JP2023164502A - Stationary object data generator, method for control, program, and storage medium - Google Patents

Abstract

To provide a stationary object data generator for generating stationary object data on a stationary object desirably used for detecting a movable object.SOLUTION: A server device 2 executes processing of: detecting a vegetation voxel Bv including a measurement point of vegetation from voxels including a measurement point related to an object measured by such a measurement device as a lidar 30; and generating a stationary obstacle flag showing a region including a stationary obstacle to a voxel from which the vegetation voxel Bv is removed, from the voxel including the measurement point.SELECTED DRAWING: Figure 9

Description

The present invention relates to data used for detecting moving objects.

Conventionally, technology has been known to estimate the vehicle's own position by matching the shape data of surrounding objects measured using a measuring device such as a laser scanner with map data in which the shapes of surrounding objects are stored in advance. There is. For example, in Patent Document 1, it is determined whether a detected object is a stationary object or a moving object in voxels where space is divided according to a predetermined rule, and matching between map data and measurement data is performed for voxels where stationary objects exist. An autonomous mobile system is disclosed. Further, Patent Document 1 also discloses that map data indicating a voxel in which a stationary object is set to exist among each voxel is stored. Further, Non-Patent Document 1 discloses a technique for recognizing a measurement point group that constitutes vegetation such as trees by analyzing a measurement point group obtained from a lidar or the like.

International publication WO2013/076829

Fabrice Monnier, Bruno Vallet, Bahman Soheilian, TREES DETECTION FROM LASER POINT CLOUDS ACQUIRED IN DENSE URBAN AREAS BY A MOBILE MAPPING SYSTEM, [online], 2012, ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, [July 2018] [Retrieved on April 14], Internet〈URL: https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/I-3/245/2012/isprsannals-I-3-245- 2012.pdf〉

When detecting moving objects, areas where stationary objects are set on the map data are excluded from the moving object detection targets to reduce the processing load and improve the detection accuracy of moving objects. It is possible to improve this. On the other hand, since vegetation is not a moving object, it is treated as a stationary object in the map data of Patent Document 1. However, since voxels containing vegetation are areas where dynamic objects can exist, there is a possibility that dynamic objects to be detected may be overlooked.

The present invention has been made to solve the above problems, and it is an object of the present invention to provide a stationary object data generation device that generates stationary object data regarding stationary objects that can be suitably used in detecting moving objects. Main purpose.

In the claimed invention, the stationary object data generation device includes a detection means for detecting an area including a measurement point of vegetation from a partitioned area of space including a measurement point related to an object measured by the measurement device. and generating means for generating stationary object data indicating that a stationary object is present in a region obtained by excluding at least a region including only the measurement point of the vegetation from a region including the measurement point of the object; has.

Further, the claimed invention is a control method executed by a stationary object data generation device, which measures vegetation from a region partitioned into space that includes measurement points related to objects measured by a measuring device. a detection step of detecting an area including a point, and a stationary area indicating that a stationary object is present in an area obtained by excluding at least an area including only the measurement point of the vegetation from an area including the measurement point of the object; and a generation step of generating object data.

Further, the claimed invention is a program executed by a computer, which divides a space into a region including measurement points of vegetation from a region including measurement points related to an object measured by a measuring device. a detection means for detecting, and generating stationary object data indicating that a stationary object is present in an area excluding at least an area including only the measurement point of the vegetation from an area including the measurement point of the object; The computer is made to function as a generating means.

This is a schematic configuration of the map update system. The block configuration of the on-vehicle device and server device is shown. An example of a schematic data structure of voxel data is shown. 3 is a flowchart showing the procedure of voxel data generation processing. It is a flowchart which shows the procedure of vegetation determination processing. It is a flowchart which shows the procedure of position estimation processing. FIG. 3 is a diagram illustrating an overview of processing of position estimation based on the first example and position estimation based on a comparative example. An example of a schematic data structure of voxel data according to a second example is shown. 7 is a flowchart showing the procedure of voxel data generation processing according to the second embodiment. 12 is a flowchart showing the procedure of dynamic object detection processing according to the second embodiment.

According to a preferred embodiment of the present invention, the stationary object data generation device detects an area including a measurement point of vegetation from a partitioned area of space that includes a measurement point related to an object measured by the measurement device. a detection means for generating stationary object data indicating that a stationary object is present in a region obtained by excluding at least a region including only the measurement point of the vegetation from a region including the measurement point of the object; and means. A "region in which space is divided" is a region in which space is divided according to a predetermined rule, and is, for example, a rectangular parallelepiped or a cube having a constant size. In this aspect, the stationary object data generation device excludes a region of vegetation whose position changes over time from targets for generating stationary object data indicating that the stationary object is an area where a stationary object exists. Thereby, the stationary object data generation device can suitably generate stationary object data that does not cause a moving object to be overlooked when used for detecting a moving object.

