Estimation device, system, estimation method, and program

Abstract

To provide a technique to accurately specify an estimated change point in which the details of map information are estimated to be different from an actual condition.SOLUTION: In an estimation device (10), a first acquisition unit (110) acquires point group data at a plurality of timings obtained from a sensor mounted on a movable body. A second acquisition unit (130) acquires map information. A division unit (150) divides the pieces of point group data acquired by the first acquisition unit (110) into a plurality of predetermined areas. A ratio value calculation unit (170) calculates, for a first area and a second area different from the first area, a ratio value indicating the association ratio between data points in the area and the map information. An estimation unit (190) specifies an estimated change point in which the details of the map information are estimated to be different from an actual condition, by using the intensity of correlation between the ratio value of the first area in which the time or position is shifted and the ratio value of the second area.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an estimation device, a system, an estimation method, and a program. [Background technology]

[0002] 2. Description of the Related Art There is known a technique for detecting objects around a moving body such as a vehicle using a radar, a camera, or the like, and estimating the vehicle's own position by comparing the detection results with map information.

[0003] Patent Document 1 describes a method for estimating the position of a moving body equipped with a lidar by associating point cloud data measured by the lidar with position information of an object for each unit area. It also describes a method for calculating a reliability index of the estimated position using the ratio of the number of associated measurement points to the number of measurement points of the point cloud data.

[0004] Patent Document 2 describes calculating an evaluation function for each voxel based on the result of matching each voxel of point cloud data measured by a lidar with a map database. It also describes extracting voxels with low evaluation functions and transmitting matching degradation information for those voxels to a server device. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] International Publication No. 2021 / 112177 [Patent Document 2] International Publication No. 2018 / 180338 Summary of the Invention [Problem to be solved by the invention]

[0006] In this method of comparing detection results with map information, the accuracy of the map information is important. If the map information does not accurately reflect the actual situation, the accuracy of location estimation may decrease.

[0007] The technology of Patent Document 1 was not able to identify change points where the map information does not correctly reflect the actual situation. The technology of Patent Document 2 had room for improvement in the accuracy of estimating change points.

[0008] One example of a problem to be solved by the present invention is to provide a technology for accurately identifying an estimated change point where the content of map information is estimated to differ from the actual situation. [Means for solving the problem]

[0009] The first invention is a first acquisition unit that acquires point cloud data at multiple timings obtained by a sensor mounted on the moving object; a second acquisition unit that acquires map information; a dividing unit that divides each point cloud data acquired by the first acquiring unit into a plurality of predetermined regions; a ratio value calculation unit that calculates a ratio value indicating a correspondence ratio between each data point in a first area and the map information for a second area different from the first area; an estimation unit that uses the strength of correlation between the ratio value of the first area shifted in time or position and the ratio value of the second area to identify an estimated change point where the content of the map information is estimated to differ from the actual situation. It is an estimation device.

[0010] The second invention is: An estimation device according to a first aspect of the present invention; a server; the estimation device transmits information indicating the identified estimated change point to a server; The server receiving information indicating a plurality of the estimated change points; The received information indicating the plurality of estimated change points is processed for each point, and the estimated change points where the content of the map information is highly likely to differ from the actual situation are extracted from the plurality of estimated change points. It is a system.

[0011] The third invention is 1. A computer-implemented estimation method comprising: a first acquisition step of acquiring point cloud data at multiple timings obtained by a sensor mounted on a moving object; a second acquisition step of acquiring map information; a dividing step of dividing each point cloud data acquired in the first acquiring step into a plurality of predetermined regions; a ratio value calculation step of calculating a ratio value indicating a correspondence ratio between each data point in a first area and the map information for a second area different from the first area; and an estimation step of identifying an estimated change point where the content of the map information is estimated to differ from the actual situation by using the strength of correlation between the ratio value of the first area shifted in time or position and the ratio value of the second area. It is an estimation method.

[0012] The fourth invention is A program that causes a computer to execute the estimation method of the third invention. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 is a block diagram illustrating a functional configuration of an estimation device according to an embodiment. [Figure 2] FIG. 1 is a diagram illustrating the matching of data obtained by a sensor mounted on a mobile object with an ND map. [Figure 3] FIG. 1 is a diagram illustrating the matching of data obtained by a sensor mounted on a mobile object with an ND map. [Figure 4] 1 is a flowchart illustrating a flow of processing performed by an estimation device according to an embodiment. [Figure 5]FIG. 1 is a block diagram illustrating a functional configuration of an estimation device according to a first embodiment. [Figure 6] 4 is a flowchart illustrating a flow of a self-position estimation process performed by the estimation device according to the first embodiment. [Figure 7] FIG. 10 is a diagram illustrating a computer for realizing the estimation device. [Figure 8] FIG. 1 is a diagram illustrating a functional configuration of a system according to a first embodiment. [Figure 9] FIG. 1 is a diagram illustrating multiple regions. [Figure 10] FIG. 10 is a diagram illustrating an association ratio. [Figure 11] FIG. 11 is a diagram showing changes over time in DDAR(1) to DDAR(8) in the example of FIG. 10. [Figure 12] FIG. 10 is a diagram illustrating an example of normalization of DDAR. [Figure 13] FIG. 10 is a diagram for explaining a process performed by an estimation unit. [Figure 14] FIG. 10 is a diagram illustrating conditions for determining correlation. [Figure 15] FIG. 10 is a diagram for explaining the meanings of the first to third conditions. [Figure 16] 4 is a flowchart illustrating a flow of a change-point detection process performed by the estimation device according to the first embodiment. [Figure 17] FIG. 2 is a block diagram illustrating a functional configuration of a server according to the first embodiment. [Figure 18] FIG. 10 is a diagram for explaining a determination based on the detection time of change point information. [Figure 19] 10 is a flowchart illustrating a flow of processing performed by a server. [Figure 20] 10 is a flowchart illustrating a flow of an ND map update process performed by the server according to the first embodiment. [Figure 21] 10 is a flowchart illustrating a flow of a change-point detection process performed by an estimation device according to a second embodiment. [Figure 22] 1 is a graph illustrating a cross-correlation function C(τ). [Figure 23]FIG. 1 shows an original ND map and an ND map with change points generated. [Figure 24] 24(a) to 24(k) are diagrams showing various data obtained by a vehicle traveling along the course shown in FIG. 23. [Figure 25] 1 is a diagram showing the relationship between a moving body (vehicle) and regions (1) to (8) in Experimental Example 1. FIG. [Figure 26] 10(a) to 10(c) are diagrams showing the results of performing the filtering process, normalization process, and limit process described in the first embodiment on each DDAR. [Figure 27] 10A to 10E are diagrams showing data in the process of change-point detection processing. [Figure 28] FIG. 10 is a diagram showing DDARs for each area on the left side of the vehicle. [Figure 29] 10A to 10E are diagrams showing data in the process of change-point detection processing. [Figure 30] FIG. 26(c) is a diagram showing a cross-correlation function C(τ) calculated using DDAR(5) and DDAR(8). [Figure 31] FIG. 29 is a diagram showing a cross-correlation function C(τ) calculated using DDAR(1) and DDAR(4) shown in FIG. 28. [Figure 32] FIG. 10 is a diagram for explaining correspondence between data points and voxels. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In all the drawings, like components are designated by like reference numerals, and the description thereof will be omitted as appropriate.

[0015] FIG. 1 is a block diagram illustrating a functional configuration of an estimation device 10 according to an embodiment. The estimation device 10 according to this embodiment includes a first acquisition unit 110, a second acquisition unit 130, a division unit 150, a ratio value calculation unit 170, and an estimation unit 190. The first acquisition unit 110 acquires point cloud data at multiple timings obtained by a sensor mounted on a moving object. The point cloud data at multiple timings is, for example, point cloud data at multiple times. The second acquisition unit 130 acquires map information. The division unit 150 divides each point cloud data acquired by the first acquisition unit 110 into multiple predetermined regions. The ratio value calculation unit 170 calculates ratio values ​​indicating the correspondence ratio between each data point in a first region and the map information for a second region different from the first region. The estimation unit 190 identifies an estimated change point where the content of the map information is estimated to differ from the actual situation using the strength of correlation between the ratio value of the first region, which is shifted in time or position, and the ratio value of the second region.

