Recognition system, recognition device, recognition method, recognition program

Abstract

To provide a recognition system that recognizes a moving body at high accuracy.SOLUTION: A processor of a recognition system that recognizes a target moving body movable in a scanning space scanned by a scanning device of a host moving body, is configured to execute: acquiring three-dimensional scanning data Dt indicating a scanning point group at which the scanning space is scanned; reading a three-dimensional dynamic map Mt indicating a mapping point group at which an object being present in the scanning space is mapped, from a storage medium; and generating recognition data Dr obtained by recognizing the target moving body among a plurality of static objects separated forward and backward in a scanning direction of the scanning device in the scanning space by clustering of the scanning point group based on identification information Ii for identifying a state of the scanning space for each of a plurality of three-dimensional voxels obtained by dividing the scanning space in the three-dimensional dynamic map Mt.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present disclosure relates to a recognition technology for recognizing a moving object. [Background technology]

[0002] Recognition techniques for recognizing movable objects in a scanned space scanned by a scanning device have become important in recent years. For example, the recognition technique disclosed in Patent Document 1 compares three-dimensional range image data with three-dimensional environmental map data, and recognizes moving objects by clustering points in the three-dimensional range image data that do not exist in the three-dimensional environmental map data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2018-173707 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the technology disclosed in Patent Document 1, there is a risk of erroneous scanning point clouds called ghost pixels occurring between multiple stationary objects spaced apart in the scanning direction of the laser range finder, which serves as a scanning device, due to overlapping of scanning echoes from each stationary object. In this case, it becomes difficult to separate the normal point cloud corresponding to a moving object that has entered between the stationary objects from the erroneous scanning point cloud by clustering, and the recognition accuracy of the entering moving object decreases. Such a problem may also occur between multiple moving objects that overlap in the scanning direction of the laser range finder.

[0005] An object of the present disclosure is to provide a recognition system that recognizes a moving object with high accuracy. Another object of the present disclosure is to provide a recognition device that recognizes a moving object with high accuracy. Yet another object of the present disclosure is to provide a recognition method that recognizes a moving object with high accuracy. Yet another object of the present disclosure is to provide a recognition program that recognizes a moving object with high accuracy. [Means for solving the problem]

[0006] The technical means of the present disclosure for solving the problems will be described below. Note that the claims and the reference characters in parentheses in this section indicate the corresponding relationship with the specific means described in the embodiments described later in detail, and do not limit the technical scope of the present disclosure.

[0007] A first aspect of the present disclosure is A recognition system having a processor (12) for recognizing a movable target moving object (9) in a scanning space (30) scanned by a scanning device (3) of a host moving object (2), comprising: The processor Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; Reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in a scanning space; and generating recognition data (Dr) that recognizes a target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0008] A second aspect of the present disclosure is A recognition device having a processor (12), configured to be mountable on a host moving body (2), and configured to recognize a target moving body (9) movable in a scanning space (30) scanned by a scanning device (3) of the host moving body, The processor Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; Reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in a scanning space; and generating recognition data (Dr) that recognizes a target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0009] A third aspect of the present disclosure is A method for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), the method comprising: Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; Reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in a scanning space; and generating recognition data (Dr) that recognizes a target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0010] A fourth aspect of the present disclosure is a recognition program including instructions stored in a storage medium (10) and executed by a processor (12) for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), The command is, Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes a target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0011] In these first to fourth aspects, recognition data is generated that recognizes a target moving object between multiple stationary objects separated forward and backward in the scanning direction of the scanning device in the scanning space. At this time, according to the first to fourth aspects, based on identification information for identifying the state of the scanning space for each of multiple three-dimensional voxels that divide the scanning space in the three-dimensional dynamic map, erroneous scanning point clouds can be excluded from the scanning point clouds that are clustered corresponding to the multiple stationary objects in the three-dimensional scanning data. Therefore, it is possible to recognize the target moving object that has entered between multiple stationary objects with high accuracy.

[0012] A fifth aspect of the present disclosure is a method for manufacturing a semiconductor device comprising: A recognition system having a processor (12) for recognizing a movable target moving object (9) in a scanning space (30) scanned by a scanning device (3) of a host moving object (2), comprising: The processor Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; Reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in a scanning space; The method is configured to generate recognition data (Dr) that recognizes a plurality of target moving objects that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) obtained by dividing the scanning space in the three-dimensional dynamic map.

[0013] A sixth aspect of the present disclosure is A recognition device having a processor (12), configured to be mountable on a host moving body (2), and configured to recognize a target moving body (9) movable in a scanning space (30) scanned by a scanning device (3) of the host moving body, The processor Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; Reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in a scanning space; The method is configured to generate recognition data (Dr) that recognizes a plurality of target moving objects that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) obtained by dividing the scanning space in the three-dimensional dynamic map.

[0014] A seventh aspect of the present disclosure is a method for manufacturing a semiconductor device comprising: A method for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), the method comprising: Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; Reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in a scanning space; and generating recognition data (Dr) that recognizes a plurality of target moving objects that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0015] An eighth aspect of the present disclosure is a method for manufacturing a semiconductor device comprising: a recognition program including instructions stored in a storage medium (10) and executed by a processor (12) for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), The command is, Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning a scanning space; reading from the storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes a plurality of target moving objects that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0016] In these fifth to eighth aspects, recognition data is generated that recognizes a plurality of target moving objects that overlap in the scanning space when viewed from the scanning device in the scanning direction. At this time, according to the fifth to eighth aspects, based on identification information for identifying the state of the scanning space for each of a plurality of three-dimensional voxels that divide the scanning space in the three-dimensional dynamic map, it is possible to exclude erroneous scanning points from the scanning point group that is clustered in correspondence with the plurality of target moving objects in the three-dimensional scanning data. Therefore, it is possible to recognize each target moving object that overlaps when viewed in the scanning direction with high accuracy. [Brief description of the drawings]

[0017] [Figure 1] 1 is a block diagram showing an overall configuration of a recognition system according to a first embodiment. [Diagram 2] 2 is a schematic diagram showing the relationship between a scanning device of a host moving body and a target moving body according to the first embodiment. FIG. [Diagram 3] 1 is a block diagram showing a functional configuration of a recognition system according to a first embodiment. [Figure 4] 4 is a flowchart showing a recognition flow according to the first embodiment. [Diagram 5] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 6] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 7] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 8] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 9] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 10] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 11] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 12] 11 is a flowchart showing a normal extraction subroutine of the recognition flow according to the first embodiment. [Figure 13] 10 is a flowchart showing a before and after extraction subroutine of the recognition flow according to the first embodiment. [Figure 14] 10 is a flowchart showing an overlap extraction subroutine of the recognition flow according to the first embodiment. [Figure 15] 10 is a flowchart showing a recognition flow fixed point extraction subroutine according to the first embodiment. [Figure 16] 10 is a flowchart showing a normal recognition subroutine of the recognition flow according to the first embodiment. [Figure 17]10 is a flowchart showing a context recognition subroutine of the recognition flow according to the first embodiment. [Figure 18] 10 is a flowchart showing an overlay recognition subroutine of the recognition flow according to the first embodiment. [Figure 19] FIG. 4 is a schematic diagram for explaining a recognition flow according to the first embodiment. [Figure 20] FIG. 11 is a schematic diagram for explaining a recognition flow according to a second embodiment. [Figure 21] FIG. 11 is a schematic diagram for explaining a recognition flow according to a second embodiment. [Figure 22] FIG. 11 is a schematic diagram for explaining a recognition flow according to a second embodiment. [Figure 23] 13 is a flowchart showing a high reflection extraction subroutine of the recognition flow according to the second embodiment. [Figure 24] 13 is a flowchart showing a highly reflective recognition subroutine of the recognition flow according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] First embodiment Hereinafter, multiple embodiments of the present disclosure will be described with reference to the drawings. In addition, by assigning the same reference numerals to corresponding components in each embodiment, duplicated descriptions may be omitted. In addition, when only a part of the configuration is described in each embodiment, the configuration of the other embodiment described above can be applied to the other parts of the configuration. Furthermore, in addition to the combination of configurations explicitly stated in the description of each embodiment, configurations of multiple embodiments can be partially combined together even if not explicitly stated, as long as there is no particular problem with the combination.

