Object detection device, object detection method, and computer program

Abstract

Provided are an object detection device, an object detection method, and a computer program that can improve the accuracy of moving body detection using a LiDAR. An object detection device (100) detects an object on the basis of measurement point data acquired by a LiDAR (11). The object detection device: extracts, from primary measurement point data that is measurement point data acquired by the LiDAR (11), a measurement point group for which the distance between neighboring measurement points is within a prescribed distance; excludes, from the primary measurement point data, a non-moving-body measurement point group that is, among the extracted measurement point groups, a measurement point group with the length in any direction being at least a prescribed length threshold value; extracts dynamic measurement points, which are measurement points obtained by excluding static measurement points that are measurement points overlapping with the primary measurement point data used in a previous processing cycle, from secondary measurement point data that is measurement point data obtained by excluding the non-moving-body measurement point group from the primary measurement point data; extracts, from the secondary measurement point data, a measurement point group for which the distance between neighboring measurement points is within a prescribed distance; identifies a first moving body measurement point group that comprises the dynamic measurement points and a second moving body measurement point group that comprises the dynamic measurement points and the static measurement points; and, through clustering processing, associates the identified first moving body measurement point group and second moving body measurement point group with moving body data for which detection was completed in or prior to the previous processing cycle.

Description

[Technical Field]

[0001] The present invention relates to an object detection device, an object detection method, and a computer program that are applied to a moving body such as a vehicle. [Background technology]

[0002] In recent years, moving objects such as vehicles have been equipped with object detection devices that use distance measurement sensors to detect the surrounding environment of the moving object. For example, vehicles execute adaptive cruise control (ACC) to automatically drive their own vehicle while maintaining a target distance from a preceding vehicle based on information about the detected surrounding environment, and collision safety functions to avoid collisions with obstacles, including preceding vehicles, or to mitigate impacts in the event of a collision. One known distance measurement sensor used in such object detection devices is LiDAR (Light Detection and Ranging), which uses optical waves such as lasers and infrared rays. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2021-163213 Summary of the Invention [Problem to be solved by the invention]

[0004] However, while LiDAR can measure the distance and direction to a detected object, it cannot measure the speed of the detected object like a radar sensor, which is also known as a distance sensor. As a result, it may mistakenly detect unevenness in the road surface or obstacles such as guardrails as moving objects, or may lose track of the moving object.

[0005] An object of the present invention is to provide an object detection device, an object detection method, and a computer program that can improve the detection accuracy of moving objects using LiDAR. [Means for solving the problem]

[0006] In order to solve the above problem, according to one aspect of the present invention, An object detection device (100) that detects an object based on data of measurement points acquired by a LiDAR (11), A filtering process is performed to extract a group of measurement points whose adjacent measurement points are within a predetermined distance from the primary measurement point data, which is data of the measurement points acquired by the LiDAR (11), and to exclude the non-moving body measurement point cloud from the primary measurement point data by defining the group of measurement points whose length in any direction is equal to or greater than a predetermined length threshold as a non-moving body measurement point cloud. a dynamic measurement point extraction process for extracting, as dynamic measurement points, measurement points from secondary measurement point data, which is data on measurement points obtained by excluding the non-moving body measurement point group from the primary measurement point data, excluding static measurement points that overlap with the primary measurement point data used in the previous processing cycle; a clustering process for extracting measurement point clouds from the secondary measurement point data where the distance between adjacent measurement points is within the predetermined distance, and identifying a first moving object measurement point cloud composed of the dynamic measurement points and a second moving object measurement point cloud composed of the dynamic measurement points and the static measurement points; a matching process for associating the first moving object measurement point cloud and the second moving object measurement point cloud identified by the clustering process with moving object data that has been detected up to the previous processing cycle; An object detection device is provided that performs the following.

