 12 inputs the input simulated travel route data and the 3D map data around the simulated travel route into the pseudo data generation model M1 learned by the learning unit 11, and pseudo GNSS data. Is output (generated) (step S22). Twice
  • the pseudo data generation unit 12 may further input a random number into the pseudo data generation model M1. As a result, the pseudo data generation unit 12 can generate different pseudo GNSS data even when the same simulated travel route data is input.
  • the pseudo data generation unit 12 may further accept the designation of the vehicle type together with the simulated travel route data in step S21. Thereby, for example, when the measured GNSS data of a specific vehicle type (for example, "large”) is insufficient for a certain traveling route, the pseudo GNSS data of this vehicle type can be generated and supplemented.
  • a specific vehicle type for example, "large”
  • FIG. 7 is a diagram showing an example of the hardware configuration of the data generation device according to the embodiment of the present disclosure. Hereinafter, the hardware configuration of the data generation device 1 according to the present embodiment will be described with reference to FIG. 7.
  • the computer 900 includes a processor 901, a main storage device 902, an auxiliary storage device 903, and an interface 904.
  • the above-mentioned data generation device 1 is mounted on one or more computers 900.
  • the operation of each of the above-mentioned functional units is stored in the auxiliary storage device 903 in the form of a program.
  • the processor 901 reads a program from the auxiliary storage device 903, deploys it to the main storage device 902, and executes the above processing according to the program. Further, the processor 901 secures a storage area corresponding to each of the above-mentioned storage units in the main storage device 902 according to the program.
  • Examples of the processor 901 include a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), a microprocessor, and the like.
  • the program may be for realizing a part of the functions exerted on the computer 900.
  • the program may exert its function in combination with another program already stored in the auxiliary storage device 903, or in combination with another program mounted on the other device.
  • the computer 900 may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or in place of the above configuration.
  • PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array).
  • PLDs Programmable Logic Device
  • PAL Programmable Array Logic
  • GAL Generic Array Logic
  • CPLD Complex Programmable Logic Device
  • FPGA Field Programmable Gate Array
  • auxiliary storage device 903 examples include HDD (Hard Disk Drive), SSD (Solid State Drive), magnetic disk, optical magnetic disk, CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disc Read Only). Memory), semiconductor memory and the like.
  • the auxiliary storage device 903 may be an internal medium directly connected to the bus of the computer 900, or an external storage device 910 connected to the computer 900 via the interface 904 or a communication line. When this program is distributed to the computer 900 via a communication line, the distributed computer 900 may expand the program to the main storage device 902 and execute the above processing.
  • the auxiliary storage device 903 is a non-temporary tangible storage medium.
  • the program may be for realizing a part of the above-mentioned functions. Further, the program may be a so-called difference file (difference program) that realizes the above-mentioned function in combination with another program already stored in the auxiliary storage device 903.
  • difference file difference program
  • the data generation device 1 is simulated by inputting the simulated travel route data and the 3D map data into the pseudo data generation model M1 learned based on the actually measured GNSS data and the 3D map data. Generate GNSS data. By doing so, the data generation device 1 learns the relationship between the actually measured GNSS data acquired when the vehicle is actually driven and the terrain around the vehicle's traveling position, the environment such as buildings, and the actual measurement. Pseudo GNSS data that closely resembles GNSS data can be generated. As a result, GNSS data for verification can be easily prepared without causing the vehicle to travel on a large number of travel routes.
  • the learning unit 11 of the data generation device 1 has a generator 110 for learning the pseudo data generation model M1 and a discriminator 111 for determining the authenticity of the input GNSS data.
  • the generator 110 further learns the pseudo data generation model M1 based on the determination result of the discriminator 111. By doing so, the data generation device 1 can generate elaborate pseudo GNSS data that is determined to be true data (actually measured GNSS data) in the discriminator 111, so that the pseudo data generation model M1 Can be further learned.
  • the learning unit 11 of the data generation device 1 uses the identification model M2 trained based on the GNSS data with a label indicating whether it is the measured GNSS data or the pseudo GNSS data and the 3D map data. , Judge the authenticity of the input GNSS data. By doing so, the data generation device 1 can improve the accuracy of determining the authenticity (accuracy of distinguishing the actually measured GNSS data from the pseudo GNSS data). Further, the data generation device 1 further trains the pseudo data generation model M1 so as to generate more elaborate pseudo GNSS data such that the discriminator 111 having high determination accuracy is determined to be true data. Can be done.
  • the learning unit 11 of the data generation device 1 may further have a travel route history creation unit that creates a travel route history of the vehicle based on the actually measured GNSS data. As a result, the data generation device 1 can more accurately learn the relationship between the traveling route and the GNSS data obtained when traveling along the traveling route.
  • the learning unit 11 of the data generation device 1 may further learn the pseudo data generation model M1 based on the vehicle type of the vehicle that has acquired the actually measured GNSS data. As a result, the data generation device 1 can generate pseudo GNSS data according to the vehicle type. As a result, when the measured GNSS data of a specific vehicle model is insufficient, the pseudo GNSS data of the vehicle model can be generated and used for verification or the like.
  • the data generator (1) is based on the measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
  • the pseudo data generation model includes a learning unit (11) that learns a pseudo data generation model, simulated travel path data representing a time series of virtual positions where the vehicle virtually moves, and 3D map data including the virtual position. It is provided with a pseudo data generation unit (12) that generates pseudo GNSS data in which GNSS data obtained at each of the virtual positions is estimated by inputting to.
  • the data generator learns the relationship between the measured GNSS data acquired when the vehicle is actually driven and the terrain around the vehicle's running position and the environment such as buildings, and the measured GNSS. Pseudo GNSS data that closely resembles the data can be generated. As a result, GNSS data for verification can be easily prepared without causing the vehicle to travel on a large number of travel routes.
  • the learning unit (11) is the pseudo data based on the actually measured GNSS data and the 3D map data.
  • a generator (110) that learns the generation model, inputs the simulated travel route data and the 3D map data to the learned pseudo data generation model, and outputs the pseudo GNSS data, and the authenticity of the input GNSS data. It has a discriminator (111) for determining the above. The generator (110) further learns the pseudo data generation model based on the determination result of the discriminator (111).
  • the data generator further learns the pseudo data generation model so that the discriminator can generate elaborate pseudo GNSS data that is determined to be true data (actually measured GNSS data). can do.
  • the discriminator (111) has a label indicating whether it is actually measured GNSS data or pseudo GNSS data. Using the identification model learned based on the attached GNSS data and the 3D map data, the authenticity of the input GNSS data is determined.
  • the data generation device can improve the authenticity determination accuracy (accuracy of distinguishing the measured GNSS data from the pseudo GNSS data). Further, the data generation device 1 can further train the pseudo data generation model so as to generate more elaborate pseudo GNSS data such that the discriminator having high determination accuracy determines that the data is true data. ..
  • the learning unit (11) is the vehicle based on the actually measured GNSS data. Further, it has a travel route history creation unit (112) that creates a travel route history representing a time series of travel positions, and further learns the pseudo data generation model based on the travel route history.
  • the data generator can more accurately learn the relationship between the traveling route and the GNSS data obtained when traveling along this traveling route.
  • the learning unit (11) is the vehicle that has acquired the measured GNSS data.
  • the pseudo data generation model is further learned based on the vehicle type, and the pseudo data generation unit (12) generates the pseudo GNSS data according to the designated vehicle type.
  • the data generation method generates pseudo data based on the actually measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the actually measured GNSS data.
  • the step of learning the model, the simulated travel route data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model, and the virtual It has a step of generating pseudo GNSS data inferring the GNSS data obtained at each of the positions.
  • the manufacturing method of the data generator (1) includes the actually measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the actually measured GNSS data.
  • the step of learning the pseudo data generation model based on the above, and the simulated travel route data representing the time series of the virtual position where the vehicle virtually moves and the 3D map data are input to the learned pseudo data generation model. It has a step of outputting the pseudo GNSS data, a step of determining the authenticity of the input GNSS data, and a step of further learning the pseudo data generation model based on the authenticity determination result.
  • the program creates a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
  • the step to be learned, the simulated travel path data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model to obtain the virtual position.
  • the computer (900) of the data generator (1) is made to execute a step of generating pseudo GNSS data in which the GNSS data obtained in each is estimated.

Abstract

A data generation device (1) comprises: a learning unit (11) that learns a pseudo-data generation model on the basis of 3D map data including actual GNSS data acquired by causing a vehicle to travel, and the position of the vehicle indicated by the actual GNSS data; and a pseudo-data generation unit (12) that inputs, to the pseudo-data generation model, imitation travel route data representing a time series of virtual positions to which the vehicle virtually moves, and 3D map data including the virtual positions, and generates pseudo GNSS data in which GNSS data obtained for each of the virtual positions is inferred.

