JP2019095569A - Map information processing device, map information processing method and map information processing program - Google Patents

Abstract

To provide a map information processing device that can easily determine road shape change.SOLUTION: A map information processing device comprises: a trajectory information acquisition unit that acquires respective trajectory information on a plurality of mobile objects; a map information storage unit that stores map information represented by vector data and including information on a location and shape of a road; a map information conversion unit that converts the map information to scalar data from the vector data; a difference extraction unit that extracts difference data representing a difference between a movement trajectory image plotted from the trajectory information and a map image of the road in the map information converted to the scalar data; and a road shape change determination unit that determines whether the difference data is a road shape change by machine learning.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a map information processing apparatus, a map information processing method, and a map information processing program for determining whether or not the road shape of map information has changed from locus information of a plurality of moving objects.

  At present, information on roads is managed in units of administrative divisions such as countries, prefectures, cities, and wards, and information on changes in road shape (hereinafter also referred to as "road shape change") is not centralized. In order to understand the road shape change in order to create a road map, it is necessary to examine where and in what shape a new road is formed for each individual road.

  Therefore, it is judged how the road shape has changed based on, for example, a digital map issued by the Geographical Survey Institute etc. and probe data (Probe Data: trajectory information) acquired from the vehicle terminal, and the road is determined. One possible way is to update the map.

  As an apparatus for generating map information of this type, there is known a configuration in which a road connection state is correctly estimated based on position information acquired by an on-vehicle terminal, and an updated map is generated (for example, Patent Document 1).

JP, 2017-97088, A

  However, since there are complicated and various patterns, such as when the road shape is changed from a crossroad shape to a rotary shape, it may be difficult to simply generate an updated map from map information and position information.

Therefore, the configuration described in Patent Document 1 is not sufficient, and further improvements are required.
The present invention has been made in view of such problems, and it is an object of the present invention to more easily determine a road shape change.

  According to one aspect of the present invention, a map information processing apparatus according to the present invention includes a locus information acquisition unit, a map information storage unit, a map information conversion unit, a difference extraction unit, and a road shape change determination unit. . The locus information acquisition unit acquires locus information of each of a plurality of moving objects. The map information storage unit stores map information represented by vector data. The map information represented by the vector data includes information indicating the position and shape of the road. The map information conversion unit converts the map information from vector data to scalar data. The difference extraction unit extracts difference data. The difference data represents the difference between the movement locus image drawn from the locus information and the map image of the road in the map information converted to the scalar data. The road shape change determination unit determines whether the difference data is a road shape change by machine learning.

  The present invention can more easily determine the road shape change.

FIG. 1 is a diagram showing an example of a map information processing system according to the present embodiment. FIG. 2 is a view showing an example of the relationship between the map information processing apparatus of the present embodiment and the on-vehicle terminal. FIG. 3 is a functional block diagram showing an example of the function of the map information processing apparatus of the present embodiment. FIG. 4A is a diagram illustrating an example of a method of drawing a moving track image of a vehicle by the moving image drawing unit. FIG. 4B is a diagram showing an example of a method of drawing a moving track image of a vehicle by the moving image drawing unit. FIG. 5A is a diagram showing an example of a method of extracting difference data from the movement trajectory image and the map image by the difference extraction unit. FIG. 5B is a diagram showing an example of a method of extracting difference data from the movement trajectory image and the map image by the difference extraction unit. FIG. 5C is a diagram illustrating an example of a method of extracting difference data from the movement trajectory image and the map image by the difference extraction unit. FIG. 5D is a diagram illustrating an example of a method of extracting difference data from the movement trajectory image and the map image by the difference extraction unit. FIG. 6 is a diagram showing an example of a method of determining a road shape change by machine learning. FIG. 7 is a flowchart showing an example of road shape change determination processing by the road shape change determination processing program. FIG. 8 is a diagram showing an example of road shape change determination processing. FIG. 9 is a flowchart showing an example of map information update processing by the road shape change determination processing program. FIG. 10A is a diagram showing an example of the map information update process. FIG. 10B is a diagram showing an example of the map information update process. FIG. 10C is a diagram showing an example of the map information update process. FIG. 10D is a diagram showing an example of the map information update process. FIG. 10E is a diagram showing an example of the map information update process.

  A map information processing system 1 according to an embodiment will be described based on FIG. The map information processing system 1 of the present embodiment is used to easily determine a road shape change when creating a map. The map information processing system 1 is comprised including the map information processing apparatus 2 and the some vehicle-mounted terminal 3, as shown in FIG. Hereinafter, each composition in map information processing system 1 of this embodiment is explained in detail.

  The on-vehicle terminal 3 is mounted on a vehicle 4 traveling on a road. The on-vehicle terminal 3 includes, for example, a car navigation device. The map information processing apparatus 2 and the on-vehicle terminal 3 are configured to be able to connect to each other by wireless communication. In FIG. 1, wireless communication between a vehicle 4 traveling in Japan and the map information processing apparatus 2 is illustrated by a single arrow. In the map information processing system 1, probe data is transmitted from the on-vehicle terminal 3 to the map information processing apparatus 2.

