JP2023174739A - Data structure, information processing device, and map data generator - Google Patents

Abstract

To provide map data containing advance information for grasping objects to be used for self-position estimation in advance.SOLUTION: An in-vehicle device 1 provided herein estimates the self-position based on measurements of objects measured using a LiDAR 2 mounted on a vehicle and position information of the objects included in a map DB 10. The map DB 10 contains object recommendation information indicative of a recommendation value of each object for the position estimation or voxel recommendation information indicative of a recommendation value of each voxel for the position estimation.SELECTED DRAWING: Figure 6

Description

The present invention relates to self-position estimation technology.

2. Description of the Related Art Conventionally, a technique has been known in which a radar or a camera is used to detect terrestrial objects installed in a destination where a vehicle is traveling, and the vehicle position is calibrated based on the detection results. For example, Patent Document 1 discloses a technique for estimating one's own position by comparing the output of a measurement sensor with position information of features registered on a map in advance. Further, Patent Document 2 discloses a vehicle position estimation technique using a Kalman filter. Furthermore, Non-Patent Document 1 discloses specifications regarding a data format for collecting data detected by a vehicle-side sensor on a cloud server.

JP2013-257742A JP2017-72422A

here company homepage, Vehicle Sensor Data Cloud Ingestion Interface Specification (v2.0.2), [searched on March 2, 2018], Internet <URL: https://lts.cms.here.com/static-cloud- content/Company_Site/2015_06/Vehicle_Sensor_Data_Cloud_Ingestion_Interface_Specification.pdf＞

When estimating the vehicle's position by comparing the measurement results of objects around the vehicle's position by an external sensor such as a lidar with the position information of the object on the map, the target object may be affected by occlusion from other vehicles or rain or snow. If it cannot be detected, the accuracy of estimating the vehicle's position will deteriorate. Similarly, if the position or shape of the object itself changes, the verification result will shift due to mismatch with the position of the object on the map, resulting in a deterioration in the accuracy of estimating the vehicle's position.

The present invention has been made to solve the above-mentioned problems, and its main purpose is to provide map data that includes prior information for understanding objects to be used for self-position estimation in advance. .

The claimed invention is a data structure of map data, wherein the position of the moving object is estimated using position information of the object and a positional relationship between the object and the object determined by a measuring device mounted on the moving object. This is a data structure of map data used for estimating the position of the mobile object, including recommended value information indicating a recommended value for what to do.

Further, the claimed invention is an information processing device that estimates the position of the moving object using position information of the object and a positional relationship between the object and the object determined by a measuring device mounted on the moving object. It has a storage unit that stores map data including recommended value information indicating recommended values for things to be done.

Further, the claimed invention provides a map data generation device that generates map data based on identification information of an object and effectiveness information regarding the effectiveness of improving the accuracy of position estimation of the moving body using the object. The apparatus further includes a generation unit that generates map data by associating recommended value information indicating a recommended value for estimating the position of the mobile object using the object with the position information of the object.

FIG. 1 is a schematic configuration diagram of a driving support system. FIG. 2 is a block diagram showing the functional configuration of an on-vehicle device and a server device. It is a diagram showing a state variable vector using two-dimensional orthogonal coordinates. FIG. 3 is a diagram showing a schematic relationship between a prediction step and a measurement update step. The functional blocks of the own vehicle position estimation section are shown. This is an example of the data structure of object recommendation information. FIG. 3 is a diagram showing an overview of the data structure of upload information. The data structure of the "object recognition event" included in the event information is shown. FIG. 2 is a bird's-eye view showing the area around the vehicle. An example of setting the valid flag is shown below. It is a graph showing changes in diagonal elements of a covariance matrix during a white line detection period. It is a graph showing changes in diagonal elements of a covariance matrix during a marker detection period. 12 is an example of a flowchart illustrating an overview of processing related to transmission and reception of upload information including a validity flag. The functional blocks of a vehicle position estimation unit that performs position estimation based on voxel data are shown. An example of a schematic data structure of voxel data is shown. This is an example of a data structure of voxel recommendation information. FIG. 2 is a bird's-eye view showing the area around the vehicle. An example of setting valid values is shown below. It is a graph showing the transition of the individual evaluation function value for voxel ID "4". It is a graph showing the transition of the individual evaluation function value for voxel ID "12". 3 is an example of a flowchart illustrating an overview of processing related to transmission and reception of upload information including valid values.

According to a preferred embodiment of the present invention, the data structure of the map data is such that the position information of the object and the positional relationship between the object and the object determined by a measuring device mounted on the moving object are used to determine the location of the moving object. and recommended value information indicating a recommended value for performing position estimation, and is a data structure of map data used for position estimation of the mobile object. By having such a data structure, the map data can suitably include prior information for understanding objects to be used for self-position estimation in advance.

In one aspect of the data structure, the recommended value information includes a recommended value for estimating the position in a direction based on the moving object. In a preferred example, the recommended value information further includes a recommended value for performing the position estimation regarding the orientation of the mobile object. According to this aspect, the degree of recommendation of position estimation for an object used for position estimation can be recorded in the map data for each state variable estimated in the position estimation process.

In another aspect of the data structure, the position information of the object and the recommended value information are provided for each unit area that partitions space, and the recommended value information is provided for each unit area of the object. Recommended values for each unit area are shown when estimating the position of the moving body by comparing the position information with the positional relationship with the object. According to this aspect, the map data suitably includes a priori information for understanding in advance the unit area to be used for position estimation in which the positional information of the object and the positional relationship with the object are checked for each unit area.

According to another preferred embodiment of the present invention, the information processing device estimates the position of the moving object using the position information of the object and the positional relationship between the object and the object determined by a measuring device mounted on the moving object. It has a storage unit that stores map data including recommended value information indicating a recommended value for performing. With this aspect, the information processing device can perform position estimation by preferably referring to prior information for understanding objects to be used for self-position estimation in advance, and can distribute the information to other devices that perform position estimation. can.