[0016] For moving objects such as vehicles, it is important to accurately determine their position in order to ensure accurate route navigation, autonomous driving, and driver assistance. When a moving object travels on a road, its position at a given point in time is estimated using its position at that point in time and its speed and direction of movement up to the next point in time. Here, by detecting objects (target objects) around the moving object using a sensor or other device and comparing them with the positions of those objects in map information, it is possible to further improve the accuracy of the estimated position of the moving object. One such correction method is NDT (Normal Distributions Transform) scan matching.

[0017] NDT scan matching is a method for calculating the self-localization of a moving vehicle by matching data detected from objects around the vehicle with a normal distribution (ND) map, which divides three-dimensional space into a grid of predetermined sizes (voxels) and expresses the normal distribution. The accuracy of the ND map is crucial for highly accurate self-localization. Since object detection data from a moving vehicle is matched with the ND map, a low accuracy of the ND map will result in poor self-localization accuracy. Therefore, the ND map must be as accurate as possible. ND maps are created, for example, by a dedicated measurement vehicle equipped with various sensors that collects high-density, high-precision three-dimensional point cloud data while traveling. Because the operation of these measurement vehicles is costly, they are not frequently driven on the same roads, and point cloud data can only be collected from the same location a few times a year. Therefore, even if structures around the road are newly constructed or demolished, it takes time for these changes to be reflected in the ND map. In such cases, the accuracy of NDT self-localization may be reduced at points where the ND map does not match the actual environment (change points). Ultimately, this could affect the accuracy of navigation, autonomous driving, driver assistance, etc.

[0018] The estimation device 10 according to this embodiment can identify estimated change points where the content of the map information is estimated to differ from the actual situation. Therefore, for such points, it is possible to adjust the handling of ND map data and promote the updating of the ND map.

[0019] In this embodiment, map information refers to data that can be used as reference data for NDT scan matching. Hereinafter, map information is also referred to as an "ND map." The ND map includes multiple voxel data. The voxel data is data that records the position information of stationary structures in each region (also referred to as a "voxel") when a three-dimensional space is divided into multiple regions. The voxel data includes data that represents measured point cloud data of stationary structures within each voxel using a normal distribution. Specifically, the ND map includes at least the voxel ID, voxel coordinates, mean vector, covariance matrix, and point cloud count information for each of multiple voxels. Each voxel is a cube obtained by dividing space into a grid, and its shape and size are predetermined. The voxel coordinates indicate the absolute three-dimensional coordinates of a reference position, such as the center position of the voxel. The mean vector and covariance matrix correspond to parameters when representing the point cloud within a voxel using a normal distribution. The point group number information is information indicating the number of point groups used to calculate the mean vector and covariance matrix of the voxel.

[0020] 2 and 3 are diagrams illustrating the matching of data obtained by the sensor 210 mounted on the mobile object 20 with an ND map. FIG. 2 shows an example in which the ND map matches the actual situation, and FIG. 3 shows an example in which the ND map does not match the actual situation. In these figures, the rectangles marked A to E indicate structures. Furthermore, each square represents a voxel included in the ND map, and the black circles indicate a point cloud obtained by the sensor 210 of the mobile object 20.

[0021] In this embodiment, the sensor 210 is, for example, a sensor that measures the distance to an object by emitting light and receiving the light reflected by the object. The sensor 210 is not particularly limited, but may be a radar or a lidar (LIDAR: Laser Imaging Detection and Ranging, Laser Illuminated Detection and Ranging, or LiDAR: Light Detection and Ranging). The light emitted from the sensor 210 is not particularly limited, but may be, for example, infrared light. The light emitted from the sensor 210 is, for example, a laser pulse. The sensor 210 calculates the distance from the sensor 210 to the object using, for example, the time from emitting the pulsed light to receiving the reflected light and the propagation speed of the pulsed light. The direction of light emitted from the sensor 210 is variable, and the measurement area is scanned by sequentially performing measurements in multiple emission directions. For example, by moving the emitted light back and forth horizontally and vertically, the distance to an object located in each angular direction can be measured, thereby obtaining three-dimensional information data within the horizontal and vertical scan ranges.

[0022] The sensor 210 outputs point cloud data in which the three-dimensional positions of light reflection points are associated with the reflection intensity (i.e., the light reception intensity at the sensor 210). The first acquisition unit 110 of the estimation device 10 acquires the point cloud data. The point cloud data output from the sensor 210 is configured in units of frames. One frame is configured with data obtained by scanning the measurement area once. The sensor 210 generates multiple consecutive frames by repeatedly scanning the measurement area. The mobile body 20 may be equipped with multiple sensors 210 so that they scan in different directions as seen from the mobile body 20. Alternatively, the sensor 210 may scan by rotating 360° around the surroundings.

[0023] In the example of Figure 2, point cloud data detecting structures A, B, and D near the road on which the vehicle (mobile body 20) is traveling is correctly associated with voxels on the ND map. As a result, accurate self-location estimation is performed. Note that structure C is located behind structure B. Therefore, when this ND map was generated, a point cloud related to structure C was not obtained, and there is no voxel corresponding to structure C on the ND map.

[0024] If a change occurs in a structure near the road on which the mobile object 20 is traveling, the actual environment and the ND map will not match, as shown in Figure 3, until the ND map is updated. Specifically, in Figure 3, a new structure E was constructed between structure D and the road, but the ND map has not yet been updated, so there is no voxel corresponding to structure E. Also, structure B was demolished, but the ND map has not yet been updated, so voxels originating from structure B remain on the ND map. In this figure, the point cloud data detecting structures C and E does not have corresponding voxels, and no association with the voxels is performed. In other words, the point cloud data detecting structures C and E is not used in the NDT calculation, and only the point cloud data detecting structure A is used in the calculation with the voxel data. Therefore, the amount of data used in the NDT calculation is reduced, which may result in a decrease in the accuracy of self-localization estimation. There is also a possibility that the voxel data of structure D may be erroneously matched with the point cloud obtained by measuring structure E, or that the voxel data of structure B may be erroneously matched with the point cloud obtained by measuring structure C. Such erroneous matching can cause the estimated self-position to deviate significantly from the actual position.

[0025] In this way, when a mismatch between the actual environment and the ND map occurs, it is important to detect that point as quickly as possible, because if such a point can be detected, processing can be performed to reduce the impact of the voxel data at that point on position estimation, and the ND map for that point can be updated.

[0026] According to the estimation device 10 of this embodiment, the ratio value calculation unit 170 calculates ratio values ​​for the first and second regions, which indicate the correlation ratio between each data point in the region and the map information. The estimation unit 190 then uses the strength of correlation between the ratio value of the first region, which is shifted in time or location, and the ratio value of the second region to identify an estimated change point where the content of the map information is estimated to differ from the actual situation. Therefore, when a state occurs in which the actual environment and the ND map do not match, the point can be identified with high accuracy.

[0027] FIG. 4 is a flowchart illustrating the flow of processing performed by the estimation device 10 according to this embodiment. The estimation method according to this embodiment is executed by a computer. The estimation method according to this embodiment includes a first acquisition step S101, a second acquisition step S102, a division step S103, a ratio value calculation step S104, and an estimation step S105. In the first acquisition step S101, point cloud data obtained at multiple times by a sensor mounted on a moving object is acquired. In the second acquisition step S102, map information is acquired. In the division step S103, each point cloud data obtained in the first acquisition step S101 is divided into a plurality of predetermined regions. In the ratio value calculation step S104, a ratio value indicating the correspondence ratio between each data point in a first region and the map information is calculated for a second region different from the first region. In estimation step S105, the strength of correlation between the ratio value of the first area shifted in time or position and the ratio value of the second area is used to identify estimated change points where the content of the map information is estimated to differ from the actual situation.

[0028] Example 1 5 is a block diagram illustrating a functional configuration of the estimation device 10 according to a first embodiment. The estimation device 10 according to this embodiment has the configuration of the estimation device 10 according to the embodiment. The estimation device 10 according to this embodiment further includes a self-location estimation unit 140 and a reliability calculation unit 180. The estimation unit 190 also includes a shift unit 191, a correlation determination unit 192, and a change-point identification unit 193.

[0029] 6 is a flowchart illustrating the flow of a self-position estimation process performed by the estimation device 10 according to this embodiment. The self-position estimation unit 140 estimates the position of the moving object 20. In the following description, the sensor 210 is described as a lidar, but the sensor 210 may be of another type.