The probe data includes, for example, information of the vehicle position (hereinafter, also referred to as “locus information”) of the vehicle 4 on which the on-vehicle terminal 3 is mounted and the time. The probe data includes, for example, information of travel history (attribute information (attribute)) of the vehicle 4 such as travel distance, travel speed, acceleration, and angular velocity, in addition to the information of the vehicle position and time. Creation of probe data is performed, for example, using the on-vehicle terminal 3 mounted on the vehicle 4. The on-vehicle terminal 3 can obtain various sensor information from a GPS (Global Positioning System) module mounted on the vehicle 4, a gyro sensor, and an acceleration sensor. The information of the own vehicle position is calculated by, for example, GPS data output from the GPS module of the car navigation device. That is, the information on the vehicle position is acquired using the function of the satellite positioning system such as GPS.
The information on the vehicle position indicates the position coordinates of the vehicle 4 at a certain time.

  The on-vehicle terminal 3 is configured to be able to communicate with the map information processing apparatus 2 via the communication terminal 5 as shown in FIG. The on-vehicle terminal 3 transmits probe data including trajectory information of the vehicle 4 to the map information processing apparatus 2 via the communication terminal 5. The on-vehicle terminal 3 may be connected to the communication terminal 5 by wire or wireless. The on-vehicle terminal 3 may be physically configured separately from the communication terminal 5 or may be integrally configured with the communication terminal 5 built therein. The communication terminal 5 is, for example, a communication unit, a mobile phone, or a smartphone. In the case of a smartphone, the communication terminal 5 is connected so as to be able to communicate with the on-vehicle terminal 3 according to, for example, the BLUETOOTH (registered trademark) or Wi-Fi communication standard. In the case of a smart phone, the communication terminal 5 is connected, for example, to be able to communicate with the map information processing apparatus 2 via a communication network.

  The map information processing apparatus 2 receives the probe data transmitted from each on-vehicle terminal 3, and acquires locus information included in the probe data. The map information processing apparatus 2 determines whether the road shape included in the map information has changed by machine learning based on the locus information. Details of this determination will be described later. The map information processing device 2 updates the map information when it is determined that the road shape is changing. Then, the map information processing device 2 transmits the updated map information to the on-vehicle terminal 3.

  In the above-mentioned embodiment, although the vehicle-mounted terminal 3 is comprised by the car navigation apparatus, it is not this limitation. The in-vehicle terminal 3 is not limited to the car navigation device, but may be various terminal devices. As a terminal device, a mobile telephone, a smart phone, a notebook computer, or a tablet personal computer is mentioned, for example. That is, the on-vehicle terminal 3 may be only a single smartphone brought into the vehicle 4. Moreover, the vehicle 4 of this embodiment is illustrated by the four-wheeled vehicle in which the vehicle-mounted terminal 3 is mounted. The vehicle 4 is not limited to a four-wheeled vehicle, and may be a three-wheeled vehicle, a motorcycle, or a bicycle. In other words, the vehicle 4 is a moving body.

Next, the map information processing apparatus 2 of the present embodiment will be described using FIG.
The map information processing apparatus 2 according to the present embodiment includes a map information storage unit 21, a trajectory information acquisition unit 22, a moving image drawing unit 23, a map information conversion unit 24, a difference extraction unit 25, and a road shape change determination unit. 26 and a map information updating unit 27. The map information storage unit 21 stores map information. The map information storage unit 21 preferably stores map information for each of a plurality of scales.

  Map information is information which comprises a map. Map information is composed of vector data. More preferably, the map information is composed of three-dimensional vector data. The vector data includes information indicating the position of the road and the shape of the road. More specifically, the map information has, for example, map image information indicating the shape of a road, information of nodes and links on the map image linked to the map image information, and attribute information indicating whether it is a general road or a highway. ing. The nodes indicate intersections and other nodules on the road network representation. A link indicates a road section between nodes. Further, the map information is configured in mesh units separated into rectangles at fixed latitude and longitude intervals. Furthermore, each mesh is composed of a plurality of layers of different scales separated by a predetermined unit. In the case of Japan, for example, the mesh can adopt the standard area mesh standard defined by the Ministry of Internal Affairs and Communications. The standard area mesh is composed of an area ratio of about 1/10 in the order of primary mesh, secondary mesh and tertiary mesh. Furthermore, the mesh can employ | adopt the division area mesh subdivided from the 1st-3rd mesh as a mesh unit. When the map information is divided into mesh units, the map information has mesh numbers and corresponding latitude and longitude information.