According to another preferred embodiment of the present invention, the map data generation device generates the map data based on the identification information of the object and the effectiveness information regarding the effectiveness of improving the accuracy of position estimation of the mobile object using the object. The apparatus further includes a generation unit that generates map data by associating recommended value information indicating a recommended value for estimating the position of the mobile body using the object with the position information of the object. With this aspect, the map data generation device can suitably include in the map data prior information for understanding in advance the object to be used for self-position estimation.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. Note that a character with "^" or "-" attached above an arbitrary symbol is expressed as "A ^{^} " or " ^A- "("A" is an arbitrary character) in this specification for convenience.

<First example>
(1-1) Overview of driving support system
FIG. 1 shows a schematic configuration of a driving support system according to a first embodiment. The driving support system includes an on-vehicle device 1 that moves together with each vehicle, which is a mobile object, and a server device 6 that communicates with each on-vehicle device 1 via a network. Then, the driving support system updates the distribution map DB 20, which is a distribution map held by the server device 6, based on the information transmitted from each on-vehicle device 1. Note that hereinafter, the term "map" includes data used in ADAS (Advanced Driver Assistance System) and automatic driving, in addition to data referenced by conventional in-vehicle equipment for route guidance.

The on-vehicle device 1 is electrically connected to a lidar 2, a gyro sensor 3, a vehicle speed sensor 4, and a GPS receiver 5, and based on these outputs, detects a predetermined object and detects a vehicle in which the on-vehicle device 1 is installed. The vehicle's position (also referred to as the "own vehicle position") is estimated. Based on the estimation result of the own vehicle position, the on-vehicle device 1 performs automatic driving control of the vehicle so that the vehicle travels along the route to the set destination. The on-vehicle device 1 stores a map database (DB: DataBase) 10 in which road data and information regarding objects such as landmarks and lane markings provided near the road are registered. The above-mentioned features that serve as landmarks are, for example, features such as kilometer posts, 100m posts, delineators, traffic infrastructure facilities (for example, signs, direction signs, traffic lights), utility poles, and streetlights that are periodically lined up on the side of the road. Based on this map DB 10, the on-vehicle device 1 compares the output from the lidar 2 and the like to estimate the own vehicle position. Furthermore, the vehicle-mounted device 1 transmits upload information “Iu” including information regarding the detected object to the server device 6. The vehicle-mounted device 1 is an example of an information transmitting device.

The lidar 2 discretely measures the distance to an object in the outside world by emitting a pulsed laser in a predetermined angular range in the horizontal and vertical directions, and generates a three-dimensional point that indicates the position of the object. Generate group information. In this case, the lidar 2 includes an irradiation section that irradiates laser light while changing the irradiation direction, a light reception section that receives reflected light (scattered light) that is reflected by the irradiated laser light on an object, and a light reception signal that the light reception section outputs. and an output section that outputs scan data based on. The scan data is point group data, and is generated based on the irradiation direction corresponding to the laser light received by the light receiving section and the distance to the object in the irradiation direction, which is specified based on the above-mentioned light reception signal. Hereinafter, objects such as marking lines and terrestrial features to be measured by the rider 2 in the vehicle position estimation process will also be referred to as "landmarks." The rider 2, the gyro sensor 3, the vehicle speed sensor 4, and the GPS receiver 5 each supply output data to the vehicle-mounted device 1. The rider 2 is an example of a "measuring device".

The server device 6 receives upload information Iu from each on-vehicle device 1 and stores it. The server device 6 updates the distribution map DB 20, for example, based on the collected upload information Iu. Further, the server device 6 transmits download information Id including update information of the distribution map DB 20 to each vehicle-mounted device 1. The server device 6 is an example of an information processing device and a map data generation device.

FIG. 2(A) is a block diagram showing the functional configuration of the vehicle-mounted device 1. The in-vehicle device 1 mainly includes an interface 11 , a storage section 12 , a communication section 13 , an input section 14 , a control section 15 , and an information output section 16 . Each of these elements is interconnected via a bus line.

The interface 11 acquires output data from sensors such as the rider 2, the gyro sensor 3, the vehicle speed sensor 4, and the GPS receiver 5, and supplies it to the control unit 15.

The storage unit 12 stores programs executed by the control unit 15 and information necessary for the control unit 15 to execute predetermined processes. In this embodiment, the storage unit 12 stores a map DB 10 that includes attribute information such as the position, size, and shape of objects serving as landmarks and object recommendation information described below. Here, the object recommendation information is information indicating, for each object, a recommendation value (also referred to as an "object recommendation value") for estimating the vehicle's position using the target object. The data structure of the object recommendation information will be described later.

The communication unit 13 transmits upload information Iu, receives download information Id, etc. under the control of the control unit 15. The input unit 14 is a button, a touch panel, a remote controller, a voice input device, etc. for user operation. The information output unit 16 is, for example, a display, a speaker, etc. that outputs based on the control of the control unit 15.

The control unit 15 includes a CPU that executes programs, and controls the entire vehicle-mounted device 1. In this embodiment, the control unit 15 includes a vehicle position estimation unit 17 and an upload control unit 18.

The own vehicle position estimating unit 17 uses the gyro sensor 3, the vehicle speed sensor 4, and/or the GPS receiver based on the distance and angle measurements made by the rider 2 with respect to the landmarks and the position information of the landmarks extracted from the map DB 10. The own vehicle position estimated from the output data of step 5 is corrected. In this embodiment, as an example, the own vehicle position estimating unit 17 performs a prediction step of predicting the own vehicle position from the output data of the gyro sensor 3, the vehicle speed sensor 4, etc. based on a state estimation method based on Bayesian estimation, and a prediction step of predicting the own vehicle position from the output data of the gyro sensor 3, the vehicle speed sensor 4, etc. A measurement update step for correcting the predicted value of the own vehicle position calculated in the prediction step is executed alternately. As the state estimation filter used in these steps, various filters developed to perform Bayesian estimation can be used, such as an extended Kalman filter, an unscented Kalman filter, and a particle filter. In the first embodiment, it is assumed that the own vehicle position estimating unit 17 performs own vehicle position estimation using an extended Kalman filter, as an example. Vehicle position estimation using the extended Kalman filter will be explained in the section "Position estimation based on extended Kalman filter".