[0019] The recognition system 1 of the first embodiment shown in Fig. 1 recognizes a movable target moving object 9 in a scanning space 30 scanned by a scanning device 3 mounted on a host moving object 2 as shown in Fig. 2. Here, the host moving object 2 to which the recognition system 1 is applied is, for example, a vehicle such as an automobile that can travel on a road with an occupant on board. The target moving object 9 to be recognized by the recognition system 1 is of multiple types, for example, other vehicles other than the host moving object 2, motorcycles, people, animals, autonomous traveling robots, and remote-controlled traveling robots.

[0020] In the host mobile body 2, an autonomous driving mode is executed so that the level is classified according to the degree of manual intervention of the occupant in the dynamic driving task. The autonomous driving mode may be realized by autonomous driving control, such as conditional driving automation, high driving automation, or full driving automation, in which the system performs all dynamic driving tasks when activated. The autonomous driving mode may be realized by advanced driving assistance control, such as driving assistance or partial driving automation, in which the occupant performs some or all of the dynamic driving tasks. The autonomous driving mode may be realized by either one of the autonomous driving control and the advanced driving assistance control, a combination of them, or a switch between them.

[0021] The host mobile object 2 is equipped with a sensor system 4, a communication system 5, and an information presentation system 6, all of which are shown in Fig. 1. The sensor system 4 acquires sensor information about the external and internal worlds of the host mobile object 2 that can be used for driving control of the host mobile object 2 in the recognition system 1. To this end, the sensor system 4 is configured to include an external sensor 40 and an internal sensor 41.

[0022] The external sensor 40 acquires information on the external world, which is the surrounding environment of the host mobile body 2, as sensor information. The external sensor 40 includes a scanning device 3 that acquires sensor information by scanning a scanning space 30 of the external world of the host mobile body 2. The scanning device 3 is at least one type of device capable of generating three-dimensional scanning data Dt, which will be described later, among, for example, three-dimensional LiDAR (Light Detection and Ranging / Laser Imaging Detection and Ranging) and three-dimensional radar. The external sensor 40 other than the scanning device 3 may include at least one type of device, for example, a camera, a sonar, etc., that senses the external world of the host mobile body 2.

[0023] Here, the scanning device 3 acquires sensor information obtained by sensing objects present in a scanning space 30 (see FIG. 2) by scanning the scanning space 30 determined according to a viewing angle set toward the outside world of the host mobile body 2. In particular, the sensor information acquired by the scanning device 3 of the first embodiment is scanning data Dt that three-dimensionally represents the state of a group of scanning points scanned in the scanning space 30. The scanning data Dt includes three-dimensional state values ​​related to at least one of distance, azimuth angle, position coordinates, speed, and beam reflection intensity. Among the state values ​​included in the scanning data Dt, the distance represents a measurement value by dTOF (direct Time Of Flight) based on the flight time from irradiation of the scanning beam to reception of the reflected beam. Among the state values ​​included in the scanning data Dt, the azimuth angle represents a scanning direction that changes to at least one of the horizontal direction and the vertical direction with respect to the scanning space 30.

[0024] The internal sensor 41 acquires information on the internal world, which is the internal environment of the host mobile body 2, as sensor information. The internal sensor 41 may include a physical quantity detection type that detects a specific physical quantity of motion in the internal world of the host mobile body 2. The physical quantity detection type internal sensor 41 is at least one of, for example, a driving speed sensor, an acceleration sensor, and a gyro sensor. The internal sensor 41 may include an occupant detection type that detects a specific state of an occupant in the internal world of the host mobile body 2. The occupant detection type internal sensor 41 is at least one of, for example, a driver status monitor (registered trademark), a biological sensor, a seating sensor, an actuator sensor, and an in-vehicle equipment sensor.

[0025] The communication system 5 acquires communication information usable for driving control of the host mobile body 2 in the recognition system 1 by wireless communication. The communication system 5 may include a V2X type that transmits and receives communication signals to and from a V2X system existing in the external world of the host mobile body 2. The V2X type communication system 5 is, for example, at least one of a Dedicated Short Range Communications (DSRC) communication device and a Cellular V2X (C-V2X) communication device. The communication system 5 may include a positioning type that receives positioning signals from a Global Navigation Satellite System (GNSS) artificial satellite existing in the external world of the host mobile body 2. The positioning type communication system 5 is, for example, a GNSS receiver. The communication system 5 may include a terminal communication type that transmits and receives communication signals to and from a terminal existing in the internal world of the host mobile body 2. The terminal communication type communication system 5 is, for example, at least one of a Bluetooth (registered trademark) device, a Wi-Fi (registered trademark) device, and an infrared communication device.

[0026] The information presentation system 6 presents notification information to the occupant in the host mobile body 2. The information presentation system 6 may be of a visual stimulation type that stimulates the occupant's vision through a display. The visual stimulation type information presentation system 6 is, for example, at least one of a head-up display (HUD), a multi-function display (MFD), a combination meter, a navigation unit, etc. The information presentation system 6 may be of an auditory stimulation type that stimulates the occupant's hearing through sound. The auditory stimulation type information presentation system 6 is, for example, at least one of a speaker, a buzzer, a vibration unit, etc.

[0027] The recognition system 1 is connected to a sensor system 4, a communication system 5, and an information presentation system 6 via at least one of, for example, a LAN (Local Area Network) line, a wire harness, an internal bus, and a wireless communication line. The recognition system 1 is configured to include at least one dedicated computer.

[0028] The dedicated computer constituting the recognition system 1 may be a recognition control ECU (Electronic Control Unit) that controls the recognition of objects present in the scanning space 30 based on the scan data Dt as sensor information by the scanning device 3. Here, the recognition control ECU may have a function of integrating sensor information from a plurality of external sensors 40 including the scanning device 3. The dedicated computer constituting the recognition system 1 may be a driving control ECU that controls the driving of the host mobile body 2. The dedicated computer constituting the recognition system 1 may be a navigation ECU that navigates the travel route of the host mobile body 2. The dedicated computer constituting the recognition system 1 may be a locator ECU that estimates the self-state quantity including the self-position of the host mobile body 2. The dedicated computer constituting the recognition system 1 may be an HCU (Human Machine Interface (HMI) Control Unit) that controls the presentation of information by the information presentation system 6 in the host mobile body 2. The dedicated computer constituting the recognition system 1 may be a computer other than the host mobile body 2 that constitutes, for example, an external center or a mobile terminal that can communicate with the communication system 5.