Description

データ生成装置、データ生成方法、データ生成装置の製造方法、及びプログラムData generator, data generation method, data generator manufacturing method, and program
　本発明は、データ生成装置、データ生成方法、データ生成装置の製造方法、及びプログラムに関する。 The present invention relates to a data generation device, a data generation method, a method for manufacturing a data generation device, and a program.
　近年では、ＧＮＳＳ（Global Navigation Satellite System）の技術を利用して、ＧＮＳＳ受信機が受信した衛星信号から車両の位置（緯度、経度、高度）を測位する技術が広く使われている（例えば、特許文献１）。 In recent years, a technique for positioning the position (latitude, longitude, altitude) of a vehicle from a satellite signal received by a GNSS receiver using GNSS (Global Navigation Satellite System) technology has been widely used (for example, a patent). Document 1).
　また、ＧＮＳＳ受信機による測位データ（以下、「ＧＮＳＳデータ」とも記載する。）をサーバで収集し、マップマッチング処理により車両の走行経路を特定するシステムが考えられている。このようなシステムでは、マップマッチング処理、及びマップマッチング処理を用いたサービス処理（例えば、有料道路の課金処理等）の検証等を行うために、多数のＧＮＳＳデータが必要となる。 Further, a system is considered in which positioning data by a GNSS receiver (hereinafter, also referred to as "GNSS data") is collected by a server and a vehicle's travel route is specified by map matching processing. In such a system, a large amount of GNSS data is required for verification of map matching processing and service processing using map matching processing (for example, toll road billing processing).
　走行経路の組み合わせは膨大であり、全ての走行経路について車両を走行させて実測したＧＮＳＳデータを収集することは費用及び時間がかかり困難である。このため、車両が走行経路に沿って走行した場合に取得されるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成し、この疑似ＧＮＳＳデータを各種処理の検証に用いることが考えられる。 The combination of travel routes is enormous, and it is costly and time-consuming to collect GNSS data measured by traveling a vehicle on all travel routes. Therefore, it is conceivable to generate pseudo GNSS data that estimates the GNSS data acquired when the vehicle travels along the traveling route, and use the pseudo GNSS data for verification of various processes.
特開２０１３－２０５３７３号公報Japanese Unexamined Patent Publication No. 2013-205373
　しかしながら、生成した疑似ＧＮＳＳデータと、実測したＧＮＳＳデータとの乖離が大きいと、検証時に正しい結果を得られない可能性がある。このため、実測したＧＮＳＳデータに精巧に似せた疑似ＧＮＳＳデータを生成する技術が求められていた。 However, if there is a large discrepancy between the generated pseudo GNSS data and the actually measured GNSS data, there is a possibility that correct results cannot be obtained at the time of verification. Therefore, there has been a demand for a technique for generating pseudo GNSS data that closely resembles the actually measured GNSS data.
　本開示の一態様によれば、データ生成装置は、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習する学習部と、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成する疑似データ生成部と、を備える。 According to one aspect of the present disclosure, the data generator creates a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data. The learning unit to be learned, simulated travel path data representing a time series of virtual positions where the vehicle virtually moves, and 3D map data including the virtual position are input to the pseudo data generation model, and the virtual position is input. It is provided with a pseudo data generation unit that generates pseudo GNSS data in which the GNSS data obtained in each of the above is estimated.
　本開示の一態様によれば、データ生成方法は、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成するステップと、を有する。 According to one aspect of the present disclosure, the data generation method creates a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data. The step to be learned, the simulated travel path data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model to obtain the virtual position. Each has a step of generating pseudo GNSS data inferring the GNSS data obtained in each.
　本開示の一態様によれば、データ生成装置（１）の製造方法は、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、学習した前記疑似データ生成モデルに、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと前記３Ｄ地図データとを入力して前記疑似ＧＮＳＳデータを出力するステップと、入力されたＧＮＳＳデータの真偽を判定するステップと、前記真偽の判定結果に基づいて、前記疑似データ生成モデルを更に学習するステップと、を有する。 According to one aspect of the present disclosure, the manufacturing method of the data generator (1) is based on the measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data. The step of learning the pseudo data generation model and the simulated travel route data representing the time series of the virtual positions where the vehicle virtually moves and the 3D map data are input to the learned pseudo data generation model. It includes a step of outputting pseudo GNSS data, a step of determining the authenticity of the input GNSS data, and a step of further learning the pseudo data generation model based on the authenticity determination result.
　本開示の一態様によれば、プログラムは、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成するステップと、をデータ生成装置のコンピュータに実行させる。 According to one aspect of the present disclosure, the program learns a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data. The step, the simulated travel route data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model, and at each of the virtual positions. The computer of the data generator is made to execute the step of generating the pseudo GNSS data inferring the obtained GNSS data.
　本開示に係るデータ生成装置、データ生成方法、データ生成装置の製造方法、及びプログラムによれば、実測したＧＮＳＳデータに精巧に似せた疑似ＧＮＳＳデータを生成することができる。 According to the data generation device, the data generation method, the manufacturing method of the data generation device, and the program according to the present disclosure, it is possible to generate pseudo GNSS data that closely resembles the actually measured GNSS data.
本開示の一実施形態に係るデータ生成装置の機能構成を示す図である。It is a figure which shows the functional structure of the data generation apparatus which concerns on one Embodiment of this disclosure. 本開示の一実施形態に係る学習部の機能構成を示す図である。It is a figure which shows the functional structure of the learning part which concerns on one Embodiment of this disclosure. 本開示の一実施形態に係る疑似データ生成部の機能構成を示す図である。It is a figure which shows the functional structure of the pseudo data generation part which concerns on one Embodiment of this disclosure. 本開示の一実施形態に係る学習部の処理の一例を示す第１のフローチャートである。It is a 1st flowchart which shows an example of the processing of the learning part which concerns on one Embodiment of this disclosure. 本開示の一実施形態に係る学習部の処理の一例を示す第２のフローチャートである。It is a 2nd flowchart which shows an example of the processing of the learning part which concerns on one Embodiment of this disclosure. 本開示の一実施形態に係る疑似データ生成部の処理の一例を示す第１のフローチャートである。It is a 1st flowchart which shows an example of the processing of the pseudo data generation part which concerns on one Embodiment of this disclosure. 本開示の一実施形態に係るデータ生成装置のハードウェア構成の一例を示す図である。It is a figure which shows an example of the hardware configuration of the data generation apparatus which concerns on one Embodiment of this disclosure.
　以下、本開示の一実施形態に係るデータ生成装置１について、図１～図７を参照しながら説明する。 Hereinafter, the data generation device 1 according to the embodiment of the present disclosure will be described with reference to FIGS. 1 to 7.
（データ生成装置の機能構成）
　図１は、本開示の一実施形態に係るデータ生成装置の機能構成を示す図である。
　図１に示すように、データ生成装置１は、ＣＰＵ１０と、記憶媒体２０とを備えている。 (Functional configuration of data generator)
FIG. 1 is a diagram showing a functional configuration of a data generation device according to an embodiment of the present disclosure.
As shown in FIG. 1, the data generation device 1 includes a CPU 10 and a storage medium 20.
　ＣＰＵ１０は、データ生成装置１全体の動作を司るプロセッサであり、所定のプログラムに従って動作することにより、学習部１１、疑似データ生成部１２としての機能を発揮する。 The CPU 10 is a processor that controls the operation of the entire data generation device 1, and by operating according to a predetermined program, it exhibits functions as a learning unit 11 and a pseudo data generation unit 12.
　学習部１１は、車両を走行させて取得した実測ＧＮＳＳデータと、実測ＧＮＳＳデータが示す車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習する。 The learning unit 11 learns the pseudo data generation model based on the actually measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the actually measured GNSS data.
　実測ＧＮＳＳデータには、例えば車両の位置を表す緯度、経度、高度等の情報が含まれている。また、実測ＧＮＳＳデータには、移動速度、移動方位、捕捉した衛星の数、捕捉した衛星を特定可能な衛星番号、信号強度等が含まれていてもよい。３Ｄ地図データは、地形、建築物等の３次元形状を特定可能な地図データである。 The measured GNSS data includes information such as latitude, longitude, and altitude that represent the position of the vehicle, for example. Further, the measured GNSS data may include the moving speed, the moving direction, the number of captured satellites, the satellite number capable of identifying the captured satellite, the signal strength, and the like. The 3D map data is map data capable of specifying a three-dimensional shape of a terrain, a building, or the like.