When the upload control unit 18 detects a predetermined object based on the output of an external sensor such as the lidar 2, it generates upload information Iu including information about the detected object, and transmits the upload information Iu to the server device 6. In addition, when the own vehicle position estimating section 17 performs own vehicle position estimation using the detected object as a landmark, the upload control section 18 determines the effectiveness of improving the accuracy of position estimation by the own vehicle position estimation as a state variable. The upload information Iu includes flag information (also referred to as a "valid flag") indicating the determination result and transmits it to the server device 6. The upload control unit 18 is an example of a “generation unit”, a “transmission unit”, and a “computer” that executes a program. Further, the validity flag is an example of "validity information".

FIG. 2(B) is a block diagram showing the functional configuration of the server device 6. As shown in FIG. The server device 6 mainly includes a communication section 61, a storage section 62, and a control section 65. Each of these elements is interconnected via a bus line.

The communication unit 61 receives upload information Iu, transmits download information Id, etc. under the control of the control unit 65. The storage unit 62 stores programs executed by the control unit 65 and information necessary for the control unit 65 to execute predetermined processes. In this embodiment, the storage unit 62 stores a distribution map DB 20 and an upload information DB 27 that stores upload information Iu received from each vehicle-mounted device 1. The distribution map DB 20 includes object recommendation information generated by the control unit 65 with reference to the upload information DB 27.

The control unit 65 includes a CPU that executes programs, and controls the entire server device 6. In this embodiment, the control unit 65 performs a process of accumulating upload information Iu received from each in-vehicle device 1 by the communication unit 61 in the upload information DB 27, a process of generating object recommendation information based on the upload information DB 27, and a process of generating object recommendation information based on the upload information DB 27. Processing such as transmitting map update information such as recommended information to each vehicle-mounted device 1 by the communication unit 61 is performed.

(1-2) Position estimation based on extended Kalman filter
FIG. 3 is a diagram showing the vehicle position to be estimated using two-dimensional orthogonal coordinates. As shown in Figure 3, the vehicle position on a plane defined on the xy two-dimensional orthogonal coordinates is expressed by the coordinates "(x, y)" and the vehicle's azimuth (yaw angle) "ψ". . Here, the yaw angle ψ is defined as the angle between the traveling direction of the vehicle and the x-axis. In this example, in addition to the coordinates (x, y) and yaw angle ψ described above, four variables (x, y, z, ψ) are used, taking into account the coordinates of the z-axis perpendicular to the x-axis and y-axis. The vehicle position is estimated using the vehicle position as a state variable. Note that since a typical road has a gentle slope, the pitch angle and roll angle of the vehicle are basically ignored in this embodiment.

FIG. 4 is a diagram showing a schematic relationship between the prediction step and the measurement update step. Further, FIG. 5 shows an example of the own vehicle position estimating section 17, which is a functional block of the control section 15. As shown in FIG. 4, by repeating the prediction step and the measurement update step, calculation and updating of the estimated value of the state variable vector "X" indicating the own vehicle position is performed sequentially. Further, as shown in FIG. 5, the vehicle position estimating section 17 includes a position estimating section 21 that performs a prediction step and a position estimating section 22 that performs a measurement updating step. The position prediction unit 21 includes a dead reckoning block 23 and a position prediction block 24, and the position estimation unit 22 includes a landmark search/extraction block 25 and a position correction block 26. Note that in FIG. 4, the state variable vector at the reference time (ie, current time) "k" to be calculated is expressed as "X ^- (k)" or "X ^{^} (k)". Here, the provisional estimated value (predicted value) estimated in the prediction step is marked with a " ^- " above the character representing the predicted value, and the more accurate estimated value updated in the measurement update step is `` ^{^} '' is added above the character representing the value.

In the prediction step, the dead reckoning block 23 of the control unit 15 calculates the moving speed "v" and the angular velocity "ω" of the vehicle (which are collectively called "control value u(k)=(v(k), ω(k))". ⁾ is used to find the travel distance and change in direction from the previous time. The position prediction block 24 of the control unit 15 adds the obtained moving distance and ^azimuth change to the state variable vector The predicted value of the own vehicle's position (also called "predicted position") X ^- (k) is calculated. At the same time, the covariance matrix "P ^- (k)" corresponding to the error distribution ^of the predicted position Calculated from ``P ^{^} (k-1)''.

In the measurement update step, the landmark search/extraction block 25 of the control unit 15 correlates the position vectors of landmarks registered in the map DB 10 with the scan data of the rider 2. Then, when this association is established, the landmark search/extraction block 25 of the control unit 15 calculates the measured value "Z(k)" by the lidar 2 of the landmark that has been associated, and the predicted position X ^- ( k) and the landmark measurement value (referred to as the "measurement predicted value") obtained by modeling the measurement process by the lidar 2 using the landmark position vector registered in the map DB 10 "Z ^- (k)" and obtain respectively. The measured value Z(k) is calculated using the vehicle coordinate system ("vehicle coordinate system"), which is converted from the landmark distance and scan angle measured by the rider 2 at time k into components with the vehicle's traveling direction and lateral direction as axes. It is also called a vector value at ). Then, the position correction block 26 of the control unit 15 applies a Kalman gain "K(k)" to the difference value between the measured value Z(k) and the measured predicted value ^Z- (k), as shown in equation (1) below. '' and adding this to the predicted position X ^- (k) to calculate the updated state variable vector (also referred to as the "estimated position") X ^{^} (k).

In addition, in the measurement update step, the position correction block 26 of the control unit 15 uses a covariance matrix P ^{^} (k) (also simply P(k)) corresponding to the error distribution of the estimated position X ^{^} (k), as in the prediction step. ) is determined from the covariance matrix P ⁻ (k). Parameters such as the Kalman gain K(k) can be calculated in the same manner as a known self-position estimation technique using, for example, an extended Kalman filter.

In this way, the prediction step and measurement update step are repeatedly performed, and the predicted position X ^{- (} k) and estimated position .

