WO2018221455A1

WO2018221455A1 - Update device, control method, program, and storage medium

Info

Publication number: WO2018221455A1
Application number: PCT/JP2018/020364
Authority: WO
Inventors: 加藤　正浩; 和紀小山
Original assignee: パイオニア株式会社
Priority date: 2017-05-31
Filing date: 2018-05-28
Publication date: 2018-12-06

Abstract

A server device 200 is a map generation device which assigns, to voxel data in a map DB 20 which is divided into a plurality of voxels, a first weighting value for each voxel, said first weighting value being associated with the presence of moving objects and stationary objects. The server device 200 acquires upload information Iu including position information related to a travel route of a measurement vehicle during measurement of point group data required for generating or updating the voxel data. The server device 200 determines the first weighting value for each voxel, on the basis of the travel route of the measurement vehicle, said travel route having been identified on the basis of the upload information Iu.

Description

Update device, control method, program, and storage medium

The present invention relates to a technology for generating a map.

Conventionally, a technique for updating map data based on the output of a sensor installed in a vehicle is known. For example, Patent Document 1 discloses that difference information representing a difference between a position of a moving object and a traveling position of the moving object on map data is associated with the position information and stored in the difference database. A map data update system configured to set information indicating whether or not to update the map data corresponding to the position information based on the difference information corresponding to the position information stored in is disclosed.

Japanese Unexamined Patent Publication No. 2016-180980

In the map data for the purpose of automatic driving or the like, there is a case where information of a point cloud measured by a distance measuring sensor or the like is stored in order to grasp a detailed position of a stationary structure around the road. Patent Document 1 does not disclose how to update the information of the point cloud stored in the map data.

The present invention has been made to solve the above-described problems, and has as its main object to provide an update device that can suitably update a map including point cloud information.

The invention according to claim 1 is an update device, wherein a first acquisition unit that acquires first point cloud information about each distance from a reference position to a plurality of positions, which is measured by a measurement unit, and a plurality of regions A second acquisition of map information in which the second point group information based on one or a plurality of position information and the weighting value based on the reliability of the second point group information are recorded for each of the plurality of regions And an updating unit that updates the weighting value so as to increase the weighting for a region in which the difference between the first point cloud information and the second point cloud information is a predetermined value or less among the plurality of regions. .

The invention according to claim 7 is a control method executed by the updating device, wherein a first acquisition step of acquiring first point cloud information relating to respective distances from the reference position to a plurality of positions measured by the measurement unit. And map information in which the second point cloud information based on one or a plurality of position information and weighting values based on the reliability of the second point cloud information are recorded for each of the plurality of areas. A second acquisition step to be acquired, and an update for updating the weighting value so as to increase weighting for a region in which a difference between the first point cloud information and the second point cloud information is a predetermined value or less among the plurality of regions. And a process.

The invention according to claim 8 is a program executed by a computer, wherein the first acquisition unit acquires first point group information about each distance from a reference position to a plurality of positions measured by the measurement unit; Obtained is map information that is divided into a plurality of regions, and second point cloud information based on one or a plurality of position information and weighting values based on the reliability of the second point cloud information are recorded for each of the plurality of regions. As an update unit that updates the weighting value so as to increase the weighting of the second acquisition unit and a region in which the difference between the first point group information and the second point group information is a predetermined value or less among the plurality of regions. Make the computer function.

1 is a schematic configuration of a map generation system. The block configuration of a vehicle equipment and a server apparatus is shown. An example of the schematic data structure of voxel data is shown. It is a flowchart which shows the procedure of the update process of the said voxel data when the voxel data of map DB already exist. It is the figure which showed roughly the road periphery and the measured point cloud when a measurement vehicle drive | worked the 1st time and the 2nd time. It is the figure which piled up the point group measured for the 1st time, and the point group measured for the 2nd time. It is a figure explaining the specific example of NDT scan matching. It is a figure explaining the specific example of the NDT scan matching which performed weighting.

According to a preferred embodiment of the present invention, the update device includes a first acquisition unit that acquires first point cloud information about each distance from a reference position to a plurality of positions, measured by the measurement unit, and a plurality of regions. A second acquisition of map information in which the second point group information based on one or a plurality of position information and the weighting value based on the reliability of the second point group information are recorded for each of the plurality of regions And an updating unit that updates the weighting value so as to increase the weighting for a region in which the difference between the first point cloud information and the second point cloud information is a predetermined value or less among the plurality of regions. . According to this aspect, the updating device regards a region in which the first point group information and the second point group information are approximated as having high reliability, and updates the weight value so as to increase the weight of the region. Thereby, the update apparatus can update the weighting value contained in map information appropriately.

In one aspect of the updating device, the updating unit is configured to perform the first operation on a region where a difference between the first point group information and the second point group information is a predetermined value or less based on accuracy information indicating the accuracy of the measuring unit. Update the two-point cloud information. According to this aspect, the update device can preferably improve the accuracy of the second point cloud information included in the map information.