[0007] In order to solve the above problems, according to another aspect of the present invention, An object detection method for detecting an object based on data of measurement points acquired by a LiDAR (11), Extracting a group of measurement points whose adjacent measurement points are within a predetermined distance from the primary measurement point data, which is data of the measurement points acquired by the LiDAR (11), and excluding the non-moving body measurement point group from the primary measurement point data by defining a group of measurement points whose length in any direction is equal to or greater than a predetermined length threshold as a non-moving body measurement point group (step S11); extracting, as dynamic measurement points, measurement points from secondary measurement point data, which is data of measurement points obtained by excluding the non-moving body measurement point group from the primary measurement point data, excluding static measurement points, which are measurement points that overlap with the primary measurement point data used in the previous processing cycle (step S15); A clustering process (step S17) extracts measurement point clouds from the secondary measurement point data where the distance between adjacent measurement points is within the predetermined distance, and identifies a first moving object measurement point cloud composed of the dynamic measurement points and a second moving object measurement point cloud composed of the dynamic measurement points and the static measurement points; a matching process (steps S19 to S29) for associating the first moving object measurement point cloud and the second moving object measurement point cloud identified by the clustering process with moving object data that has been detected up to the previous processing cycle; An object detection method is provided, including:

[0008] In order to solve the above problems, according to another aspect of the present invention, A computer program that causes a computer to execute a process of detecting an object based on data of measurement points acquired by a LiDAR (11), Extracting a group of measurement points whose adjacent measurement points are within a predetermined distance from the primary measurement point data, which is data of the measurement points acquired by the LiDAR (11), and excluding the non-moving body measurement point group from the primary measurement point data by defining a group of measurement points whose length in any direction is equal to or greater than a predetermined length threshold as a non-moving body measurement point group (step S11); extracting, as dynamic measurement points, measurement points from secondary measurement point data, which is data of measurement points obtained by excluding the non-moving body measurement point group from the primary measurement point data, excluding static measurement points, which are measurement points that overlap with the primary measurement point data used in the previous processing cycle (step S15); A clustering process (step S17) extracts measurement point clouds from the secondary measurement point data where the distance between adjacent measurement points is within the predetermined distance, and identifies a first moving object measurement point cloud composed of the dynamic measurement points and a second moving object measurement point cloud composed of the dynamic measurement points and the static measurement points; a matching process (steps S19 to S29) for associating the first moving object measurement point cloud and the second moving object measurement point cloud identified by the clustering process with moving object data that has been detected up to the previous processing cycle; A computer program is provided to cause the computer to perform a process including: [Effects of the Invention]

[0009] As described above, according to the present invention, it is possible to improve the detection accuracy of a moving object using LiDAR. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a schematic diagram illustrating an example of the configuration of a vehicle as an example of a moving body according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram illustrating an example of the configuration of an object detection device according to the embodiment. [Figure 3] 10 is a flowchart showing a processing operation performed by the object detection device according to the embodiment. [Figure 4] 10 is a flowchart showing a moving object detection processing operation performed by the object detection device according to the embodiment. [Figure 5] FIG. 10 is an explanatory diagram showing a filtering process performed by the object detection device according to the embodiment. [Figure 6] FIG. 10 is an explanatory diagram showing a filtering process performed by the object detection device according to the embodiment. [Figure 7] 10A and 10B are explanatory diagrams showing a dynamic measurement point extraction process and a clustering process performed by the object detection device according to the embodiment. [Figure 8] 10A and 10B are explanatory diagrams showing a dynamic measurement point extraction process and a clustering process performed by the object detection device according to the embodiment. [Figure 9] FIG. 2 is an explanatory diagram showing a matching process performed by the object detection device according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0012] <1. Example of mobile configuration> First, a configuration example of a moving body to which the object detection device according to this embodiment can be applied will be described.

[0013] The object detection device according to this embodiment can be applied to various moving bodies such as vehicles such as four-wheeled automobiles and motorcycles, as well as ships, aircraft, robots, etc. In this embodiment, an example in which the object detection device is applied to a four-wheeled automobile as a moving body will be described.

[0014] FIG. 1 is a schematic diagram showing an example of the configuration of a vehicle 1 as an example of a moving body. The vehicle 1 shown in FIG. 1 is configured as a two-wheel drive vehicle 1 that transmits drive torque output from a drive power source 20, such as an internal combustion engine or a drive motor, to the left and right front wheels. The vehicle 1 may also be a four-wheel drive vehicle that transmits drive power to the front, rear, left and right wheels. Furthermore, if the vehicle 1 is an electric vehicle or a hybrid electric vehicle, the vehicle 1 is equipped with a secondary battery that stores power supplied to the drive motor, a generator that generates power to be supplied to the drive motor and to charge the battery, an inverter device that controls the drive of the drive motor, and the like.

[0015] The vehicle 1 is equipped with a driving force source 20, an electric steering device 25, and a brake fluid pressure control unit 30 as devices for controlling the traveling of the vehicle 1. The driving force source 20 outputs driving torque that is transmitted to the wheels via a transmission and a differential mechanism (not shown). The operation of the driving force source 20 and the transmission is controlled by a vehicle control device 40.