Here, whether or not the accuracy of position estimation has improved can be determined based on the values of the diagonal elements of the covariance matrix P. Here, if the covariance matrix calculated based on the measured value for the landmark at time "k" is P(k), the covariance matrix P(k) is expressed by the following equation (2).

In this case, the diagonal elements “σ _x ² (k)”, “σ _y ² (k)”, “σ _z ² (k)”, and “σ _ψ ² (k)” of the covariance matrix P(k) are If each becomes smaller, it can be determined that the accuracy of estimating the vehicle's position using the landmark at time k has improved.

Note that if the state variables x, y, and z are the global coordinate system adopted in the map, the matrix “C _ψ (k) ” is converted into the vehicle coordinate system (X, Y, Z).

The matrix C _ψ (k) is expressed by the following equation (4).

In this case, the onboard device 1 converts the diagonal elements σ _X ² (k), σ _Y ² (k), σ _Z ² (k), σ _Ψ ² ( k), it is possible to determine the estimation accuracy of each state variable regarding the traveling direction, lateral direction, height direction, and orientation of the own vehicle position estimation.

(1-3) Data structure of object recommendation information
Next, object recommendation information commonly included in the map DB 10 and distribution map DB 20 will be explained.

FIG. 6 is an example of the data structure of object recommendation information. As shown in FIG. 6, the object recommendation information is information in which "object ID", "position", and "object recommendation value" are associated with each other. An object ID, which is an identification number assigned to each object serving as a landmark, is specified in "object ID." The object ID is an example of "object identification information." The latitude, longitude, and altitude of the target landmark are each specified in the "position" field. "Object recommendation value" indicates the object recommendation value given to the target object as the state variable for vehicle position estimation (here, traveling direction (x), lateral direction (y), height direction (z)) , direction (ψ)). The object recommendation value is set to a value between 0 and 1, and approaches 1 as the effectiveness of improving the estimation accuracy of vehicle position estimation for the target state variable increases.

In this manner, in this embodiment, the map DB 10 includes object recommendation information indicating the effectiveness (suitability) of position estimation for each object. Thereby, the in-vehicle device 1 can refer to the object recommendation information and specify an object that improves the accuracy of estimating the own vehicle position. Note that the in-vehicle device 1 transmits a predetermined request signal including current position information to the server device 6, so that even if the in-vehicle device 1 receives download information Id including object recommendation information regarding objects around the vehicle's location from the server device 6, good. In this case, the vehicle-mounted device 1 controls the vehicle to improve the accuracy of estimating the own vehicle position based on the object recommendation information included in the received download information Id.

Here, a specific example of vehicle control based on object recommendation information will be described.

For example, if the accuracy of the current self-vehicle position estimation result in the lateral direction is poor (that is, σ _y ² (k) is high), the onboard device 1 searches for objects existing within a predetermined distance from the current position from the object recommendation information. However, the moving route of the vehicle may be determined based on the position of an object that has a high object recommendation value of the target state variable (in this case, lateral direction) among the searched objects. In this case, for example, the vehicle-mounted device 1 moves the vehicle to a lane where the target object is easily detected (for example, a lane closest to the target object). In addition to or instead of controlling the vehicle described above, when estimating the own vehicle position using a plurality of objects, the on-vehicle device 1 also measures the object with a high object recommendation value of the target state variable. A Kalman filter or the like may be calculated by increasing the weighting of the values. In addition, when there is no object with a high object recommendation value of the target state variable around the current position, the onboard device 1 switches the position estimation method (for example, position estimation by dead reckoning or the second embodiment described later). (switching to the position estimation method) may also be performed.

(1-4) Data structure of upload information
Next, a specific example of the data structure of the upload information Iu will be explained.

FIG. 7 is a diagram showing an outline of the data structure of upload information Iu transmitted by the on-vehicle device 1. As shown in FIG. 7, the upload information Iu includes header information, travel route information, event information, and media information.

The header information includes each item of "version", "sender", and "vehicle metadata". The in-vehicle device 1 specifies the version information of the data structure of the upload information Iu to be used in "Version", and the name of the company (OEM name of the vehicle or system vendor) that sends the upload information Iu in "Sender". (first name) information. Furthermore, the on-vehicle device 1 specifies vehicle attribute information (for example, vehicle type, vehicle ID, vehicle width, and vehicle height) in "vehicle metadata." The driving route information includes an item of "position estimation". In this "position estimation", the on-vehicle device 1 specifies, in addition to time stamp information indicating the estimated position time, information on latitude, longitude, and altitude indicating the estimated position of the vehicle, and information regarding the accuracy of these estimates. .

The event information includes an item of "object recognition event." When the in-vehicle device 1 detects an object recognition event, it designates information representing the detection result as an "object recognition event." Media information is a data type used when transmitting raw data that is output data (detection information) of an external sensor such as the lidar 2.

FIG. 8 shows the data structure of the "object recognition event" included in the event information. FIG. 8 shows, for each element (subitem) included in the "object recognition event", information as to whether designation of information corresponding to each element is essential or optional.

As shown in FIG. 8, the "object recognition event" includes "time stamp", "object ID", "offset position", "object type", "object size", "object size precision", "media ID", Contains each element of "valid flag".

When the control unit 15 of the vehicle-mounted device 1 detects an object based on the output of an external sensor such as the lidar 2, it generates event information of an "object recognition event" having a data structure shown in FIG. Here, the in-vehicle device 1 specifies the time at which the object was detected in the "time stamp" and specifies the object ID of the detected object in the "object ID".

Further, in the "offset position", the vehicle-mounted device 1 specifies information about the relative position of the detected object from the vehicle (for example, latitude difference, longitude difference, etc.). In the "object type", the in-vehicle device 1 specifies information indicating the type of the detected object. Furthermore, when the vehicle-mounted device 1 is able to generate size information of the detected object and accuracy information of the size, it specifies these information as each element of "object size" and "object size accuracy". In addition, when it is necessary to transmit raw data such as images, videos, point cloud data, etc. output by the sensor unit 7, the in-vehicle device 1 stores the identification information given to the raw data in the “media ID”. specify. Detailed information on the medium (raw data) specified by the "media ID" element is stored separately in the "media information" item.