　疑似データ生成部１２は、車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、仮想位置を含む３Ｄ地図データとを疑似データ生成モデルに入力して、仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成する。即ち、疑似データ生成部１２は、実際に車両を走行させることなく、模擬走行経路において得られると推測されるＧＮＳＳデータを疑似的に生成する。 The pseudo data generation unit 12 inputs simulated travel route data representing a time series of virtual positions where the vehicle virtually moves and 3D map data including the virtual positions into the pseudo data generation model, and at each of the virtual positions. Pseudo GNSS data is generated by estimating the obtained GNSS data. That is, the pseudo data generation unit 12 pseudo-generates GNSS data that is presumed to be obtained in the simulated travel route without actually driving the vehicle.
　模擬走行経路データに含まれる仮想位置は、例えばリンク（道路）を特定可能なリンクＩＤと、リンク上の位置を示す情報（リンク始点からの距離等）とにより表される。また、模擬走行経路データは、例えばデータ生成装置１のユーザにより、マップマッチング処理及び／又はマップマッチング処理を用いたサービス処理の検証対象となる走行経路が予め入力される。 The virtual position included in the simulated travel route data is represented by, for example, a link ID that can identify the link (road) and information indicating the position on the link (distance from the link start point, etc.). Further, as the simulated travel route data, for example, the user of the data generation device 1 inputs in advance a travel route to be verified for the map matching process and / or the service process using the map matching process.
　記憶媒体２０は、いわゆる補助記憶装置であって、ＨＤＤ（Hard Disk Drive）、ＳＳＤ（Solid State Drive）などである。記憶媒体２０には、データ生成装置１が取得、生成した各種データが保存される。例えば、記憶媒体２０には、実測ＧＮＳＳデータ、３Ｄ地図データ、学習部１１により学習された疑似データ生成モデル、疑似データ生成部１２により生成された疑似ＧＮＳＳデータ等が記憶される。 The storage medium 20 is a so-called auxiliary storage device, such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). Various data acquired and generated by the data generation device 1 are stored in the storage medium 20. For example, the storage medium 20 stores actually measured GNSS data, 3D map data, a pseudo data generation model learned by the learning unit 11, pseudo GNSS data generated by the pseudo data generation unit 12, and the like.
　車両の周辺環境（衛星信号を遮蔽、反射する地形、建築物の有無等）により、ＧＮＳＳ受信機による衛星信号の受信状態（捕捉する衛星の数、捕捉した衛星から受信する信号の強度等）が変化する。このため、ＧＮＳＳ受信機が計測する実測ＧＮＳＳデータには、衛星信号の受信状態に応じた測位誤差が含まれる。 Depending on the surrounding environment of the vehicle (the terrain that shields and reflects satellite signals, the presence or absence of buildings, etc.), the reception status of satellite signals by the GNSS receiver (number of satellites to be captured, strength of signals received from captured satellites, etc.) Change. Therefore, the measured GNSS data measured by the GNSS receiver includes a positioning error according to the reception state of the satellite signal.
　この測位誤差を考慮せずに、単に車両の仮想位置（緯度、経度）を含むＧＮＳＳデータを疑似的に生成しただけでは、実測ＧＮＳＳデータとは大きく乖離したデータが生成される可能性がある。 If GNSS data including the virtual position (latitude, longitude) of the vehicle is simply generated without considering this positioning error, there is a possibility that data that deviates significantly from the measured GNSS data will be generated.
　このため、本実施形態に係るデータ生成装置１は、車両の走行経路に応じた測位誤差が反映された疑似データ生成モデルを学習し、学習された疑似データ生成モデルを用いて実測ＧＮＳＳデータを精巧に模倣した疑似ＧＮＳＳデータを生成する。以下、疑似データ生成モデルの学習を行う学習部１１と、学習された疑似データ生成モデルＭ１を用いて疑似ＧＮＳＳデータを生成する疑似データ生成部１２の機能構成について説明する。 Therefore, the data generation device 1 according to the present embodiment learns a pseudo data generation model that reflects the positioning error according to the traveling route of the vehicle, and uses the learned pseudo data generation model to elaborate the measured GNSS data. Generate pseudo GNSS data that imitates. Hereinafter, the functional configurations of the learning unit 11 that learns the pseudo data generation model and the pseudo data generation unit 12 that generates pseudo GNSS data using the learned pseudo data generation model M1 will be described.
（学習部の機能構成）
　図２は、本開示の一実施形態に係る学習部の機能構成を示す図である。
　図２に示すように、学習部１１は、ジェネレータ１１０と、ディスクリミネータ１１１と、走行経路履歴作成部１１２とを有している。本実施形態にかかる学習部１１は、敵対的生成ネットワーク（ＧＡＮ：Generative Adversarial Networks）の技術を利用して、疑似ＧＮＳＳデータを生成する疑似データ生成モデルＭ１を学習する。 (Functional configuration of the learning department)
FIG. 2 is a diagram showing a functional configuration of a learning unit according to an embodiment of the present disclosure.
As shown in FIG. 2, the learning unit 11 includes a generator 110, a discriminator 111, and a traveling route history creating unit 112. The learning unit 11 according to the present embodiment learns the pseudo data generation model M1 that generates pseudo GNSS data by using the technology of the hostile generation network (GAN: Generative Adversarial Networks).
　ジェネレータ１１０は、実測ＧＮＳＳデータと、３Ｄ地図データとに基づいて疑似データ生成モデルＭ１を学習し、学習した疑似データ生成モデルＭ１に模擬走行経路データと３Ｄ地図データとを入力して疑似ＧＮＳＳデータを出力する。 The generator 110 learns the pseudo data generation model M1 based on the actually measured GNSS data and the 3D map data, and inputs the simulated travel route data and the 3D map data into the learned pseudo data generation model M1 to input the pseudo GNSS data. Output.
　ディスクリミネータ１１１は、入力されたＧＮＳＳデータの真偽を判定する。具体的には、ディスクリミネータ１１１は、実測ＧＮＳＳデータと、疑似ＧＮＳＳデータであることを示すラベルを付した疑似ＧＮＳＳデータと、３Ｄ地図データとに基づいて学習された識別モデルＭ２を用いて真偽を判定する。 The discriminator 111 determines the authenticity of the input GNSS data. Specifically, the discriminator 111 is true using the discriminative model M2 learned based on the actually measured GNSS data, the pseudo GNSS data labeled to indicate that it is the pseudo GNSS data, and the 3D map data. Judge false.
　走行経路履歴作成部１１２は、実測ＧＮＳＳデータに基づいて、車両の走行位置の時系列を表す走行経路履歴を作成する。 The travel route history creation unit 112 creates a travel route history representing the time series of the vehicle's travel position based on the actually measured GNSS data.
（疑似データ生成部の機能構成）
　図３は、本開示の一実施形態に係る疑似データ生成部の機能構成を示す図である。
　図３に示すように、疑似データ生成部１２は、学習部１１により学習された疑似データ生成モデルＭ１に模擬走行経路データ及び３Ｄ地図を入力して、疑似ＧＮＳＳデータを生成する。このとき、疑似データ生成部１２は、疑似データ生成モデルＭ１に乱数を更に入力してもよい。これにより、疑似データ生成部１２は、同一の模擬走行経路データから、乱数により変化をつけられた複数種類の疑似ＧＮＳＳデータを生成することができる。 (Functional configuration of pseudo data generator)
FIG. 3 is a diagram showing a functional configuration of a pseudo data generation unit according to an embodiment of the present disclosure.
As shown in FIG. 3, the pseudo data generation unit 12 inputs the simulated travel route data and the 3D map into the pseudo data generation model M1 learned by the learning unit 11 to generate pseudo GNSS data. At this time, the pseudo data generation unit 12 may further input a random number into the pseudo data generation model M1. As a result, the pseudo data generation unit 12 can generate a plurality of types of pseudo GNSS data changed by random numbers from the same simulated travel route data.
（学習部の処理フロー）
　図４は、本開示の一実施形態に係る学習部の処理の一例を示す第１のフローチャートである。
　以下、図４を参照しながら、学習部１１のジェネレータ１１０の学習処理の流れについて説明する。 (Processing flow of learning department)
FIG. 4 is a first flowchart showing an example of processing of the learning unit according to the embodiment of the present disclosure.
Hereinafter, the flow of the learning process of the generator 110 of the learning unit 11 will be described with reference to FIG.
　図４に示すように、走行経路履歴作成部１１２は、車両を実際に走行させて得た実測ＧＮＳＳデータから、当該車両の走行経路履歴を作成する（ステップＳ０１）。例えば、走行経路履歴作成部１１２は、マップマッチング処理を行い、実測ＧＮＳＳデータの時刻歴から、車両の走行位置（リンク上の位置）の時系列を示す走行経路履歴を作成する。 As shown in FIG. 4, the travel route history creation unit 112 creates a travel route history of the vehicle from the actually measured GNSS data obtained by actually traveling the vehicle (step S01). For example, the travel route history creation unit 112 performs map matching processing and creates a travel route history indicating a time series of the vehicle's travel position (position on the link) from the time history of the measured GNSS data.
　次に、ジェネレータ１１０は、実測ＧＮＳＳデータと、実測ＧＮＳＳデータから特定される車両の走行位置周辺の３Ｄ地図と基づいて、疑似データ生成モデルＭ１を学習する（ステップＳ０２）。疑似データ生成モデルＭ１は、例えばリカレントニューラルネットワークを利用したモデルであって、走行経路及び走行経路周辺の３Ｄ地図データを入力すると、この走行経路を走行した場合に得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを出力する。ジェネレータ１１０は、車両の走行位置周辺の３Ｄ地図データを学習に用いることにより、車両周辺の地形、建築物等の影響に応じてどのような実測ＧＮＳＳデータが取得できるかを疑似データ生成モデルＭ１に学習させる。学習した疑似データ生成モデルＭ１は、記憶媒体２０に記憶される。 Next, the generator 110 learns the pseudo data generation model M1 based on the actually measured GNSS data and the 3D map around the traveling position of the vehicle specified from the actually measured GNSS data (step S02). The pseudo data generation model M1 is, for example, a model using a recurrent neural network, and when 3D map data of the travel route and the vicinity of the travel route is input, the pseudo GNSS that estimates the GNSS data obtained when traveling on this travel route is estimated. Output data. The generator 110 uses the 3D map data around the traveling position of the vehicle for learning, and uses the pseudo data generation model M1 to determine what kind of measured GNSS data can be acquired according to the influence of the terrain, buildings, etc. around the vehicle. Let them learn. The learned pseudo data generation model M1 is stored in the storage medium 20.