  The locus information acquisition unit 22 includes a vehicle data reception unit 28 and a data storage unit 29. The vehicle data receiving unit 28 receives probe data transmitted from the on-vehicle terminals 3 of the plurality of vehicles 4. The data storage unit 29 stores probe data received by the vehicle data receiving unit 28. The locus information acquisition unit 22 can acquire locus information of each vehicle 4 included in the probe data by receiving the probe data from each on-vehicle terminal 3. Trajectory information is coordinates expressed by latitude and longitude.

  The moving image drawing unit 23 draws movement locus images together based on the locus information of the plurality of vehicles 4 in the predetermined range acquired by the locus information acquisition unit 22. The predetermined range adopts, for example, a mesh of any scale that constitutes road information.

  Hereinafter, an example in which the moving image drawing unit 23 draws the movement trajectory image of the vehicle 4 will be described. The moving image drawing unit 23 is continuous as shown in FIG. 4A based on information of coordinates, time, traveling speed, acceleration, and angular velocity of locus information included in a plurality of probe data received by the vehicle data receiving unit 28. Acquire m sets of probe data. For example, the moving image drawing unit 23 acquires, as probe data groups, probe data groups having continuous coordinates at predetermined time intervals among the probe data. The moving image drawing unit 23 acquires, from the data storage unit 29, the coordinates of m pieces of locus information included in a group of probe data. The moving image drawing unit 23 plots the acquired coordinates as points P1, P2, and P3 to Pm. As illustrated in FIG. 4B, the moving image drawing unit 23 draws the plotted points P1, P2, and P3 to Pm as lines of one moving locus image L by connecting them. As a result, the moving image drawing unit 23 can generate a moving locus image in scalar form from probe data in vector form including information on time, traveling distance, traveling speed, acceleration, and angular velocity in addition to locus information. .

  In addition, when the vehicle data receiving unit 28 receives the probe data, the moving image drawing unit 23 repeatedly draws the movement locus image. When the moving image drawing unit 23 draws a plurality of moving trajectory images in a superimposed manner, it calculates a median or a square average value of each moving trajectory image, and the width direction of each moving trajectory image drawn in a superimposed manner A line connecting center values (center coordinates) can be drawn as an averaged movement trajectory image. As a result, the moving image drawing unit 23 suppresses an error in GPS data due to multipath and fluctuation in coordinates of trajectory information caused by communication failure. The moving image drawing unit 23 can improve the drawing accuracy by suppressing the fluctuation of the coordinates.

  The moving image drawing unit 23 may draw a movement locus image by omitting probe data having a traveling speed equal to or higher than a predetermined speed with reference to information on traveling speed included in the probe data. When calculating the traveling speed from the GPS data, the on-vehicle terminal 3 may be calculated, for example, to 300 km or more where the vehicle speed can not normally be obtained due to an error in GPS data due to multipath or communication failure. The moving image drawing unit 23 can suppress erroneous detection of traveling information caused by an error in GPS data due to multipath or communication failure. The moving image drawing unit 23 improves the drawing accuracy by suppressing the erroneous detection of the traveling information.

  The moving image drawing unit 23 may draw a movement locus image by omitting probe data in which the traveling speed is not continuous, with reference to information on the traveling speed included in the probe data. As a result, the moving image drawing unit 23 can omit, for example, probe data obtained from the on-vehicle terminal 3 of the vehicle 4 parked halfway in the parking lot. The moving image drawing unit 23 can improve the drawing accuracy of the movement locus image by omitting unnecessary probe data.

  The moving image drawing unit 23 can estimate that it is moving on a general road and probe data that can be estimated to move on a freeway with reference to locus information and traveling speed information included in the probe data It is also possible to draw the movement trajectory image by mutually identifying the probe data. Thereby, the map information processing apparatus 2 can set only the road type arbitrarily selected as the determination processing target of the road shape change. The map information processing apparatus 2 can improve the processing speed.

  In addition, the moving image drawing unit 23 may draw a movement locus image based on the locus information of the vehicle 4 acquired in a predetermined unit period. The moving image drawing unit 23 normalizes data, for example, based on locus information accumulated for a predetermined unit period. The moving image drawing unit 23 can create probe data for each unit period for each mesh unit from the normalized data. The predetermined unit period can be set, for example, for the past 30 days based on the current time point. In the present embodiment, the moving image drawing unit 23 creates probe data of secondary mesh as probe data for each unit period. As described above, the map information processing apparatus 2 generates the probe data based on the trajectory information accumulated over a predetermined period, thereby improving the drawing accuracy of the movement trajectory image.

  The moving image drawing unit 23 may accumulate past probe data for reference. The moving image drawing unit 23 stores probe data of the past serving as a reference for a specific period in advance before probe data of a secondary mesh created based on locus information acquired in a predetermined unit period is created. It can be created based on the trajectory information. It is preferable that past probe data be created for each mesh unit. The moving image drawing unit 23 may be set in advance as probe data in the past, for example, 30 days as a specific period. In the present embodiment, the moving image drawing unit 23 creates probe data of secondary mesh accumulated for 30 days as probe data in the past.