[0016] The electric steering device 25 includes an electric motor and a gear mechanism (not shown), and adjusts the steering angle of the left and right front wheels by being controlled by the vehicle control device 40. During manual driving, the vehicle control device 40 controls the electric steering device 25 based on the steering angle of the steering wheel operated by the driver. During automatic driving, the vehicle control device 40 controls the electric steering device 25 based on a target steering angle.

[0017] The brake fluid pressure control unit 30 adjusts the hydraulic pressure supplied to the brake calipers provided on each wheel to generate braking force. The operation of the brake fluid pressure control unit 30 is controlled by a vehicle control device 40. If the vehicle 1 is an electric vehicle or a hybrid electric vehicle, the brake fluid pressure control unit 30 is used in conjunction with regenerative braking using a drive motor.

[0018] The vehicle control device 40 includes one or more electronic control devices that control the operation of the driving force source 20, the electric steering device 25, and the brake fluid pressure control unit 30. The vehicle control device 40 is configured to be able to acquire signals transmitted from the object detection device 50, and is configured to be able to execute automatic driving control of the vehicle 1. Note that automatic driving control includes emergency brake control and ACC (Adaptive Cruise Control). Furthermore, when the vehicle 1 is being manually driven, the vehicle control device 40 acquires information on the amount of driving operation by the driver, and controls the operation of the driving force source 20, the electric steering device 25, and the brake fluid pressure control unit 30.

[0019] The vehicle 1 is equipped with a LiDAR 11 and a position detection sensor 13. In the vehicle 1 shown in FIG. 1, the LiDAR 11 is installed in the center of the front part of the vehicle 1. The LiDAR 11 is installed with its illumination axis facing forward in the vehicle length direction. The LiDAR 11 is installed in an aligned state, for example, so that a plane formed by the illumination axis and an axis perpendicular to the illumination axis is perpendicular to the vehicle height direction. However, the installation position of the LiDAR 11 and the axial direction of the illumination axis are not limited to this example, and the LiDAR 11 may be installed at any position and facing any direction. Furthermore, the number of LiDARs 11 mounted on the vehicle 1 is not limited to one, and each LiDAR 11 is set to face its illumination axis in a predetermined direction.

[0020] The LiDAR 11 transmits optical waves such as laser or infrared light toward a predetermined angle range (angular resolution), receives reflected waves of the optical waves, and calculates the position of the reflection point (hereinafter referred to as the "measurement point") based on information about the transmitted and received waves. The information about the position of the measurement point includes information about the distance from the LiDAR 11 to the measurement point, and information about the angle (hereinafter also referred to as the "azimuth angle") between the irradiation axis of the LiDAR 11 and the direction in which the measurement point is located.

[0021] Specifically, the LiDAR 11 emits optical waves within a predetermined angular range centered on the irradiation axis and receives reflected waves during each processing cycle set at a predetermined time interval. For example, the LiDAR 11 emits optical waves by scanning multiple laser emitters arranged to form a predetermined angular range in the horizontal direction in a vertical direction. Alternatively, the LiDAR 11 may emit optical waves by scanning multiple laser emitters arranged to form a predetermined angular range in the vertical direction in a horizontal direction. The LiDAR 11 calculates the distance to the measurement point based on the time from transmitting the optical waves to receiving the reflected waves. The LiDAR 11 also calculates the azimuth angle of the measurement point based on the receiving direction of the measurement point. The LiDAR 11 performs conventionally known processing on all reflected waves received during each processing cycle to calculate the position and velocity of the measurement point and transmits the results to the object detection device 50.

[0022] The position detection sensor 13 is, for example, a GNSS (Global Navigation Satellite System) sensor represented by a GPS (Global Positioning System) sensor, and receives satellite signals transmitted from satellites to acquire position information of the longitude and latitude of the position detection sensor 13 on a world coordinate system. The position detection sensor 13 transmits the acquired position information and information on a reference direction set in the position detection sensor 13 to the object detection device 50 as position data. The reference direction is a direction that defines one axis of a vehicle coordinate system whose origin is the installation position of the position detection sensor 13, and is set to, for example, a direction that coincides with the forward direction in the vehicle length direction of the vehicle 1.