[0029] The dedicated computer constituting the recognition system 1 has at least one memory 10 and one processor 12. The memory 10 is at least one type of non-transitory tangible storage medium, such as a semiconductor memory, a magnetic medium, an optical medium, etc., that non-temporarily stores computer-readable programs and data, etc. The processor 12 includes at least one type of core, such as a central processing unit (CPU), a graphics processing unit (GPU), a reduced instruction set computer (RISC)-CPU, a data flow processor (DFP), and a graph streaming processor (GSP).

[0030] In the recognition system 1, the memory 10 stores map information that can be used for driving control of the host mobile body 2. The memory 10 acquires and stores the latest map information by communication with an external center through, for example, a V2X type communication system 5. In particular, the map information in the first embodiment is a three-dimensional dynamic map Mt as digital data of a high-precision map that is three-dimensionally digitized so as to represent the driving environment of the host mobile body 2. The dynamic map Mt represents the state of a mapping point cloud in which objects at fixed positions existing in the scanning space 30 of the scanning device 3 are mapped. The dynamic map Mt includes three-dimensional state values ​​related to at least one of, for example, the position coordinates, distance, azimuth angle, and shape of the target. Objects to be mapped by the dynamic map Mt are, for example, a plurality of types of objects located at at least fixed points among roads, signs, traffic lights, structures, railroad crossings, vegetation, space division objects, space division lines, and marking lines.

[0031] In the recognition system 1, the processor 12 executes a plurality of instructions included in a recognition program stored in the memory 10 to recognize a target moving object 9 movable in a scanning space 30 scanned by the scanning device 3 of the host moving object 2. In this way, the recognition system 1 constructs a plurality of functional blocks for recognizing the target moving object 9 in the scanning space 30. The plurality of functional blocks constructed in the recognition system 1 include a matching block 100 and a clustering block 110 as shown in FIG.

[0032] The flow of the recognition method (hereinafter referred to as the recognition flow) in which the recognition system 1 recognizes the target moving object 9 in the scanning space 30 by cooperation of these blocks 100, 110 will be described below with reference to Fig. 4. The algorithm cycle of the recognition flow is repeatedly executed while the host moving object 2 is running. Note that each "S" in the recognition flow means multiple steps executed by multiple commands included in the recognition program.

[0033] In S101, the matching block 100 acquires scan data Dt from the scanning device 3. In S102, the matching block 100 reads out a dynamic map Mt corresponding to the acquisition point of the scan data Dt by S101 from the memory 10. At this time, the acquisition point of the scan data Dt may be determined by self-position estimation based on information acquired by the sensor system 4 and the communication system 5, for example.

[0034] In the dynamic map Mt read out in S101, three-dimensional voxels 300 obtained by dividing the scanning space 30 into a plurality of voxels are set as shown in FIG. 5. In the dynamic map Mt, each voxel 300 is defined as a three-dimensional lattice shape of a cube or a rectangular parallelepiped having six sides along the three-dimensional absolute coordinate axis assigned to the scanning space 30. However, in the vicinity of the outermost edge of the scanning space 30 determined according to the viewing angle of the scanning device 3, the voxel 300 may be defined as a partial shape of the three-dimensional lattice that is a part of a cube or a rectangular parallelepiped. The lattice size of each voxel 300 is set to either the same size or a plurality of different sizes as long as no gap space is generated between each voxel 300 in the scanning space 30.

[0035] As shown in Fig. 6, in the dynamic map Mt, identification information Ii for identifying the state of the scanning space 30 is assigned to each set voxel 300 in an individual correspondence. The identification information Ii includes an index number Iii as an identification ID associated with the position coordinates of each voxel 300. The identification information Ii includes distribution information Iis representing a distribution feature of a mapping point cloud corresponding to an object in the scanning space 30 for each voxel 300. The distribution information Iis may represent the presence or absence of a point cloud in the scanning space 30 for each voxel 300 by a value exceeding 0 representing a feature of the point cloud distribution, such as a Mahalanobis distance, and a value of 0.

[0036] Furthermore, in the dynamic map Mt, the identification information Ii includes type information Iik representing the type of the scanning space 30 for each voxel 300. As shown in Figs. 6 and 7, the type information Iik represents a type that identifiably characterizes the environmental state of the scanning space 30 by a type number as an identification ID.

[0037] Here, the type information Iik in the first embodiment includes at least one type number (number 5 in the example of Figs. 6 and 7) to identify an environmental state in which the probability of occurrence of erroneous scanning points due to the occurrence of ghost pixels in the scan data Dt exceeds a threshold or is outside the allowable range above the threshold. Here, the environmental state in which the probability of occurrence of erroneous scanning points is outside the allowable range corresponds to a front-rear recognition state Sg in which multiple stationary objects 8 exist in the front and rear of the scanning direction of the scanning device 3 in the scanning space 30 as shown in Fig. 8, and thus erroneous recognition due to layered ghost pixels (GP in Fig. 8) is likely to occur. Note that Fig. 8 illustrates a guardrail and a standing wall as multiple stationary objects 8 in the front and rear of the scanning direction.

[0038] In addition, the type information Iik of the first embodiment includes at least one type number (numbers 6 to 8, 10 to 15 in the examples of Figs. 6 and 7) in order to identify an environmental state that may cause ghost pixels or point cloud overlap in the scan data Dt. Here, the environmental state that may cause ghost pixels or point cloud overlap corresponds to an overlap recognition state Ss in which multiple target moving objects 9 are likely to overlap in the scanning space 30 as viewed in the scanning direction of the scanning device 3 as shown in Figs. 9 and 10, and thus erroneous recognition is likely to occur. Note that Fig. 9 illustrates other vehicles passing each other in different directions as multiple target moving objects 9 overlapping as viewed in the scanning direction on a narrow path that can be traveled in both directions. Meanwhile, Fig. 10 illustrates multiple people waiting at a railroad crossing as multiple target moving objects 9 overlapping as viewed in the scanning direction around the railroad crossing.

[0039] Furthermore, the type information Iik in the first embodiment includes at least one type number (number 9 in the example of Figs. 6 and 7) to identify an environmental state in which an object at a fixed position may move in the scan data Dt. Here, the environmental state in which an object at a fixed position may move corresponds to a fixed-point recognition state Sf in which a vegetation 7 at a fixed position moves in the wind as shown in Fig. 11 and is likely to be erroneously recognized as a target moving object 9. The fixed-point recognition state Sf can also be said to be a state in which a vegetation 7 at a fixed position is likely to be erroneously recognized as a target moving object 9 due to growth or felling of the vegetation 7 at the fixed position.