In the "valid flag", the on-vehicle device 1 includes a state variable to be estimated in estimating the vehicle's position. Specify each. For example, the in-vehicle device 1 calculates the diagonal elements (also referred to as "pre-estimation diagonal elements") of the covariance matrix P before estimating the own vehicle position using the target object. σ _x ² , σ _y ² , σ _z ² , σ _ψ ² , and the diagonal elements of the covariance matrix P after estimating the own vehicle position using the target object (also referred to as "post-estimation diagonal elements") σ _x ² , σ _y ² , σ _z ² , and σ _ψ ² are compared for each state variable. Then, the onboard device 1 sets the valid flag of the state variable whose post-estimation diagonal element is smaller than the pre-estimation diagonal element (that is, the accuracy has improved) to "1", and the post-estimation diagonal element is smaller than the pre-estimation diagonal element. The valid flag of a state variable with a large element (that is, accuracy has decreased) is set to "0".

Here, specific examples of setting the valid flag will be explained with reference to FIGS. 9 to 12.

FIG. 9 is a bird's-eye view showing the vicinity of a vehicle on which the on-vehicle device 1 is mounted. Further, FIG. 10 shows an example of the setting value of the valid flag specified by the on-vehicle device 1 in the upload information Iu indicating the detection results of each object with object IDs "1", "2", and "3". Further, FIG. 11 is a graph showing changes in the diagonal element σ _y ² during a predetermined period including execution periods “Tw1” to “Tw4” of vehicle position estimation based on the white line 51 of object ID “1”, FIG. 12 is a graph showing changes in the diagonal element σ _x ² during a predetermined period including the execution period “Tw5” of vehicle position estimation based on the sign 52 with object ID “2”.

First, the in-vehicle device 1 refers to the position information, size information, etc. of objects registered in the map DB 10, and identifies the white line 51 of the object ID "1" as a landmark, the sign 52 of the object ID "2", and the object ID. Prediction windows "Wp1" to "Wp3" that define the range in which each object is detected are set as shown in FIG. 9 for the region where the marker 53 "3" is estimated to exist.

Then, the vehicle-mounted device 1 detects the white line 51 with the object ID “1” within the prediction window Wp1, and estimates the own vehicle position based on the white line 51. In this case, as shown in FIG. 11, the diagonal element σ _y ² decreases at the end of each of the execution periods Tw1 to Tw4 of vehicle position estimation based on the white line 51 compared to the corresponding beginning. That is, after the own vehicle position is estimated based on the white line 51, the position estimation accuracy in the lateral direction (y) is improved. Therefore, in this case, the vehicle-mounted device 1 sets the validity flag in the lateral direction (y) to "1", as shown in FIG. On the other hand, if the other diagonal elements σ _x ² , σ _z ² , and σ _ψ ² do not decrease before and after the execution period Tw1 to Tw4, the on-vehicle device 1 determines the direction of travel, height, and azimuth. Set each valid flag to "0". Note that the onboard device 1 sets the valid flag for the target state variable to "1" when the decrease width of the post-estimation diagonal element from the pre-estimation diagonal element for the target state variable is greater than or equal to a predetermined threshold. However, when the above-mentioned reduction width is less than a predetermined threshold value, the validity flag for the target state variable may be set to "0". Then, the vehicle-mounted device 1 generates upload information Iu including the object ID of the white line 51 and the set validity flag, and transmits it to the server device 6.

Furthermore, the vehicle-mounted device 1 detects the sign 52 with the object ID “2” within the prediction window Wp2, and estimates the own vehicle position based on the sign 52. In this case, as shown in FIG. 12, at the end of the vehicle position estimation execution period Tw5 based on the sign 52, the diagonal element σ _x ² is decreased compared to the corresponding start period. That is, after estimating the vehicle's position based on the sign 52, the accuracy of estimating the position in the traveling direction (x) is improved. Therefore, in this case, the vehicle-mounted device 1 sets the valid flag for the traveling direction (x) to "1", as shown in FIG. Similarly, when the diagonal element σ _z ² and the diagonal element σ _ψ ² decrease before and after the execution period Tw5, the onboard device 1 sets the effective flags for the height direction and the direction to "1". do. On the other hand, if the diagonal element σ _y ² has not decreased before and after the execution period Tw5, the vehicle-mounted device 1 sets the horizontal direction validity flag to “0”. Then, the vehicle-mounted device 1 generates upload information Iu including the object ID of the sign 52 and the set validity flag, and transmits it to the server device 6.

On the other hand, the vehicle-mounted device 1 cannot detect the sign 53 with the object ID "3" within the prediction window Wp3 due to occlusion caused by the obstacles 54 and 55. In this case, the vehicle-mounted device 1 sets the validity flag of each state variable of the object ID "3" to "0". Then, the vehicle-mounted device 1 generates upload information Iu including the object ID of the sign 53 and the set validity flag, and transmits it to the server device 6.

(1-5) Calculation of object recommendation value
Next, a method for calculating object recommendation values for each object will be explained. The server device 6 calculates an object recommendation value for each object and for each state variable based on the validity flag included in the upload information Iu received from each on-vehicle device 1.

For example, the server device 6 sets the total number (number of samples) of valid flag information for each state variable of each object ID to "S1", and sets the total number of valid flag information (0 or 1) for each state variable of each object ID. is set to "S2", the object recommended value for each state variable of each object ID is set to "S2/S1". For example, if upload information Iu including a valid flag for the direction of travel of each object ID "1" is received 100 times from each on-vehicle device 1, and the sum of these valid flags is 70, the on-vehicle device 1 The object recommendation value for the traveling direction of "1" is set to "0.7" (=70/100). Object recommendation values in the horizontal direction, height direction, and orientation can be calculated in the same way.

By doing so, the server device 6 can suitably set the object recommendation value for each state variable of each object ID to be in the range from 0 to 1. In addition, in this case, the object recommendation value for objects that were not detected due to occlusion or the effects of rain or snowfall will be small, and even for objects that were detected, the proportion of objects that were judged to be effective for position estimation would be low. The recommended value will also be smaller. On the other hand, the object recommendation value of an object with a high proportion of objects determined to be effective for position estimation becomes high.