[0023] The object detection device 50 executes a process of detecting objects present around the vehicle 1 based on data of measurement points transmitted from the LiDAR 11. In particular, the object detection device 50 according to this embodiment is configured to be able to detect a moving object moving at a predetermined speed based on data of measurement points of the LiDAR 11 that does not include speed information. The object detection device 50 according to this embodiment will be described in detail below.

[0024] <2. Object detection device> The object detection device 50 is configured to be able to perform the process of detecting a moving object moving at a predetermined speed based on the data of the measurement point transmitted from the LiDAR 11, and the process of detecting a stationary object that is stopped or fixed or installed and has no speed.

[0025] (2-1. Configuration example) 2 is a block diagram showing an example configuration of the object detection device 50. The object detection device 50 is configured as a microcomputer or a microprocessor unit including an arithmetic processing device such as a CPU (Central Processing Unit). Some or all of these devices may be configured with updatable firmware or the like, or may be program modules executed by commands from the CPU or the like.

[0026] The object detection device 50 includes a communication unit 51, a processing unit 53, and a storage unit 55. The communication unit 51 is an interface for transmitting and receiving signals or messages to and from the LiDAR 11, the position detection sensor 13, and the vehicle control device 40, and is configured to comply with one or more communication protocol standards such as CAN (Controller Area Network) or LIN (Local Internet). The processing unit 53 includes an arithmetic processing device and executes various types of arithmetic processing.

[0027] The storage unit 55 includes a storage element such as a random access memory (RAM) or a read only memory (ROM), or a recording medium such as a hard disk drive (HDD) or a solid state drive (SSD). The storage unit 55 stores data such as a computer program executed by the processing unit 53, various parameters used in the arithmetic processing, information acquired from the LiDAR 11 and the position detection sensor 13, and the results of the arithmetic processing by the processing unit 53.

[0028] Below, we will briefly explain the functions of the processing unit 53, and then explain a specific example of the operation of the processing unit 53. The processing unit 53 includes an acquisition unit 61 and an object detection processing unit 63. Some or all of the acquisition unit 61 and the object detection processing unit 63 are functions realized by the execution of a program by the arithmetic processing device.

[0029] The acquisition unit 61 acquires data transmitted from the LiDAR 11 and the position detection sensor 13. Specifically, the acquisition unit 61 acquires data of measurement points transmitted from the LiDAR 11 and position data transmitted from the position detection sensor 13 for each processing cycle set at a predetermined time interval.

[0030] The object detection processing unit 63 executes a process of detecting objects around the vehicle 1 based on the data of the measurement points acquired by the acquisition unit 61. Specifically, the ... sensor coordinate system C R (0 R ,X R ,YR ) coordinates of each measurement point (x r ,y r ) is a fixed coordinate system C that does not change even if the position of the LiDAR 11 changes. W (0 W ,X W ,Y W ) coordinates (x w ,y w In this embodiment, the position (L o ,L a ) to the origin (x w ,y w =0 W ) and the direction along the longitude on the world map is X W The direction along the axis and latitude is Y W Axes of fixed coordinate system C W (0 W ,X W ,Y W ) is used.

[0031] However, the fixed coordinate system C W (0 W ,X W ,Y W ) is the direction along the latitude, W axis, and the direction along the longitude is Y W It may also be an X axis. W axis and Y W The axis may be set arbitrarily without being related to longitude and latitude as long as it can remain unchanged regardless of the movement of the vehicle 1. For example, the longitudinal direction of the vehicle 1 indicated by the reference direction of the position detection sensor 13 at the time of system startup of the vehicle 1 is set as X W axis, and the vehicle width direction is Y W Axes of fixed coordinate system C W (0 W ,X W ,Y W ) can also be used.

[0032] Furthermore, the object detection processing unit 63 extracts one or more measurement point groups in which the distance between adjacent measurement points is within a predetermined distance from the data of measurement points acquired by the acquisition unit 61, and detects stationary and moving objects based on the information of the measurement point group extracted for each processing cycle. The object detection processing unit 63 transmits information on the detected objects to the vehicle control device 40. The vehicle control device 40 executes automatic driving control of the vehicle 1 based on the received object information, and avoids a collision with the object or alleviates the impact of a collision.

[0033] (2-2. Example of operation) Next, a specific example of the operation of the process executed by the object detection device 50 according to this embodiment will be described.