  As a first step, the moving image drawing unit 23 compares the probe data of the newly created secondary mesh with the probe data of the secondary mesh created in the past to determine whether there is a difference. Can. Then, the moving image drawing unit 23 compares the new probe data with the map information as the second step only when there is a difference. Thereby, the map information processing apparatus 2 can detect the road shape change in two steps. Further, the map information processing apparatus 2 makes only the probe data having a difference from the past probe data as a processing target of the extraction process of the difference data by the difference extracting unit 25. The map information processing apparatus 2 processes only the probe data having a difference with the past probe data as a processing target of the difference data, so that the processing target in the difference extracting unit 25 can be reduced. The map information processing apparatus 2 can reduce the processing target in the difference extracting unit 25, so that the processing load can be reduced and the processing speed can be improved.

  In the above embodiment, the predetermined unit period is set to one day, three days, seven days, and thirty days, but the present invention is not limited to this. The predetermined unit period may be set not only on a daily basis, but also on an hourly basis, a monthly basis, or an annual basis, and can be set to any period. Moreover, in the above-mentioned embodiment, although the specific period was made into 30 days, it is not restricted to this. The specific period can be set to any period. Furthermore, in the above-described embodiment, the moving image drawing unit 23 is not limited to only the configuration for creating probe data of secondary mesh. The moving image drawing unit 23 may create probe data in accordance with a primary mesh, a tertiary mesh, or another unit mesh.

  The map information conversion unit 24 converts map information in vector format into map information in scalar format. Specifically, among the map information of vector data stored in the map information storage unit 21, the map information conversion unit 24 is information of a node linked to the map image information, information of a link, and whether it is a general road or an expressway Omit attribute information indicating. In other words, the map information conversion unit 24 takes out only map image information indicating the shape of the road. The map information conversion unit 24 converts the map information from three-dimensional vector data into two-dimensional scalar data by using only map image information. That is, the map information of the three-dimensional vector data is only map image information after the node information, the link information, and the attribute information included in the map information are removed. The map information of the three-dimensional vector data is converted into the map information of the two-dimensional scalar data having a simple data configuration by being only the map image information. As a result, the map information processing apparatus 2 can convert the movement locus image obtained by the probe data and the map information into data of the same simple dimension, and it becomes easy to extract the difference in the difference extraction unit 25 described later. Can be Further, the map information conversion unit 24 converts the two-dimensional map information converted into scalar data into scalar data divided into mesh units of arbitrary scale.

  In the above embodiment, the map information conversion unit 24 is not limited to the case of converting the two-dimensional map information converted into scalar data into scalar data divided into mesh units of arbitrary scale. For example, when the road shape change determination unit 26 determines that the probability that there is a road shape change from the difference data is equal to or less than a predetermined probability, the map information conversion unit 24 is stored in the map information storage unit 21 Among the map information, map information with a large scale is converted from vector data to scalar data.

  As a result, the map information processing apparatus 2 can improve the determination speed as to whether or not there is a change in the road shape by roughly determining the whole in the initial stage. In other words, when it is determined that the probability that there is a road shape change from the difference data is less than the predetermined probability or the certain probability, the map information processing apparatus 2 changes the processing target to map information with a larger scale. The processing target can be limited to improve the processing speed. As a result, the map information processing apparatus 2 can further improve the determination accuracy.

  The difference extraction unit 25 can extract difference data between the movement trajectory image drawn by the movement image drawing unit 23 and the map image of the map information converted into scalar data by the map information conversion unit 24. It is configured.

  In the map information processing apparatus 2 according to the present embodiment, the map information storage unit 21 and the data storage unit 29 can be configured by, for example, a hard disk drive or a memory of semiconductor memory. The memory may store a program for driving a CPU (Central Processing Unit). The CPU causes the moving image drawing unit 23, the map information conversion unit 24, the difference extraction unit 25, the road shape change determination unit 26, and the map information update unit 27 to function by executing the program stored in the memory. It is configured to be able to. The vehicle data receiving unit 28 can be configured by an appropriate communication module.

Hereinafter, an example of extracting difference data by the difference extracting unit 25 will be described with reference to FIGS. 5A to 5D.
The difference extraction unit 25 acquires the movement trajectory image drawn by the movement image drawing unit 23 and the map image of the map information converted into scalar data by the map information conversion unit 24. FIG. 5A exemplifies the moving track image drawn by the moving image drawing unit 23. FIG. 5B exemplifies the map image of the map information converted into scalar data. Next, as shown in FIG. 5C, the difference extraction unit 25 creates an image of composite data obtained by combining the acquired movement trajectory image and the map image. As a result of creating the image of the composite data, as shown in FIG. 5D, the difference extraction unit 25 uses an image that does not overlap between the movement locus image shown in FIG. 5A and the map image shown in FIG. Extract. As a result, among the movement locus images, it is possible to eliminate the fluctuation of the movement locus and to extract only the image of the movement locus that has run out of the map image. The image of the difference data extracted by the difference extraction part 25 is illustrated in FIG. 5D. The image of the difference data is considered to represent the difference between the road at the time of generation of the map information and the current road. That is, the image of difference data is considered to represent road shape change. In the example shown in FIGS. 5A-5D, a new road may have been formed in the circled area in FIG. Next, the map information processing apparatus 2 determines whether or not the extracted difference data is a road shape change, including the circled portion in FIG.