Preferably, the server device 6 calculates the object recommendation value using the validity flag of the upload information Iu received within a predetermined amount of time (for example, 10 minutes) in the past. Thereby, the server device 6 can accurately set or update the object recommendation value based on the latest collected valid flags.

Furthermore, the larger the traffic volume on the road where the target object can be detected, the more upload information Iu including the valid flag can be collected in a short time. It is a good idea to set this. For example, the server device 6 sets the above-mentioned predetermined time to 10 minutes for an object that can be detected on a road with heavy traffic, and sets the above-mentioned predetermined time to 10 minutes for an object that can be detected on a road with low traffic. It is preferable to set the predetermined time to one hour. In another example, the server device 6 may calculate the object recommendation value of the target object based on the validity flag included in the latest predetermined number of upload information Iu acquired in the past regarding the target object. In this case, since the value S1 is the predetermined number described above, the vehicle-mounted device 1 can suitably calculate the object recommendation value using a fixed number of samples.

(1-6) Processing flow
FIG. 13 is an example of a flowchart illustrating an overview of processing related to transmission and reception of upload information Iu including a validity flag and download information Id including object recommendation information.

First, the vehicle-mounted device 1 refers to the map DB 10, and if an object serving as a landmark exists, sets a prediction window for detecting the object (step S101). Then, the vehicle-mounted device 1 determines whether or not the target object has been detected (step S102). When the in-vehicle device 1 detects the target object (step S102; Yes), the in-vehicle device 1 determines for each state variable whether or not the accuracy of position estimation has improved before and after estimating the own vehicle position based on the object. A valid flag is set for each state variable (step S103). On the other hand, if the vehicle-mounted device 1 cannot detect the target object (step S102; No), it sets the validity flag of each state variable for the object to 0 (step S104). Then, the vehicle-mounted device 1 transmits upload information Iu including the object ID of the target object and the validity flag to the server device 6 (step S105).

The server device 6 receives the upload information Iu transmitted in step S105, and stores the upload information Iu in the upload information DB 27 (step S201). Then, the server device 6 determines whether it is time to update the distribution map DB 20 (step S202). The above update timing may be determined based on the length of time since the last update of the distribution map DB 20, or may be determined based on the cumulative number of received upload information Iu received since the last update of the distribution map DB 20. good.

Then, when it is time to update the distribution map DB 20 (step S202; Yes), the server device 6 generates object recommendation information etc. with reference to the upload information DB 27, and uses the generated object recommendation information etc. to update the distribution map DB 20. is updated (step S203). Then, the server device 6 transmits the download information Id including the object recommendation information generated in step S203 to each in-vehicle device 1 (step S204). Note that the server device 6 may transmit the download information Id only to the in-vehicle device 1 that has received a request to transmit the download information Id. On the other hand, if it is not time to update the distribution map DB 20 (step S202; No), the server device 6 continues to execute step S201.

When the in-vehicle device 1 receives the download information Id (step S106; Yes), it updates the map DB 10 using the download information Id (step S107). As a result, the latest object recommendation information is recorded in the map DB 10. On the other hand, if the in-vehicle device 1 has not received the download information Id from the server device 6 (step S106; No), the process returns to step S101.

<Second example>
In the driving support system according to the second embodiment, voxel data in which position information of stationary structures and the like are recorded for each unit area (also called "voxel") when a three-dimensional space is divided into a plurality of areas is stored in a map DB 10. and is recorded in the distribution map DB 20, and the in-vehicle device 1 uses the voxel data to estimate the own vehicle position. In this case, the map DB 10 and distribution map DB 20 include information (also referred to as "voxel recommendation information") in which a recommendation value for use in position estimation is indicated for each voxel, instead of object recommendation information. . Hereinafter, components similar to those of the driving support system of the first embodiment will be designated by the same reference numerals as appropriate, and their explanation will be omitted.

(2-1) Position estimation based on voxel data
The voxel data includes data representing measured point cloud data of stationary structures within each voxel using a normal distribution, and is used for scan matching using NDT (Normal Distributions Transform).

FIG. 14 shows an example of the own vehicle position estimation unit 17 in position estimation based on voxel data. The difference between the own vehicle position estimating unit 17 in position estimation based on the extended Kalman filter shown in FIG. A point cloud data matching block 27 is provided as a process for matching the data.

FIG. 15 shows an example of a schematic data structure of voxel data. Voxel data includes information on parameters when expressing a point group within a voxel with a normal distribution, and in this example, as shown in FIG. including. Here, "voxel coordinates" indicates the absolute three-dimensional coordinates of a reference position such as the center position of each voxel. Note that each voxel is a cube obtained by dividing the space into a grid pattern, and has a predetermined shape and size, so it is possible to specify the space of each voxel using voxel coordinates. Voxel coordinates may be used as voxel IDs. The voxel ID is an example of "object identification information."

"Mean vector" and "covariance matrix" indicate the mean vector and covariance matrix that correspond to the parameters when the point group within the target voxel is represented by a normal distribution, and The coordinates of point “i” in voxel ⁿ are defined as “X _n (i) = [x _n (i), y _n (i), z _n (i) _] '', the mean vector "μ _n " and covariance matrix "V _n " at voxel n are expressed by the following equations (5) and (6), respectively.

Scan matching by NDT assuming a vehicle uses estimated parameters P = [t _x , _ty , _tz , t ψ ^{] T} using the amount of movement in the road plane (here xy coordinates) and the direction of the vehicle _as elements. will be estimated. Here, "t _x " indicates the amount of movement in the x direction, " _ty " indicates the amount of movement in the y direction, " _tz " indicates the amount of movement in the z direction, and "t _ψ " indicates the amount of movement in the z direction. , indicating the yaw angle. Although the pitch angle and roll angle are caused by the road slope and vibration, they are small enough to be ignored.

Furthermore, the point cloud data obtained by the lidar 2 is associated with the voxel to be matched, and the coordinates of an arbitrary point at the corresponding voxel n are calculated as X _L (i) = [x _n (i) , y _n (i), z _n (i)] ^T , the average value “L _{′ n} ” of X _L (i) at voxel n is expressed by the following equation (7).