[0034] 3 is a flowchart showing the processing operation of the object detection device 50. The processing described below may be executed continuously while the system of the vehicle 1 is running, or may be executed continuously from the start of driving to the end of driving.

[0035] First, the acquisition unit 61 acquires data transmitted from the LiDAR 11 and the position detection sensor 13 (step S1). Specifically, the acquisition unit 61 acquires measurement point data transmitted from the LiDAR 11 and position data transmitted from the position detection sensor 13. The position information of the measurement point includes information on the angle (azimuth angle) formed by the direction of the measurement point as seen from the LiDAR 11 with respect to the irradiation axis of the LiDAR 11, and information on the distance from the LiDAR 11 to the measurement point. Furthermore, the position data includes position information of the position detection sensor 13 on a fixed coordinate system and information on the reference direction of the position detection sensor 13 on a vehicle coordinate system.

[0036] Next, the object detection processing unit 63 executes a process of converting the coordinates of each measurement point of the data of the measurement point acquired from the LiDAR 11 into coordinates in a fixed coordinate system (step S3). That is, the coordinates of the sensor coordinate system, which has the position of the LiDAR 11 that moves together with the vehicle 1 as its origin, are converted into coordinates in a fixed coordinate system in which the origin is maintained regardless of the position of the LiDAR 11.

[0037] The storage unit 55 of the object detection device 50 stores in advance information indicating the relative relationship between the installation position of the position detection sensor 13 and the installation position of the LiDAR 11. In this embodiment, the storage unit 55 stores information on the coordinates of the installation position of the LiDAR 11 on a vehicle coordinate system having the installation position of the position detection sensor 13 as the origin and two axes in the vehicle length direction and the vehicle width direction. The storage unit 55 also stores information on the angle between the illumination axis of the LiDAR 11 and one axis of the vehicle coordinate system. The position and reference direction of the position detection sensor 13 on the fixed coordinate system are constantly acquired from the position detection sensor 13. Therefore, the object detection processing unit 63 can convert the coordinates of the sensor coordinate system of the measurement point measured by the LiDAR 11 into coordinates in the vehicle coordinate system, and further convert the coordinates of the vehicle coordinate system into coordinates in the fixed coordinate system.

[0038] Next, the object detection processing unit 63 executes a process of detecting a moving object based on the data of the measurement point group converted into coordinates in the fixed coordinate system (step S5). Fig. 4 shows a specific flowchart of the process of detecting a moving object.

[0039] (filtering process) The object detection processing unit 63 acquires data (primary measurement point data) mpts of measurement points detected by the LiDAR 11 and performs filtering processing to remove non-moving object measurement point clouds from the primary measurement point data (step S11). Specifically, the object detection processing unit 63 extracts one or more measurement point clouds where the distance between adjacent measurement points is within a predetermined distance from the primary measurement point data mpts. For example, the object detection processing unit 63 calculates the distance between each measurement point converted into coordinates in a fixed coordinate system, and extracts a measurement point cloud by combining multiple measurement points where the distance between adjacent measurement points is within a predetermined distance set in advance. As the distance, Euclidean distance may be typically used, but other distances may also be used.

[0040] Furthermore, the object detection processor 63 excludes from the primary measurement point data any measurement point group whose length in any direction is equal to or greater than a predetermined length threshold, as a non-moving object measurement point group. For example, the object detection processor 63 calculates the maximum length along an appropriate direction for each of the extracted measurement point groups. The appropriate direction may be any direction in a fixed coordinate system. The object detection processor 63 then excludes from the primary measurement point data any measurement point group whose calculated maximum length is equal to or greater than a predetermined length threshold set to a value exceeding the maximum length expected for a moving object, as a non-moving object measurement point group. The predetermined length threshold may be a value exceeding the maximum length expected for a detected vehicle, such as a bus or truck, but may be set arbitrarily depending on the size of the moving object to be detected. The predetermined length threshold may be set to a value within a range of 10 to 12 meters, for example.

[0041] 5 and 6 are explanatory diagrams showing the filtering process of step S11. FIG. 5 shows a two-dimensional image of primary measurement point data detected when multiple LiDARs are installed on the front, rear, left, and right sides of the vehicle 1. The object detection processing unit 63 extracts a measurement point cloud from the primary measurement point data by combining multiple measurement points whose adjacent measurement points are within a predetermined distance, and excludes from the primary measurement point data any measurement point cloud whose maximum length is equal to or greater than a predetermined length threshold as a non-moving body measurement point cloud. FIG. 6 shows measurement point data (secondary measurement point data) obtained by excluding the non-moving body measurement point cloud from the primary measurement point data.