In another aspect of the updating device, the map information further includes accuracy information related to measurement accuracy of the second point cloud information, and the updating unit includes the first point cloud information and the second point cloud. The second point cloud information is updated based on the first point cloud information when the accuracy of the first point cloud information is higher than that of the second point cloud information for an area where the difference in information is a predetermined value or less. Alternatively, the second point group information is updated by weighted averaging the first point group information and the second point group information based on accuracy. According to this aspect, the update device can suitably update the map information based on the accuracy information so that the accuracy of the second point cloud information is increased.

In another aspect of the updating apparatus, the updating unit corresponds to a position indicated by the first point group information in a region where a difference between the first point group information and the second point group information is greater than a predetermined value. When it is farther from the measurement position than the position indicated by the second point group information, the map information is updated with the first point group information. In this case, the update device regards the first point cloud information as information that has not been measured by occlusion when measuring the second point cloud information, and updates the map information with the first point cloud information. Thereby, the point group information of the stationary structure that was not measured at the time of measuring the second point group information can be suitably added to the map information.

In another aspect of the updating apparatus, the updating unit corresponds to a position indicated by the first point group information in a region where a difference between the first point group information and the second point group information is greater than a predetermined value. When the position is closer to the measurement position than the position indicated by the second point group information, the weight value is updated so as to reduce the weight for the region of the position indicated by the second point group information. In this aspect, the updating device can preferably update the weighting value so as to reduce the weighting for the region where the occlusion may occur.

In another aspect of the updating device, the updating unit determines whether or not a difference between the first point group information and the second point group information is equal to or less than a predetermined value, the first point group information and the second point point. A determination is made based on the average or / and variance of each group information. According to this aspect, the update device can suitably perform an analogy determination between the first point group information and the second point group information.

According to another preferred embodiment of the present invention, the control method is executed by the updating device, and first point cloud information relating to each distance from the reference position to a plurality of positions, which is measured by the measurement unit, is acquired. The first acquisition step is divided into a plurality of areas, and the second point group information based on one or a plurality of position information and the weighting value based on the reliability of the second point group information are recorded for each of the plurality of areas. The second acquisition step of acquiring the map information, and the weighting value so as to increase the weighting of the plurality of regions where the difference between the first point cloud information and the second point cloud information is a predetermined value or less. And an update process for updating. The update device can appropriately update the weighting value included in the map information by executing this control method.

According to still another preferred embodiment of the present invention, there is provided a program executed by a computer for acquiring first point cloud information about each distance from a reference position to a plurality of positions, which is measured by a measurement unit. One acquisition unit and a plurality of areas, and the second point group information based on one or a plurality of position information and a weighting value based on the reliability of the second point group information are recorded for each of the plurality of areas. A second acquisition unit for acquiring map information; and the weighting value is set so as to increase weighting for an area in which a difference between the first point cloud information and the second point cloud information is a predetermined value or less among the plurality of areas. The computer is caused to function as an updating unit for updating. The computer can appropriately update the weighting value included in the map information by executing this program. Preferably, the program is stored in a storage medium.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

[System Overview]
FIG. 1 is a schematic configuration diagram of a map generation system according to the present embodiment. The map generation system shown in FIG. 1 is a system for generating position information of features around a road necessary for automatic driving or the like, and mainly includes a measurement unit 100 mounted on a measurement vehicle, a server device 200, Have

The measurement unit 100 is a system that generates highly accurate 3D point cloud data. The measurement unit 100 mainly includes an in-vehicle device 1, a lidar (Lida: Light Detection and Ranging, or Laser Illuminated Detection And Ranging) 2, an RTK-GPS3, and the like. , IMU (Internal Measurement Unit) 4.

The lidar 2 emits a pulse laser in a predetermined angular range in the horizontal direction and the vertical direction, thereby discretely measuring the distance to an object existing in the outside world and indicating a three-dimensional point indicating the position of the object Generate group information. In this case, the lidar 2 includes an irradiation unit that emits laser light while changing the irradiation direction, a light receiving unit that receives reflected light (scattered light) of the irradiated laser light, and scan data based on a light reception signal output by the light receiving unit. Output unit. The scan data is generated based on the irradiation direction corresponding to the laser beam received by the light receiving unit and the response delay time of the laser beam specified based on the above-described received light signal. Generally, the accuracy of the distance measurement value of the lidar is higher as the distance to the object is shorter, and the accuracy is lower as the distance is longer.

RTK-GPS3 generates highly accurate position information indicating the absolute position (for example, the three-dimensional position of latitude, longitude, and altitude) of the measurement vehicle based on the RTK positioning method (that is, the interference positioning method). The RTK-GPS 3 outputs the generated position information and information on the accuracy (accuracy) of the position information (also referred to as “position accuracy information”) to the in-vehicle device 1. The IMU (inertial measurement device) 4 outputs the acceleration and angular velocity (or angle) of the measurement vehicle in the three-axis directions to the in-vehicle device 1.