[0030] When the operation of the estimation device 10 is started, the estimation device 10 first determines its initial estimated self-location (S10). Specifically, the self-location estimation unit 140 sets the positioning result obtained by a Global Navigation Satellite System (GNSS) provided in the moving body 20 or the estimation device 10 as the initial estimated self-location. Next, the self-location estimation unit 140 calculates the latest predicted self-location using the previous estimated self-location (S20). Specifically, the self-location estimation unit 140 acquires the speed and yaw angular velocity of the moving body 20 from a speed sensor and a gyro sensor provided in the moving body 20, and determines the direction and amount of movement of the moving body 20 from the previous estimated self-location. Then, the self-location estimation unit 140 calculates the position after movement in the determined direction and amount of movement from the previous estimated self-location as the predicted self-location.

[0031] Next, the self-location estimation unit 140 determines whether an ND map of the vicinity of the predicted self-location has already been acquired (S30). If the ND map has already been acquired (Yes in S30), the self-location estimation unit 140 skips the process of S40 and proceeds to the process of S50. If the ND map has not yet been acquired (No in S30), the second acquisition unit 130 acquires an ND map of the vicinity of the predicted self-location (S40). The second acquisition unit 130 may acquire the ND map by reading it from a storage unit accessible from the second acquisition unit 130, or may acquire it from an external device via a network. The storage unit accessible from the second acquisition unit 130 may be provided inside the estimation device 10 or may be provided outside the estimation device 10.

[0032] In S40, the second acquisition unit 130 sets voxels in the acquired ND map that are assigned a non-recommended flag so as not to be used in NDT matching. Alternatively, the second acquisition unit 130 reduces the weight of the voxels that are assigned a non-recommended flag. Note that the weight is a value that indicates the magnitude of the influence that information about the voxel has on the calculation of self-location estimation in NDT matching. The non-recommended flag will be described in detail later.

[0033] Next, the first acquisition unit 110 determines whether or not point cloud data has been acquired (S50). If point cloud data has not been acquired (No in S50), such as when traveling in an open area with no surrounding structures, the process returns to S20. If point cloud data has been acquired by the first acquisition unit 110 (Yes in S50), downsampling of the point cloud data is performed in S60. Specifically, the first acquisition unit 110 downsamples the acquired point cloud data so as to obtain a point cloud with a predetermined number of data points.

[0034] Next, in S70, the self-location estimation unit 140 performs an NDT matching process to calculate an estimated self-location. Specifically, the self-location estimation unit 140 sets the predicted self-location as an initial value and performs an NDT matching process using an ND map around the predicted self-location and the downsampled point cloud data obtained by the first acquisition unit 110. The NDT matching process and self-location estimation using the results thereof can be performed by an existing method. The self-location estimation unit 140 regards the obtained estimated self-location as the latest estimated self-location. The estimated self-location may be output to a device other than the estimation device 10 or may be stored in a storage device accessible from the self-location estimation unit 140. The estimated self-location may be used for functions such as route navigation, automatic driving, or driving assistance.

[0035] Next, in S80, a change-point detection process is performed by the dividing unit 150, the ratio value calculation unit 170, and the estimation unit 190. The change-point detection process will be described in detail later with reference to FIG.

[0036] When the change point detection process of S80 ends, the self-position estimation unit 140 determines whether or not the termination condition is satisfied (S90). The termination condition is satisfied, for example, when the movement of the moving body 20 stops or when an operation to stop the estimation process by the estimation device 10 is performed. If the termination condition is satisfied (Yes in S90), the estimation device 10 terminates the process. If the termination condition is not satisfied (No in S90), the process returns to S20.

[0037] The hardware configuration of the estimating device 10 will be described below. Each functional component of the estimating device 10 may be realized by hardware that realizes the functional component (e.g., a hardwired electronic circuit, etc.), or may be realized by a combination of hardware and software (e.g., a combination of an electronic circuit and a program that controls it). Below, a case where each functional component of the estimating device 10 is realized by a combination of hardware and software will be further described.

[0038] FIG. 7 is a diagram illustrating a computer 1000 for realizing the estimation device 10. The computer 1000 is any computer. For example, the computer 1000 is a system on chip (SoC), a personal computer (PC), a server machine, a tablet terminal, a smartphone, or the like. The computer 1000 may be a dedicated computer designed to realize the estimation device 10, or may be a general-purpose computer.

[0039] The computer 1000 includes a bus 1020, a processor 1040, a memory 1060, a storage device 1080, an input / output interface 1100, and a network interface 1120. The bus 1020 is a data transmission path through which the processor 1040, the memory 1060, the storage device 1080, the input / output interface 1100, and the network interface 1120 transmit and receive data to and from each other. However, the method of interconnecting the processor 1040 and other components is not limited to bus connection. The processor 1040 may be any of various processors, such as a central processing unit (CPU), a graphics processing unit (GPU), or a field-programmable gate array (FPGA). The memory 1060 is a main storage device implemented using a random access memory (RAM) or the like. The storage device 1080 is an auxiliary storage device implemented using a hard disk, a solid state drive (SSD), a memory card, a read-only memory (ROM), or the like.

[0040] The input / output interface 1100 is an interface for connecting the computer 1000 to an input / output device. For example, the input / output interface 1100 is connected to an input device such as a keyboard and an output device such as a display.

[0041] The network interface 1120 is an interface for connecting the computer 1000 to a network. This communication network is, for example, a LAN (Local Area Network) or a WAN (Wide Area Network). The network interface 1120 may be connected to the network wirelessly or by wire.

[0042] The storage device 1080 stores program modules that realize the respective functional components of the estimation device 10. The processor 1040 reads each of these program modules into the memory 1060 and executes them to realize the function corresponding to each program module.

[0043] 8 is a diagram illustrating an example of the functional configuration of a system 50 according to this embodiment. The system 50 according to this embodiment includes an estimation device 10 and a server 40. The estimation device 10 transmits information indicating identified estimated change points (hereinafter also referred to as "change point information") to the server 40. The server 40 receives a plurality of pieces of change point information and processes the received plurality of change point information for each point, thereby extracting estimated change points (hereinafter also referred to as "highly reliable change points") where the content of the map information is highly likely to differ from the actual situation from the plurality of estimated change points indicated in the plurality of change point information.

[0044] The estimation device 10 and the server 40 can communicate wirelessly. The estimation device 10 is provided in, for example, a mobile body 20. The server 40 may receive change-point information from a plurality of estimation devices 10. The processing performed by the server 40 will be described in detail later.

[0045] Each component of the estimation device 10 will be described in detail below with reference to FIG. 5. The first acquisition unit 110 acquires point cloud data generated by a LIDAR. The first acquisition unit 110 may acquire point cloud data generated by a LIDAR and temporarily stored in a storage unit, or may acquire point cloud data directly from the LIDAR. The first acquisition unit 110 acquires point cloud data for multiple frames in the order in which they were generated. As described above, the first acquisition unit 110 downsamples the acquired point cloud data.

[0046] <Split> FIG. 9 is a diagram illustrating multiple regions. The dividing unit 150 divides point cloud data obtained about the periphery of the moving object 20 into data for multiple regions based on the position of the moving object 20. The position and size of each region relative to the moving object 20 are predetermined. In the example shown in this figure, multiple sensors 210 are provided on the moving object 20, and each sensor repeatedly generates point cloud data within the measurement region 220 at the same time. The first acquisition unit 110 acquires point cloud data from all sensors 210. The dividing unit 150 combines the point cloud data obtained at the same time on a coordinate system based on the moving object 20 to generate point cloud data about the periphery of the moving object 20. The same time means, for example, the same time. However, the same time does not have to be exactly the same time and may include some error. Then, by identifying which region the position of each data point falls within, each data point around the moving object 20 is assigned to multiple regions. In this way, point cloud data for each of the multiple regions is obtained.

[0047] In the example of this figure, the periphery of the moving body 20 is divided into eight regions, region (1) to region (8). The interval between adjacent regions is L. The region interval L is, for example, the distance between the centers of the regions, and is the distance in the traveling direction of the moving body 20. In the example of this figure, the point cloud data is divided at least into a region on the right side and a region on the left side of the moving body 20. Also, in the example of this figure, the point cloud data is divided at least into a region on the front side and a region on the rear side of the moving body 20 in the traveling direction. It is preferable that the point cloud data is divided into four or more regions in the traveling direction of the moving body 20. The sizes of the multiple regions do not need to be the same.