Then, when the average value L' is coordinate-transformed using the estimated parameter P described above, the coordinates "L _n " after the transformation are expressed by the following equation (8).

In this embodiment, the on-vehicle device 1 uses the coordinate-transformed point group, the average vector μ _n and the covariance matrix V _n included in the voxel data, and uses The evaluation function value “E _n ” and the comprehensive evaluation function value “E(k)” (also referred to as “comprehensive evaluation function value”) targeting all voxels to be matched as shown by formula (10). Calculate.

Hereinafter, the evaluation function value _En for each voxel will also be referred to as an "individual evaluation function value." Thereafter, the vehicle-mounted device 1 calculates the estimated parameter P that maximizes the comprehensive evaluation function value E(k) using an arbitrary root-finding algorithm such as Newton's method. Then, the on-vehicle device 1 applies the estimation parameter P to the own vehicle position X ⁻ (k) predicted by the position prediction unit 21 etc. shown in FIG. Estimate the accurate vehicle position X ^{^} (k).

The individual evaluation function value E _n is an example of a "verification result."

(2-2) Data structure
Next, voxel recommendation information commonly included in the map DB 10 and distribution map DB 20 will be explained.

FIG. 16 is an example of the data structure of voxel recommendation information. As shown in FIG. 16, the voxel recommendation information is information in which "voxel ID", "position", and "voxel recommended value" are associated with each other. A voxel ID assigned to each voxel is specified in “Voxel ID”. The voxel coordinates (latitude, longitude, and altitude, or xyz coordinates from a reference point) of the target voxel are specified in the "position". A voxel recommended value that is a recommended value for using a target voxel for position estimation is specified in the "voxel recommended value". Here, the recommended voxel value is set to a value between 0 and 1, and the value approaches 1 as the effectiveness in improving position estimation accuracy increases.

As described above, in this embodiment, the map DB 10 includes voxel recommendation information indicating the effectiveness (suitability) of position estimation for each voxel. Thereby, the vehicle-mounted device 1 can control the vehicle so as to improve the accuracy of estimating the own vehicle position by referring to the voxel recommendation information, as in the case of using the object recommendation information in the first embodiment. For example, if the accuracy of the current vehicle position estimation is low (that is, the comprehensive evaluation function value E(k) is low), the on-vehicle device 1 may move the vehicle to a lane where voxels with high recommended voxel values are easier to detect, In NDT matching, a voxel with a high voxel recommendation value is weighted high. Further, the vehicle-mounted device 1 may switch the position estimation method when there are no voxels with high recommended voxel values in the vicinity.

Note that even if the in-vehicle device 1 receives download information Id including voxel recommendation information regarding voxels around the vehicle's location from the server device 6 by transmitting a predetermined request signal including current location information to the server device 6, good. In this case, the on-vehicle device 1 controls the vehicle based on the received download information Id to improve the accuracy of estimating the own vehicle position.

Furthermore, in the second embodiment, the on-vehicle device 1 determines a value (also referred to as an "effective value") indicating the effectiveness of the target voxel in improving the accuracy of estimating the vehicle's position, instead of the effectiveness flag. Below, the vehicle-mounted device 1 determines the individual evaluation function value _En of the target voxel as the above-mentioned effective value. Then, the in-vehicle device 1 transmits upload information Iu, which includes at least information in which a valid value is associated with each voxel ID, to the server device 6 as the upload information Iu. The validity value is an example of "validity information."

Here, the individual evaluation function value E _n approaches 1 when the degree of matching is high, and approaches 0 when the degree of matching is low. It can be said that the larger the individual evaluation function value E _n of a voxel is, the more effective it is because it contributes to increasing the overall evaluation function value E(k). Therefore, the individual evaluation function value _En of each voxel is suitable as the effective value of each voxel.

Here, specific examples of setting valid values will be explained with reference to FIGS. 17 to 20.

FIG. 17 is a bird's-eye view showing the vicinity of a vehicle on which the on-vehicle device 1 is mounted. Here, voxels with voxel IDs "1" to "35" exist within a predetermined distance from the vehicle. Note that voxels with voxel IDs “1” to “35” are located on the surfaces of objects 56 to 58. FIG. 18 shows an example of setting valid values specified by the on-vehicle device 1 in the upload information Iu indicating the detection results of voxel IDs "1" to "35". Furthermore, FIG. 19 is a graph showing the transition of the individual evaluation function value _E4 for voxel ID "4" in a predetermined period including the detection period "Tw6" of voxel ID "4", and FIG. 12 is a graph showing the transition of the individual evaluation function value _E12 for the voxel ID "12" during a predetermined period including the detection period "Tw7" of "12".

First, the on-vehicle device 1 acquires voxel data of voxel IDs "1" to "35" existing within a predetermined distance from the vehicle from the map DB 10, and performs measurement by the lidar 2. As a result, the in-vehicle device 1 selects voxels with voxel IDs “4” to “11” located on the surface of the object 56 and voxels with voxel IDs “12” to “19” located on the surface of the object 57 to be the same or Detection is performed at different timings, and the own vehicle position is estimated by NDT matching. Then, the vehicle-mounted device 1 sets the individual evaluation function value E _n of each voxel calculated in the vehicle position estimation using NDT matching as the valid value of the corresponding voxel ID.

At this time, the vehicle-mounted device 1 preferably determines the effective value corresponding to each voxel ID to be the average value of the individual evaluation function values _En calculated during the period in which the voxel of each voxel ID was detected. For example, in the case of voxel ID "4", as shown in FIG. 19, in-vehicle device 1 sets the average value of individual evaluation function values _E4 calculated in detection period Tw6 as the effective value for voxel ID "4". do. Similarly, in the case of voxel ID " ₁₂ ", as shown in FIG. Set. Note that the above-mentioned detection period may be the entire time when the target voxel is detected, or only the time when the target voxel is detected within a predetermined distance range.