The in-vehicle device 1 specifies the absolute position and orientation of the measurement vehicle based on the output supplied from the RTK-GPS 3 and depends on the position and orientation of the measurement vehicle for each point in the point cloud detected by the lidar 2. Absolute 3D position information is calculated from relative 3D position information. Note that the absolute position and direction of the measurement vehicle may be specified based on the output of the IMU 4 in addition to the output from the RTK-GPS 3. The in-vehicle device 1 supplies the calculated three-dimensional point cloud data to the server device 200 as upload information “Iu” together with the position information and position accuracy information of the measurement vehicle at the time of measurement output from the RTK-GPS 3. In this case, the in-vehicle device 1 may immediately transmit the upload information Iu to the server device 200 by wireless communication, or may store it in a storage medium or the like that can be read later by the server device 200. The point cloud data included in the upload information Iu is an example of “first point cloud information” in the present invention.

The server device 200 stores the upload information Iu acquired from the in-vehicle device 1 and updates a map DB (DB: DataBase) 20 based on the stored upload information Iu. Here, the map DB 20 includes voxel data which is data in which position information of a stationary structure is recorded for each region (also referred to as “voxel”) when the three-dimensional space is divided into a plurality of regions. ing. The voxel data includes data representing point cloud data measured for stationary structures in each voxel by a normal distribution, and is used for scan matching using NDT (Normal Distributions Transform) as will be described later. As will be described later, the server device 200 updates (including generation) voxel data corresponding to voxels within the measurement range of the measurement vehicle based on the upload information Iu. The server device 200 is an example of the “update device” in the present invention, and the voxel data is an example of “map information” in the present invention.

FIG. 2A is a block diagram showing a functional configuration of the in-vehicle device 1. The in-vehicle device 1 mainly includes an interface 11, a storage unit 12, an input unit 14, a control unit 15, and an information output unit 16. Each of these elements is connected to each other via a bus line.

The interface 11 acquires output data from sensors such as the lidar 2, the RTK-GPS 3, and the IMU 4, and supplies the output data to the control unit 15. The storage unit 12 stores a program executed by the control unit 15 and information necessary for the control unit 15 to execute a predetermined process. The input unit 14 is a button, a touch panel, a remote controller, a voice input device, or the like for a user to operate. The information output unit 16 is, for example, a display or a speaker that outputs based on the control of the control unit 15. The control unit 15 includes a CPU that executes a program, and controls the entire vehicle-mounted device 1.

FIG. 2B is a block diagram illustrating a functional configuration of the server apparatus 200. The server device 200 includes an interface 21, a storage unit 22, and a control unit 25.

The interface 21 acquires the upload information Iu generated by the in-vehicle device 1 based on the control of the control unit 25. The interface 21 may be a wireless interface for performing wireless communication with the in-vehicle device 1 or a hardware interface for reading the upload information Iu from a storage medium or the like that stores the upload information Iu.

The storage unit 22 stores a program for the control unit 25 to execute a predetermined process and information necessary for the process of the control unit 25. In the present embodiment, the storage unit 22 stores the upload information Iu acquired by the interface 21 based on the control of the control unit 25. In addition, the storage unit 22 stores a map DB 20 including voxel data updated by the upload information Iu.

The control unit 25 executes a program or the like stored in the storage unit 22 or the like, and controls the entire server device 200. In the present embodiment, the control unit 25 acquires the upload information Iu via the interface 21 and stores it in the storage unit 22. Thereafter, the control unit 25 updates the voxel data in the map DB 20 based on the upload information Iu stored in the storage unit 22. The control unit 25 is an example of a “first acquisition unit”, a “second acquisition unit”, an “update unit”, and a “computer” that executes a program in the present invention.

[Data structure of voxel data]
Next, voxel data used for scan matching based on NDT will be described. FIG. 3 shows an example of a schematic data structure of voxel data.

The voxel data includes parameter information when the point group in the voxel is expressed by a normal distribution. In this embodiment, as shown in FIG. 3, the voxel ID, voxel coordinates, average vector, and covariance matrix are used. , Point cloud number information, and reliability information. Here, “voxel coordinates” indicate absolute three-dimensional coordinates of a reference position such as the center position of each voxel. Each voxel is a cube obtained by dividing the space into a lattice shape, and since the shape and size are determined in advance, the space of each voxel can be specified by the voxel coordinates. The voxel coordinates may be used as a voxel ID.

“Average vector” and “covariance matrix” indicate an average vector and a covariance matrix corresponding to parameters when a point group in the target voxel is expressed by a normal distribution, and an arbitrary vector in any voxel “k” The coordinates of the point "i"

Is defined as, when a point number set in the voxel k to "N _k", mean vector "mu _k" and the covariance matrix "V _k" at the voxel k, respectively the following formulas (1) and ( 2).

The average vector and the covariance matrix included in the voxel data are an example of “second point group information” in the present invention.

“Point cloud number information” is information indicating the number of point clouds used to calculate the corresponding mean vector and covariance matrix. The point group number information may be information indicating the number of specific point groups, or information indicating the level of the number of point groups (for example, large, medium, small, etc.).