In one aspect of the stationary object data generation device, the detection means includes a first determination means for determining a first region in which the measurement point forms a plane based on the calculation result of the calculation means; a second determination means for determining a second region, which is the region in which the measurement point forms a columnar body, based on the calculation result of the calculation means; and third determining means for determining an area that does not fall under any of the second areas as an area that includes the vegetation measurement point. With this aspect, the stationary object data generation device can accurately detect an area including a vegetation measurement point.

In another aspect of the stationary object data generation device, the generation means divides a region including measurement points of the object into a region including measurement points of a dynamic object and a region including only measurement points of the vegetation; Stationary object data indicating that the stationary object is present is generated for the area where at least the stationary object is excluded. With this aspect, the stationary object data generation device can suitably generate stationary object data.

In another aspect of the stationary object data generation device, the generation means detects that a stationary object exists in a region that includes at least a region that includes a measurement point of a moving object from a region that includes a measurement point of the object. Generate stationary object data indicating the area. With this aspect, the stationary object data generation device can more reliably exclude areas where a moving object may appear from the targets for which stationary object data indicating areas where stationary objects exist are generated. .

In another aspect of the stationary object data generation device, the stationary object data generation device further includes a transmitting unit that transmits the stationary object data to an information processing device that detects a moving object using a measurement unit. According to this aspect, the stationary object data generation device can supply stationary object data suitable for detecting a moving object to an information processing device (for example, a vehicle that performs automatic driving).

According to another embodiment of the present invention, there is provided a control method executed by a stationary object data generation device, the control method being a control method executed by a stationary object data generation device, in which a vegetation a detection step of detecting an area including the measurement point; and indicating that the area including only the measurement point of the vegetation is excluded from the area including the measurement point of the object as an area where a stationary object exists; and a generation step of generating stationary object data. By executing this control method, the stationary object data generation device can suitably generate stationary object data that will not cause a moving object to be overlooked when used for detecting a moving object.

According to another embodiment of the present invention, there is provided a program executed by a computer, which is a region in which a space is divided from a region including a measurement point regarding an object measured by a measuring device, and a region including a measurement point of vegetation. and generating stationary object data indicating that a stationary object is present in a region obtained by excluding at least a region including only the measurement point of the vegetation from a region including the measurement point of the object. The computer is made to function as a generating means for generating the information. By executing this program, the stationary object data generation device can suitably generate stationary object data that will not cause a moving object to be overlooked when used for detecting a moving object. Preferably, the program is stored in a storage medium.

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

<First example>
The first embodiment relates to position estimation based on voxel data.

(1) Overview of map update system
FIG. 1 shows a schematic configuration of a map update system according to a first embodiment. The map update system includes an on-vehicle device 1 that moves with a vehicle, and a server device 2 that distributes map information. Although FIG. 1 shows only one set of the on-vehicle device 1 and vehicle that communicate with the server device 2, in reality, a plurality of sets of the on-vehicle device 1 and the vehicle exist at different positions.

The in-vehicle device 1 is electrically connected to external sensors such as lidar (Light Detection and Ranging or Laser Illuminated Detection and Ranging), internal sensors such as a gyro sensor and a vehicle speed sensor, and based on the outputs of these sensors, The position of the vehicle in which the on-vehicle device 1 is mounted (also referred to as "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 including voxel data. Voxel data is data in which position information of a stationary structure, etc. is recorded for each region (also referred to as "voxel") when a three-dimensional space is divided into a plurality of regions. The voxel data includes data representing point cloud data of measured measurement points of a stationary structure within each voxel using a normal distribution, and is used for scan matching using NDT (Normal Distributions Transform), as described later. .

The on-vehicle device 1 performs scan matching based on NDT based on point cloud data output by the lidar, which is obtained by converting measurement points on the surface of an object into an absolute coordinate system, and voxel data corresponding to voxels to which the point cloud data belongs. This allows the vehicle's position to be estimated. Furthermore, the on-vehicle device 1 transmits measurement data “D1” including the above-mentioned point cloud data to the server device 2. Furthermore, the vehicle-mounted device 1 updates the map DB 10 by receiving update data “D2” regarding the map DB 10 from the server device 2 . The on-vehicle device 1 is an example of an "information processing device" and a "position estimating device."

The server device 2 performs data communication with the on-vehicle device 1 corresponding to a plurality of vehicles. The server device 2 stores a distribution map DB 23 for distribution to the on-vehicle device 1 corresponding to a plurality of vehicles, and the distribution map DB 23 includes voxel data corresponding to each voxel. Further, the server device 2 stores a measurement point group DB 24 in which measurement data D1 received from the on-vehicle device 1 is accumulated. Then, the server device 2 generates voxel data based on the measurement data D1 accumulated in the measurement point group DB 24, and updates the distribution map DB 23 based on the generated voxel data. Further, the server device 2 transmits update data D2 including the generated voxel data to the vehicle-mounted device 1. The server device 2 is an example of a "storage device" and a "map generation device."