　また、ジェネレータ１１０は、走行経路履歴作成部１１２が作成した走行経路履歴を利用して、車両の走行経路に応じた実測ＧＮＳＳデータを疑似データ生成モデルＭ１に学習させてもよい。 Further, the generator 110 may make the pseudo data generation model M1 learn the measured GNSS data according to the travel route of the vehicle by using the travel route history created by the travel route history creation unit 112.
　更に、ジェネレータ１１０は、実測ＧＮＳＳデータを取得した車両の車種に基づいて、疑似データ生成モデルＭ１を更に学習させてもよい。車両の車種によってＧＮＳＳ受信機の取り付け位置が異なる場合がある。そうすると、衛星信号の受信状態が変化し、測位誤差の傾向も変わる可能性がある。このため、ジェネレータ１１０は、疑似データ生成モデルＭ１に実測ＧＮＳＳデータに加えて、当該実測ＧＮＳＳデータを取得した車両の車種を入力して、疑似データ生成モデルＭ１を学習する。なお、実測ＧＮＳＳデータを取得した車両の車種は、例えばデータ生成装置１のユーザにより入力される。 Further, the generator 110 may further learn the pseudo data generation model M1 based on the vehicle type of the vehicle for which the actually measured GNSS data has been acquired. The mounting position of the GNSS receiver may differ depending on the vehicle type. Then, the reception state of the satellite signal may change, and the tendency of the positioning error may also change. Therefore, the generator 110 learns the pseudo data generation model M1 by inputting the vehicle type of the vehicle that has acquired the actually measured GNSS data into the pseudo data generation model M1 in addition to the actually measured GNSS data. The vehicle type of the vehicle for which the actually measured GNSS data has been acquired is input by, for example, the user of the data generation device 1.
　次に、例えば所定データ数の学習が完了すると、ジェネレータ１１０は、学習された疑似データ生成モデルＭ１に模擬走行経路データと、模擬走行経路周辺の３Ｄ地図とを入力して、模擬走行経路を走行した場合に得られると推測される疑似ＧＮＳＳデータを出力（生成）させる（ステップＳ０３）。 Next, for example, when the learning of a predetermined number of data is completed, the generator 110 inputs the simulated travel route data and the 3D map around the simulated travel route into the learned pseudo data generation model M1 and travels on the simulated travel route. Pseudo-GNSS data presumed to be obtained in this case is output (generated) (step S03).
　このとき、ジェネレータ１１０は、疑似データ生成モデルＭ１に乱数を更に入力してもよい。これにより、ジェネレータ１１０は、同一の模擬走行経路データを入力した場合であっても、異なる疑似ＧＮＳＳデータを生成することができる。 At this time, the generator 110 may further input a random number into the pseudo data generation model M1. As a result, the generator 110 can generate different pseudo GNSS data even when the same simulated travel route data is input.
　また、ジェネレータ１１０は、車両の車種を入力して疑似データ生成モデルＭ１を学習した場合、任意の車種を疑似データ生成モデルＭ１に入力して、予め指定された車種に応じた疑似ＧＮＳＳデータを生成してもよい。 Further, when the generator 110 inputs the vehicle type of the vehicle and learns the pseudo data generation model M1, the generator 110 inputs an arbitrary vehicle type into the pseudo data generation model M1 and generates pseudo GNSS data according to the vehicle type specified in advance. You may.
　ジェネレータ１１０は、生成した疑似ＧＮＳＳデータをディスクリミネータに出力し、ディスクリミネータ１１１による真偽判定結果を受領する。なお、このディスクリミネータ１１１は、識別モデルＭ２を学習済みであるとする。そうすると、ジェネレータ１１０は、ディスクリミネータ１１１から出力された判定結果に基づいて、疑似データ生成モデルＭ１を更に学習する（ステップＳ０４）。 The generator 110 outputs the generated pseudo GNSS data to the discriminator, and receives the authenticity determination result by the discriminator 111. It is assumed that the discriminator 111 has already learned the discriminative model M2. Then, the generator 110 further learns the pseudo data generation model M1 based on the determination result output from the discriminator 111 (step S04).
　ジェネレータ１１０は、生成した疑似ＧＮＳＳデータがディスクリミネータ１１１により「真（実測ＧＮＳＳデータ）」と判定されるように、ステップＳ０３～Ｓ０４を繰り返し実行して、疑似データ生成モデルＭ１を学習する。 The generator 110 repeatedly executes steps S03 to S04 to learn the pseudo data generation model M1 so that the generated pseudo GNSS data is determined to be "true (actually measured GNSS data)" by the discriminator 111.
　図５は、本開示の一実施形態に係る学習部の処理の一例を示す第２のフローチャートである。
　以下、図５を参照しながら、学習部１１のディスクリミネータ１１１の学習処理の流れについて説明する。 FIG. 5 is a second flowchart showing an example of processing of the learning unit according to the embodiment of the present disclosure.
Hereinafter, the flow of the learning process of the discriminator 111 of the learning unit 11 will be described with reference to FIG.
　図５に示すように、ディスクリミネータ１１１は、真偽ラベル付きのＧＮＳＳデータと、ＧＮＳＳデータから特定される車両の走行位置周辺の３Ｄ地図と基づいて、識別モデルＭ２を学習する（ステップＳ１１）。具体的には、ディスクリミネータ１１１は、識別モデルＭ２に、実測データ（真）であることを示すラベルを付した実測ＧＮＳＳデータと、疑似データ（偽）であることを示すラベルを付した疑似ＧＮＳＳデータと、３Ｄ地図データとを入力して、ＧＮＳＳデータの真偽の識別を学習させる。学習した識別モデルＭ２は、記憶媒体２０に記憶される。 As shown in FIG. 5, the discriminator 111 learns the discriminative model M2 based on the GNSS data with the truth label and the 3D map around the traveling position of the vehicle identified from the GNSS data (step S11). .. Specifically, the discriminator 111 uses the identification model M2 with actually measured GNSS data with a label indicating that it is actually measured data (true) and pseudo data (false) with a label indicating that it is pseudo data (false). The GNSS data and the 3D map data are input to learn the authenticity identification of the GNSS data. The learned discriminative model M2 is stored in the storage medium 20.
　次に、ディスクリミネータ１１１は、例えば所定データ数の学習が完了すると、真偽ラベルを外したＧＮＳＳデータと、３Ｄ地図データとを識別モデルＭ２に入力して、入力されたＧＮＳＳデータが実測ＧＮＳＳデータ（真）であるか、疑似ＧＮＳＳデータ（偽）であるかを判定する（ステップＳ１２）。このとき、ディスクリミネータ１１１は、識別モデルＭ２に実測ＧＮＳＳデータ及び疑似ＧＮＳＳデータの何れか一方をランダムで入力する。 Next, the discriminator 111 inputs, for example, the GNSS data from which the truth label has been removed and the 3D map data into the identification model M2 when the learning of a predetermined number of data is completed, and the input GNSS data is actually measured GNSS. It is determined whether the data is data (true) or pseudo GNSS data (false) (step S12). At this time, the discriminator 111 randomly inputs either the actually measured GNSS data or the pseudo GNSS data into the discriminative model M2.
　次に、ディスクリミネータ１１１は、自身の判定結果に基づいて識別モデルＭ２を更に学習する（ステップＳ１３）。例えば、ディスクリミネータ１１１は、識別モデルＭ２に入力するＧＮＳＳデータが実測ＧＮＳＳデータであるか、疑似ＧＮＳＳデータであるかを記憶しておき、識別モデルＭ２の判定結果と照らし合わせて、判定の成否に応じて識別モデルＭ２を学習する。 Next, the discriminator 111 further learns the discriminative model M2 based on its own determination result (step S13). For example, the discriminator 111 stores whether the GNSS data input to the identification model M2 is the measured GNSS data or the pseudo GNSS data, and compares it with the determination result of the identification model M2 to determine the success or failure of the determination. The discriminative model M2 is trained according to the above.
　ディスクリミネータ１１１は、入力されたＧＮＳＳデータの真偽を正しく判定できるように、ステップＳ１２～Ｓ１３を繰り返し実行して、識別モデルＭ２を学習する。 The discriminator 111 repeatedly executes steps S12 to S13 to learn the discriminative model M2 so that the authenticity of the input GNSS data can be correctly determined.
　データ生成装置１は、学習部１１のジェネレータ１１０及びディスクリミネータ１１１に上述の学習を交互に繰り返し実行させることにより、疑似データ生成モデルＭ１を用いて実測ＧＮＳＳデータに精巧に似せた疑似ＧＳＮＮデータを生成することができる。 The data generation device 1 uses the pseudo data generation model M1 to generate pseudo GSNN data that closely resembles the measured GNSS data by causing the generator 110 and the discriminator 111 of the learning unit 11 to alternately and repeatedly execute the above-mentioned learning. Can be generated.
（疑似データ生成部の処理フロー）
　図６は、本開示の一実施形態に係る疑似データ生成部の処理の一例を示す第１のフローチャートである。
　以下、図６を参照しながら、疑似データ生成部１２の処理の流れについて説明する。 (Processing flow of pseudo data generator)
FIG. 6 is a first flowchart showing an example of processing of the pseudo data generation unit according to the embodiment of the present disclosure.
Hereinafter, the processing flow of the pseudo data generation unit 12 will be described with reference to FIG.
　図６に示すように、疑似データ生成部１２は、まず、データ生成装置のユーザによる模擬走行経路データの入力を受け付ける（ステップＳ２１）。 As shown in FIG. 6, the pseudo data generation unit 12 first accepts the input of the simulated travel route data by the user of the data generation device (step S21).
　次に、疑似データ生成部１２は、学習部１１により学習された疑似データ生成モデルＭ１に、入力された模擬走行経路データと、模擬走行経路周辺の３Ｄ地図データとを入力して、疑似ＧＮＳＳデータを出力（生成）させる（ステップＳ２２）。　 Next, the pseudo data generation unit 12 inputs the input simulated travel route data and the 3D map data around the simulated travel route into the pseudo data generation model M1 learned by the learning unit 11, and pseudo GNSS data. Is output (generated) (step S22). Twice
　このとき、疑似データ生成部１２は、疑似データ生成モデルＭ１に乱数を更に入力してもよい。これにより、疑似データ生成部１２は、同一の模擬走行経路データを入力した場合であっても、異なる疑似ＧＮＳＳデータを生成することができる。 At this time, the pseudo data generation unit 12 may further input a random number into the pseudo data generation model M1. As a result, the pseudo data generation unit 12 can generate different pseudo GNSS data even when the same simulated travel route data is input.
　また、疑似データ生成部１２は、ステップＳ２１において、模擬走行経路データとともに、車種の指定を更に受け付けてもよい。これにより、例えば、ある走行経路について、特定の車種（例えば、「大型」）の実測ＧＮＳＳデータが不足している場合は、この車種の疑似ＧＮＳＳデータを生成して補うことができる。 Further, the pseudo data generation unit 12 may further accept the designation of the vehicle type together with the simulated travel route data in step S21. Thereby, for example, when the measured GNSS data of a specific vehicle type (for example, "large") is insufficient for a certain traveling route, the pseudo GNSS data of this vehicle type can be generated and supplemented.