The “reliability information” is a weighting value based on the first weighting value that is a weighting value based on the possibility of occlusion (occlusion by an obstacle) and the accuracy of the voxel data (particularly the mean vector and covariance matrix) of the target voxel. And a second weighting value. As will be described later, the second weighting value is set based on the position accuracy at the time of measurement of the measurement vehicle and the measurement accuracy of the lidar 2. In the present embodiment, it is assumed that the first weighting value is set to a larger value as the voxel is less likely to cause occlusion, and the second weighting value is set to a larger value as the data accuracy is higher. A specific method for setting the first weight value and the second weight value will be described later. The first weighting value is an example of the “weighting value” in the present invention. The second weighting value is an example of “accuracy information” in the present invention.

[Update voxel data]
Next, a process for updating voxel data in the map DB 20 based on the upload information Iu will be described. The control unit 25 of the server device 200 divides point cloud data (hereinafter also referred to as “latest measurement data”) included in the upload information Iu supplied from the in-vehicle device 1 for each voxel, and includes each of the latest measurement data. It is determined whether voxel data already exists for the voxel.

(1) When there is no voxel data When the voxel data for the target voxel does not exist, the control unit 25 generates voxel data for the target voxel based on the upload information Iu including the latest measurement data. In this case, the control unit 25 generates NDT data (that is, an average vector, a covariance matrix, and the like) for the target voxel based on the latest measurement data, and the first weighting value and the second weighting value as described later. To decide.

In this case, for example, the control unit 25 sets the first weighting value based on the height (that is, altitude) at which each voxel is located. Specifically, the control unit 25 determines that the higher the altitude of each voxel is, the lower the possibility of occurrence of occlusion, and increases the first weighting value. Thereby, the control unit 25 can set the first weighting value that accurately reflects the possibility of occlusion for each voxel.

Further, the control unit 25 sets the second weighting value based on the position accuracy at the time of measurement of the measurement vehicle and the measurement accuracy of the lidar 2. Here, the upload information Iu includes position information and position accuracy information based on RTK-GPS3 together with point cloud data used to generate voxel data. The control unit 25 calculates the position accuracy used for calculating the second weighting value of each voxel based on the position accuracy information included in the upload information Iu. Generally, the accuracy of the distance measurement value of the lidar 2 is higher as the distance to the object is shorter, and the accuracy is lower as the distance is longer. Therefore, for example, the control unit 25 calculates the measurement accuracy of the lidar 2 based on the distance between the position indicated by the position information of the measurement vehicle included in the upload information Iu and the position indicated by the voxel coordinates of each voxel. And the control part 25 calculates | requires the precision of each voxel from the square root of the square sum of the position precision at the time of measurement of a measurement vehicle, and the measurement accuracy of the lidar 2, for example, and carries out the 2nd weighting of the reciprocal of the square of the precision. Set as a value.

(2) When Voxel Data Exists When the voxel data for the target voxel exists, the control unit 25 determines whether the voxel data needs to be updated based on the difference between the voxel data in the map DB 20 and the latest measurement data. judge.

FIG. 4 is a flowchart showing the procedure for updating the voxel data when the voxel data in the map DB 20 already exists. The server device 200 executes the process of the flowchart shown in FIG. 4 for each voxel including the latest measurement data.

First, the control unit 25 calculates the difference between the voxel data in the map DB 20 and the latest measurement data for the target voxel (step S101). For example, the control unit 25 calculates an average vector and a covariance matrix (that is, NDT data) when each point of the latest measurement data is regarded as a normal distribution in the voxel, and calculates the average vector of the voxel data in the map DB 20 and The difference from the covariance matrix is calculated. In this case, for example, the control unit 25 calculates the distance difference between the average vector of the latest measurement data and the average vector of the voxel data in the map DB 20 as the first difference, and the covariance matrix of the latest measurement data and the voxel in the map DB 20 A difference (for example, a difference between eigenvalues) from the data covariance matrix is calculated as a second difference. In another example, the control unit 25 may perform NDT matching based on the voxel data of the map DB 20 and the latest measurement data, and calculate the reciprocal of the evaluation function as a difference.

And the control part 25 determines whether the difference calculated by step S101 is below a predetermined threshold value (step S102). This threshold is a threshold for determining whether or not the difference is small enough to determine that the voxel data in the map DB 20 and the latest measurement data are substantially the same, and are set in advance based on, for example, experiments. In addition, the control part 25 sets the threshold value with respect to each difference, when the difference (namely, the above-mentioned 1st difference and 2nd difference) was each calculated by the average vector and the covariance matrix, When the difference of 1 and the second difference are each equal to or less than the threshold value, it may be determined that the difference calculated in step S101 is equal to or less than the predetermined threshold value.