[0040] In S102 shown in Fig. 4, the matching block 100 may resize the grid size of each voxel 300 set in the dynamic map Mt in the memory 10 to a value optimized for recognizing the target moving object 9 according to the driving scene based on, for example, information acquired by the sensor system 4 and the communication system 5. At this time, the index number Iii of the identification information Ii is reassigned to each voxel 300 after resizing. The distribution information Iis of the identification information Ii is reassigned by converting the distribution feature amount for each voxel 300 before resizing into the distribution feature amount for each voxel 300 after resizing. The type information Iik of the identification information Ii is reassigned a type number corresponding to the environmental state after resizing by sorting or integrating the environmental state corresponding to the type number for each voxel 300 before resizing as the environmental state for each voxel 300 after resizing. Here, particularly in the case of integration, it is preferable to set a priority in advance for each type of environmental state, so that different environmental states in the multiple voxels 300 before resizing are integrated into an environmental state with a high priority.

[0041] The setting of each voxel 300 in the dynamic map Mt and the assignment of the identification information Ii described above may be performed by the generator of the latest map information. The setting of each voxel 300 in the dynamic map Mt and the assignment of the identification information Ii may be performed based on the past or current scan data Dt acquired in S101 by a voxel setting flow separate from this recognition flow.

[0042] 4, in S103, the matching block 100 matches the scan data Dt acquired in S101 with the dynamic map Mt read out in S102. At this time, the matching block 100 may perform map matching for the scan data Dt for each voxel 300 set in the dynamic map Mt based on, for example, an NDT (Normal Distribution Transform) algorithm.

[0043] In S104, the matching block 100 identifies, in the dynamic map Mt read out in S102, a plurality of voxels 300 corresponding to each position coordinate of the scanning point cloud in the scan data Dt map-matched in S103. In S105, the matching block 100 acquires, from the dynamic map Mt read out in S102, the identification information Ii assigned to each of the voxels 300 identified in S104 (hereinafter, referred to as the specific voxels 300). In S106, the matching block 100 performs a candidate extraction process for each specific voxel 300, according to at least the type information Iik of the identification information Ii acquired in S105, for the scanning point cloud in the scan data Dt map-matched in S103. A specific example of the candidate extraction process will be described below.

[0044] For a specific voxel 300 whose type information Iik acquired in S105 indicates a type number other than the states Sg, Ss, and Sf, a normal extraction subroutine is executed in S106 to perform candidate extraction processing of S111 to S113 shown in Fig. 12. Specifically, in S111, the matching block 100 compares the distribution feature represented by the distribution information Iis of the identification information Ii acquired in S105 with the distribution feature in the specific voxel 300 of the scanning point group in the scan data Dt map-matched in S103. As a result, if it is determined that the distribution features are deviated from each other, the normal extraction subroutine proceeds to S112. Here, the determination of the deviation between the distribution features may be made when the deviation between the distribution features exceeds or is equal to or greater than a threshold value, and the threshold value may be set to a value equal to or greater than 0.

[0045] In S112, the matching block 100 extracts a group of points within the specific voxel 300 as a group of candidate points Pc from among the group of scanning points in the scan data Dt map-matched in S103. Then, the matching block 100 in S112 stores list data Dc of the extracted group of candidate points Pc in the memory 10. With the above, the normal extraction subroutine ends.

[0046] On the other hand, if it is determined in S111 that the distribution features are not deviated from each other, the normal extraction subroutine proceeds to S113. In S113, the matching block 100 recognizes the point group within the specific voxel 300 among the scanning point group in the scan data Dt map-matched in S103 as the mapping point group of the object at the fixed point position, and excludes it from the candidate point group Pc stored in the memory 10. With the above, the normal extraction subroutine ends.

[0047] For the specific voxel 300 whose type information Iik acquired in S105 represents the type number of the front-rear recognition state Sg, the front-rear extraction subroutine is executed in S106 to perform the candidate extraction process of S121 to S123 shown in FIG. 13. Specifically, in S121, the matching block 100 compares the distribution feature represented by the distribution information Iis of the identification information Ii acquired in S105 with the distribution feature in the specific voxel 300 of the scanning point group in the scan data Dt map-matched in S103. As a result, if it is determined that the distribution feature values ​​are deviated from each other, the front-rear extraction subroutine proceeds to S122. Here, the determination of the deviation between the distribution feature values ​​is based on the normal extraction subroutine. However, the threshold value serving as the criterion for the deviation determination in the front-rear extraction subroutine may be adjusted so as to allow a deviation greater than the threshold value in the normal recognition subroutine.

[0048] In S122, the matching block 100 extracts a group of points within the specific voxel 300 as a group of candidate points Pc from among the group of scanning points in the scan data Dt map-matched in S103. Then, the matching block 100 in S122 stores list data Dc of the extracted group of candidate points Pc in the memory 10. This ends the before and after extraction subroutine.

[0049] On the other hand, if it is determined in S121 that the distribution features are not deviated from each other, the front-back extraction subroutine proceeds to S123. In S123, the matching block 100 recognizes the point group within the specific voxel 300 among the scanning point group in the scan data Dt map-matched in S103 as the mapping point group of the object at the fixed point position, and excludes it from the candidate point group Pc stored in the memory 10. With the above, the front-back extraction subroutine ends.

[0050] For the specific voxel 300 in which the type information Iik acquired in S105 represents the type number of the superposition recognition state Ss, the superposition extraction subroutine is executed in S106 to perform the candidate extraction process of S131 to S133 shown in FIG. 14. Specifically, in S131, the matching block 100 compares the distribution feature represented by the distribution information Iis of the identification information Ii acquired in S105 with the distribution feature in the specific voxel 300 of the scanning point group in the scan data Dt map-matched in S103. As a result, if it is determined that the distribution feature values ​​are deviated from each other, the superposition extraction subroutine proceeds to S132. Here, the determination of the deviation between the distribution feature values ​​is based on the normal extraction subroutine. However, the threshold value serving as the criterion for the deviation determination in the superposition extraction subroutine may be adjusted so as to allow a deviation greater than the threshold value in the normal recognition subroutine.

[0051] In S132, the matching block 100 extracts a group of points within the specific voxel 300 as a group of candidate points Pc from among the group of scanning points in the scan data Dt map-matched in S103. Then, the matching block 100 in S132 stores list data Dc of the extracted group of candidate points Pc in the memory 10. With the above, the overlap extraction subroutine is completed.

[0052] On the other hand, if it is determined in S131 that the distribution features are not deviated from each other, the overlap extraction subroutine proceeds to S133. In S133, the matching block 100 recognizes the point group within the specific voxel 300 among the scanning point group in the scan data Dt map-matched in S103 as the mapping point group of the object at the fixed point position, and excludes it from the candidate point group Pc stored in the memory 10. With the above, the overlap extraction subroutine ends.

[0053] For the specific voxel 300 whose type information Iik acquired in S105 represents the type number of the fixed-point recognition state Sf, a fixed-point extraction subroutine is executed in S106 to perform the candidate extraction process of S141 shown in Fig. 15. Specifically, in S141, the matching block 100 recognizes the points in the specific voxel 300 among the scanning points in the scan data Dt map-matched in S103 as a mapping point group of an object that may move at the fixed-point position, and excludes them from the candidate point group Pc stored in the memory 10. With the above, the fixed-point extraction subroutine is completed.