[0042] The non-moving object measurement point cloud excluded in step S11 is considered to be road surface irregularities such as curbs or road installations such as guardrails, and is not considered to be measurement points for detecting moving objects. Therefore, the object detection processing unit 63 labels and records the non-moving object measurement point cloud excluded in step S11 as road surface irregularities or road installations (step S13).

[0043] (Dynamic measurement point extraction processing) The object detection processing unit 63 executes a dynamic measurement point extraction process on the secondary measurement point data from which the non-first moving object measurement point cloud was excluded in step S11 (step S15). For example, the object detection processing unit 63 excludes static measurement points that overlap with the primary measurement point data acquired from the LiDAR 11 in the previous processing cycle from the secondary measurement point data, which is data of measurement points from which the non-moving object measurement point cloud was excluded from the primary measurement point data, and extracts the remaining measurement points as dynamic measurement points. More specifically, the object detection processing unit 63 uses the secondary measurement point data and the primary measurement point data acquired in the previous processing cycle to identify and exclude static measurement points whose positions have not changed from the previous processing cycle to the current processing cycle. The object detection processing unit 63 extracts measurement points from the secondary measurement point data whose positions have changed from the previous processing cycle to the current processing cycle as dynamic measurement points.

[0044] (clustering process) Next, the object detection processing unit 63 extracts measurement point clouds from the secondary measurement point data in which the distance between adjacent measurement points is within a predetermined distance, and performs clustering processing to identify a first moving object measurement point cloud composed of dynamic measurement points and a second moving object measurement point cloud composed of dynamic measurement points and static measurement points (step S17). Specifically, the object detection processing unit 63 extracts one or more measurement point clouds from the secondary measurement point data in which the distance between adjacent measurement points is within a predetermined distance. The predetermined distance may be the same as the distance used in the filtering in step S11. However, the predetermined distance may be different from the distance used in the filtering in step S11. Of the extracted measurement point clouds, the object detection processing unit 63 identifies the measurement point cloud composed of the dynamic measurement points extracted in step S15 as the first moving object measurement point cloud, and the measurement point cloud composed of the dynamic measurement points and static measurement point cloud as the second moving object measurement point cloud.

[0045] In other words, in the dynamic measurement point extraction process (S15) and the clustering process (S17), the object detection processing unit 63 classifies the measurement point groups extracted from the secondary measurement point data into a first moving body measurement point group that has undergone a large change in position from the previous processing cycle, and a second moving body measurement point group that has undergone a small change in position.

[0046] 7 and 8 are explanatory diagrams showing the dynamic measurement point extraction process (S15) and the clustering process (S17). Each black circle indicates a measurement point cloud extracted in the current processing cycle, and each white circle indicates a primary measurement point data acquired in the previous processing cycle. In the example shown in FIG. 7, the measurement point cloud extracted from the secondary measurement point data does not include static measurement points that overlap with the primary measurement point data of the previous processing cycle. Therefore, this measurement point cloud is classified as a first moving object measurement point cloud composed of dynamic measurement points. On the other hand, in the example shown in FIG. 8, the measurement point cloud extracted from the secondary measurement point data is composed of static measurement points (measurement points within dashed line H1) that overlap with the primary measurement point data of the previous processing cycle, and measurement points excluding the static measurement points are dynamic measurement points (measurement points within dashed line H2). Therefore, this measurement point cloud is classified as a second moving object measurement point cloud composed of dynamic measurement points and static measurement points.

[0047] (Matching process) Next, the object detection processing unit 63 executes a matching process to associate the first moving body measurement point cloud and the second moving body measurement point cloud identified by the clustering process with moving bodies that have been detected up to the previous processing cycle (steps S19 to S29). In this embodiment, the matching process includes a first matching process (steps S19 to S23) that is a simple process, and a second matching process (steps S25 to S29) that associates the first moving body measurement point clouds that could not be associated by the first matching process.

[0048] "Associating the first and second moving object measurement point clouds with moving objects already detected up to the previous processing cycle" means identifying and recording measurement point clouds from the first and second moving object measurement point clouds that have detected a moving object identical to the data of the detected moving object. For example, if data of detected moving objects labeled with identification information that identifies each moving object along with information estimating a predetermined moving speed and moving direction has been recorded up to the previous processing cycle, a process is executed to identify measurement point clouds from the first and second moving object measurement point clouds identified in the current processing cycle that have detected a moving object identical to the data of the detected moving object, label them with the same identification information, and record them together with information on the moving speed and moving direction.