When the difference calculated in step S101 is equal to or smaller than a predetermined threshold (step S102; Yes), the control unit 25 increases the first weighting value of the voxel data in the map DB 20 for the target voxel by a predetermined value or a predetermined rate ( Step S103). In this case, the control unit 25 obtains similar data between the previous measurement and the current measurement for the target voxel, so that the occlusion is unlikely to occur and the voxel data in the map DB 20 And the first weighting value is increased. Thereby, in the NDT matching performed by the general vehicle to which the map data is distributed from the server device 200, the weight of the target voxel becomes relatively high. Note that the first weight value of the voxel data further increases as the difference from the latest measurement data continues to be below a predetermined threshold.

In step S103, the control unit 25 may update voxel data other than the first weight value in addition to the change of the first weight value. In the first example, the control unit 25 has a second weight value calculated based on the update information Iu including the latest measurement data larger than the second weight value registered in the map DB 20 (that is, indicates high accuracy). In this case, the voxel data in the map DB 20 is updated with the average vector and the covariance matrix generated based on the latest measurement data, and the updated first weighting value is increased. Thereby, the control part 25 can improve the precision of the voxel data registered into map DB20 suitably. In the second example, the control unit 25 calculates the weighted average based on the second weighted value registered in the map DB 20 and the second weighted value for the latest measured data, from the voxel data and the latest measured data in the map DB 20. Voxel data such as an average vector and a covariance matrix is calculated, and the voxel data in the map DB 20 is updated with the calculated voxel data. In this case, the control unit 25 performs, for example, weighted averaging based on the second weight value, so that the voxel data for updating is weighted so that the voxel data with higher accuracy indicated by the second weight value becomes larger. calculate.

On the other hand, when the difference calculated in step S101 is larger than the predetermined threshold (step S102; No), the control unit 25 determines that the point group indicated by the latest measurement data is the measurement position than the point group indicated by the voxel data in the map DB 20. It is determined whether or not it exists far from (that is, the position on the road) (step S104). For example, in this case, the control unit 25 specifies the measurement position based on the position information included in the upload information Iu, and the point group (for example, the average vector) indicated by the latest measurement data for the measurement position is the voxel data in the map DB 20. It is determined whether or not it exists farther than the point cloud indicated by (for example, average vector).

Then, when the point cloud indicated by the latest measurement data exists farther than the point cloud indicated by the voxel data of the map DB 20 (Step S104; Yes), the control unit 25 updates the map DB 20 based on the latest measurement data (Step S104). S105). In this case, the control unit 25 determines that there is a high possibility that the occlusion generated in the point cloud data used for generating the voxel data registered in the map DB 20 did not occur during the current measurement, and the target voxel The voxel data in the map DB 20 is updated with the voxel data calculated based on the latest measurement data.

On the other hand, when the point cloud indicated by the latest measurement data does not exist farther than the point cloud indicated by the voxel data in the map DB 20 (step S104; No), the control unit 25 determines that the point cloud indicated by the latest measurement data is the voxel in the map DB 20. It is determined whether or not the data is present closer to the point cloud (step S106). For example, in this case, as in step S104, the control unit 25 determines whether or not the point group indicated by the latest measurement data is closer to the point group indicated by the voxel data in the map DB 20 within the target voxel or between voxels. Judge with. In this case, when the control unit 25 determines in step S104 that the point cloud indicated by the latest measurement data does not exist farther than the point cloud indicated by the voxel data in the map DB 20, the point cloud indicated by the latest measurement data becomes the map DB 20 It may be automatically determined that it exists closer to the point cloud indicated by the voxel data.

When the point cloud indicated by the latest measurement data is present closer to the point cloud indicated by the voxel data in the map DB 20 (step S106; Yes), the control unit 25 first weights the voxel data in the target map DB 20. The value is lowered (step S107). In this case, the control unit 25 may have generated an occlusion that did not occur during the previous measurement during the current measurement, and verifies whether it is a new stationary object or a moving object based on a plurality of measurement data separated by date and time. It judges that it is necessary, and lowers the first weighting value of the voxel data in the map DB 20 of the target voxel. When the control unit 25 determines that a new stationary object has occurred in the target voxel based on further measurement or the like, the control unit 25 updates the map DB 20 based on the latest measurement data as in step S105.

Next, a specific example based on the flowchart of FIG. 4 will be described with reference to FIGS.

FIG. 5 (A) is a diagram schematically showing the surroundings of the road 31 and the measured point cloud when the measurement vehicle first travels on the road 31. Along the road 31, stationary structures 32 to 34 exist, and a parked vehicle 37 exists in the vicinity of the stationary structure 32. Dashed

lines

42A, 43A, 44A, and 47A indicate the positions of the point groups measured by the lidar 2 at a predetermined height. In the example of FIG. 5A, occlusion by the parked vehicle 37 occurs, and one side surface of the stationary structure 32 is not measured by the lidar 2. The control unit 25 generates voxel data of voxels around the road 31 based on the upload information Iu generated when the measurement vehicle first travels on the road 31 and registers it in the map DB 20.