(2) Configuration of onboard equipment
FIG. 2(A) shows a block diagram showing the functional configuration of the vehicle-mounted device 1. As shown in FIG. 2(A), the on-vehicle device 1 mainly includes a communication section 11, a storage section 12, a sensor section 13, an input section 14, a control section 15, and an output section 16. The communication section 11, the storage section 12, the sensor section 13, the input section 14, the control section 15, and the output section 16 are interconnected via a bus line.

The communication unit 11 transmits measurement data D1 generated by the control unit 15 to the server device 2 and receives update data D2 distributed from the server device 2 under the control of the control unit 15. The communication unit 11 also transmits signals for controlling the vehicle to the vehicle, and receives signals regarding the state of the vehicle from the vehicle.

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 including voxel data.

The sensor section 13 includes a lidar 30, a camera 31, a GPS receiver 32, a gyro sensor 33, and a speed sensor 34. The lidar 30 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 indicating the position of the object. Generate group data. In this case, the lidar 30 includes an irradiation section that irradiates laser light while changing the irradiation direction, a light reception section that receives reflected light (scattered light) of the irradiated laser light, and scan data based on the light reception signal output by the light reception section. and an output section that outputs. The scan data is generated based on the irradiation direction corresponding to the laser light received by the light receiving section and the response delay time of the laser light specified based on the above-mentioned light reception signal. The rider 30 is an example of a "measuring device".

The input unit 14 is a button, a touch panel, a remote controller, a voice input device, etc. for the user to operate, and accepts inputs such as specifying a destination for route searching, inputs specifying on and off of automatic driving, etc. , supplies the generated input signal to the control section 15. The output unit 16 is, for example, a display, a speaker, or the like 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. For example, the control unit 15 may perform scan matching based on NDT based on point cloud data obtained by converting measurement points output by the lidar 30 into an absolute coordinate system and voxel data corresponding to voxels to which the point cloud data belongs. Then, the vehicle's position is estimated. Further, the control unit 15 updates the map DB 10 based on the update data D2 that the communication unit 11 receives from the server device 2.

Further, the control unit 15 transmits measurement data D1 generated based on the point cloud data output from the lidar 30 to the server device 2 through the communication unit 11. In this case, the control unit 15 converts, for example, point cloud data of measurement points output from the rider 30 into an absolute coordinate system based on the estimated own vehicle position and information on the position and orientation of the rider 30 with respect to the vehicle, and The subsequent point cloud data is included in the measurement data D1 and transmitted to the server device 2 by the communication unit 11. In another example, the control unit 15 outputs point cloud data (so-called raw data) output from the lidar 30 and data necessary to convert the point cloud data into an absolute coordinate system (such as the above-mentioned own vehicle position). is included in the measurement data D1 and transmitted to the server device 2 by the communication unit 11. In the latter example, the server device 2 generates point cloud data of measurement points expressed in an absolute coordinate system based on the measurement data D1.

(3) Configuration of server device
FIG. 2(B) shows a schematic configuration of the server device 2. As shown in FIG. As shown in FIG. 2(B), the server device 2 includes a communication section 21, a storage section 22, and a control section 25. The communication section 21, the storage section 22, and the control section 25 are interconnected via a bus line.

The communication unit 21 communicates various data with the vehicle-mounted device 1 under the control of the control unit 25 . The storage unit 22 stores programs for controlling the operation of the server device 2 and holds information necessary for the operation of the server device 2. The storage unit 22 also stores a distribution map DB 23 and a measurement point cloud DB 24 that records point cloud data of measurement points of objects based on the measurement data D1 transmitted from the plurality of in-vehicle devices 1.

The control unit 25 includes a CPU, ROM, RAM, etc. (not shown), and performs various controls on each component within the server device 2. In this embodiment, the control unit 25 stores the measurement data D1 that the communication unit 21 receives from the on-vehicle device 1 in the measurement point group DB 24, and generates voxel data based on the stored measurement data D1. In this case, the control unit 25 determines for each voxel whether or not the voxel includes vegetation, and based on the determination result, generates a weighting value and a vegetation flag, which will be described later, and includes them in the voxel data. Further, the control unit 25 transmits update data D2 based on the generated voxel data to the vehicle-mounted device 1 through the communication unit 21. The control unit 25 is an example of a "computer" that executes a program.

(4) Scan matching based on NDT
Next, scan matching based on NDT in this embodiment will be explained.

First, voxel data used for scan matching based on NDT will be explained. FIG. 3 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. , a weighting value, and a vegetation flag. 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.

"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 (1) and (2), respectively.

The "weighting value" is set to a value according to the reliability of voxel data (in particular, the mean vector and covariance matrix) of the target voxel, and represents the weight set for the target voxel in scan matching. Further, in this embodiment, the weighting value for a voxel determined to be a voxel containing vegetation (also referred to as a "vegetation voxel Bv") is set lower than the weighting value for other voxels. The "vegetation flag" is flag information indicating whether or not the target voxel includes vegetation.