  The road shape change determination unit 26 determines by machine learning whether the difference data extracted by the difference extraction unit 25 is a road shape change. Deep learning is used as machine learning. Specifically, the road shape change determination unit 26 performs machine learning using the difference data extracted in the past as teacher data. The road shape change determination unit 26 determines whether the newly extracted difference data is a road shape change based on the result of the machine learning.

  Next, with reference to FIG. 6, an example of a method in which the road shape change determination unit 26 determines the road shape change by deep learning will be described. Deep learning is a type of machine learning method using a multi-layered neural network 60. The neural network 60 has input data and output data, and internal arithmetic processing is performed based on a plurality of artificial neurons. The neural network 60 used in machine learning includes three layers. The three layers are illustrated as an input layer 61, an intermediate layer 62, and an output layer 63. The middle layer 62 is also called a hidden layer, and may include two or more layers. The middle layer 62 is unlearned if the number of layers is too small. The middle layer 62 is overfit if there are too many layers. The road shape change determination unit 26 may appropriately set the number of layers in order to determine whether the difference data is a road shape change. An artificial neuron outputs the sum of the output of the previous layer multiplied by a parameter. The output data of the artificial neuron is controlled by the activation function to add nonlinearity. The activation function for machine learning used in the present embodiment can employ, for example, a soft max function, a sigmoid function, or a Gaussian function.

  In order to perform machine learning, the neural network 60 is first given teaching data to the input layer 61 as input data. The teacher data is, for example, difference data extracted in the past by the difference extraction unit 25. The road shape change determination unit 26 processes teacher data by the input layer 61, the intermediate layer 62, and the output layer 63. That is, the road shape change determination unit 26 dynamically generates and learns an optimal feature amount for the input difference data, and performs forward propagation in which calculation processing is performed by forward information propagation. In FIG. 6, the direction of forward propagation is indicated by a thick single arrow. The output result indicates the prediction result as to whether the input difference data is an image of road shape change or noise.

  When learning is performed, the road shape change determination unit 26 reversely performs arithmetic processing by information propagation in the reverse direction by giving information on whether the output result is a road shape change image or noise. Propagation is done. In FIG. 6, the direction of back propagation is indicated by a thick single arrow. In the machine learning, the road shape change determination unit 26 evaluates an output error between input data used for learning as teacher data and output data. The road shape change determination unit 26 preferably optimizes the parameters of each layer and each node in machine learning sequentially from the output error by back propagation.

  By this learning, the road shape change determination unit 26 can gradually bring the parameters of each layer closer to the optimal value. Then, when the road shape change determination unit 26 inputs the difference data extracted by the difference extraction unit 25 to the input layer 61, the difference data is changed according to the road shape change using the parameter adjusted based on the result of machine learning. It can be judged whether it is an image or not.

  For example, as shown in FIG. 6, when the road shape change determination unit 26 propagates n pieces of difference data d1 and d2 forward as teacher data, information on M pieces of output data y1 to yM is output to the output layer 63. Is obtained. In the present embodiment, the output data represents a predicted value for predicting whether the input difference data is an image of road shape change or another image.

  The road shape change determination unit 26 is provided with information of M pieces of correct data t1 to tM indicating whether it is an image of road shape change or noise, to the obtained output data y1 to yM. When information on the correct data t1 to tM is given and back propagation is performed, the road shape change determination unit 26 performs machine learning to adjust the parameter to an optimal value one by one. In other words, the road shape change determination unit 26 evaluates the deviation between the output data and the correct data by back propagation to optimize the parameters. As software used for machine learning, for example, OpenCV, Numpy, Matplotlib, or Chainer can be adopted.

  The map information updating unit 27 updates the map information stored in the map information storage unit 21. Specifically, the map information updating unit 27 converts the map information of the vector data from the map information converted into the scalar data corresponding to the difference data determined by the road shape change determination unit 26 to be the road shape change. Generate Then, the map information updating unit 27 updates the map information stored in the map information storage unit 21 with the generated map information. The process of updating the map information will be described in the later-described map information update process.

By the way, in the map information processing apparatus, it is conceivable to simply identify a change in road shape from probe data and a digital map by machine learning.
However, in general, digital map data is three-dimensional including map image information indicating the shape of a road, information on nodes and links on the map image linked to the map image information, and attribute information indicating whether it is a general road or an expressway. It is composed of vector data.