[0048] <Ratio value> The ratio value calculation unit 170 calculates a ratio value indicating the association ratio between each data point in the region and the map information. Specifically, the ratio value calculation unit 170 calculates time-series ratio values ​​for the first region and the second region. The ratio value may be the association ratio itself, or may be a value obtained by inverting the ratio or performing a predetermined calculation on the ratio. Hereinafter, the ratio value for each region will be referred to as DDAR (Divided Data Association Ratio), and the DDAR for region k will be referred to as DDAR(k). The ratio value calculation unit 170 can calculate DDAR(k) using the relationship "DDAR(k) = number of data points associated in region k / number of data points included in region k."

[0049] FIG. 10 is a diagram illustrating the correspondence ratio. In this figure, the rectangles marked A and C represent structures. Each square represents a voxel included in the ND map, and the black circle represents a point cloud obtained by the sensor 210 of the mobile object 20. The self-localization unit 140 performs correspondence between the ND map and the point cloud data. The ND map includes voxels corresponding to positions where stationary structures and the like exist. If a data point of the point cloud is located at a position corresponding to the position of a certain voxel, that data point is associated with that voxel. On the other hand, since no voxel exists at a position where no stationary structures and the like exist, a data point corresponding to a position where no voxel exists is not associated with any voxel. Therefore, at points where the actual situation and the ND map do not match, the number of unassociated data points increases, and the correspondence ratio decreases.

[0050] 32 is a diagram for explaining the association between data points and voxels. First, the self-location estimation unit 140 converts the coordinates of each point included in the point cloud data into a world coordinate system based on the predicted self-location. Then, for example, if the voxel size of the ND map is 1 m square, the x, y, and z coordinates of each point are rounded to the nearest integer. The self-location estimation unit 140 then compares each point with the ND map shown in the same world coordinate system, identifies which voxel each point is in, and associates the points.

[0051] For ease of explanation, this figure shows voxels in two dimensions. In this figure, each solid-line rectangle represents a voxel. The voxel number is indicated in the upper left corner of each voxel. The black circles represent data points in the point cloud. For example, data point 60 has rounded x-y coordinates of (x,y) = (2,1). Therefore, it is associated with voxel 1, whose voxel coordinates are (x,y) = (2,1). Note that in this example, the voxel coordinates are the coordinates of the center position of the voxel, but voxel coordinates are not limited to this and may be, for example, the coordinates of any vertex of the voxel. On the other hand, data point 61 has rounded coordinates of (x,y) = (4,1). Since there is no voxel whose voxel coordinates are (x,y) = (4,1), data point 61 is not associated with any voxel. The self-position estimation unit 140 performs NDT matching processing using the associated data points and voxels.

[0052] For example, when the mobile object 20 travels near a point where there is a difference between the actual environment and the ND map, as shown in Figure 10, the data point where structure C is detected will not be associated, and the association ratio will decrease. Furthermore, the decrease in DDAR for each region is expected to progress in the order of DDAR(5), DDAR(6), DDAR(7), and DDAR(8) in the example of region division shown in Figure 9. That is, in the state on the left side of Figure 10, there is a high possibility that DDAR will decrease from region (5) to region (6), while in the state on the right side of the figure, there is a high possibility that DDAR will decrease from region (7) to region (8).

[0053] FIG. 11 is a diagram showing the changes over time of DDAR(1) to DDAR(8) in the example of FIG. 10. DDAR(1) to DDAR(4), which correspond to the left side of the moving object 20 where there is no change point, maintain high values. On the other hand, DDAR(5) to DDAR(8), which correspond to the right side of the moving object 20 where there is a change point, each have a timing at which their values ​​decrease. Such timings become later in the order of DDAR(5) to DDAR(8). Therefore, by understanding how the timing at which the DDAR decreases changes in multiple regions, it is possible to accurately estimate the change point.

[0054] The first region and the second region are two different regions among the multiple regions obtained by division. While the relationship between the first region and the second region is not particularly limited, it is preferable that the first region and the second region are regions that are offset in the traveling direction of the moving body 20. In this case, it is preferable that the amount of offset in the traveling direction is large. Specifically, for example, it is preferable that the first region is one of the frontmost region and the rearmost region among the multiple regions in the traveling direction, and the second region is the other of the frontmost region and the rearmost region. Furthermore, it is preferable that the first region and the second region are both regions on the right side of the traveling direction relative to the moving body 20, or both regions on the left side of the traveling direction relative to the moving body 20. In the following, an example will be described in which region (5) is the first region and region (8) is the second region, but the first region and the second region are not limited to this example.

[0055] The estimation device 10 does not need to use the DDARs of all regions obtained by division to estimate the estimated change points. The processing performed by the ratio value calculation unit 170 and the estimation unit 190 only needs to be performed for at least the first region and the second region. The estimation device 10 may also estimate the estimated change points by providing multiple pairs of first regions and second regions. For example, a first region and a second region may be set on the right side of the moving body 20 in the traveling direction, and a first region and a second region may be set on the left side of the moving body 20 in the traveling direction, and the estimated change points may be estimated on the left and right sides of the moving body 20, respectively.

[0056] The ratio value calculation unit 170 acquires information indicating the correspondence between the point cloud data of the entire region and the ND map from the self-position estimation unit 140. The ratio value calculation unit 170 also counts the number of data points included in each divided region. Then, for each region, the ratio value calculation unit 170 counts the number of data points associated with voxels. Then, the ratio value calculation unit 170 calculates the DDAR for each region by dividing the number of associated data points by the number of data points included in the region.

[0057] Furthermore, the ratio value calculation unit 170 performs several processes on the DDAR to improve the estimation accuracy of the change point. Below, we will explain the filtering process and normalization process. Although it is not necessary to perform all of these processes, it is preferable to perform all of them to improve the estimation accuracy. Note that these processes may be performed on the ratio value before shifting, which will be described later, or on the ratio value after shifting.

[0058] <<Filtering>> When point cloud data is divided into multiple regions, the amount of data in each region is reduced. Therefore, changes due to increases or decreases in DDAR tend to appear drastically. If there is an instantaneous increase or decrease in DDAR, it may become noise in the subsequent correlation determination. Therefore, it is preferable to suppress instantaneous fluctuations by applying appropriate filtering to the DDAR.

[0059] For example, it is preferable that the ratio value calculation unit 170 performs first-order lag filter processing on the DDAR of each region. Specifically, it is preferable that the ratio value calculation unit 170 performs first-order lag filter processing on the DDAR of each region, with a time constant that is 10 to 20 times the period of the NDT processing. This makes it possible to grasp the trend of change while suppressing instantaneous fluctuations in the DDAR. This in turn improves the stability of correlation determination.

[0060] <<Normalization>> Figure 12 shows an example of DDAR normalization. If an error occurs in the self-location estimation by NDT due to some influence and the estimated self-location deviates from the correct position, the correspondence ratio between the point cloud data and voxels will decrease overall. In such a situation, the DDAR values ​​of each region will decrease at the same time. Since this is not due to a change point, it will result in an error in the correlation determination for change point detection.

[0061] Therefore, in this embodiment, the ratio value calculation unit 170 calculates an overall ratio value DAR (Data Association Ratio) that indicates the association ratio between each data point of the point cloud data before division and the map information, and normalizes the ratio value of the first region and the ratio value of the second region by the overall ratio value DAR.The estimation unit 190 then identifies the estimated change point using the normalized ratio value.This makes it possible to reduce the influence of deviation in the estimated self-location.

[0062] Specifically, the ratio value calculation unit 170 calculates the DAR using information indicating the result of associating the point cloud data of the entire region with the ND map, acquired from the self-position estimation unit 140. The DAR can be calculated using the relationship "DAR = number of associated data in the entire region / number of data included in the entire region." Furthermore, the ratio value calculation unit 170 normalizes the DDAR(k) by dividing the DDAR(k) of each region by the DAR.

[0063] Note that DDAR may be larger than DAR in some cases, and therefore may exceed 1 after normalization. In correlation determination, which will be described later, it is desirable that the upper limit of DDAR is 1, and therefore ratio value calculation unit 170 may further perform limit processing to limit the value of DDAR to a range of 0 to 1. In other words, if the value of DDAR after normalization exceeds 1, ratio value calculation unit 170 replaces that value with 1.