On the other hand, the on-vehicle device 1 measures voxels with voxel IDs "22" to "31" located on the surface of the object 58 corresponding to a building on which a curing sheet is pasted for exterior wall repair, due to the presence of the curing sheet described above. It cannot be detected because the value has shifted. Therefore, in this case, the vehicle-mounted device 1 sets the valid values corresponding to the voxel IDs "22" to "31" to zero. Then, the vehicle-mounted device 1 generates upload information Iu indicating the combination of the voxel ID and valid value shown in FIG. 18, for example, and transmits it to the server device 6.

(2-3) Calculation of recommended voxel values
Next, a method for calculating recommended voxel values for each voxel will be described. The server device 6 calculates a voxel recommended value for each voxel ID based on the valid value for each voxel ID included in the upload information Iu received from each on-vehicle device 1.

For example, if the total number (number of samples) of valid value information for each voxel ID is "S3" and the sum of valid values for each voxel ID is "S4", the server device 6 sets the voxel recommended value for each voxel ID to "S4". , set to "S4/S3". For example, if upload information Iu including the valid value of each voxel ID "1" is received 100 times from each on-vehicle device 1, and the sum of these valid values is 70.0, the on-vehicle device 1 receives the voxel ID "1". The recommended voxel value for "1" is set to "0.7" (=70.0/100).

By doing so, the server device 6 can suitably set the voxel recommended value of each voxel ID to be in the value range from 0 to 1. In addition, in this case, the recommended voxel values for voxels that were not detected due to occlusion, rain, snow, etc. were small, and even for voxels that were detected, the individual evaluation function value E _n was statistically low (in other words, the The recommended voxel value for voxels (which were not effective for estimation) will also be smaller. On the other hand, the recommended voxel value of a voxel whose individual evaluation function value _En is statistically high (in other words, effective for position estimation) becomes high.

Preferably, the server device 6 calculates the recommended voxel value using the effective value of the upload information Iu received for a predetermined amount of time in the past (for example, 10 minutes). More preferably, the server device 6 may set the predetermined time to be shorter as the traffic volume of the road on which the target voxel can be detected is greater. In another example, the server device 6 may calculate the recommended voxel value of the target voxel based on valid values included in the latest predetermined number of upload information Iu acquired in the past regarding the target voxel. In this case, since the value S3 is equal to the predetermined number described above, the vehicle-mounted device 1 can suitably calculate the recommended voxel value using a fixed number of samples.

(2-4) Processing flow
FIG. 21 is an example of a flowchart illustrating an overview of processing related to transmission and reception of upload information Iu including valid values and download information Id including voxel recommendation information.

First, the vehicle-mounted device 1 refers to the map DB 10 and acquires voxel data of voxels existing at positions detectable by the lidar 2 (step S111). Then, the on-vehicle device 1 sets a valid value for each voxel ID based on the presence or absence of detection by the lidar 2 of each voxel for which voxel data was acquired in step S111 and the individual evaluation function value _En for each voxel (step S112). . In this case, the on-vehicle device 1 sets the effective value for the voxel ID of the voxel for which the lidar 2 could not acquire point cloud data to 0, and sets the effective value for the voxel ID of the voxel for which the lidar 2 could acquire the point cloud data to the voxel concerned. is set to the individual evaluation function value E _n . Then, the vehicle-mounted device 1 transmits upload information Iu in which a valid value is associated with each voxel ID to the server device 6 (step S113).

The server device 6 receives the upload information Iu transmitted in step S113, and stores the upload information Iu in the upload information DB 27 (step S211). Then, when it is time to update the distribution map DB 20 (step S212; Yes), the server device 6 generates voxel recommendation information etc. with reference to the upload information DB 27, and uses the generated voxel recommendation information etc. to update the distribution map DB 20. is updated (step S213). Then, the server device 6 transmits the download information Id including the voxel recommendation information generated in step S213 to each vehicle-mounted device 1 (step S214). Note that the server device 6 may transmit the download information Id only to the in-vehicle device 1 that has received a request to transmit the download information Id. On the other hand, if it is not time to update the distribution map DB 20 (step S212; No), the server device 6 continues to execute step S211.

When the in-vehicle device 1 receives the download information Id (step S114; Yes), it updates the map DB 10 using the download information Id (step S115). As a result, the latest voxel recommendation information is recorded in the map DB 10. On the other hand, if the in-vehicle device 1 has not received the download information Id from the server device 6 (step S114; No), the process returns to step S111.

<Modified example>
Modifications suitable for the first and second embodiments will be described below. The following variations may be applied to these embodiments in combination.

(Modification 1)
Instead of the onboard device 1 determining the vehicle route (target trajectory) by referring to the map DB 10 that includes object recommendation information or voxel recommendation information, the server device 6 that has received the route search request from the onboard device 1 refers to the object recommendation information or voxel recommendation information. Processing may be performed to determine the route (target trajectory) of the vehicle of the requesting vehicle-mounted device 1 by referring to the distribution map DB 20 including voxel recommended information. In this way, the object recommendation information and voxel recommendation information are suitably used by the server device 6 as well.

(Modification 2)
The configuration of the driving support system shown in FIG. 1 is an example, and the configuration of the driving support system to which the present invention is applicable is not limited to the configuration shown in FIG. 1. For example, instead of having the on-vehicle device 1 in the driving support system, the electronic control device of the vehicle may execute the processing of the vehicle position estimating section 17 and the upload control section 18 of the on-vehicle device 1. In this case, the map DB 10 is stored, for example, in a storage section in the vehicle, and the electronic control device of the vehicle sends and receives upload information Iu and download information Id to and from the server device 6 via the on-vehicle device 1 or through communication (not shown). It may also be done through the department.

1 On-vehicle device 2 Lidar 3 Gyro sensor 4 Vehicle speed sensor 5 GPS receiver 6 Server device 10 Map DB
20 Distribution map DB

Claims (1)

Recommended values for estimating the position of the moving object using the object are generated based on identification information of the object and effectiveness information regarding the effectiveness of improving accuracy of estimating the position of the moving object using the object. A map data generation device including a generation unit that generates map data by associating recommended value information with position information of the object.