[0054] 4, in S107, the clustering block 110 performs a recognition process by clustering, in particular, according to the type information Iik of the identification information Ii acquired in S105, on the candidate point group Pc extracted in S106 and stored in the memory 10, for each specific voxel 300. A specific example of the recognition process will be described below.

[0055] For a specific voxel 300 whose type information Iik acquired in S105 indicates a type number other than the states Sg, Ss, and Sf, a normal recognition subroutine is executed in S107 to perform candidate extraction processing in S151 to S153 shown in FIG. 16. Specifically, in S151, the clustering block 110 sets a clustering range Rc for clustering the candidate point group Pc in the specific voxel 300 for the scan data Dt map-matched in S103 to a default range. Here, the clustering range Rc for clustering the candidate point group Pc is defined as a range with the same distance in each dimensional direction along the three-dimensional absolute coordinate axis assigned to the scanning space 30. Therefore, the default range of the clustering range Rc is defined as a fixed range (for example, 1 m) necessary to search for and properly recognize the target moving object 9 in an environmental state other than the states Sg, Ss, and Sf. However, the default range of the clustering range Rc may be adjusted to a variable range that increases as the distance to the specific voxel 300 increases.

[0056] In S152, the clustering block 110 clusters the scanning points whose Euclidean distance falls within the clustering range Rc set in S151 from the candidate point group Pc in the specific voxel 300 in the scan data Dt map-matched in S103. In S153, the clustering block 110 recognizes the target moving object 9 based on the clustered scanning point group. Here, the clustering block 110 in S153 may filter the clustered scanning point group in order to recognize the target moving object 9 by presumably distinguishing it from ghost pixels caused by sensor noise, etc. The estimation model used for filtering at this time may be a rule-based model or an AI model. In particular, when a rule-based model is used, filtering may be performed based on judgment indices such as the number of clustering points, clustering size, and spatial position of the scanning point group combined in S152. With the above, the normal recognition subroutine is terminated.

[0057] For the specific voxel 300 whose type information Iik acquired in S105 represents the type number of the front-rear recognition state Sg, a front-rear recognition subroutine is executed in S107 to perform the recognition process of S161 to S163 shown in Fig. 17. Specifically, in S161, the clustering block 110 adjusts the clustering range Rc for the scan data Dt map-matched in S103 to a smaller range (for example, 0.2 m or less) than the default range of the normal recognition subroutine. At this time, if a candidate point group Pc exists corresponding to a plurality of stationary objects 8 in front and behind in the scanning space 30 in the specific voxel 300, the range Rc in which the candidate point group Pc between the stationary objects 8 is clustered is reduced from the default range by the adjustment in S161.

[0058] In S162, the clustering block 110 clusters the scanning points whose Euclidean distance falls within the clustering range Rc set in S161 from the candidate points Pc in the specific voxel 300 in the scan data Dt map-matched in S103. In S163, the clustering block 110 recognizes the target moving object 9 based on the clustered scanning points. In this case, if the scanning points corresponding to the plurality of stationary objects 8 that are in front and behind in the scanning space 30 in the specific voxel 300 are clustered, it is possible to recognize the target moving object 9 (see FIG. 8) between the stationary objects 8. In S163, filtering similar to S153 may be performed to recognize the target moving object 9. However, in S163 of the front and rear recognition subroutine, in particular, in order to distinguish the target moving object 9 between the stationary objects 8 from the ghost pixels using a rule-based model, the judgment threshold for the above-mentioned judgment index may be adjusted to a side that increases the probability of excluding ghost pixels compared to the default value of the normal recognition subroutine. This completes the front-back recognition subroutine.

[0059] For the specific voxel 300 in which the type information Iik acquired in S105 indicates the type number of the superposition recognition state Ss, the superposition recognition subroutine is executed in S107 to perform the recognition process of S171 to S173 shown in Fig. 18. Specifically, in S171, the clustering block 110 adjusts the clustering range Rc for the scan data Dt map-matched in S103 to a smaller side (for example, 0.2 m or less) than the default range. At this time, when the candidate point groups Pc corresponding to the multiple target moving bodies 9 in the scanning space 30 in the specific voxel 300 are superimposed as viewed in the scanning direction of the scanning device 3, the range Rc in which the candidate point groups Pc of the target moving bodies 9 are clustered is reduced from the default range by the adjustment in S161.

[0060] In S172, the clustering block 110 clusters the scanning points whose Euclidean distance falls within the clustering range Rc set in S171 from the candidate points Pc in the specific voxel 300 in the scan data Dt map-matched in S103. Then, in S173, the clustering block 110 recognizes the target moving object 9 based on the clustered scanning points. At this time, when the scanning points corresponding to the multiple target moving objects 9 in the scanning space 30 in the specific voxel 300 are clustered in a superimposed manner as viewed in the scanning direction of the scanning device 3, it is possible to separately recognize the target moving objects 9 (see FIGS. 9 and 10). Therefore, in S173, filtering similar to that in S153 may be performed to recognize the target moving object 9. However, in S173 of the overlap recognition subroutine, in order to distinguish between the multiple target moving objects 9 overlapping in the scanning direction and ghost pixels using a rule-based model, the judgment threshold for the judgment index described above may be adjusted to a side that increases the probability of excluding ghost pixels compared to the default value in the normal recognition subroutine. With the above, the overlap recognition subroutine is completed.

[0061] For the specific voxel 300 whose type information Iik acquired in S105 represents the type number of the fixed-point recognition state Sf, the fixed-point extraction subroutine in S106 does not extract the candidate point group Pc. As a result, the recognition process in S107 is not performed for the specific voxel 300 in the fixed-point recognition state Sf.

[0062] As shown in FIG. 4, in S108, the clustering block 110 generates recognition data Dr to represent the result of the recognition processing of the target moving object 9 by the clustering in S107. Then, in S109, the clustering block 110 stores the generated recognition data Dr in the memory 10. At this time, the clustering block 110 may control the display of the generated or further stored recognition data Dr so that the generated or further stored recognition data Dr is displayed by the information presentation system 6 in the host moving object 2, for example, as shown in FIG. 19. The clustering block 110 may control the transmission of the generated or further stored recognition data Dr to the outside of the host moving object 2 (for example, an external center or another vehicle, etc.) by the communication system 5. With the above, the current execution of the recognition flow is completed. The recognition data Dr stored in the memory 10 is used for driving control of the host moving object 2.

[0063] (Action and effect) The effects of the first embodiment described above will be described below.

[0064] In the first embodiment, recognition data Dr is generated that recognizes a target moving object 9 between multiple stationary objects 8 separated forward and backward in the scanning direction of the scanning device 3 in the scanning space 30. At this time, according to the first embodiment, based on the identification information Ii for identifying the state of the scanning space 30 for each of multiple voxels 300 into which the scanning space 30 is divided in the dynamic map Mt, it is possible to exclude erroneous scanning points from the scanning points clustered in correspondence with the multiple stationary objects 8 in the scanning data Dt. Therefore, it is possible to recognize the target moving object 9 that has entered between the multiple stationary objects 8 with high accuracy.