  Therefore, when the map information processing apparatus detects a change in the road shape newly created using machine learning, it simply performs information processing with the three-dimensional vector data of the digital map as the machine learning process target. The amount of information to be processed may be so large that the burden of information processing may become too large. Furthermore, since the map information processing apparatus has a large amount of information to be compared as a machine learning object, it may be difficult to determine which information is to be compared.

Since the map information processing apparatus 2 of the present embodiment creates a road map by machine learning, determination of road shape change is easily performed by making the processing object as simple as possible.
Moreover, since the shape of the earth is not a true sphere but an oblate sphere, the length and width of the mesh may differ depending on the place. When performing machine learning, it is preferable that the map information processing apparatus 2 of the present embodiment performs normalization in which the vertical and horizontal lengths of the meshes are unified in advance. The map information processing apparatus 2 can perform the determination of the road shape change more accurately by restoring the normalized mesh and using it after machine learning.

  Next, road shape change determination processing by the map information processing apparatus 2 according to the present embodiment will be described with reference to FIGS. 7 and 8. When the map information processing apparatus 2 receives an instruction to start road shape change processing, the map information processing apparatus 2 starts the determination processing of step 11 to step 19 of FIG. 7. In the following, the steps are indicated by S.

  The map information processing apparatus 2 performs parameter optimization based on teacher data in advance before determining whether or not there is a road shape change. The road shape change determination unit 26 performs machine learning based on the teacher data (S11).

  The road shape change determination unit 26 receives information prepared in advance as teacher data. The information prepared in advance is used to set parameters for mutually identifying whether the difference data extracted in the past is an image showing a road shape change or an image such as other noise. Then, for example, the set parameter is held by the road shape change determination unit 26 or stored in a memory accessible by the road shape change determination unit 26.

  Next, in the locus information acquisition unit 22, the vehicle data reception unit 28 receives the probe data transmitted from each on-vehicle terminal 3. The locus information acquisition unit 22 stores the received probe data in the data storage unit 29. Thereby, the locus information acquisition unit 22 acquires locus information of each vehicle 4 included in the received probe data (S12).

  The map information conversion unit 24 extracts map image information indicating the shape of the road from the information included in the map information of the vector data stored in the map information storage unit 21. The map information conversion unit 24 converts map information from three-dimensional vector data into two-dimensional scalar data by extracting map image information (S13).

  In other words, the map information conversion unit 24 converts the vector type map image into a raster type map image. In the present embodiment, the map information conversion unit 24 converts the two-dimensional map information converted into scalar data into scalar data of a secondary mesh divided into secondary mesh units as mesh units of an arbitrary scale. The moving image drawing unit 23 creates secondary mesh probe data by normalizing the accumulated data of the probe data acquired by the locus information acquisition unit 22 for a predetermined unit period.

The moving image drawing unit 23 draws a moving trace image of the secondary mesh based on the normalized probe data of the secondary mesh (S14).
The moving image drawing unit 23 accumulates the past probe data acquired by the locus information acquiring unit 22 for a specific period before creating the movement locus image based on the probe data acquired in a predetermined unit period. Perform data normalization. The moving image drawing unit 23 creates probe data of a normalized past secondary mesh of data. Then, the moving image drawing unit 23 draws the moving trajectory image of the past secondary mesh based on the normalized past probe data of the secondary mesh. The movement trajectory image of the past secondary mesh is referred to for reference.

  As a first step, the moving image drawing unit 23 draws a movement trajectory image of a secondary mesh drawn based on probe data of a secondary mesh, and a past secondary mesh drawn based on probe data of a past secondary mesh. It is determined whether or not there is a difference by comparing the movement locus image of (S15).

  In the above-described embodiment, the moving image drawing unit 23 draws the movement trajectory image of the secondary mesh, but the present invention is not limited to this. The moving image drawing unit 23 can draw a movement locus image of a mesh of any scale, such as a primary mesh and a tertiary mesh. The moving image drawing unit 23 is not limited to the case of determining whether there is a difference by comparing the movement trajectory image of the secondary mesh and the movement trajectory image of the secondary mesh in the past. For example, the moving image drawing unit 23 directly compares the information of the latitude and longitude included in the probe data of the secondary mesh with the information of the latitude and longitude included in the probe data of the past secondary mesh, and the difference is It can also be determined whether there is any.

  In the case of NO in S15 where there is no difference between the movement locus image and the past movement locus image, the map information processing apparatus 2 skips S16 to S19 shown in FIG. 7, and the road shape change determination processing ends. That is, the map information processing apparatus 2 executes S16 to S17 only when there is a difference in the determination of S15. Therefore, the map information processing apparatus 2 can reduce the execution frequency of S17. In addition, the map information processing apparatus 2 may process only the movement locus image having a difference from the movement locus image created based on the probe data accumulated in the past as the processing object of the difference data extraction process by the difference extraction unit 25. it can. Therefore, the map information processing apparatus 2 can reduce the processing load and reduce the processing speed by reducing the processing target in the difference extracting unit 25.