FIG. 5 (B) is a diagram schematically showing the surroundings of the road 31 and the measured point cloud when the measurement vehicle travels on the road 31 for the second time. In FIG. 5B, dotted

lines

42B, 43B, 44B, and 48B indicate the positions of the point groups measured by the lidar 2 at a predetermined height. At the time of the second measurement, another parked vehicle 38 is present in the vicinity of the

stationary structures

33 and 34 instead of the parked vehicle 37 present at the time of the first measurement. And the occlusion by the parked vehicle 38 occurs, and some of the

stationary structures

33 and 34 measured at the time of the first measurement are not measured by the lidar 2.

FIG. 6 is a diagram in which the point cloud measured for the first time and the point cloud measured for the second time are superimposed. Here, the frame 50 encloses a point group obtained by the second measurement of the stationary structure 32 that was not measured by the first measurement. The frame 51 surrounds a point group obtained by the first measurement of the stationary structure 33 that was not measured in the second measurement, and the frame 52 is a stationary structure that was not measured in the second measurement. The point cloud by 34 1st measurement is enclosed.

First, for the voxel existing at the position where the

broken lines

42A, 43A, 44A and the

dotted lines

42B, 43B, 44B overlap (that is, the position not surrounded by the frames 50 to 52), the control unit 25 performs step S102 in FIG. For the target voxel, the difference between the voxel data in the map DB 20 and the latest measurement data is determined to be equal to or less than the threshold. Therefore, in this case, the control unit 25 increases the first weighting value of these voxels (see step S103).

Moreover, since the difference between the voxel data in the map DB 20 and the latest measurement data is greater than the threshold for the voxel including the dotted line 42B in the frame 50 (see step S102), the control unit 25 determines the point group indicated by the latest measurement data. It is determined whether or not the object exists farther from the measurement position than the point cloud indicated by the voxel data in the map DB 20 (see step S104). In this case, for example, the control unit 25 determines that the point group in the dotted line 42B in the frame 50 exists farther than the point group indicated by the broken line 47A. Therefore, in this case, the control unit 25 updates the map DB 20 based on the latest measurement data indicating the point group in the dotted line 42B in the frame 50 (see step S105).

On the other hand, in step S106, the control unit 25 determines that the point cloud indicated by the latest measurement data is closer to the measurement position for the voxel including the point cloud indicated by the dotted line 48B. Therefore, in this case, the control unit 25 lowers the first weighting value for the voxel including the point group indicated by the

broken lines

43A and 44A in the

frames

51 and 52 based on step S107.

[Example of scan matching using voxel data]
Next, scan matching by NDT using voxel data including reliability information will be described. Here, an example in which a vehicle equipped with a positioning device such as a gyro sensor and a GPS receiver and a lidar (also referred to simply as “general vehicle”) refers to map data distributed from the server device 200 and estimates its own vehicle position. explain.

Scan matching by NDT assuming a vehicle is to estimate the following estimated parameter “P” with the amount of movement in the road plane (here, xy coordinates) and the direction of the vehicle as elements.

“T _x ” indicates the amount of movement in the x direction, “t _y ” indicates the amount of movement in the y direction, and “Ψ” indicates the rotation angle (ie, yaw angle) in the xy plane. The vertical movement amount, pitch angle, and roll angle are small enough to be ignored, although they are caused by road gradients and vibrations.

When the coordinates [x _k (i), y _k (i), z _k (i)] ^T of the arbitrary point of the point cloud data obtained by the lidar 2 are coordinate-transformed using the estimation parameter P described above, The coordinates “X ′ _k (i)” are expressed by the following equation (3).

In this embodiment, the general vehicle uses the coordinate-converted point group, the average vector μ _k and the covariance matrix V _k included in the voxel data, and evaluates the voxel k expressed by the following equation (4). A comprehensive evaluation function “E” (also referred to as “total evaluation function”) for all voxels to be matched indicated by the function “E _k ” and Expression (5) is calculated.

“M” indicates the number of voxels to be matched, “w _k ” indicates a first weighting value for voxel k, and “1 / σ using accuracy information“ σ _k ”for voxel k. “ _k ² ” indicates a second weighting value for voxel k. Here, as the second weighting value 1 / σ _k ² is larger, the accuracy is higher (that is, higher reliability). Therefore, according to the equation (4), the evaluation function E _k becomes a larger value as the first weighting value w _k is larger and the second weighting value 1 / σ _k ² is larger. In addition, since normalization is performed using the number of point groups N _k , differences due to the number of point groups are reduced. Note that the coordinates of the point cloud data obtained by the lidar are relative coordinates with respect to the vehicle position, and the average vector of the voxel data is an absolute coordinate. Therefore, when calculating the equation (4), for example, by the lidar The coordinates of the obtained point cloud data are converted based on the vehicle position predicted from the output of the GPS receiver or the like.

On the other hand, the evaluation function E _k of the voxel k used in the conventional NDT matching is expressed by the following equation (6).