[0064] <Identifying estimated change points> In this embodiment, the estimation unit 190 identifies an estimated change point using the strength of correlation between a post-shift ratio value, which is a time-shifted version of a ratio value of a first region, and a ratio value of a second region. Specifically, the estimation unit 190 generates a post-shift ratio value by shifting the time axis of the ratio value of the first region by a time τ so that the time axis of the ratio value of the second region is aligned with the time axis of the ratio value of the second region. The estimation unit 190 then identifies an estimated change point using the generated post-shift ratio value of the first region and the ratio value of the second region. Here, the time τ is a time calculated using the speed of the moving object. To identify an estimated change point, the estimation unit 190 calculates a first index by multiplying the post-shift ratio value of the first region by the ratio value of the second region. The estimation unit 190 also calculates a second index by subtracting one of the post-shift ratio value of the first region and the ratio value of the second region from the other. The estimation unit 190 then identifies an estimated change point using the first index and the second index. Identifying an estimated change point will be described in detail below.

[0065] 13 is a diagram for explaining the processing performed by the estimation unit 190. When a change point exists, a portion where the value drops occurs in the time series DDAR, and the drop appears slightly shifted in multiple regions. In other words, it is estimated that there is a strong correlation between the DDARs in multiple regions shifted in time according to the movement of the moving object 20. Therefore, it is possible to identify the estimated change point by evaluating the strength of the correlation.

[0066] For example, if region (5) is the first region and region (8) is the second region, the region interval between these regions is 3L. The estimation unit 190 divides the region interval 3L by the speed of the moving object 20 to obtain the time τ for the moving object 20 to travel the distance of 3L. Next, the estimation unit 190 shifts the time axis of DDAR(5) by the time τ. In this way, the shifted DDAR(5), i.e., the shifted ratio value, is obtained.

[0067] The correlation determination unit 192 determines the correlation between the post-shift ratio value of region (5) and the ratio value of region (8). Specifically, the correlation determination unit 192 first inverts the post-shift DDAR(5) and DDAR(8). That is, the DDAR(k) before inversion is subtracted from 1 to obtain the post-inversion DDAR(k). The correlation determination unit 192 then multiplies the post-inversion DDAR(5) and the post-inversion DDAR(8) together to obtain a first index M. The correlation determination unit 192 also subtracts the DDAR(8) from the post-shift DDAR(5) to obtain a second index S. The time-series first index M and second index S are obtained using the time-series DDAR(5) and the time-series DDAR(8).

[0068] 14 is a diagram illustrating the conditions for determining the correlation. The correlation determination unit 192 determines whether the following first to third conditions are satisfied for the first index M and the second index S. If all of the first to third conditions are satisfied, the correlation determination unit 192 determines that the correlation between DDAR(5) and DDAR(8) is strong.

[0069] The first condition is W d =W t × v≧Th2. W t is the time duration during which the first index M is continuously equal to or greater than the threshold value Th1. The threshold value Th1 is, for example, equal to or greater than 0.01 and equal to or less than 0.1, and preferably 0.04. v is the speed of the moving object 20. The threshold value Th2 is, for example, equal to or greater than 1 m and equal to or less than 8 m, and preferably 5 m.

[0070] The second condition is expressed by the following formula (1): In other words, the second condition is that the average value M of the first index M(M(k)) within a period in which the first index M is continuously equal to or greater than the threshold value Th1. ave is greater than threshold value Th3. In formula (1), N is the number of data items in a period in which first index M is continuously greater than or equal to threshold value Th1. Threshold value Th3 is, for example, greater than or equal to 0.01 and less than or equal to 0.1, and is preferably 0.05.

[0071]

number

[0072] The third condition is expressed by the following formula (2): In other words, the third condition is the deviation S of the second index S(S(k)) within a period in which the first index M is continuously equal to or greater than the threshold value Th1. dev is smaller than a threshold value Th4. The threshold value Th4 is, for example, equal to or greater than 0.1 and equal to or less than 0.5, and is preferably 0.2.

[0073]

number

[0074] Figure 15 is a diagram to explain the meaning of the first to third conditions. As shown in the leftmost column of this figure, when two DDARs are relatively large, have similar sizes, and are positioned, the multiplication result is large and the subtraction result is small. In such a case, it can be said that there is a strong correlation between the two DDARs.

[0075] As shown in the second column from the left, if the two DDARs are small, the multiplication result will be small even if the peak positions match. As shown in the third column from the left, if there is a difference in peak positions, even if the two DDARs are relatively large, the multiplication result will be small and the deviation in the subtraction result will be large. Also, as shown in the fourth column from the left, if one DDAR is small, the multiplication result will be large, but the deviation in the subtraction result will also be large.

[0076] Therefore, by detecting when the multiplication result (first index M) is large and the deviation of the subtraction result (second index S) is small, it is possible to detect a state in which the correlation between the two DDARs is strong and identify the estimated change point.

[0077] However, the correlation determination unit 192 may evaluate the correlation between two DDARs using a method other than the above. Also, the correlation determination unit 192 may evaluate the correlation between two DDARs using only one of the first index and the second index.

[0078] If it is determined that all of the first to third conditions are satisfied, the correlation determination unit 192 identifies a period during which the above-mentioned first index M is continuously equal to or greater than the threshold value Th1 as a period reflecting a change point. The change point identification unit 193 then identifies the timing at which data for this period was obtained in the DDAR(8), and identifies the direction of the DDAR(8) as seen from the moving object 20 as the change point direction. The change point identification unit 193 then identifies a position in the change point direction as the position of the estimated change point, using the estimated self-location at that timing as a reference.

[0079] <Reliability> The estimation device 10 according to this embodiment further calculates the reliability of the estimated change point and transmits the calculated reliability to the server 40. Using the received reliability, the server 40 can extract, from the plurality of estimated change points, estimated change points where the content of the map information is likely to differ from the actual situation.

[0080] When the change point identification unit 193 identifies an estimated change point, the reliability calculation unit 180 calculates the reliability of the change point estimation. d The larger is, the more ave The larger is, and S dev The smaller the value, the higher the reliability of the change point (or the importance of the change point).

[0081] The change point identification unit 193 calculates the reliability R cp a, b, and c are predetermined coefficients. R, which is the calculation result of Equation (3), Equation (4), and Equation (5), can be calculated. cp1 , R cp2 , and R cp3 are all values ​​in the range of 0 to 1. Therefore, R cp1 , R cp2 , and R cp3 The larger both of these are, the higher the reliability R cp also increases and approaches 1. Also, R cp1 , R cp2 , and R cp3If either of these is close to 0, the reliability R cp becomes smaller.

[0082]

number

[0083] The estimation device 10 associates information indicating the position of the estimated change point with the time when the estimated change point was identified and the reliability, and transmits the information to the server 40. The server 40 receives the information transmitted from the estimation device 10.

[0084] 16 is a flowchart illustrating the flow of the change-point detection process performed by the estimation device 10 according to this embodiment. When S80 shown in FIG. 6 starts, in S801, the dividing unit 150 divides the point cloud data into data of a plurality of regions. Then, the ratio value calculation unit 170 calculates the DDAR of each region. In addition, the ratio value calculation unit 170 performs the above-described filtering process and normalization on the DDAR.

[0085] Next, in S802, the shift unit 191 calculates the amount of shift based on the speed of the moving object 20, and shifts the DDAR of the first region. Then, in S803, the correlation determination unit 192 inverts the DDAR of the first region after the shift with the DDAR of the second region, and calculates the first index and the second index using both the inverted DDARs. In addition, in S804, the correlation determination unit 192 calculates the W d , M ave , and S dev Calculate.

[0086] Next, the correlation determination unit 192 calculates the correlation d , M ave , and S devThe process then proceeds to step S805, where it is determined whether or not the first to third conditions are satisfied (S805). If at least one of the first to third conditions is not satisfied (No in S805), no change point is detected, and the change point detection process ends. On the other hand, if all of the first to third conditions are satisfied (Yes in S805), the change point identification unit 193 identifies the position of the estimated change point using the estimated self-position identified by the self-position estimation unit 140. Furthermore, the reliability calculation unit 180 calculates the W d , M ave , and S dev The reliability of the estimated change point is generated using (S806). Then, in S807, information indicating the position, detection time, and reliability of the estimated change point is transmitted from the estimation device 10 to the server 40. Then, the change point detection process ends.

[0087] <Processing by the server> 17 is a block diagram illustrating the functional configuration of the server 40 according to this embodiment. The server 40 according to this embodiment includes a change point information acquisition unit 410, an information expansion unit 420, an evaluation unit 430, a change point extraction unit 440, a maintenance point registration unit 450, and a flag assignment unit 460.