  On the other hand, in the case of YES in S15 where there is a difference between the movement locus image and the past movement locus image, the difference extraction unit 25 converts the movement locus image drawn in S14 into scalar data in S13. Difference data representing the difference between the map information and the map image is extracted (S16).

The road shape change determination unit 26 determines whether the difference data newly extracted in S16 is a road shape change based on the result of the machine learning learned in S11 (S17).
If the extracted difference data is YES at S18, which is a road shape change, the road shape change determination unit 26 outputs the extracted difference data as a road shape change (S19).

In this case, since the difference data indicates a road shape change, it is stored for use in the later map information creation process.
On the other hand, if the extracted difference data is NO in S18, which is not a road shape change, S19 is skipped, and the road shape change determination process ends. In this case, the difference data is not information indicating a road shape change, but is likely to be a simple noise. Thus, the extracted difference data may be discarded.

  As described above, the map information processing apparatus 2 extracts the map image information indicating the shape of the road from the map information of the vector data stored in the map information storage unit 21 so that the map of the three-dimensional vector data is extracted. Information is converted to simple two-dimensional scalar data map information of data configuration. Thereby, in the map information processing apparatus 2, it is possible to make the movement locus image obtained by the probe data and the map information the same simple two-dimensional data.

  As a result, when the map information processing apparatus 2 detects a change in the road shape newly created using machine learning, it is possible to reduce the processing load of the machine learning by reducing the processing target of the machine learning. The map information processing apparatus 2 can easily determine the road shape change by improving the processing speed and further improving the determination accuracy.

Next, map information update processing by the map information processing apparatus 2 according to the present embodiment will be described with reference to FIG. 9 and FIGS. 10A and 10B.
When the map information processing apparatus 2 receives an instruction to start the map information update process, the following map information update process is started. In the above-described embodiment, when the instruction to start the map information update process is received, the map information update process is started, but the present invention is not limited to this. For example, when a predetermined number of road shape changes are output in S19 of the above-described road shape change determination process, the map information update process may be automatically started.

  First, the map information update unit 27 accumulates N pieces of difference data output in S19 of the road shape change determination process, and averages the error of the probe data (S31). Here, N is an arbitrary natural number. Error averaged probe data is shown in FIGS. 10A and 10B. FIG. 10B is an enlarged view of a region F1 of FIG. 10A. The map information updating unit 27 performs a smoothing process on difference data obtained by averaging the errors, that is, data representing the road shape determined to be a road shape change (S32). The road shape on which the smoothing process has been performed corresponds to the shape of the newly created road.

  The map information updating unit 27 acquires, among the map information of the vector data stored in the map information storage unit 21, the map information of the mesh including the latitude and longitude of the road shape on which the smooth processing has been performed in S32. The road shape determined based on the difference data has latitude and longitude information because it is created based on the probe data. Therefore, the map information updating unit 27 can acquire the map information of the corresponding mesh from the map information storage unit 21 by referring to the latitude and longitude of the probe data of the road shape.

  As shown in FIG. 10C and FIG. 10D, the map information updating unit 27 superimposes the two-dimensional road shape D2 which has been smoothed in S32 on the road shape D3 included in the map information of the mesh of the acquired vector data S33). FIG. 10D is an enlarged view of a region F2 of FIG. 10C.

  As shown in FIG. 10D, the map information updating unit 27 specifies a contact point between the two-dimensional road shape D2 which has been smoothed in S32 and the road shape D3 included in the map information of the mesh of vector data. Here, since the map information is represented by vector data, it includes information in the Z-axis direction. The Z-axis direction indicates a direction perpendicular to the ground. That is, the map information includes information representing Z-axis coordinates of each road. The Z-axis coordinate is, for example, the elevation. Therefore, the contact point between the road shape D2 generated from the difference data and the road in the map information is represented by latitude, longitude, and altitude. And the map information update part 27 adds a node to each identified contact point (S34). In FIG. 10D and FIG. 10E, they are illustrated as the node N1 and the node N2. Note that the nodes added to the road shape D2 generated from the difference data include information of contact points representing the Z-axis direction of each road, so the road shape D2 is converted into vector data.

The map information updating unit 27 generates an axis of the road shape D2 of the vector data using the specified node (S35).
The map information updating unit 27 corrects the road width of the road shape D2 obtained from the difference data so as to match the road width of the road shape D3 included in the map information. Then, the road shape D2 of the vector data in which the road width is corrected is connected to the road shape D3 included in the map information by the node (S36).

  The map information updating unit 27 deletes the unnecessary road location D4 and the added node from the road shape D3 included in the map information (S37). For example, the map information updating unit 27 deletes, as the unnecessary road location D4, a location where no probe data exists between the nodes added to the original two-dimensional road shape D2 before conversion.