Next, scan matching by NDT using voxel data will be explained. In this embodiment, as will be described later, the vehicle-mounted device 1 weights and calculates the value of the evaluation function (evaluation value) obtained by NDT scan matching using the weighting value included in the voxel data. Thereby, the vehicle-mounted device 1 suitably improves the position estimation accuracy based on NDT scan matching.

Scan matching by NDT assuming a vehicle uses an estimated parameter 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. , indicates the yaw angle (the amount of change in angle in the yaw direction). Although the pitch angle and roll angle are caused by the road slope and vibration, they are small enough to be ignored.

When the point cloud data obtained by converting the measurement points obtained by lidar 2 into an absolute coordinate system is associated with the voxel to be matched, the coordinates of an arbitrary point at the corresponding voxel n are expressed as X _L Let (i)=[x _n (i), y _n (i), z _n (i)] ^T. Then, when the coordinates of X _L (i) = [x _n (i), y _n (i), z _n (i)] ^T are transformed using the estimated parameter P described above, the coordinates after the transformation are "X' _n ". is expressed by the following equation (3).

そして、本実施例では、車載機１は、座標変換した点群と、ボクセルデータに含まれる平均ベクトルμ_ｎと共分散行列Ｖ_ｎとを用い、以下の式（４）により示されるボクセルｎの評価関数値「Ｅ_ｎ」を算出する。

In this embodiment, the on-vehicle device 1 uses the coordinate-transformed point group, the mean vector μ _n and the covariance matrix V _n included in the voxel data, and calculates the value of the voxel n expressed by the following equation (4). An evaluation function value “E _n ” is calculated.

ここで、「ｗ_ｎ」は、ボクセルｎに対する重み付け値を示す。式（４）により、重み付け値ｗ_ｎが大きいほど、評価関数Ｅ_ｎは大きい値となる。

Here, “w _n ” indicates a weighting value for voxel n. According to equation (4), the larger the weighting value w _n is, the larger the evaluation function E _n becomes.

In addition, after calculating the evaluation function value for each voxel (also referred to as "individual evaluation function value") shown by equation (4), the in-vehicle device 1 calculates all voxels to be matched shown by equation (5). A comprehensive evaluation function value "E" (also referred to as "comprehensive evaluation function value") is calculated.

「Ｍ」は、マッチングの対象となるボクセルの数を示す。総合評価関数値Ｅは「評価値」の一例である。

"M" indicates the number of voxels to be matched. The comprehensive evaluation function value E is an example of an "evaluation value."

Thereafter, the vehicle-mounted device 1 calculates the estimated parameter P that maximizes the comprehensive evaluation function value E using an arbitrary root-finding algorithm such as Newton's method. Then, the onboard device 1 applies the estimated parameter P to the predicted own vehicle position (also referred to as the "predicted own vehicle position") "X ^- " based on the output of the autonomous positioning device, etc. The own vehicle position (also referred to as the "estimated own vehicle position") "X ^{^} " is estimated with higher accuracy than the vehicle position.

In this way, the vehicle-mounted device 1 multiplies each voxel data (average vector, covariance matrix) by a weighting value depending on whether the voxel is a vegetation voxel Bv or not. As a result, the evaluation function E _n of the vegetation voxel Bv becomes relatively small, and the accuracy of position estimation by NDT matching is suitably improved.

Note that the weighting value for the vegetation voxel Bv may be 0. In this case, since the estimated parameter P is estimated based on the evaluation function E _n of voxels other than the vegetation voxel Bv, it is possible to perform position estimation that completely eliminates the influence of vegetation.

(5) Generation of voxel data
Next, a description will be given of the vegetation voxel Bv determination process that the server device 2 executes when generating voxel data.

FIG. 4 is a flowchart regarding the voxel data generation process executed by the server device 2. The server device 2 executes the process of the flowchart shown in FIG. 4 at a predetermined timing.

First, the server device 2 recognizes the voxel to which each measurement point recorded in the measurement point group DB 24 belongs, and calculates the average vector and covariance of the measurement point for each voxel based on the above-mentioned formulas (1) and (2). A matrix is calculated (step S101).

Next, the server device 2 executes a vegetation determination process for determining whether each voxel including the measurement point is a vegetation voxel Bv (step S102). A specific example of this vegetation determination process will be described later with reference to the flowchart of FIG.

Then, the server device 2 sets a weighting value and a vegetation flag for each voxel based on the determination result in step S102 (step S103). For example, the server device 2 stores in advance setting information indicating weighting values and vegetation flag values to be assigned to the vegetation voxel Bv and other voxels, respectively. Then, the server device 2 refers to the above-mentioned setting information and assigns different weighting values and vegetation flags to the voxel determined to be the vegetation voxel Bv in step S103 and the voxel other than the vegetation voxel Bv. In this case, the weighting value for the vegetation voxel Bv is set lower than the weighting value for other voxels.

Note that the server device 2 may perform a process of excluding a measurement point group indicating a dynamic object from the measurement point group recorded in the measurement point group DB 24 as a preprocessing of the process in the flowchart of FIG. For example, the server device 2 identifies a measurement point group forming a specific dynamic object such as a pedestrian or vehicle based on processing such as pattern matching based on shape and size, and measures the identified measurement point group. Delete from the point cloud DB 24.