[0065] According to the first embodiment, the clustering range Rc in which the scanning points corresponding to the plurality of stationary objects 8 in the scan data Dt are clustered is adjusted to the reduced side based on the identification information Ii. According to this, among the scanning points corresponding to the plurality of stationary objects 8, the erroneous scanning points with a low point cloud density are excluded from the clustering range Rc, while the scanning points of the target moving object 9 with a high point cloud density are included in the clustering range Rc and appropriately clustered. Therefore, it is possible to ensure reliability for high-precision recognition of the target moving object 9 that has entered between the plurality of stationary objects 8.

[0066] According to the first embodiment, the clustering range Rc for the scanning point cloud between each stationary object 8 is adjusted based on the type information Iik representing the type of the scanning space 30 among the identification information Ii. According to this, when the type represented by the type information Iik in the dynamic map Mt is a type that is highly likely to cause an erroneous scanning point cloud between stationary objects 8 in the scanning direction, it is possible to easily exclude the erroneous scanning point cloud from the clustering range Rc by adjusting to the reduction side according to the possibility. Therefore, it is possible to realize clustering that narrows down to the scanning point cloud of the target moving object 9 that has entered between multiple stationary objects 8, and to increase the reliability of the high-precision recognition of the target moving object 9.

[0067] In the first embodiment, recognition data Dr is generated by recognizing a plurality of target moving objects 9 overlapping in the scanning space 30 as viewed from the scanning device 3 in the scanning direction. At this time, according to the first embodiment, based on the identification information Ii for identifying the state of the scanning space 30 for each of a plurality of voxels 300 into which the scanning space 30 is divided in the dynamic map Mt, it is possible to exclude erroneous scanning point clouds from the scanning point clouds clustered in correspondence with the plurality of target moving objects 9 in the scanning data Dt. Therefore, it is possible to recognize each target moving object 9 overlapping in the scanning direction with high accuracy.

[0068] According to the first embodiment, the clustering range Rc in which the scanning point clouds corresponding to each target moving body 9 in the scan data Dt are clustered is adjusted to the reduced side based on the identification information Ii. According to this, among the scanning point clouds corresponding to the multiple target moving bodies 9, the erroneous scanning point clouds with low point cloud density are excluded from the clustering range Rc, while the scanning point clouds of each target moving body 9 with high point cloud density can be appropriately clustered by placing them in their own clustering ranges Rc. Therefore, it is possible to ensure reliability for high-precision recognition of each target moving body 9 that overlaps when viewed in the scanning direction.

[0069] According to the first embodiment, the clustering range Rc for the scanning point group of each target moving body 9 is adjusted based on the type information Iik representing the type of the scanning space 30, particularly, among the identification information Ii. According to this, when the type represented by the type information Iik in the dynamic map Mt is a type that is highly likely to be overlapped by the target moving body 9 when viewed in the scanning direction, it is possible to easily exclude the erroneous scanning point group from the clustering range Rc by adjusting to the reduction side according to the possibility. In addition, it is possible to easily fit the scanning point group of each target moving body 9 into each individual clustering range Rc by adjusting to the reduction side according to such a possibility. From these points, it is possible to increase the reliability of high-precision recognition of each target moving body 9 overlapping when viewed in the scanning direction.

[0070] According to the first embodiment, the candidate points Pc to be the scanning points to be clustered are extracted based on the type information Iik representing the type of the scanning space 30 among the identification information Ii. According to this, when the type represented by the type information Iik in the dynamic map Mt is a type such as a vegetation environment in which an object may move at a fixed point position, the type may be excluded from the candidate points Pc to be clustered. Therefore, it is possible to suppress a situation in which the recognition accuracy of the target moving object 9 is reduced due to the clustering of the scanning points corresponding to the object moving at the fixed point position.

[0071] According to the first embodiment, the candidate point group Pc to be the scanning point group to be clustered is extracted based on the distribution information Iis representing the feature amount of the point cloud distribution in the scanning space 30, particularly, from the identification information Ii. According to this, when the feature amount represented by the distribution information Iis in the dynamic map Mt deviates from the feature amount of the point cloud distribution in the scan data Dt, the candidate point group Pc of the specific voxel 300 corresponding to the point cloud distribution can be subjected to clustering. Therefore, it is possible to realize clustering limited to the scanning point group of the target moving object 9, which does not have a matching point group in the dynamic map Mt, and to increase the reliability of the high-precision recognition of the target moving object 9.

[0072] Second Embodiment The second embodiment is a modification of the first embodiment.

[0073] In the recognition flow of the second embodiment, the type information Iik of the dynamic map Mt read out by S101 includes at least one type number (number 16 in the example of FIG. 20) for identifying an environmental state that may cause ghost pixels in the scan data Dt, as shown in FIG. 20. Here, the environmental state that may cause ghost pixels corresponds to a high reflection recognition state Sr in which a highly reflective stationary object 80 whose reflectance to the scanning beam of the scanning device 3 is higher than the allowable range exists in the scanning space 30 as shown in FIG. 21 and FIG. 22, and erroneous recognition is likely to occur in front of the highly reflective stationary object 80 in the scanning direction. Therefore, the reflectance on the high reflection side is defined as the reflectance outside the allowable range of the reflectance that can be predicted not to cause erroneous recognition due to ghost pixels in front of the highly reflective stationary object 80. Note that FIG. 21 illustrates a sign and a mirror as examples of the highly reflective stationary object 80. Meanwhile, FIG. 22 illustrates a window glass in a building as an example of the highly reflective stationary object 80.

[0074] In the recognition flow of the second embodiment, for a specific voxel 300 whose type information Iik acquired in S105 represents a type number other than states Sg, Ss, Sf, and Sr, a normal extraction subroutine similar to that of the first embodiment is executed in S106. In the recognition flow of the second embodiment, for a specific voxel 300 whose type information Iik represents a type number of the front-back recognition state Sg, a front-back extraction subroutine similar to that of the first embodiment is executed in S106. In the recognition flow of the second embodiment, for a specific voxel 300 whose type information Iik represents a type number of the superimposed recognition state Ss, a superimposed extraction subroutine similar to that of the first embodiment is executed in S106. In the recognition flow of the second embodiment, for a specific voxel 300 whose type information Iik represents a type number of the fixed-point recognition state Sf, a fixed-point extraction subroutine similar to that of the first embodiment is executed in S106.

[0075] Further, in the recognition flow of the second embodiment, the high reflection extraction subroutine is executed in S106 to perform the candidate extraction process of S181 to S183 shown in FIG. 23 for the specific voxel 300 whose type information Iik represents the type number of the high reflection recognition state Sr. Specifically, in S181, the matching block 100 compares the distribution feature represented by the distribution information Iis of the identification information Ii acquired in S105 with the distribution feature in the specific voxel 300 of the scanning point group in the scan data Dt map-matched in S103. As a result, if it is determined that the distribution feature values ​​are deviated from each other, the high reflection extraction subroutine proceeds to S182. Here, the determination of the deviation between the distribution feature values ​​is similar to the normal extraction subroutine described in the first embodiment. However, the threshold value serving as the criterion for the deviation determination in the high reflection extraction subroutine may be adjusted so as to allow a deviation greater than the threshold value in the normal recognition subroutine.