[0088] As described above, when an estimated change point is identified by the estimation device 10, change point information is transmitted from the estimation device 10 to the server 40. The change point information acquisition unit 410 acquires change point information from the estimation device 10. The change point information is transmitted from the estimation device 10 each time an estimated change point is identified. The server 40 may acquire change point information from multiple estimation devices 10. In this way, the change point information acquisition unit 410 acquires multiple pieces of change point information. The information expansion unit 420 expands the change point information acquired by the change point information acquisition unit 410 into information for each point. For example, if positions indicated by multiple pieces of change point information are within a predetermined area on the map, the information expansion unit 420 groups the change point information as information relating to the same point. The information expansion by the information expansion unit 420 may be performed each time the change point information acquisition unit 410 acquires change point information, or may be performed at a predetermined interval.

[0089] The evaluation unit 430 evaluates the reliability of the estimated change point for each point, that is, for each of the grouped information. For example, for each point, the evaluation unit 430 evaluates the detection time associated with the change point information, the number of acquired change point information, and the reliability R associated with the change point information. cp The evaluation unit 430 extracts highly reliable change points where the content of the map information is highly likely to differ from the actual situation by making a judgment using the above. cp It is also possible to make the determination using only one or two of the above. Note that if the change point information is not associated with a detection time, the time when the change point information acquisition unit 410 acquires the change point information may be used instead of the detection time.

[0090] For example, if the fourth, fifth, and sixth conditions described below are all satisfied, the evaluation unit 430 determines that the point is a highly reliable change point. This allows the change point extraction unit 440 to extract highly reliable change points from the multiple estimated change points indicated by the multiple pieces of change point information acquired by the change point information acquisition unit 410.

[0091] FIG. 18 is a diagram for explaining a determination based on the detection time of change-point information. For example, as shown in this figure, if a change-point is detected due to occlusion by a parked vehicle, it will be detected at a certain time and then will no longer be detected. This is because if the parked vehicle moves and disappears, it will no longer be detected as a change-point. Furthermore, the change in the number of detections will be steep. If the detection time is temporary in this way, it can be determined that the content of the map information is unlikely to differ from the actual situation. On the other hand, since the construction and demolition of structures do not occur instantaneously, the increase in the number of detections is gradual, and the number of detections will remain high. In this case, it can be determined that the content of the map information is likely to differ from the actual situation.

[0092] The evaluation unit 430 identifies the number of detections for each detection time (i.e., the number of change point information acquired by the change point information acquisition unit 410). Then, it determines whether the state in which the number of detections is equal to or greater than a predetermined number continues for a predetermined time or more. The fourth condition is a condition related to the duration, and requires that the state in which the number of detections is equal to or greater than a predetermined number N1 continues for a predetermined time or more.

[0093] Furthermore, the evaluation unit 430 counts the number of acquired change-point information for each location. Then, the evaluation unit 430 determines whether the number of acquired change-point information is equal to or greater than a predetermined number N2. The fifth condition is a condition related to the number of detections, and requires that the number of acquired change-point information is equal to or greater than the predetermined number N2.

[0094] Furthermore, the evaluation unit 430 calculates the reliability R associated with the acquired plurality of pieces of change-point information. cp The average value of each point is calculated. Then, it is determined whether the calculated average value is equal to or greater than a predetermined value. The sixth condition is a condition related to the reliability, and the reliability R cp The average value of is equal to or greater than a predetermined value.

[0095] The change point extraction unit 440 extracts highly reliable change points from the multiple estimated change points based on the evaluation results by the evaluation unit 430. The maintenance point registration unit 450 then registers the extracted highly reliable change points in a database as points requiring maintenance of the ND map. At this time, points with a greater number of acquired change point information may be assigned a higher maintenance priority. By checking this database during the ND map update process, it is possible to identify points that require remeasurement as a priority, leading to the realization of an accurate ND map. The database is stored in a storage unit 470 that can be accessed by the maintenance point registration unit 450.

[0096] The flag assigning unit 460 assigns a non-recommended flag to voxels corresponding to the highly reliable change points extracted by the change point extracting unit 440 in the ND map.

[0097] The hardware configuration of the computer that realizes the server 40 is shown in, for example, Fig. 7, similar to that of the estimating device 10. However, a storage device 1080 of the computer 1000 that realizes the server 40 stores program modules that realize each functional component of the server 40 according to this embodiment. Also, the storage device 1080 realizes the memory unit 470.

[0098] 19 is a flowchart illustrating the flow of processing performed by the server 40. When the change point information acquisition unit 410 acquires change point information in S410, the information expansion unit 420 expands the change point information into information for each location (S420). Then, the evaluation unit 430 determines whether or not the fourth condition is satisfied for the location to which the change point information belongs (S430). If the fourth condition is not satisfied (No in S430), the location is not extracted as a highly reliable change point, and the ND map update process (S490) is started.

[0099] If the fourth condition is met (Yes in S430), the evaluation unit 430 determines whether the fifth condition is met for the same point (S440). If the fifth condition is not met (No in S440), the point is not extracted as a highly reliable change point, and the ND map update process (S490) is started.

[0100] If the fifth condition is met (Yes in S440), the evaluation unit 430 determines whether the sixth condition is met for the same point (S450). If the sixth condition is not met (No in S450), the point is not extracted as a highly reliable change point, and the ND map update process (S490) is started.

[0101] If the sixth condition is met (Yes in S450), the change point extraction unit 440 extracts the point as a highly reliable change point (S460). Next, the maintenance point registration unit 450 registers the extracted point as a maintenance point in the database (S470). Furthermore, the flag assignment unit 460 determines that a non-recommended flag should be assigned to a voxel in the ND map that corresponds to the highly reliable change point extracted by the change point extraction unit 440 (S480). Then, the server 40 performs an ND map update process (S490).

[0102] FIG. 20 is a flowchart illustrating the flow of the ND map update process performed by the server 40 according to this embodiment. When the ND map update process starts, it is determined whether or not there are any voxels to which a deprecated flag should be assigned (S491). If it was determined in the previous S480 that a deprecated flag should be assigned, it is determined that there are voxels to which a deprecated flag should be assigned (Yes in S491). Then, the flag assigning unit 460 reads the ND map stored in the storage unit 470 and assigns a deprecated flag to the corresponding voxels (S492). Then, the flag assigning unit 460 updates the ND map stored in the storage unit 470 (S493). Then, the process proceeds to S494. On the other hand, if the previous S480 is not performed and it is not determined that there are any voxels to which a deprecated flag should be assigned (No in S491), S492 and S493 are not performed, and the process proceeds to S494.

[0103] In S494, it is determined whether new point cloud data for updating the ND map has been acquired. For example, when new point cloud data is acquired by a measurement and maintenance vehicle or the like, the data is stored in the storage unit 470. Therefore, whether new point cloud data has been acquired can be determined based on whether new data is stored in the storage unit 470. If new data has been acquired (Yes in S494), the server 40 uses the acquired point cloud data to create voxel data in the ND map. Then, the server 40 clears the deprecated flag of the newly created voxel (S495). The flag assignment unit 460 updates the ND map stored in the storage unit 470 (S496), and the ND map update process ends. If new data has not been acquired (No in S494), new voxel data is not created, and the ND map update process ends.

[0104] The ND map stored in the storage unit 470 is acquired by the estimation device 10 and used for estimating the self-position of the moving body 20.

[0105] In this embodiment, an example has been shown in which the estimation device 10 identifies an estimated change point and transmits the information to the server 40, but the allocation of processing between the estimation device 10 and the server 40 is not particularly limited. For example, point cloud data obtained by the sensor 210 of the moving object 20 and movement information of the moving object 20 may be transmitted from the estimation device 10 to the server 40, and the server 40 may identify an estimated change point. In this case, the server 40 also functions as the estimation device 10. Alternatively, one of the stages of the change point detection processing may be performed by the estimation device 10, and necessary information may be transmitted from the estimation device 10 to the server 40, and the remaining processing may be performed by the server 40.

[0106] In this embodiment, an example has been shown in which the change point detection process is performed in conjunction with the self-location estimation by the self-location estimation unit 140, but the change point detection process may be performed after the fact separately from the self-location estimation process. In that case, the first acquisition unit 110 may acquire the point cloud data by reading it out, which has been acquired in advance by the sensor 210 and stored in the storage unit. In addition, information indicating the speed of the moving object 20 at each time point, etc. may also be temporarily stored in the storage unit, and the estimation device 10 may read and use it.