FIG. 5 is a flowchart showing an example of the vegetation determination process executed in step S102 of FIG. In the example shown in FIG. 5, the server device 2 determines voxels representing a plane or columnar structure by performing principal component analysis for each voxel on the measurement point group recorded in the measurement point group DB 24, Voxels that do not correspond to any planar or columnar structures are determined as vegetation voxels Bv.

First, the server device 2 performs principal component analysis for each voxel on the measurement point group recorded in the measurement point group D24 (step S201). Specifically, the server device 2 calculates three sets of eigenvalues and eigenvectors corresponding to the first to third principal components from the covariance matrix for each voxel calculated in step S101 of FIG. Here, the eigenvalue indicates the variance (that is, the magnitude) of each principal component, and the eigenvector indicates the direction of each principal component. Therefore, in the server device 2, the largest eigenvalue and its corresponding eigenvector correspond to the first principal component, the second largest eigenvalue and its corresponding eigenvector correspond to the second principal component, and the remaining eigenvalues and eigenvectors correspond to the first principal component. are assumed to correspond to the third principal component, respectively. Note that the first to third principal components are orthogonal to each other.

Then, the server device 2 extracts voxels forming a plane (also referred to as "plane voxels Bf") based on the result of the principal component analysis in step S201 (step S202). For example, the server device 2 regards a voxel in which the size of the third principal component (that is, the variance indicated by the eigenvalue of the third principal component) is less than or equal to a predetermined value as a planar voxel Bf. In addition, the server device 2 may consider a voxel in which the direction indicated by the eigenvector corresponding to the third principal component substantially coincides with the vertical direction or the horizontal direction to be a planar voxel Bf.

Further, the server device 2 extracts voxels forming a column (also referred to as "column voxel Bp") based on the result of the principal component analysis in step S201 (step S203). For example, the server device 2 extracts a voxel in which the contribution rate of the first principal component is equal to or greater than a predetermined value and the variance of the first principal component is close to a uniform distribution as a columnar voxel Bp. In addition, the server device 2 may consider a voxel whose direction indicated by the eigenvector corresponding to the first principal component substantially coincides with the vertical direction or the horizontal direction as a columnar voxel Bp.

Then, the server device 2 determines the remaining voxels that do not correspond to either the planar voxel Bf or the columnar voxel Bp among the voxels including the measurement points as vegetation voxels Bv (step S204). Here, most of the static structures that are not vegetation are composed of a planar shape or a columnar shape, and the applicant has also found in experiments etc. that a group of measurement points forming a planar shape or a columnar shape By regarding the voxels excluding the voxels as the vegetation voxels Bv, good recognition results for the vegetation voxels Bv have been obtained.

(6) Position estimation using voxel data
Next, position estimation processing using voxel data will be explained. FIG. 6 is an example of a flowchart showing the procedure of position estimation processing using voxel data.

First, the vehicle-mounted device 1 sets an initial value of the own vehicle position based on the output of the GPS receiver 32 or the like (step S301). Next, the vehicle-mounted device 1 acquires the vehicle body speed from the speed sensor 34 and the angular velocity in the yaw direction from the gyro sensor 33 (step S302). Then, the vehicle-mounted device 1 calculates the travel distance of the vehicle and the change in the direction of the vehicle based on the obtained result in step S302 (step S303).

After that, the on-vehicle device 1 adds the travel distance and azimuth change calculated in step S303 to the estimated own vehicle position one time ago, and calculates the predicted own vehicle position (step S304). Then, the vehicle-mounted device 1 refers to the map DB 10 and obtains voxel data of voxels existing around the vehicle position based on the predicted vehicle position calculated in step S304 (step S305). Further, the on-vehicle device 1 divides the point cloud data obtained by converting the measurement points obtained from the rider 30 into an absolute coordinate system into voxel units based on the predicted vehicle position calculated in step S304 (step S306).

Then, the vehicle-mounted device 1 calculates NDT scan matching using the average, covariance matrix, and weighting value included in the voxel data acquired in step S305 (step S307). Specifically, the in-vehicle device 1 uses the average, covariance matrix, and weighting value included in the corresponding voxel data for each voxel to which a measurement point is assigned in step S306, and calculates the value shown in equation (4). The individual evaluation function value E _n is calculated. Then, the vehicle-mounted device 1 calculates the comprehensive evaluation function value E shown in equation (5) based on the individual evaluation function value E _n for each voxel. In this case, the comprehensive evaluation function value E becomes a nonlinear equation including variables of each element of the estimated parameter P.

Then, the vehicle-mounted device 1 determines the estimated parameter P that gives the maximum overall evaluation function value, using a numerical solution method of nonlinear equations such as Newton's method (step S308). After that, the vehicle-mounted device 1 calculates the estimated own vehicle position by applying the estimated parameter P calculated in step S308 to the predicted own vehicle position calculated in step S304 (step S309).