As is apparent from the comparison between the equations (4) and (6), in this embodiment, the general vehicle uses the first weighting value w _k and the second weighting value 1 / σ _k ² , so that each voxel On the other hand, weighting is performed according to the reliability of each voxel data (average vector, covariance matrix). As a result, the general vehicle relatively reduces the weighting of the evaluation function E _k of the voxel with low reliability, and suitably improves the position estimation accuracy by NDT matching.

After that, the general vehicle calculates an estimation parameter P that maximizes the comprehensive evaluation function E by an arbitrary root finding algorithm such as Newton's method. Then, the general vehicle estimates the own vehicle position with high accuracy by applying the estimation parameter P to the own vehicle position predicted from the output of the GPS receiver or the like.

Next, a specific example of NDT scan matching will be described. Hereinafter, for convenience of explanation, the case of a two-dimensional plane will be described as an example.

FIG. 7A shows the point groups measured by the measuring vehicle in four adjacent voxels “B1” to “B4” with circles, and based on these point groups, Expressions (1) and (2) It is the figure which showed the two-dimensional normal distribution created from the above by gradation. The average and variance of the normal distribution shown in FIG. 7A correspond to the average vector and covariance matrix in the voxel data, respectively.

FIG. 7B is a diagram showing the point cloud acquired by the lidar 2 while the general vehicle is traveling in FIG. The position of the lidar point cloud indicated by the asterisk is aligned with the voxels B1 to B4 based on the estimated position based on the output of the GPS receiver 5 or the like. In the example of FIG. 7B, there is a deviation between the point cloud (circle) measured by the measurement vehicle and the point cloud (star) acquired by the general vehicle.

FIG. 7C is a diagram illustrating a state after the point cloud (star) acquired by the general vehicle is moved based on the matching result of the NDT scan matching. In FIG. 7C, a parameter P that maximizes the evaluation function E shown in the equations (4) and (5) is calculated based on the mean and variance of the normal distribution shown in FIGS. 7 (A) and (B). The calculated parameter P is applied to the star point cloud shown in FIG. In this case, the deviation between the point cloud (circle) measured by the measurement vehicle and the point cloud (star) acquired by the general vehicle is suitably reduced.

Here, when the evaluation functions “E1” to “E4” and the comprehensive evaluation function E corresponding to the voxels B1 to B4 are calculated by the general formula (6) conventionally used, these values are as follows: Become.
E1 = 1.3290
E2 = 1.1365
E3 = 1.1100
E4 = 0.9686
E = 4.5441
In this example, there is no significant difference in the evaluation functions E1 to E4 of each voxel, but there are some differences depending on the number of point groups included in the voxel.

In this embodiment, a first weighting value and a second weighting value are set for each voxel. Therefore, it is possible to increase the degree of matching of voxels by increasing the weighting of voxels with high reliability. Hereinafter, as an example, a specific example in which the first weighting value is set for each voxel will be described with reference to FIG.

FIG. 8A is a diagram showing a matching result when the first weight values for voxels B1 to B4 are all equal (that is, the same diagram as FIG. 7C). FIG. 8B is a diagram showing a matching result when the first weighting value of the voxel B1 is 10 times the weighting value of the other voxels. FIG. 8C is a diagram showing a matching result when the first weighting value of the voxel B3 is set to 10 times the weighting values of the other voxels. In any example, it is assumed that the second weighting values are all set to the same value.

In the example of FIG. 8B, the values of the evaluation functions E1 to E4 and the comprehensive evaluation function E corresponding to the voxels B1 to B4 are as follows.
E1 = 0.3720
E2 = 0.0350
E3 = 0.0379
E4 = 0.0373
E = 0.4823

As described above, in the example of FIG. 8B, matching is performed so that the value of the evaluation function E1 corresponding to the voxel B1 is high, and the degree of matching in the voxel B1 is increased. Therefore, the deviation between the circle mark and the star mark of the voxel B1 is reduced. Moreover, since the value of the evaluation function is small because it is normalized by the number of point groups, each evaluation function value has a ratio similar to the weighting value.

In the example of FIG. 8C, the values of the evaluation functions E1 to E4 and the comprehensive evaluation function E corresponding to the voxels B1 to B4 are as follows.
E1 = 0.0368
E2 = 0.0341
E3 = 0.3822
E4 = 0.0365
E = 0.4896

In the example of FIG. 8C, matching is performed such that the value of the evaluation function E3 corresponding to the voxel B3 is high, and the degree of matching in the voxel B3 is increased. Therefore, the deviation between the circle mark and the star mark of the voxel B3 is reduced. As described above, by appropriately setting the first weighting value, the degree of matching for voxels with low possibility of occurrence of occlusion is preferably increased, in other words, the degree of matching for voxels with high possibility of occurrence of occlusion is preferable. Can be lowered. Similarly, for the second weighting value, by appropriately setting the second weighting value, the degree of matching for voxels with relatively high measurement accuracy is increased, and the degree of matching for voxels with relatively low measurement accuracy is reduced. Can do. Then, according to the present embodiment, the server device 200 calculates the average vector, covariance matrix, first weight value, second weight value, and the like for each voxel that is preferably used for the above-described NDT matching based on the update information Iu. It can be suitably updated.