（データ生成装置のハードウェア構成）
　図７は、本開示の一実施形態に係るデータ生成装置のハードウェア構成の一例を示す図である。
　以下、図７を参照しながら、本実施形態に係るデータ生成装置１のハードウェア構成について説明する。 (Hardware configuration of data generator)
FIG. 7 is a diagram showing an example of the hardware configuration of the data generation device according to the embodiment of the present disclosure.
Hereinafter, the hardware configuration of the data generation device 1 according to the present embodiment will be described with reference to FIG. 7.
　コンピュータ９００は、プロセッサ９０１、主記憶装置９０２、補助記憶装置９０３、インタフェース９０４を備える。 The computer 900 includes a processor 901, a main storage device 902, an auxiliary storage device 903, and an interface 904.
　上述のデータ生成装置１は、一つ又は複数のコンピュータ９００に実装される。そして、上述した各機能部の動作は、プログラムの形式で補助記憶装置９０３に記憶されている。プロセッサ９０１は、プログラムを補助記憶装置９０３から読み出して主記憶装置９０２に展開し、当該プログラムに従って上記処理を実行する。また、プロセッサ９０１は、プログラムに従って、上述した各記憶部に対応する記憶領域を主記憶装置９０２に確保する。プロセッサ９０１の例としては、ＣＰＵ（Central Processing Unit）、ＧＰＵ（Graphic Processing Unit）、マイクロプロセッサなどが挙げられる。 The above-mentioned data generation device 1 is mounted on one or more computers 900. The operation of each of the above-mentioned functional units is stored in the auxiliary storage device 903 in the form of a program. The processor 901 reads a program from the auxiliary storage device 903, deploys it to the main storage device 902, and executes the above processing according to the program. Further, the processor 901 secures a storage area corresponding to each of the above-mentioned storage units in the main storage device 902 according to the program. Examples of the processor 901 include a CPU (Central Processing Unit), a GPU (Graphic Processing Unit), a microprocessor, and the like.
　プログラムは、コンピュータ９００に発揮させる機能の一部を実現するためのものであってもよい。例えば、プログラムは、補助記憶装置９０３に既に記憶されている他のプログラムとの組み合わせ、または他の装置に実装された他のプログラムとの組み合わせによって機能を発揮させるものであってもよい。なお、他の実施形態においては、コンピュータ９００は、上記構成に加えて、または上記構成に代えてＰＬＤ（Programmable Logic Device）などのカスタムＬＳＩ（Large Scale Integrated Circuit）を備えてもよい。ＰＬＤの例としては、ＰＡＬ(Programmable Array Logic)、ＧＡＬ(Generic Array Logic)、ＣＰＬＤ(Complex Programmable Logic Device)、ＦＰＧＡ（Field Programmable Gate Array）が挙げられる。この場合、プロセッサ９０１によって実現される機能の一部または全部が当該集積回路によって実現されてよい。このような集積回路も、プロセッサの一例に含まれる。 The program may be for realizing a part of the functions exerted on the computer 900. For example, the program may exert its function in combination with another program already stored in the auxiliary storage device 903, or in combination with another program mounted on the other device. In another embodiment, the computer 900 may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or in place of the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions realized by the processor 901 may be realized by the integrated circuit. Such integrated circuits are also included as an example of a processor.
　補助記憶装置９０３の例としては、ＨＤＤ（Hard Disk Drive）、ＳＳＤ（Solid State Drive）、磁気ディスク、光磁気ディスク、ＣＤ－ＲＯＭ（Compact Disc Read Only Memory）、ＤＶＤ－ＲＯＭ（Digital Versatile Disc Read Only Memory）、半導体メモリ等が挙げられる。補助記憶装置９０３は、コンピュータ９００のバスに直接接続された内部メディアであってもよいし、インタフェース９０４または通信回線を介してコンピュータ９００に接続される外部記憶装置９１０であってもよい。また、このプログラムが通信回線によってコンピュータ９００に配信される場合、配信を受けたコンピュータ９００が当該プログラムを主記憶装置９０２に展開し、上記処理を実行してもよい。少なくとも１つの実施形態において、補助記憶装置９０３は、一時的でない有形の記憶媒体である。 Examples of the auxiliary storage device 903 include HDD (Hard Disk Drive), SSD (Solid State Drive), magnetic disk, optical magnetic disk, CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disc Read Only). Memory), semiconductor memory and the like. The auxiliary storage device 903 may be an internal medium directly connected to the bus of the computer 900, or an external storage device 910 connected to the computer 900 via the interface 904 or a communication line. When this program is distributed to the computer 900 via a communication line, the distributed computer 900 may expand the program to the main storage device 902 and execute the above processing. In at least one embodiment, the auxiliary storage device 903 is a non-temporary tangible storage medium.
　また、当該プログラムは、前述した機能の一部を実現するためのものであってもよい。
さらに、当該プログラムは、前述した機能を補助記憶装置９０３に既に記憶されている他のプログラムとの組み合わせで実現するもの、いわゆる差分ファイル（差分プログラム）であってもよい。 Further, the program may be for realizing a part of the above-mentioned functions.
Further, the program may be a so-called difference file (difference program) that realizes the above-mentioned function in combination with another program already stored in the auxiliary storage device 903.
（作用効果）
　以上のように、本実施形態に係るデータ生成装置１は、実測ＧＮＳＳデータ及び３Ｄ地図データに基づいて学習した疑似データ生成モデルＭ１に、模擬走行経路データ及び３Ｄ地図データを入力することにより、疑似ＧＮＳＳデータを生成する。このようにすることで、データ生成装置１は、実際に車両を走行させたときに取得した実測ＧＮＳＳデータと、車両の走行位置周辺の地形、建築物等の環境との関連を学習し、実測ＧＮＳＳデータに精巧に似せた疑似ＧＮＳＳデータを生成することができる。これにより、車両に多数の走行経路を走行させることなく、検証用のＧＮＳＳデータを容易に準備することができる。 (Action effect)
As described above, the data generation device 1 according to the present embodiment is simulated by inputting the simulated travel route data and the 3D map data into the pseudo data generation model M1 learned based on the actually measured GNSS data and the 3D map data. Generate GNSS data. By doing so, the data generation device 1 learns the relationship between the actually measured GNSS data acquired when the vehicle is actually driven and the terrain around the vehicle's traveling position, the environment such as buildings, and the actual measurement. Pseudo GNSS data that closely resembles GNSS data can be generated. As a result, GNSS data for verification can be easily prepared without causing the vehicle to travel on a large number of travel routes.
　また、データ生成装置１の学習部１１は、疑似データ生成モデルＭ１を学習するジェネレータ１１０と、入力されたＧＮＳＳデータの真偽を判定するディスクリミネータ１１１とを有する。ジェネレータ１１０は、ディスクリミネータ１１１の判定結果に基づいて、疑似データ生成モデルＭ１を更に学習する。このようにすることで、データ生成装置１は、ディスクリミネータ１１１に真のデータ（実測ＧＮＳＳデータ）であると判定されるような精巧な疑似ＧＮＳＳデータを生成できるように、疑似データ生成モデルＭ１を更に学習することができる。 Further, the learning unit 11 of the data generation device 1 has a generator 110 for learning the pseudo data generation model M1 and a discriminator 111 for determining the authenticity of the input GNSS data. The generator 110 further learns the pseudo data generation model M1 based on the determination result of the discriminator 111. By doing so, the data generation device 1 can generate elaborate pseudo GNSS data that is determined to be true data (actually measured GNSS data) in the discriminator 111, so that the pseudo data generation model M1 Can be further learned.
　また、データ生成装置１の学習部１１は、実測ＧＮＳＳデータであるか疑似ＧＮＳＳデータであるかを示すラベルを付したＧＮＳＳデータと、３Ｄ地図データとに基づいて学習された識別モデルＭ２を用いて、入力されたＧＮＳＳデータの真偽を判定する。このようにすることで、データ生成装置１は、真偽の判定精度（実測ＧＮＳＳデータと疑似ＧＮＳＳデータとを識別する精度）を向上させることができる。また、データ生成装置１は、判定精度が高いディスクリミネータ１１１に真のデータであると判定されるような、より精巧な疑似ＧＮＳＳデータを生成するように疑似データ生成モデルＭ１を更に学習させることができる。 Further, the learning unit 11 of the data generation device 1 uses the identification model M2 trained based on the GNSS data with a label indicating whether it is the measured GNSS data or the pseudo GNSS data and the 3D map data. , Judge the authenticity of the input GNSS data. By doing so, the data generation device 1 can improve the accuracy of determining the authenticity (accuracy of distinguishing the actually measured GNSS data from the pseudo GNSS data). Further, the data generation device 1 further trains the pseudo data generation model M1 so as to generate more elaborate pseudo GNSS data such that the discriminator 111 having high determination accuracy is determined to be true data. Can be done.
　また、データ生成装置１の学習部１１は、実測ＧＮＳＳデータに基づいて車両の走行経路履歴を作成する走行経路履歴作成部を更に有していてもよい。これにより、データ生成装置１は、走行経路と、この走行経路に沿って走行したときに得られるＧＮＳＳデータとの関連性をより精度よく学習することができる。 Further, the learning unit 11 of the data generation device 1 may further have a travel route history creation unit that creates a travel route history of the vehicle based on the actually measured GNSS data. As a result, the data generation device 1 can more accurately learn the relationship between the traveling route and the GNSS data obtained when traveling along the traveling route.
　また、データ生成装置１の学習部１１は、実測ＧＮＳＳデータを取得した車両の車種に基づいて、疑似データ生成モデルＭ１を更に学習してもよい。これにより、データ生成装置１は、車種に応じた疑似ＧＮＳＳデータを生成することができる。これにより、特定の車種の実測ＧＮＳＳデータが不足している場合は、当該車種の疑似ＧＮＳＳデータを生成して、検証等に利用することができる。 Further, the learning unit 11 of the data generation device 1 may further learn the pseudo data generation model M1 based on the vehicle type of the vehicle that has acquired the actually measured GNSS data. As a result, the data generation device 1 can generate pseudo GNSS data according to the vehicle type. As a result, when the measured GNSS data of a specific vehicle model is insufficient, the pseudo GNSS data of the vehicle model can be generated and used for verification or the like.
＜付記＞
　上述の実施形態に記載のデータ生成装置、データ生成方法、データ生成装置の製造方法、及びプログラムは、例えば以下のように把握される。 <Additional notes>