Here, the weighting value used for calculating the individual evaluation function value _En in step S307 is set lower for the vegetation voxel Bv than for other voxels. Thereby, the vehicle-mounted device 1 relatively reduces the degree of influence of the vegetation voxel Bv representing vegetation on position estimation, and suitably improves position estimation accuracy.

Figure 7 shows position estimation based on the flowchart in Figure 6 using weighted values (also referred to as "regular position estimation"), and generation based on a measurement point group from which vegetation measurement points are removed instead of using weighted values. FIG. 2 is a diagram showing an overview of processing for position estimation with reference to voxel data (also referred to as "position estimation based on a comparative example"). Here, consideration will be given to NDT scan matching targeting voxel B1, which includes a portion of vegetation and a structure, respectively, and voxel B2, which is adjacent to voxel B1, includes a portion of the structure, and does not include vegetation.

Voxel data used in this position estimation is schematically shown in voxels B1 and B2 indicated by an arrow A1 labeled "weighting." Here, the hatched area 50 indicates the distribution of the measurement point group of vegetation used to generate the voxel data, and the hatched areas 51 and 52 indicate the distribution of the measurement point group of the structure used to generate the voxel data. show. Furthermore, a broken line frame 60 indicates the variance of a point group within voxel B1 indicated by a covariance matrix recorded as voxel data, and a broken line frame 61 indicates a point group within voxel B2 indicated by a covariance matrix recorded as voxel data. Shows the group variance. Furthermore, voxel data used in position estimation based on the comparative example is schematically shown in voxels B1 and B2 indicated by an arrow A2 labeled "removal of point group." Here, the hatched areas 54 and 55 indicate the distribution of the measurement point group of the structure used to generate the voxel data. Furthermore, a broken line frame 63 indicates the variance of a point group within voxel B1 indicated by a covariance matrix recorded as voxel data, and a broken line frame 64 indicates a point group within voxel B2 indicated by a covariance matrix recorded as voxel data. Shows the group variance.

Here, in the position estimation based on the comparative example, since voxel data is generated after removing the measurement point group representing vegetation, the broken line frame 63 representing the dispersion of the point group in voxel B1 is indicated by the hatched area 54. It exists at a position centered on the structure. On the other hand, in this position estimation, the dashed line frame 60 representing the dispersion of the point group within the voxel B1 exists across the vegetation indicated by the hatched region 50 and the structure indicated by the hatched region 51. Furthermore, in position estimation based on the comparative example, voxel B1 and voxel B2 are treated with the same weight. On the other hand, in this position estimation, voxel B1 is regarded as a vegetation voxel Bv, and the weighting value of voxel B1 is set to a smaller value than the weighting value of voxel B2, which is not a vegetation voxel Bv.

Here, when the on-vehicle device 1 acquires the measurement point group 40 represented by x marks in voxels B1 and B2 by the lidar 30, in this position estimation, the weighting value of voxel B1 is lower than the weighting value of voxel B2. Therefore, the voxel data in voxel B2 and the measurement point group 40 are matched preferentially. On the other hand, in the position estimation based on the comparative example, since no weighting value is set for each voxel, matching between the voxel data and the measurement point group 40 is performed by uniformly weighting the entire voxel data (here, voxels B1 and B2).

The matching result obtained when this position estimation is executed is indicated by an arrow A3, and the matching result obtained when the position estimation based on the comparative example is executed is indicated by an arrow A4. Here, in the position estimation based on the comparative example, since the measurement point group of the vegetation in voxel B1 is excluded when generating voxel data, the measurement point group 40 in voxel B1 that includes vegetation as a measurement target is excluded. Matching with voxel data becomes inaccurate. Further, in the position estimation based on the comparative example, since weighting is not performed, the matching accuracy as a whole decreases due to the influence of mismatch between the voxel data at voxel B1 and the measurement point group 40.

On the other hand, in this position estimation, since the measurement point group of the vegetation in voxel B1 is not excluded when generating voxel data, the measurement point group 40 in voxel B1 that includes vegetation as a measurement target and the voxel data are Matching is preferably performed. Note that since the vegetation in voxel B1 sways due to the influence of wind, etc., the matching accuracy in voxel B1 is lower than the matching accuracy in voxel B2. Therefore, if NDT matching is performed without weighting each voxel, the matching accuracy as a whole will decrease due to the influence of the matching accuracy of voxel B1. On the other hand, in this position estimation, since the weighting value of voxel B1 is set relatively low, matching can be performed while suitably reducing the influence of the swaying movement of vegetation.

As explained above, the data structure of the map data according to this embodiment includes, for each voxel, the average vector, covariance matrix, etc. related to the position of the object surface, and the weighting value related to the weight when using these for position estimation. The weighting value for a voxel Bv that contains vegetation is set to a lower value than the weighting value for a voxel that does not contain vegetation. The map data having this data structure is created by weighting the measurement information of the object measured by the measuring device such as the lidar 30 mounted on the vehicle and the voxel data for each voxel. It is suitably referred to by the vehicle-mounted device 1 that estimates the position.