[0107] As described above, according to this example, the same functions and effects as those of the embodiment can be obtained.

[0108] Example 2 FIG. 21 is a flowchart illustrating the flow of a change-point detection process performed by the estimation device 10 according to the second embodiment. The estimation device 10 according to the second embodiment is the same as the estimation device 10 according to the first embodiment, except for the points described below. In the second embodiment, the estimation unit 190 calculates the strength of correlation between the ratio value of the first region after the shift and the ratio value of the second region for multiple shift amounts, and identifies the shift amount that maximizes the strength of the correlation. The estimation unit 190 then identifies an estimated change point by comparing the identified shift amount with a reference value. Here, the reference value is determined based on the distance between the first region and the second region.

[0109] The estimation unit 190 according to this embodiment identifies an estimated change point using a cross-correlation function instead of using the first index and the second index described in the first embodiment. In the estimation device 10 according to this embodiment, when the change point detection process is started, the division unit 150 divides the point cloud data into data of a plurality of regions, similar to S801 described in the first embodiment. Then, the ratio value calculation unit 170 calculates the DDAR of each region (S811). In addition, the ratio value calculation unit 170 performs the above-mentioned filtering process and normalization on the DDAR.

[0110] Next, in S182, the shift unit 191 calculates a cross-correlation function between the DDAR in the first region and the DDAR in the second region. The shift unit 191 may or may not invert the DDAR in the first region and the DDAR in the second region before calculating the cross-correlation function. Specifically, the shift unit 191 calculates the cross-correlation function C(τ) using equation (7). Here, f(t) is the DDAR in the first region, g(t) is the DDAR in the second region, and τ is the time shift amount. The time shift amount τ can be converted into a position shift amount D by multiplying it by the velocity v of the moving object 20. The shift unit 191 can acquire the velocity v at the timing (e.g., time) when the point cloud data that is the basis of the DDAR is obtained from a velocity sensor provided on the moving object 20.

[0111]

number

[0112] Figure 22 is a graph showing an example of the cross-correlation function C(τ). In this figure, a high correlation value is obtained at the circled point. If the shift amount at this time is appropriate in light of the interval between the first and second regions, it can be determined that the result is due to a change point.

[0113] Returning to FIG. 21, once the cross-correlation function is calculated, in the next step S813, the correlation determination unit 192 identifies the maximum correlation value Cmax in the cross-correlation function, and further determines whether the correlation value is the maximum correlation value C max Next, in S814, the correlation determining section 192 determines whether or not the seventh and eighth conditions described below are satisfied.

[0114] The seventh condition is the maximum correlation value C maxis greater than a threshold value Th5. The eighth condition is that the following |(DL) / L| is smaller than a threshold value Th6. Here, D is the position shift amount D described above, and L is the area distance between the first area and the second area. The seventh condition indicates that the magnitude of correlation is large enough to allow the existence of a change point to be estimated. The eighth condition indicates that the difference in the characteristics of the DDAR of the first area and the DDAR of the second area is reasonable in light of the actual positional relationship between the areas. The area distance L between the first area and the second area is predetermined.

[0115] If at least one of the seventh and eighth conditions is not satisfied (No in S814), the change-point detection process ends. On the other hand, if both the seventh and eighth conditions are satisfied (Yes in S814), the change-point identification unit 193 identifies the position of the estimated change point. In addition, the reliability calculation unit 180 calculates the reliability R of the estimated change point. cp (S815). The change point identification unit 193 identifies the peak timing of the DDAR in the first region or the second region. Then, based on the estimated self-position of the moving object 20 at that timing, the change point identification unit 193 can identify a position in the direction of that region from the moving object 20 as the position of the estimated change point.

[0116] The reliability calculation unit 180 calculates the maximum correlation value C max and the reliability R using the value of |(DL) / L| cp Specifically, the reliability calculation unit 180 calculates the reliability R cp Here, d and e are predetermined coefficients. d is, for example, 0.1 or more and 1.0 or less, and e is, for example, 5.0 or more and 10.0 or less. That is, the reliability calculation unit 180 calculates the maximum correlation value C max The larger the reliability R cp The smaller the value of |(DL) / L|, the higher the reliability R cp The calculation results of Equation (8) and Equation (9) are cp4 , and R cp5 are all values ​​in the range of 0 to 1. Therefore, Rcp4 , and R cp5 The larger both of these are, the higher the reliability R cp also increases and approaches 1. Also, R cp4 , and R cp5 If either of these is close to 0, the reliability R cp becomes smaller.

[0117]

number

[0118] Next, in S816, information indicating the position, detection time, and reliability of the estimated change point is transmitted from the estimation device 10 to the server 40. Then, the change point detection process ends.

[0119] The method according to this embodiment also makes it possible to estimate the change point with high accuracy.

[0120] As described above, according to this example, the same functions and effects as those of the embodiment can be obtained.

[0121] Hereinafter, the embodiments and examples will be described with reference to experimental examples, but the embodiments and examples are not limited to the descriptions of these experimental examples.

[0122] (Experimental Example 1) In order to confirm the effect of change-point detection according to Example 1, voxel data of a certain building was deleted from the ND map, and an ND map that did not match the actual environment was generated.

[0123] Figure 23 shows the original ND map and the ND map with change points generated. The information about the building that existed in the area surrounded by the white ellipse on the original ND map has been deleted. Since the building actually exists, measurement data for that building is obtained. However, since the ND map does not contain voxel data for the building, the correspondence rate decreases. This situation is equivalent to a situation where a new building has been constructed but is not reflected in the ND map.

[0124] Figures 24(a) to 24(k) are diagrams showing various data obtained by a vehicle traveling counterclockwise around the course of the road shown in Figure 23. Figures 24(a) to 24(d) are graphs showing the difference between the estimated self-position, which is the result of NDT scan matching, and the vehicle position and orientation obtained for evaluation using an RTK-GPS (Real Time Kinematic Global Positioning System) as a reference. These results show that the accuracy of self-position estimation is hardly degraded even though the voxel data of buildings is deleted from the ND map.

[0125] Figures 24(e) to 24(g) show the number of all data points after downsampling of the point cloud data generated by the vehicle-mounted LIDAR, the number of data points associated with voxels on the ND map, and the overall ratio value DAR, respectively. It can be seen that the DAR decreases slightly in the 30-50 s range.

[0126] Figure 25 shows the relationship between the moving body 20 (vehicle) and areas (1) to (8) in this experimental example. Figures 24(h) to 24(k) show the DDAR for each area on the right side as seen from the vehicle. It can be seen that the area where the value is decreasing is gradually moving.

[0127] 26(a) to 26(c) are diagrams showing the results of performing the filtering, normalization, and limiting processes described in Example 1 on each DDAR. Specifically, FIG. 26(a) shows data obtained by performing filtering on each of FIGS. 24(h) to 24(k), FIG. 26(b) shows data obtained by performing normalization on each of the data in FIG. 26(a), and FIG. 26(c) shows data obtained by performing limiting on each of the data in FIG. 26(b). It can be seen that these processes relatively enhance the contrast of the DDAR-decreased areas due to differences between the point cloud data and the voxel data. Of these, DDAR(5) and DDAR(8), which have a large number of data points and clear change characteristics, were used for the subsequent processing.

[0128] Figures 27(a) to 27(e) show data from the change-point detection process. Figure 27(a) is a graph in which the time axis of DDAR(5) in Figure 26(c) is shifted based on the vehicle speed and the area interval (30 m). Figure 27(b) is a graph inverted from Figure 27(a). Figure 27(c) is a graph inverted from Figure 26(c) of DDAR(8). Figures 27(d) and 27(e) are graphs showing the first index M and the second index S calculated using the data shown in Figures 27(b) and 27(c), respectively.

[0129] W calculated based on the first indicator M t is 10.792 [s], and W d The threshold value Th1 was set to 0.04. N was 131, and M ave was calculated as 0.128. In addition, S calculated based on the second index S dev The result was 0.071. When the threshold value Th2 was set to 5.0, the threshold value Th3 was set to 0.05, and the threshold value Th4 was set to 0.2, the result was d ≥ Th2, M ave ≥ Th3, and S dev All the conditions of ≦Th4 were met, and the result was estimated as the change point. d , M ave , and S dev Using the reliability R cp was calculated. R cp1 =0.999, R cp2 =0.923, R cp3 =0.931, R cp =0.858. Note that a=0.1, b=20.0, c=1.0. R cp = 0.858 is a value close to 1, and can be said to be a high result for the change-point reliability. In addition, the values ​​of thresholds Th1, Th2, Th3, Th4, coefficient a, coefficient b, and coefficient c were set based on the results of multiple experiments.