[0076] After S106, for a specific voxel 300 whose type information Iik acquired in S105 in the recognition flow of the second embodiment indicates a type number other than states Sg, Ss, Sf, and Sr, a normal recognition subroutine similar to that of the first embodiment is executed in S107. For a specific voxel 300 whose type information Iik indicates a type number of the front-back recognition state Sg in the recognition flow of the second embodiment, a front-back recognition subroutine similar to that of the first embodiment is executed in S107. For a specific voxel 300 whose type information Iik indicates a type number of the superimposed recognition state Ss in the recognition flow of the second embodiment, a superimposed recognition subroutine similar to that of the first embodiment is executed in S107.

[0077] Furthermore, in the recognition flow of the second embodiment, a high-reflection recognition subroutine is executed in S107 to perform the recognition process of S191 to S193 shown in Fig. 24 for a specific voxel 300 whose type information Iik indicates the type number of the high-reflection recognition state Sr. Specifically, in S191, the clustering block 110 adjusts the clustering range Rc for the scan data Dt map-matched in S103 to a smaller side (for example, 0.2 m or less) than the default range. At this time, if a candidate point group Pc exists corresponding to an area in front of the highly reflective stationary object 80 in the scanning space 30 in the specific voxel 300, the range Rc in which the candidate point group Pc of the front area is clustered is reduced from the default range by the adjustment in S191.

[0078] In S192, the clustering block 110 clusters the scanning points whose Euclidean distance falls within the clustering range Rc set in S191 from the candidate point group Pc in the specific voxel 300 in the scan data Dt map-matched in S103. Then, in S193, the clustering block 110 recognizes the target moving object 9 based on the clustered scanning point group. At this time, when the scanning point group corresponding to the area in front of the highly reflective stationary object 80 in the scanning space 30 in the specific voxel 300 is clustered, it becomes possible to recognize the target moving object 9 in the front area (see Figs. 21 and 22). Then, in S193, filtering similar to S153 described in the first embodiment may be performed to recognize the target moving object 9. However, in S193 of the high reflection recognition subroutine, in order to distinguish between the target moving object 9 in the area in front of the highly reflective stationary object 80 and ghost pixels using a rule-based model, the judgment threshold for the judgment index explained in the first embodiment may be adjusted to a side that increases the probability of excluding ghost pixels compared to the default value in the normal recognition subroutine. With the above, the high reflection recognition subroutine is completed.

[0079] In this way, in the second embodiment, recognition data Dr is generated in which the target moving body 9 is recognized as being in front of the highly reflective stationary object 80, which has a reflectance to the scanning beam of the scanning device 3 that is outside the allowable range, as viewed in the scanning direction. At this time, according to the second embodiment, based on the identification information Ii for identifying the state of the scanning space 30 for each of the multiple voxels 300 into which the scanning space 30 is divided in the dynamic map Mt, it is possible to exclude erroneous scanning points from the scanning points clustered in front of the highly reflective stationary object 80 in the scanning data Dt. Therefore, it is possible to recognize the target moving body 9 that has entered in front of the highly reflective stationary object 80 with high accuracy.

[0080] According to the second embodiment, the clustering range Rc in which the scanning point cloud corresponding to the area in front of the highly reflective stationary object 80 in the scan data Dt is clustered is adjusted to the reduced side based on the identification information Ii. According to this, among the scanning point clouds corresponding to the area in front of the highly reflective stationary object 80, the erroneous scanning point cloud with a low point cloud density is excluded from the clustering range Rc, while the scanning point cloud of the target moving body 9 with a high point cloud density is included in the clustering range Rc and appropriately clustered. Therefore, it is possible to ensure reliability for high-precision recognition of the target moving body 9 that has entered in front of the highly reflective stationary object 80.

[0081] According to the second embodiment, the clustering range Rc for the scanning point group in front of the highly reflective stationary object 80 is adjusted based on the type information Iik representing the type of the scanning space 30, particularly, of the identification information Ii. According to this, when the type represented by the type information Iik in the dynamic map Mt is a type that is highly likely to cause an erroneous scanning point group in front of the highly reflective stationary object 80, it is possible to easily exclude the erroneous scanning point group from the clustering range Rc by adjusting to the reduction side according to the possibility. Therefore, it is possible to realize clustering that is limited to the scanning point group of the target moving body 9 that has entered in front of the highly reflective stationary object 80, and to increase the reliability of the high-precision recognition of the target moving body 9.

[0082] (Other embodiments) Although several embodiments have been described above, the present disclosure should not be construed as being limited to those embodiments, and can be applied to various embodiments within the scope not departing from the gist of the present disclosure.

[0083] In a modified example, the dedicated computer constituting the recognition system 1 may have at least one of a digital circuit and an analog circuit as a processor. Here, the digital circuit is at least one of ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), SOC (System on a Chip), PGA (Programmable Gate Array), CPLD (Complex Programmable Logic Device), etc. Furthermore, such a digital circuit may have a memory that stores a program.

[0084] In the modified example, in S106, either one of the before-and-after extraction subroutine and the overlap extraction subroutine may not be executed, and the normal extraction subroutine may be executed for the environmental state corresponding to either one of the before-and-after recognition state Sg and the overlap recognition state Ss. In this case, in S107, either one of the before-and-after recognition subroutine and the overlap recognition subroutine may not be executed, and the normal recognition subroutine may be executed for the environmental state corresponding to either one of the before-and-after recognition state Sg and the overlap recognition state Ss.

[0085] In the modified example, in S106, the fixed point extraction subroutine is not executed, and the normal extraction subroutine may be executed for the fixed point recognition state Sf corresponding to the fixed point extraction subroutine. In this case, in S107, the normal recognition subroutine may be executed for the fixed point recognition state Sf.

[0086] In the modified example, the host moving body to which the recognition system 1 is applied may be, for example, a traveling robot capable of autonomously traveling or remotely traveling to transport luggage or collect information, etc. In addition to the forms described so far, the above-mentioned embodiment and modified example may be implemented in the form of a processing circuit (e.g., a processing ECU, etc.) or a semiconductor device (e.g., a semiconductor chip, etc.) as a recognition device configured to be mountable on a host moving body and having at least one processor 12 and one memory 10.

[0087] (Additional remarks) This specification discloses the following technical ideas and combinations thereof.

[0088] (Technical thought 1) A recognition system having a processor (12) for recognizing a movable target moving object (9) in a scanning space (30) scanned by a scanning device (3) of a host moving object (2), comprising: The processor, Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes the target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0089] (Technical thought 2) generating the recognition data The recognition system described in technical idea 1 includes adjusting a clustering range (Rc) in which the scanning point group corresponding to the plurality of stationary objects in the 3D scan data is clustered to the reduced side based on the identification information.

[0090] (Technical Thought 3) generating the recognition data The recognition system according to technical idea 1 or 2 includes generating the recognition data by recognizing a plurality of target moving bodies that overlap when viewed in the scanning direction in the scanning space by clustering the scanning point cloud based on the identification information (Ii).