Further, the server device 2 according to the present embodiment performs a process of recognizing vegetation voxels Bv including measurement points of vegetation from voxels including measurement points of objects measured by a measurement device such as the lidar 30, and an average vector for each voxel. and a process of generating map data in which covariance matrices and the like are associated with weighting values when these are used for position estimation. Here, the server device 2 makes the weighting value for the vegetation voxel Bv smaller than the weighting value for the voxel that does not include vegetation.

In addition, the on-vehicle device 1 according to the present embodiment stores voxel data including an average vector, covariance matrix, etc. for each voxel that divides space, and weighting values for determining weights when using these for position estimation. By performing a process of storing and acquiring measurement information regarding an object measured by a measurement device such as the lidar 30, comparing the measurement information and voxel data for each voxel, and weighting the results of the comparison based on a weighting value. , and a process of estimating the vehicle's position. In this case, the weighting value of the vegetation voxel Bv is smaller than the weighting value of a voxel that does not include vegetation.

<Second example>
The second embodiment differs from the first embodiment in that the vehicle-mounted device 1 detects a moving object based on voxel data instead of or in addition to position estimation based on voxel data. The configurations of the map update system, in-vehicle device 1, and server device 2 in the second embodiment are the same as those shown in FIGS. 1 and 2, and therefore their descriptions will be omitted.

FIG. 8 shows an example of a schematic data structure of voxel data in the second embodiment.

The voxel data shown in FIG. 8 includes a voxel ID, voxel coordinates, a stationary obstacle flag, a vegetation flag, an average vector, a covariance matrix, and a weighting value. Here, the stationary obstacle flag indicates that the target voxel is a voxel where stationary obstacles mainly exist and there is no need to detect a moving object (also called a "stationary obstacle voxel"). This is a flag indicating whether or not there is. For stationary obstacle voxels, the entire or most part of the target voxel is composed of stationary obstacles, so there is no need to detect moving objects that should be detected during driving (there is no room for moving objects to be included). ) indicates the area. Therefore, when detecting a moving object based on the output of the lidar 30, the onboard device 1 identifies the stationary obstacle voxel by referring to the stationary obstacle flag, and targets voxels other than the stationary obstacle voxel. Detection of moving objects is performed as follows. By limiting the area to be detected for moving objects in this manner, the load of processing for detecting moving objects is suitably reduced, and the accuracy of detecting moving objects is improved. The stationary obstacle flag is an example of "stationary object data."

Furthermore, in the present embodiment, when generating a stationary obstacle flag, the server device 2 does not consider at least a voxel determined to be a vegetation voxel Bv to be a stationary obstacle voxel. This ensures that dynamic objects that appear near vegetation are detected. The server device 2 is an example of a "stationary object data generation device."

FIG. 9 is a flowchart showing the procedure of voxel data generation processing according to the second embodiment. The server device 2 executes the process of the flowchart shown in FIG. 9 at a predetermined timing.

First, the server device 2 refers to the measurement point group DB 24 in which the measurement data D1 transmitted from each vehicle-mounted device 1 is recorded, and calculates the average vector and covariance matrix of the measurement points for each voxel (step S401). Next, the server device 2 executes a vegetation determination process for determining whether each voxel including the measurement point is a vegetation voxel Bv (step S402). This vegetation determination process is executed, for example, based on the flowchart of FIG. 5 described above.

Next, the server device 2 generates a stationary obstacle flag based on the vegetation determination result in step S402 (step S403). At this time, the server device 2 generates a stationary obstacle flag indicating that the voxel determined to be a vegetation voxel Bv is not a stationary obstacle voxel. Thereby, the server device 2 suitably suppresses overlooking of a moving object due to regarding a region where a moving object would originally exist as a stationary obstacle voxel.

Note that the server device 2 may execute a process of excluding a measurement point group indicating a moving object from the measurement point group recorded in the measurement point group DB 24 as a preprocessing of the process in the flowchart of FIG. For example, the server device 2 identifies a measurement point group forming a specific dynamic object such as a pedestrian or vehicle based on processing such as pattern matching based on shape and size, and measures the identified measurement point group. Delete from the point cloud DB 24. Note that the server device 2 also sets a stationary obstacle flag indicating that the voxel including the measurement point indicating the moving object to be excluded is not a stationary obstacle voxel so that it is included in the detection target area of the moving object. It is recommended to set this.

FIG. 10 is a flowchart showing the procedure of a dynamic object detection process according to the second embodiment. The server device 2 repeatedly executes the process shown in the flowchart shown in FIG. 10, for example, while the vehicle is running.