[0049] The first matching process is a matching process that can be performed more easily (in a shorter time) than the second matching process described below, and is performed with the aim of completing the association for the first moving body measurement point group and the second moving body measurement point group, which can be easily and reliably associated without having to perform the second matching process.

[0050] The first matching process may be performed using an existing matching technology. For example, the object detection processing unit 63 searches for a first moving object measurement point cloud and a second moving object measurement point cloud that can be associated with the detected moving object data, using the first moving object measurement point cloud and the second moving object measurement point cloud and moving object data (detected moving object data) that have been labeled as moving objects up to the previous processing cycle. At this time, the object detection processing unit 63 may determine whether a moving object in the detected moving object data whose speed is equal to or greater than a predetermined speed threshold can be associated with any of the first moving object measurement point cloud. For example, the object detection processing unit 63 determines whether first detected moving object data having a speed equal to or greater than a predetermined threshold (e.g., 15 km / h) can be associated with only the first moving object measurement point cloud. Furthermore, in the first matching process, the object detection processing unit 63 may determine whether a moving object in the detected moving object data whose speed is less than a predetermined speed threshold can be associated with any of the first moving object measurement point cloud and the second moving object measurement point cloud. For example, the object detection processing unit 63 determines whether the second detected moving body data having a speed less than a predetermined threshold (e.g., 15 km / h) can be associated with the first moving body measurement point group and the second moving body measurement point group.

[0051] The object detection processing unit 63 then compares the first moving object measurement point cloud with the first detected moving object data or the second detected moving object data associated with the first moving object measurement point cloud, calculates the speed by dividing the moving distance by the sampling time, and labels and records the associated first moving object measurement point cloud (steps S19 and S23).The object detection processing unit 63 also compares the second moving object measurement point cloud with the second detected moving object data associated with the second moving object measurement point cloud, calculates the speed by dividing the moving distance by the sampling time, and labels and records the associated second moving object measurement point cloud (steps S21 and S23).

[0052] In this way, by dividing the detected moving object data into first detected moving object data and second detected moving object data according to the speed, and narrowing down the objects to be associated, the execution time of the first matching process can be shortened. In this case, not only the speeds of the first moving object measurement point cloud and the second moving object measurement point cloud, but also the moving direction and the orientation of the moving object may be calculated and labeled together.

[0053] On the other hand, for the first moving body measurement point group and the second moving body measurement point group that were not associated by the first matching process, a more detailed second matching process is performed to associate them with detected moving body data that were not associated with the first moving body measurement point group and the second moving body measurement point group by the first matching process.

[0054] In the second matching process, the object detection processing unit 63 uses a scan matching technique such as ICP (Iterative Closest Point) and a machine learning model to search for detected moving body data that is associated with the first moving body measurement point group and the second moving body measurement point group that were not associated by the first matching process from among the detected moving body data that were not associated by the first matching process (step S25).

[0055] Specifically, the object detection processing unit 63 determines the center of gravity of each of the unassociated first moving object measurement point cloud and second moving object measurement point cloud. The object detection processing unit 63 also determines the center of gravity of each unassociated detected moving object data. Then, the object detection processing unit 63 calculates the distance D between the center of gravity of each measurement point cloud and the center of gravity of each detected moving object data, ... W (0 W ,X W ,Y W ) and the velocity components Vx_cl(n), Vy_cl(n) in the Xw-axis direction and the Yw-axis direction of each measurement point group on the fixed coordinate system C W (0 W ,X W ,Y WThe velocity components Vx_mo(m) and Vy_mo(m) in the Xw-axis direction and the Yw-axis direction of each detected moving object data on the ordinate are calculated. Note that "n" is an identifier (n=1, 2, 3...) that distinguishes between measurement point groups, and "m" is an identifier (m=1, 2, 3...) that distinguishes between detected moving object data.