As described above, the server apparatus 200 according to the present embodiment uses the map DB 20 in which voxel data including at least the first weight value in addition to the average vector and covariance matrix of the point cloud for each voxel is recorded for each voxel. Prepare. And the server apparatus 200 acquires the upload information Iu containing the point cloud data which the measurement vehicle measured. And the server apparatus 200 raises the 1st weighting value with respect to the voxel from which the difference of the latest measurement data divided | segmented for every voxel based on the upload information Iu, and the voxel data of map DB20 becomes below a predetermined threshold value. By doing in this way, the server apparatus 200 increases the first weighting value for the voxel in which the measurement result similar to the latest measurement data is recorded in the map DB 20, and can relatively increase the matching degree for the voxel. it can.

[Modification]
Hereinafter, modified examples suitable for the embodiments will be described. The following modifications may be applied to the embodiments in combination.

(Modification 1)
As shown in FIG. 3, the first weighting value and the second weighting value are recorded in the voxel data included in the map DB 20 as reliability information. Instead, only the first weight value may be recorded in the voxel data. In this case, the server device 200 determines or updates the first weighting value for each voxel based on the upload information Iu.

(Modification 2)
The voxel data is not limited to a data structure including an average vector and a covariance matrix as shown in FIG. For example, the voxel data may include point cloud data measured by a measurement vehicle used when calculating an average vector and a covariance matrix. Further, the voxel data generated or updated by the server apparatus 200 is not limited to a case where only scan matching by NDT is targeted, and is voxel data for use in other scan matching such as ICP (Iterative Closest Point). Also good.

1 In-vehicle device 20 Map DB
DESCRIPTION OF

SYMBOLS

11, 21

Interface

12, 22 Storage part 14

Input part

15, 25 Control part 16 Information output part 200 Server apparatus

Claims

A first acquisition unit that acquires first point cloud information about each distance from the reference position to a plurality of positions measured by the measurement unit;
Obtained is map information that is divided into a plurality of regions, and second point cloud information based on one or a plurality of position information and weighting values based on the reliability of the second point cloud information are recorded for each of the plurality of regions. A second acquisition unit;
An updating unit that updates the weighting value so as to increase the weighting for a region in which a difference between the first point group information and the second point group information is a predetermined value or less among the plurality of regions;
An update device comprising:
The update unit updates the second point group information for a region where a difference between the first point group information and the second point group information is a predetermined value or less based on accuracy information indicating the accuracy of the measurement unit. Item 2. The updating device according to Item 1.
The map information further includes accuracy information related to the measurement accuracy of the second point cloud information,
The update unit, for a region where the difference between the first point cloud information and the second point cloud information is a predetermined value or less,
When the first point cloud information is more accurate than the second point cloud information, update the second point cloud information based on the first point cloud information, or
The updating apparatus according to claim 2, wherein the second point group information is updated by weighted averaging the first point group information and the second point group information based on accuracy.
The update unit is configured such that the position indicated by the first point cloud information in a region where the difference between the first point cloud information and the second point cloud information is greater than a predetermined value is greater than the position indicated by the corresponding second point cloud information. The updating apparatus according to any one of claims 1 to 3, wherein the map information is updated with the first point group information when the position is far from the measurement position.
The update unit is configured such that the position indicated by the first point cloud information in a region where the difference between the first point cloud information and the second point cloud information is greater than a predetermined value is greater than the position indicated by the corresponding second point cloud information. The updating device according to any one of claims 1 to 4, wherein when the position is close to the measurement position, the weighting value is updated so as to reduce the weighting for the region of the position indicated by the second point group information.
The updating unit determines whether a difference between the first point group information and the second point group information is equal to or less than a predetermined value, and an average or / and variance of each of the first point group information and the second point group information. The update device according to any one of claims 1 to 5, wherein the update device is determined based on:
A control method executed by the update device,
A first acquisition step of acquiring first point cloud information about each distance from the reference position to a plurality of positions measured by the measurement unit;
Obtained is map information that is divided into a plurality of regions, and second point cloud information based on one or a plurality of position information and weighting values based on the reliability of the second point cloud information are recorded for each of the plurality of regions. A second acquisition step;
An update step of updating the weighting value so as to increase the weighting for the region where the difference between the first point cloud information and the second point cloud information is a predetermined value or less among the plurality of regions,
A control method.
A program executed by a computer,
A first acquisition unit that acquires first point cloud information about each distance from the reference position to a plurality of positions measured by the measurement unit;
Obtained is map information that is divided into a plurality of regions, and second point cloud information based on one or a plurality of position information and weighting values based on the reliability of the second point cloud information are recorded for each of the plurality of regions. A second acquisition unit;
A program that causes the computer to function as an update unit that updates the weighting value so as to increase the weighting for a region in which the difference between the first point cloud information and the second point cloud information is a predetermined value or less among the plurality of regions. .
A storage medium storing the program according to claim 8.