  The map information updating unit 27 deletes the unnecessary road location D4. The map information updating unit 27 updates the map information in the map information storage unit 21 based on the road shape D3 to which the road shape D2 of the vector data is connected (S38). When this process ends, the map information update process ends.

  Thereby, the two-dimensional difference data extracted as the road shape change can be returned to the three-dimensional vector data again to update the map information. As a result, the map information can be automatically formatted periodically, and the time required for updating the map information can be reduced.

  The map information processing apparatus 2 of the present embodiment has a control unit, a temporary storage unit, a storage unit, and a wireless communication unit. The control unit executes the program read into the storage unit. The vehicle data reception unit 28, the moving image drawing unit 23, the map information conversion unit 24, the difference extraction unit 25, the road shape change determination unit 26, and the map information update unit 27 according to the present embodiment included. The map information storage unit 21 and the data storage unit 29 of the present embodiment are included in the storage unit of the map information processing apparatus 2. The temporary storage unit is a working area for expanding programs and various data read from the storage unit. The parts such as the control unit, the temporary storage unit, the storage unit, and the wireless communication unit are mutually connected.

  The map information processing apparatus 2 of the present embodiment can be used, for example, as a technology for automatically formatting map data used in a car navigation system or an automatic driving support system.

  The present invention is not limited to the above-described embodiment as it is, and constituent elements can be modified and embodied without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriate combinations of a plurality of components disclosed in the above embodiments. For example, all components shown in the embodiments may be combined as appropriate. Furthermore, components in different embodiments may be combined as appropriate. Of course, various modifications and applications are possible without departing from the scope of the invention.

DESCRIPTION OF SYMBOLS 1 map information processing system 2 map information processing apparatus 3 vehicle-mounted terminal 4 vehicle 5 communication terminal 21 map information storage part 22 locus information acquisition part 23 moving image drawing part 24 map information conversion part 25 difference extraction part 26 road shape change determination part 27 map Information updating unit 28 Vehicle data receiving unit 29 Data storage unit

Claims (9)

A trajectory information acquisition unit that acquires trajectory information of each of a plurality of moving objects;
A map information storage unit that stores map information represented by vector data including information indicating the position and shape of a road;
A map information conversion unit that converts the map information from vector data to scalar data;
A difference extraction unit that extracts difference data representing a difference between the movement locus image drawn from the locus information and the map image of the road in the map information converted into the scalar data;
A road shape change determination unit that determines whether the difference data is a road shape change by machine learning;
A map information processing apparatus comprising: It further comprises a moving image drawing unit that draws the movement trajectory image based on trajectory information of each of the plurality of moving objects,
The map information processing apparatus according to claim 1, wherein the moving image drawing unit draws the movement trajectory images within a predetermined range together. The moving image drawing unit draws a past movement locus image based on locus information acquired in a specific period in advance, before drawing the movement locus image based on locus information acquired in a predetermined period, and draws it. Comparing the moved trajectory image with the past moved trajectory image to determine whether there is a difference;
The difference extraction unit is characterized in that only the movement locus image determined to have a difference compared to the past movement locus image is a processing target of the extraction process of the difference data. Map information processing apparatus as described. The said map information conversion part takes out only the map image information which shows the shape of a road among the said map information, Converts into vector data from scalar data, It is characterized by the above-mentioned. Map Information Processing Device. The road shape change determination unit performs machine learning based on the difference data extracted in the past, and determines whether the difference data newly extracted is a road shape change based on the result of the machine learning. The map information processing apparatus according to any one of claims 1 to 4, characterized in that: It further comprises a map information update unit for updating the map information stored in the map information storage unit,
The map information updating unit
Converting the difference data determined to be road shape change by the road shape change determination unit into vector data;
The map information processing apparatus according to any one of claims 1 to 5, wherein the map information stored in the map information storage unit is updated using difference data converted into vector data. The map information storage unit stores map information including information indicating positions and shapes of a plurality of roads for each of a plurality of scales.
The map information conversion unit stores the map information storage unit when the road shape change determination unit determines that the probability that the difference data is a road shape change is less than or equal to a certainty by the machine learning. The map information processing apparatus according to any one of claims 1 to 6, wherein map information having a large scale is converted from vector data to scalar data among the stored map information. Get trajectory information for each of multiple mobiles,
Convert map information including information indicating the position and shape of roads from vector data to scalar data,
Extracting difference data representing a difference between the movement locus image drawn from the locus information and the map image of the map information converted into the scalar data;
A map information processing method in which a computer executes a process of determining whether the difference data is a road shape change by machine learning. Get trajectory information for each of multiple mobiles,
Convert map information including information indicating the position and shape of roads from vector data to scalar data,
Extracting difference data representing a difference between the movement locus image drawn from the locus information and the map image of the map information converted into the scalar data;
A map information processing program that causes a computer to execute a process of determining whether the difference data is a road shape change by machine learning.