The data generation device, the data generation method, the manufacturing method of the data generation device, and the program described in the above-described embodiment are grasped as follows, for example.
　本開示の第１の態様によれば、データ生成装置（１）は、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習する学習部（１１）と、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成する疑似データ生成部（１２）と、を備える。 According to the first aspect of the present disclosure, the data generator (1) is based on the measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data. The pseudo data generation model includes a learning unit (11) that learns a pseudo data generation model, simulated travel path data representing a time series of virtual positions where the vehicle virtually moves, and 3D map data including the virtual position. It is provided with a pseudo data generation unit (12) that generates pseudo GNSS data in which GNSS data obtained at each of the virtual positions is estimated by inputting to.
　このようにすることで、データ生成装置は、実際に車両を走行させたときに取得した実測ＧＮＳＳデータと、車両の走行位置周辺の地形、建築物等の環境との関連を学習し、実測ＧＮＳＳデータに精巧に似せた疑似ＧＮＳＳデータを生成することができる。これにより、車両に多数の走行経路を走行させることなく、検証用のＧＮＳＳデータを容易に準備することができる。 By doing so, the data generator learns the relationship between the measured GNSS data acquired when the vehicle is actually driven and the terrain around the vehicle's running position and the environment such as buildings, and the measured GNSS. Pseudo GNSS data that closely resembles the data can be generated. As a result, GNSS data for verification can be easily prepared without causing the vehicle to travel on a large number of travel routes.
　本開示の第２の態様によれば、第１の態様に係るデータ生成装置（１）において、前記学習部（１１）は、前記実測ＧＮＳＳデータと、前記３Ｄ地図データとに基づいて前記疑似データ生成モデルを学習し、学習した前記疑似データ生成モデルに前記模擬走行経路データと前記３Ｄ地図データとを入力して前記疑似ＧＮＳＳデータを出力するジェネレータ（１１０）と、入力されたＧＮＳＳデータの真偽を判定するディスクリミネータ（１１１）と、を有する。前記ジェネレータ（１１０）は、前記ディスクリミネータ（１１１）の判定結果に基づいて、前記疑似データ生成モデルを更に学習する。 According to the second aspect of the present disclosure, in the data generation device (1) according to the first aspect, the learning unit (11) is the pseudo data based on the actually measured GNSS data and the 3D map data. A generator (110) that learns the generation model, inputs the simulated travel route data and the 3D map data to the learned pseudo data generation model, and outputs the pseudo GNSS data, and the authenticity of the input GNSS data. It has a discriminator (111) for determining the above. The generator (110) further learns the pseudo data generation model based on the determination result of the discriminator (111).
　このようにすることで、データ生成装置は、ディスクリミネータに真のデータ（実測ＧＮＳＳデータ）であると判定されるような精巧な疑似ＧＮＳＳデータを生成できるように、疑似データ生成モデルを更に学習することができる。 By doing so, the data generator further learns the pseudo data generation model so that the discriminator can generate elaborate pseudo GNSS data that is determined to be true data (actually measured GNSS data). can do.
　本開示の第３の態様によれば、第２の態様に係るデータ生成装置（１）において、前記ディスクリミネータ（１１１）は、実測ＧＮＳＳデータであるか疑似ＧＮＳＳデータであるかを示すラベルを付したＧＮＳＳデータと、前記３Ｄ地図データとに基づいて学習された識別モデルを用いて、入力されたＧＮＳＳデータの真偽を判定する。 According to the third aspect of the present disclosure, in the data generator (1) according to the second aspect, the discriminator (111) has a label indicating whether it is actually measured GNSS data or pseudo GNSS data. Using the identification model learned based on the attached GNSS data and the 3D map data, the authenticity of the input GNSS data is determined.
　このようにすることで、データ生成装置は、真偽の判定精度（実測ＧＮＳＳデータと疑似ＧＮＳＳデータとを識別する精度）を向上させることができる。また、データ生成装置１は、判定精度が高いディスクリミネータに真のデータであると判定されるような、より精巧な疑似ＧＮＳＳデータを生成するように疑似データ生成モデルを更に学習させることができる。 By doing so, the data generation device can improve the authenticity determination accuracy (accuracy of distinguishing the measured GNSS data from the pseudo GNSS data). Further, the data generation device 1 can further train the pseudo data generation model so as to generate more elaborate pseudo GNSS data such that the discriminator having high determination accuracy determines that the data is true data. ..
　本開示の第４の態様によれば、第１から第３の何れか一の態様に係るデータ生成装置（１）において、前記学習部（１１）は、前記実測ＧＮＳＳデータに基づいて、当該車両の走行位置の時系列を表す走行経路履歴を作成する走行経路履歴作成部（１１２）を更に有し、前記走行経路履歴に基づいて前記疑似データ生成モデルを更に学習する。 According to the fourth aspect of the present disclosure, in the data generation device (1) according to any one of the first to third aspects, the learning unit (11) is the vehicle based on the actually measured GNSS data. Further, it has a travel route history creation unit (112) that creates a travel route history representing a time series of travel positions, and further learns the pseudo data generation model based on the travel route history.
　このようにすることで、データ生成装置は、走行経路と、この走行経路に沿って走行したときに得られるＧＮＳＳデータとの関連性をより精度よく学習することができる。 By doing so, the data generator can more accurately learn the relationship between the traveling route and the GNSS data obtained when traveling along this traveling route.
　本開示の第５の態様によれば、第１から第４の何れか一の態様に係るデータ生成装置（１）において、前記学習部（１１）は、前記実測ＧＮＳＳデータを取得した前記車両の車種に基づいて前記疑似データ生成モデルを更に学習し、前記疑似データ生成部（１２）は、指定された車種に応じた前記疑似ＧＮＳＳデータを生成する。 According to the fifth aspect of the present disclosure, in the data generation device (1) according to any one of the first to fourth aspects, the learning unit (11) is the vehicle that has acquired the measured GNSS data. The pseudo data generation model is further learned based on the vehicle type, and the pseudo data generation unit (12) generates the pseudo GNSS data according to the designated vehicle type.
　このようにすることで、特定の車種の実測ＧＮＳＳデータが不足している場合は、当該車種の疑似ＧＮＳＳデータを生成して、検証等に利用することができる。 By doing so, if the actual measurement GNSS data of a specific vehicle model is insufficient, it is possible to generate pseudo GNSS data of the vehicle model and use it for verification or the like.
　本開示の第６の態様によれば、データ生成方法は、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成するステップと、を有する。 According to the sixth aspect of the present disclosure, the data generation method generates pseudo data based on the actually measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the actually measured GNSS data. The step of learning the model, the simulated travel route data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model, and the virtual It has a step of generating pseudo GNSS data inferring the GNSS data obtained at each of the positions.
　本開示の第７の態様によれば、データ生成装置（１）の製造方法は、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、学習した前記疑似データ生成モデルに、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと前記３Ｄ地図データとを入力して前記疑似ＧＮＳＳデータを出力するステップと、入力されたＧＮＳＳデータの真偽を判定するステップと、前記真偽の判定結果に基づいて、前記疑似データ生成モデルを更に学習するステップと、を有する。 According to the seventh aspect of the present disclosure, the manufacturing method of the data generator (1) includes the actually measured GNSS data acquired by traveling the vehicle and the 3D map data including the position of the vehicle indicated by the actually measured GNSS data. The step of learning the pseudo data generation model based on the above, and the simulated travel route data representing the time series of the virtual position where the vehicle virtually moves and the 3D map data are input to the learned pseudo data generation model. It has a step of outputting the pseudo GNSS data, a step of determining the authenticity of the input GNSS data, and a step of further learning the pseudo data generation model based on the authenticity determination result.
　本開示の第８の態様によれば、プログラムは、車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成するステップと、をデータ生成装置（１）のコンピュータ（９００）に実行させる。 According to the eighth aspect of the present disclosure, the program creates a pseudo data generation model based on the measured GNSS data acquired by driving the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data. The step to be learned, the simulated travel path data representing the time series of the virtual position where the vehicle virtually moves, and the 3D map data including the virtual position are input to the pseudo data generation model to obtain the virtual position. The computer (900) of the data generator (1) is made to execute a step of generating pseudo GNSS data in which the GNSS data obtained in each is estimated.
　上述のデータ生成装置、データ生成方法、データ生成装置の製造方法、及びプログラムによれば、実測したＧＮＳＳデータに精巧に似せた疑似ＧＮＳＳデータを生成することができる。 According to the above-mentioned data generation device, data generation method, manufacturing method of the data generation device, and program, it is possible to generate pseudo GNSS data that closely resembles the actually measured GNSS data.
１　データ生成装置
１０　ＣＰＵ
１１　学習部
１１０　ジェネレータ
１１１　ディスクリミネータ
１１２　走行経路履歴作成部
１２　疑似データ生成部
２０　記憶媒体
９００　コンピュータ 1 Data generator 10 CPU
11 Learning unit 110 Generator 111 Discriminator 112 Travel route history creation unit 12 Pseudo data generation unit 20 Storage medium 900 Computer