[0091] (Technical Thought 4) A recognition system having a processor (12) for recognizing a movable target moving object (9) in a scanning space (30) scanned by a scanning device (3) of a host moving object (2), comprising: The processor, Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes the target moving bodies that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

[0092] (Technical Thought 5) generating the recognition data The recognition system described in technical idea 3 or 4 includes adjusting a clustering range (Rc) in which the scanning point group corresponding to each of the target moving bodies in the 3D scanning data is clustered to a reduced side based on the identification information.

[0093] (Technical Thought 6) generating the recognition data The recognition system according to technical idea 2 or 5, further comprising adjusting the clustering range based on type information (Iik) representing the type of the scanning space among the identification information.

[0094] (Technical Thought 7) generating the recognition data The recognition system according to any one of technical ideas 1 to 6 includes generating recognition data in which the target moving object is recognized as being closer in the scanning direction than a highly reflective stationary object (80) in the scanning space whose reflectivity to the scanning beam of the scanning device is unacceptably high, by clustering the scanning point cloud based on the identification information.

[0095] (Technical Thought 8) generating the recognition data The recognition system described in Technical Idea 7 includes adjusting a clustering range (Rc) in which the scanning point group corresponding to the position in front of the highly reflective stationary object in the three-dimensional scanning data is clustered to a reduced side based on the identification information.

[0096] (Technical Thought 9) generating the recognition data The recognition system according to technical idea 8 includes adjusting the clustering range based on type information (Iik) representing the type of the scanning space among the identification information.

[0097] (Technical Thought 10) generating the recognition data The recognition system according to any one of technical ideas 1 to 9, comprising extracting the scanning point group to be clustered based on type information (Iik) representing the type of the scanning space among the identification information.

[0098] (Technical Thought 11) generating the recognition data The recognition system according to any one of technical ideas 1 to 10, comprising extracting the scanning point cloud to be clustered based on distribution information (Iis) representing a feature of the point cloud distribution in the scanning space from among the identification information.

[0099] (Technical Thought 12) The processor, The recognition system according to any one of Technical Ideas 1 to 11, further configured to store the generated recognition data in a storage medium (10).

[0100] (Technical Thought 13) The processor, The recognition system according to any one of Technical Ideas 1 to 12, further configured to control display of the generated recognition data.

[0101] The above-described technical ideas 1 to 13 may be realized in the form of an apparatus, a method, or a program. [Explanation of symbols]

[0102] 1: Recognition system, 2: Host mobile object, 3: Scanning device, 8: Stationary object, 9: Target mobile object, 10: Memory, 12: Processor, 30: Scanning space, 80: Highly reflective stationary object, 300: Three-dimensional voxel, Dr: Recognition data, Dt: Scanning data, Ii: Identification information, Iik: Classification information, Iis: Distribution information, Mt: Dynamic map, Rc: Clustering range

Claims

1. A recognition system having a processor (12) for recognizing a movable target moving object (9) in a scanning space (30) scanned by a scanning device (3) of a host moving object (2), comprising: The processor, acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes the target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

2. generating the recognition data The recognition system according to claim 1 , further comprising: adjusting a clustering range (Rc) in which the scanning points corresponding to a plurality of stationary objects in the three-dimensional scanning data are clustered to a reduced side based on the identification information.

3. generating the recognition data The recognition system of claim 1, further comprising: generating the recognition data by clustering the scanning point cloud based on the identification information (Ii) to recognize a plurality of target moving bodies that overlap when viewed in the scanning direction in the scanning space.

4. A recognition system having a processor (12) for recognizing a movable target moving object (9) in a scanning space (30) scanned by a scanning device (3) of a host moving object (2), comprising: The processor, acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes a plurality of target moving bodies that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

5. generating the recognition data The recognition system according to claim 4 , further comprising: adjusting a clustering range (Rc) in which the scanning points corresponding to each of the target moving bodies in the three-dimensional scanning data are clustered to a reduced side based on the identification information.

6. generating the recognition data The recognition system according to claim 2 or 5, further comprising adjusting the clustering range based on type information (Iik) representing a type of the scanning space among the identification information.

7. generating the recognition data The recognition system of claim 1 or 4, further comprising: generating the recognition data in which the target moving object is recognized as being closer in the scanning direction than a highly reflective stationary object (80) in the scanning space whose reflectivity to the scanning beam of the scanning device is unacceptably high, by clustering the scanning point cloud based on the identification information.

8. generating the recognition data The recognition system according to claim 7, further comprising adjusting a clustering range (Rc) in which the scanning points corresponding to a position in front of the highly reflective stationary object in the three-dimensional scanning data are clustered to a reduced side based on the identification information.

9. generating the recognition data The recognition system according to claim 8 , further comprising adjusting the clustering range based on type information (Iik) representing a type of the scanning space among the identification information.

10. generating the recognition data The recognition system according to claim 1 or 4, further comprising: extracting the scanning points to be clustered based on type information (Iik) representing a type of the scanning space among the identification information.

11. generating the recognition data The recognition system according to claim 1 or 4, further comprising: extracting the scanning point cloud to be clustered based on distribution information (Iis) representing a feature of a point cloud distribution in the scanning space from among the identification information.

12. The processor, The recognition system of claim 1 or 4, further configured to store the generated recognition data in a storage medium (10).

13. The processor, The recognition system of claim 1 or 4, further configured to control a display of the generated recognition data.

14. A recognition device having a processor (12), configured to be mountable on a host moving body (2), and configured to recognize a target moving body (9) movable in a scanning space (30) scanned by a scanning device (3) of the host moving body, comprising: The processor, acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes the target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

15. A recognition device having a processor (12), configured to be mountable on a host moving body (2), and configured to recognize a target moving body (9) movable in a scanning space (30) scanned by a scanning device (3) of the host moving body, comprising: The processor, acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes a plurality of target moving bodies that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying the state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

16. A method for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), the method comprising: acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) in which the target moving object is recognized among a plurality of stationary objects (8) spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) obtained by dividing the scanning space in the three-dimensional dynamic map.

17. A method for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), the method comprising: acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes a plurality of target moving bodies that overlap when viewed in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

18. A recognition program including instructions stored in a storage medium (10) and executed by a processor (12) for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), The instruction: Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes the target moving object among a plurality of stationary objects (8) that are spaced forward and backward in the scanning direction of the scanning device in the scanning space by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

19. A recognition program including instructions stored in a storage medium (10) and executed by a processor (12) for recognizing a movable target object (9) in a scanning space (30) scanned by a scanning device (3) of a host object (2), The instruction: Acquiring three-dimensional scanning data (Dt) representing a scanning point cloud obtained by scanning the scanning space; Reading from a storage medium (10) a three-dimensional dynamic map (Mt) representing a mapping point cloud obtained by mapping an object present in the scanning space; and generating recognition data (Dr) that recognizes a plurality of target moving bodies that overlap in the scanning space when viewed in the scanning direction of the scanning device by clustering the scanning point cloud based on identification information (Ii) for identifying a state of the scanning space for each of a plurality of three-dimensional voxels (300) that divide the scanning space in the three-dimensional dynamic map.

Images