[0130] Figure 28 shows the DDAR for each area on the left side of the vehicle. All data are after filtering, normalization, and limiting. Using this data, we also examined the area on the left side of the vehicle. Because the actual environment and the ND map match on the left side, the correspondence rate for areas (1) to (4) is high.

[0131] Figures 29(a) to 29(e) show data from the change-point detection process. Figure 29(a) is a graph in which the time axis of DDAR(1) in Figure 28 is shifted based on the vehicle speed and the area interval (30 m). Figure 29(b) is a graph inverted from Figure 29(a). Figure 29(c) is a graph inverted from Figure 28 (DDAR(4)). Figures 29(d) and 29(e) are graphs showing the first index M and the second index S calculated using the data shown in Figures 29(b) and 29(c), respectively.

[0132] In this result, the first index M was almost zero throughout, and there was no portion exceeding the threshold value Th1, resulting in no change point being detected.

[0133] As described above, it was confirmed that the change point can be detected using the method of Example 1.

[0134] (Experimental Example 2) The change-point detection process of Example 2 was carried out using the same ND map and point cloud data as in Experimental Example 1.

[0135] FIG. 30 is a diagram showing the cross-correlation function C(τ) calculated using the DDAR(5) and DDAR(8) shown in FIG. 26(c). This diagram shows the inverted DDAR(5) and DDAR(8) together. In the cross-correlation function C(τ) of FIG. 30, the maximum correlation value C max In the results of this figure, when D=30.505[m], C max= 17.758. The area interval L between area (5) and area (8) was 30 m, and |(DL) / L| was calculated to be 0.0168. When the threshold values ​​Th5 = 10.0 and Th6 = 0.2 were used for the judgment, C max ≧Th5 and |(DL) / L|≦Th6 were both satisfied, and the change point was detected.

[0136] Also, the calculated C max and |(DL) / L| to obtain the reliability R cp was calculated. R cp4 =0.999, R cp5 =0.874, R cp The result was =0.873. Note that d=0.4 and e=8.0. R cp = 0.873 is a value close to 1, and can be said to be a high result for the change-point reliability. The values ​​of the threshold value Th5, the threshold value Th6, the coefficient d, and the coefficient e were set based on the results of investigations conducted through multiple experiments.

[0137] FIG. 31 is a diagram showing the cross-correlation function C(τ) calculated using the DDAR(1) and DDAR(4) shown in FIG. 28. This diagram shows the inverted DDAR(1) and DDAR(4) together. In the cross-correlation function C(τ) of FIG. 31, the maximum correlation value C max In the results of this figure, when D=34.572[m], C max = 0.347. The area interval L between area (1) and area (4) was 30 m, and |(DL) / L| was calculated to be 0.1524. When the thresholds Th5 = 10.0 and Th6 = 0.2 were used for the judgment, |(DL) / L| ≦ Th6 was satisfied, but C max ≧Th5 was not satisfied. Therefore, the result was that the change point was not detected.

[0138] As described above, it was confirmed that the change point can be detected using the method of Example 2.

[0139] Although the embodiments and examples have been described above with reference to the drawings, these are merely examples of the present invention, and various configurations other than those described above can also be adopted. Below, examples of reference forms are added. 1. a first acquisition unit that acquires point cloud data at multiple timings obtained by a sensor mounted on the moving object; a second acquisition unit that acquires map information; a dividing unit that divides each point cloud data acquired by the first acquiring unit into a plurality of predetermined regions; a ratio value calculation unit that calculates a ratio value indicating a correspondence ratio between each data point in a first area and the map information for a second area different from the first area; an estimation unit that uses the strength of correlation between the ratio value of the first area shifted in time or position and the ratio value of the second area to identify an estimated change point where the content of the map information is estimated to differ from the actual situation. Estimation device. 2. In the estimation device according to 1. The ratio value calculation unit calculates the ratio value of the first region and the second region in time series. Estimation device. 3. In the estimation device according to 2., The estimation unit identifies the estimated change point using a strength of correlation between a post-shift ratio value obtained by shifting the time of the ratio value of the first region and a ratio value of the second region. Estimation device. 4. In the estimation device according to 3., The estimation unit identifies the estimated change point using the ratio value in the second region and the shifted ratio value obtained by shifting the ratio value in the first region by a time τ so as to align the time axis of the ratio value in the first region with the time axis of the ratio value in the second region. Estimation device. 5. In the estimation device according to 4., The time τ is calculated using the speed of the moving object. Estimation device. 6. 6. The estimation device according to any one of 3. to 5., The estimation unit calculating a first index by multiplying the post-shift ratio value by the ratio value of the second region; Calculating a second index by subtracting the post-shift ratio value of the first region and the ratio value of the second region from each other; The estimated change point is identified using the first index and the second index. Estimation device. 7. In the estimation device according to 1. or 2., The estimation unit Calculating the strength of the correlation for a plurality of shift amounts, and identifying the shift amount that maximizes the strength of the correlation; The estimated change point is identified by comparing the identified shift amount with a reference value. Estimation device. 8. 7. The estimation device according to claim 7, The reference value is determined based on the distance between the first area and the second area. Estimation device. 9. 9. The estimation device according to any one of 1. to 8., the ratio value calculation unit calculates an overall ratio value indicating a correspondence ratio between each data point of the point cloud data before division and the map information, and normalizes the ratio value of the first region and the ratio value of the second region by the overall ratio value; The estimation unit identifies the estimated change point using the normalized ratio value. Estimation device. 10. 10. The estimation device according to any one of 1. to 9., The map information is data that can be used as reference data for NDT scan matching. Estimation device. 11. an estimation device according to any one of 1. to 10.; a server; the estimation device transmits information indicating the identified estimated change point to a server; The server receiving information indicating a plurality of the estimated change points; The received information indicating the plurality of estimated change points is processed for each point, and the estimated change points where the content of the map information is highly likely to differ from the actual situation are extracted from the plurality of estimated change points. system. 12. 11. The system according to claim 11, the estimation device further calculates a reliability of the estimated change point and transmits the reliability to the server; The server further uses the received reliability to extract, from the plurality of estimated change points, the estimated change points at which the content of the map information is highly likely to differ from the actual situation. system. 13. 1. A computer-implemented estimation method comprising: a first acquisition step of acquiring point cloud data at multiple timings obtained by a sensor mounted on a moving object; a second acquisition step of acquiring map information; a dividing step of dividing each point cloud data acquired in the first acquiring step into a plurality of predetermined regions; a ratio value calculation step of calculating a ratio value indicating a correspondence ratio between each data point in a first area and the map information for a second area different from the first area; and an estimation step of identifying an estimated change point where the content of the map information is estimated to differ from the actual situation by using the strength of correlation between the ratio value of the first area shifted in time or position and the ratio value of the second area. Estimation method. 14. 13. A program for causing a computer to execute the estimation method described above.

[0140] This application claims priority based on Japanese Patent Application No. 2022-038992, filed on March 14, 2022, the disclosure of which is incorporated herein in its entirety. [Explanation of symbols]

[0141] 10 Estimation device 20 Mobile 40 servers 50 systems 110 First acquisition part 130 Second acquisition part 140 Self-position estimation part 150 Division 170 Ratio value calculation unit 180 Reliability calculation unit 190 Estimation Department 410 Change point information acquisition unit 420 Information Development Department 430 Evaluation Department 440 Change point extraction unit 450 Maintenance Points Registration Department 460 Flagging section 470 Storage section 1000 calculator

Claims

[Claim 1] a first acquisition unit that acquires point cloud data at a plurality of timings obtained by a sensor mounted on the moving object; a second acquisition unit that acquires map information; a dividing unit that divides each point cloud data acquired by the first acquiring unit into a plurality of predetermined regions; a ratio value calculation unit that calculates a ratio value indicating a correspondence ratio between each data point in each of the plurality of regions and the map information; an estimation unit that identifies an estimated change point where the content of the map information is estimated to differ from the actual situation based on the ratio value within each of the areas; Estimation device.

Images