Claims (8)

  1. 　車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習する学習部と、
    　前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成する疑似データ生成部と、
    　を備えるデータ生成装置。     A learning unit that learns a pseudo data generation model based on the measured GNSS data acquired by running the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
    GNSS obtained at each of the virtual positions by inputting simulated travel route data representing a time series of virtual positions where the vehicle virtually moves and 3D map data including the virtual positions into the pseudo data generation model. A pseudo data generator that generates pseudo GNSS data that infers the data,
    A data generator comprising.
  2. 　前記学習部は、
    　前記実測ＧＮＳＳデータと、前記３Ｄ地図データとに基づいて前記疑似データ生成モデルを学習し、学習した前記疑似データ生成モデルに前記模擬走行経路データと前記３Ｄ地図データとを入力して前記疑似ＧＮＳＳデータを出力するジェネレータと、
    　入力されたＧＮＳＳデータの真偽を判定するディスクリミネータと、
    　を有し、
    　前記ジェネレータは、前記ディスクリミネータの判定結果に基づいて、前記疑似データ生成モデルを更に学習する、
    　請求項１に記載のデータ生成装置。     The learning unit
    The pseudo data generation model is learned based on the actually measured GNSS data and the 3D map data, and the simulated travel route data and the 3D map data are input to the learned pseudo data generation model to obtain the pseudo GNSS data. With a generator that outputs
    A discriminator that determines the authenticity of the input GNSS data, and
    Have,
    The generator further learns the pseudo data generation model based on the determination result of the discriminator.
    The data generator according to claim 1.
  3. 　前記ディスクリミネータは、実測ＧＮＳＳデータであるか疑似ＧＮＳＳデータであるかを示すラベルを付したＧＮＳＳデータと、前記３Ｄ地図データとに基づいて学習された識別モデルを用いて、入力されたＧＮＳＳデータの真偽を判定する、
    　請求項２に記載のデータ生成装置。     The discriminator is input GNSS data using a GNSS data labeled as indicating whether it is actually measured GNSS data or pseudo GNSS data, and an identification model learned based on the 3D map data. Judging the authenticity of
    The data generator according to claim 2.
  4. 　前記学習部は、
    　前記実測ＧＮＳＳデータに基づいて、当該車両の走行位置の時系列を表す走行経路履歴を作成する走行経路履歴作成部を更に有し、
    　前記走行経路履歴に基づいて前記疑似データ生成モデルを更に学習する、
    　請求項１から３の何れか一項に記載のデータ生成装置。     The learning unit
    Further, it has a travel route history creation unit that creates a travel route history representing a time series of the travel positions of the vehicle based on the measured GNSS data.
    Further learning the pseudo data generation model based on the travel route history.
    The data generator according to any one of claims 1 to 3.
  5. 　前記学習部は、前記実測ＧＮＳＳデータを取得した前記車両の車種に基づいて前記疑似データ生成モデルを更に学習し、
    　前記疑似データ生成部は、指定された車種に応じた前記疑似ＧＮＳＳデータを生成する、
    　請求項１から４の何れか一項に記載のデータ生成装置。     The learning unit further learns the pseudo data generation model based on the vehicle type of the vehicle that has acquired the measured GNSS data, and further learns the pseudo data generation model.
    The pseudo data generation unit generates the pseudo GNSS data according to the designated vehicle type.
    The data generator according to any one of claims 1 to 4.
  6. 　車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、
    　前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成するステップと、
    　を有するデータ生成方法。     A step of learning a pseudo data generation model based on the measured GNSS data acquired by running the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
    GNSS obtained at each of the virtual positions by inputting simulated travel route data representing a time series of virtual positions where the vehicle virtually moves and 3D map data including the virtual positions into the pseudo data generation model. Steps to generate pseudo GNSS data that infers the data,
    Data generation method having.
  7. 　車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、
    　学習した前記疑似データ生成モデルに、前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと前記３Ｄ地図データとを入力して前記疑似ＧＮＳＳデータを出力するステップと、
    　入力されたＧＮＳＳデータの真偽を判定するステップと、
    　前記真偽の判定結果に基づいて、前記疑似データ生成モデルを更に学習するステップと、
    　を有するデータ生成装置の製造方法。     A step of learning a pseudo data generation model based on the measured GNSS data acquired by running the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
    A step of inputting simulated traveling route data representing a time series of virtual positions where the vehicle virtually moves and the 3D map data into the learned pseudo data generation model and outputting the pseudo GNSS data.
    Steps to determine the authenticity of the input GNSS data,
    A step of further learning the pseudo data generation model based on the truth determination result, and
    A method of manufacturing a data generator having.
  8. 　車両を走行させて取得した実測ＧＮＳＳデータと、前記実測ＧＮＳＳデータが示す前記車両の位置を含む３Ｄ地図データとに基づいて疑似データ生成モデルを学習するステップと、
    　前記車両が仮想的に移動する仮想位置の時系列を表す模擬走行経路データと、前記仮想位置を含む３Ｄ地図データとを前記疑似データ生成モデルに入力して、前記仮想位置のそれぞれにおいて得られるＧＮＳＳデータを推測した疑似ＧＮＳＳデータを生成するステップと、
    　をデータ生成装置のコンピュータに実行させるプログラム。     A step of learning a pseudo data generation model based on the measured GNSS data acquired by running the vehicle and the 3D map data including the position of the vehicle indicated by the measured GNSS data.
    GNSS obtained at each of the virtual positions by inputting simulated travel route data representing a time series of virtual positions where the vehicle virtually moves and 3D map data including the virtual positions into the pseudo data generation model. Steps to generate pseudo GNSS data that infers the data,
    A program that causes the computer of the data generator to execute.
PCT/JP2020/014606 2020-03-30 2020-03-30 Data generation device, data generation method, method for manufacturing data generation device, and program WO2021199175A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/JP2020/014606 WO2021199175A1 (en) 2020-03-30 2020-03-30 Data generation device, data generation method, method for manufacturing data generation device, and program
JP2022512915A JP7381719B2 (en) 2020-03-30 2020-03-30 Data generation device, data generation method, and program