First, the vehicle-mounted device 1 acquires a group of measurement points obtained by measuring the surface of an object around the vehicle based on the output of the lidar 30 (step S501). Next, the vehicle-mounted device 1 refers to the map DB 10 based on the vehicle position and obtains voxel data of voxels existing around the vehicle position (step S502). The above-mentioned vehicle position is obtained by adding the travel distance and azimuth change calculated based on the output of the sensor unit 13 to the estimated vehicle position one time ago, for example, similar to the position estimation process of the first embodiment shown in FIG. This is the predicted vehicle position obtained by Then, the vehicle-mounted device 1 identifies stationary obstacle voxels from the voxels including the measurement point acquired in step S501 by referring to the stationary obstacle flag included in the voxel data of each voxel (step S503). Then, the vehicle-mounted device 1 performs a moving object detection process excluding stationary obstacle voxels from the detection target area (step S504). In other words, the vehicle-mounted device 1 regards stationary obstacle voxels as areas where no moving object appears, skips the moving object detection process, and performs the moving object detection process on the remaining voxels.

As explained above, the server device 2 according to the second embodiment performs a process of detecting a vegetation voxel Bv including a measurement point of vegetation from voxels including a measurement point related to an object measured by a measurement device such as the lidar 30; A process of generating a stationary obstacle flag indicating that the area includes a stationary obstacle is executed for voxels including the measurement point excluding at least the vegetation voxel Bv.

<Modified example>
Modifications suitable for the first and second embodiments will be described below. The following modifications may be combined and applied to each of the above embodiments.

(Modification 1)
Voxel data is not limited to a data structure including an average vector and a covariance matrix, as shown in FIGS. 3 and 8. For example, the voxel data may directly include point cloud data obtained by converting measurement data of a measurement maintenance vehicle into an absolute coordinate system, which is used when calculating the average vector and covariance matrix. In this case, the vehicle-mounted device 1 is not limited to scan matching using NDT, and may apply other scan matching such as ICP (Iterative Closest Point) to estimate the own vehicle position.

(Modification 2)
In the first example, the voxel data shown in FIG. 3 had both a weighting value and a vegetation flag, but it may have only one of them.

For example, if the voxel data has a data structure that does not include a weighting value, the in-vehicle device 1 sets the weighting value used in step S307 of the position estimation process shown in FIG. 6 to a predetermined value according to the value of the vegetation flag. . In this case, the on-vehicle device 1 sets a weighting value to be set when the vegetation flag is a value indicating that the vegetation voxel Bv is a vegetation voxel, and a weighting value to be set when the vegetation flag is a value indicating that the vegetation voxel is not a vegetation voxel Bv. Set to a value smaller than . Thereby, the vehicle-mounted device 1 relatively reduces the degree of influence of the vegetation voxel Bv on the position estimation, and suitably improves the position estimation accuracy, as in the case where the voxel data has a data structure having a weighting value. Note that in this case, the vegetation flag is an example of "information for determining weights."

(Modification 3)
A function corresponding to the on-vehicle device 1 may be built into the vehicle. In this case, the electronic control unit (ECU) of the vehicle executes processing equivalent to the control unit 15 of the on-vehicle device 1 by executing a program stored in the memory of the vehicle.

(Modification 4)
In the second embodiment, instead of excluding the vegetation voxel Bv (i.e., the voxel that includes vegetation) from the stationary obstacle voxels, the on-vehicle device 1 excludes the voxel that only includes vegetation (i.e., the voxel that includes vegetation and other stationary structures). ) may be excluded from the stationary obstacle voxels. In this case, the vehicle-mounted device 1 regards voxels that include stationary structures in addition to vegetation as areas where no moving objects appear, and skips the moving object detection process.

(Modification 5)
The on-vehicle device 1 may execute the vegetation determination process instead of the server device 2. In this case, the on-vehicle device 1 performs the processing in step S101 in FIG. 4 and the vegetation determination in step S102 on the point cloud data obtained by converting the measurement points output by the lidar 30 into an absolute coordinate system in the first embodiment, for example. By executing the process, vegetation voxel Bv is determined. Then, the vehicle-mounted device 1 includes the determined vegetation voxel Bv information in the measurement data D1 and transmits it to the server device 2. In this case, the server device 2 generates and updates weighting values of voxel data and vegetation flags based on the measurement data D1 received from the on-vehicle device 1. Similarly, in the second embodiment, the on-vehicle device 1 performs the process of step S401 and the vegetation determination process of step S402 in FIG. By executing this, the vegetation voxel Bv is determined. Then, the vehicle-mounted device 1 includes the determined vegetation voxel Bv information in the measurement data D1 and transmits it to the server device 2. In this case, the server device 2 generates a stationary obstacle flag based on the measurement data D1 received from the on-vehicle device 1.

1 On-vehicle device 2 Server device 10 Map DB
23 Distribution map DB
24 Measurement point group DB
11, 21 communication section 12, 22 storage section 15, 25 control section 13 sensor section 14 input section 16 output section

Claims (1)

a detection means for detecting an area including a measurement point of vegetation from an area that is a partitioned space and includes a measurement point related to an object measured by the measurement device;
generating means for generating stationary object data indicating that a stationary object is present in a region obtained by excluding at least a region including only the measurement point of the vegetation from a region including the measurement point of the object;
A stationary object data generation device having:

Images