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2020/014606 WO2021199175A1 (en) 2020-03-30 2020-03-30 Data generation device, data generation method, method for manufacturing data generation device, and program

Publications (1)

Publication Number Publication Date
WO2021199175A1 true WO2021199175A1 (en) 2021-10-07

Family

ID=77930142

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/014606 WO2021199175A1 (en) 2020-03-30 2020-03-30 Data generation device, data generation method, method for manufacturing data generation device, and program

Country Status (2)

Country Link
JP (1) JP7381719B2 (en)
WO (1) WO2021199175A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220035972A1 (en) * 2020-07-28 2022-02-03 Verizon Patent And Licensing Inc. Systems and methods for denoising gps signals using simulated models

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2011174771A (en) * 2010-02-24 2011-09-08 Clarion Co Ltd Position estimation device and position estimation method
JP2014002103A (en) * 2012-06-20 2014-01-09 Clarion Co Ltd Vehicle position detection device and vehicle position detection method
JP2019159462A (en) * 2018-03-08 2019-09-19 日産自動車株式会社 Traveling support method of traveling support device, and traveling support device

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2011174771A (en) * 2010-02-24 2011-09-08 Clarion Co Ltd Position estimation device and position estimation method
JP2014002103A (en) * 2012-06-20 2014-01-09 Clarion Co Ltd Vehicle position detection device and vehicle position detection method
JP2019159462A (en) * 2018-03-08 2019-09-19 日産自動車株式会社 Traveling support method of traveling support device, and traveling support device

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220035972A1 (en) * 2020-07-28 2022-02-03 Verizon Patent And Licensing Inc. Systems and methods for denoising gps signals using simulated models

Also Published As

Publication number Publication date
JPWO2021199175A1 (en) 2021-10-07
JP7381719B2 (en) 2023-11-15

Similar Documents

Publication Publication Date Title
US9536146B2 (en) Determine spatiotemporal causal interactions in data
Essa et al. Simulated traffic conflicts: do they accurately represent field-measured conflicts?
CN111406278A (en) Autonomous vehicle simulation system
JP2002236444A (en) Method for transmitting information on position on digital map and device used for the same
CN104380044B (en) The estimation of the order of usage history state in real-time positioning or navigation system
JP6944472B2 (en) Methods, devices, and systems for detecting reverse-way drivers
CN113848855B (en) Vehicle control system test method, device, equipment, medium and program product
AU2021202991B2 (en) Method and system for vehicle speed profile generation
US11150099B2 (en) Detecting vehicular deviation from a specified path
CN108780605A (en) Automatic running auxiliary device, roadside equipment and automatic running auxiliary system
WO2021199175A1 (en) Data generation device, data generation method, method for manufacturing data generation device, and program
Dogramadzi et al. Accelerated map matching for GPS trajectories
Griffin et al. Routing-based map matching for extracting routes from GPS trajectories
EP3470790B1 (en) Information processing device and travel control system
CN110726414B (en) Method and apparatus for outputting information
CN114689074B (en) Information processing method and navigation method
CN113008246B (en) Map matching method and device
KR100502423B1 (en) Link Travel Time Deduction Method of Vehicle using Global Positioning System
van Steenbergen et al. Network generation for simulation of multimodal logistics systems
JP7467756B2 (en) Anomaly detection system, on-board device, anomaly detection method, and program
JP6587529B2 (en) Driving support device and driving support method
JP6214826B1 (en) Fuel consumption estimation system, fuel consumption estimation method, and fuel consumption estimation program
WO2021192082A1 (en) Position error prediction device, model creation device, prediction error prediction method, model creation method, and program
JP6054808B2 (en) Parallel road judgment device
US20230055974A1 (en) Method of Selecting a Route for Recording Vehicle

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20929541

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022512915

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20929541

Country of ref document: EP

Kind code of ref document: A1