[0056] 9, the object detection processing unit 63 inputs the distance D calculated for each measurement point cloud and detected movement data, the velocity components Vx_cl(n) and Vy_cl(n) of each measurement point cloud, and the velocity components Vx_mo(m) and Vy_mo(m) of each detected moving object data into a machine learning model 70 to calculate measurement point clouds and detected moving object data that can be correlated with each other. The machine learning model 70 is a model that learns the movement patterns of moving objects using, as input data, learning measurement point cloud data that has been collected or generated in advance for various types of moving objects traveling at various speeds, turning speeds, and orientations. A support vector machine may be used as the machine learning model, but other machine learning models such as a random forest or a neural network may also be used.

[0057] This allows for the determination of detected moving object data associated with the first moving object measurement point cloud and the second moving object measurement point cloud that were not associated by the first matching process. The object detection processing unit 63 records the first moving object measurement point cloud and the second moving object measurement point cloud for which associated detected moving object data exists, together with information on the speed, moving direction, and orientation, in association with the detected moving object data (step S27).

[0058] Meanwhile, the object detection processing unit 63 generates and records detected moving object data as new moving objects for the first moving object measurement point cloud and the second moving object measurement point cloud for which no associated detected moving object data exist (step S29). When generating new detected moving object data, the object detection processing unit 63 may generate new detected moving object data using the first moving object measurement point cloud and the second moving object measurement point cloud that could not be associated with the detected moving object data in each processing cycle. Specifically, the object detection processing unit 63 identifies the first moving object measurement point cloud and the second moving object measurement point cloud in multiple frames that are estimated to be measurement point clouds of the same object based on the distance between the first moving object measurement point cloud and the second moving object measurement point cloud between each processing cycle (between frames) and the moving speed and moving velocity in each processing cycle, and calculates the position, speed, and size (length and width) of the first moving object measurement point cloud and the second moving object measurement point cloud for each processing cycle. The newly generated detected moving object data is used in the matching process from the next processing cycle onwards.

[0059] The object detection processing unit 63 repeats the processes from step S11 to step S29 for each processing cycle, and associates the measurement point group extracted from the data of the measurement points detected by the LiDAR 11 with the detected moving object data.

[0060] As described above, in this embodiment, the object detection processing unit 63 excludes non-moving object measurement point clouds whose size clearly exceeds the size of a moving object from the measurement point data (primary measurement point data) of the LiDAR 11, and extracts dynamic measurement points by excluding static measurement points whose position has not changed since the previous processing cycle. The object detection processing unit 63 also clusters the secondary measurement point data obtained by excluding non-moving object measurement point clouds from the primary measurement point data, classifying them into a first moving object measurement point cloud consisting of dynamic measurement point clouds, and a second moving object measurement point cloud consisting of dynamic measurement point clouds and static measurement point clouds, and performs a matching process with already recorded detected moving object data.

[0061] Therefore, it is possible to separate the first moving object measurement point cloud, which is estimated to have a relatively fast moving speed, from the second moving object measurement point cloud, which is estimated to have a relatively slow moving speed, and perform a matching process with already recorded detected moving object data. Therefore, it is possible to narrow down the matching targets to the first moving object measurement point cloud or the second moving object measurement point cloud depending on the speed of the detected moving object data, thereby improving the detection accuracy of moving objects. In this way, the object detection device according to this embodiment can associate and detect the same moving object from the primary measurement point data detected in a series of processing cycles, even though the data of the measurement points of the LiDAR 11 does not contain speed information.

[0062] In this embodiment, the object detection processing unit 63 executes a first matching process to determine whether or not a moving object whose speed is equal to or greater than a predetermined speed threshold among the detected moving object data can be associated with any of the first moving object measurement point clouds, and to determine whether or not a moving object whose speed is less than the predetermined speed threshold among the detected moving object data can be associated with any of the first moving object measurement point cloud and the second moving object measurement point cloud. This allows the matching targets to be narrowed down according to the speed of the detected moving object data, thereby reducing the load of the matching process.

[0063] Furthermore, in this embodiment, the object detection processing unit 63 executes a second matching process in which the data of the first moving object measurement point cloud and the second moving object measurement point cloud, as well as the moving object data detected up to the previous processing cycle, are input into a machine learning model, and the second moving object measurement point cloud data is determined to be able to be associated with the moving object data detected up to the previous processing cycle. This allows the first moving object measurement point cloud data and the second moving object measurement point cloud data to be accurately matched with the moving object data detected up to the previous processing cycle, based on the learning results of the motion patterns of various moving objects.

[0064] Furthermore, in this embodiment, the object detection processing unit 63 performs a second matching process using a machine learning model after removing the first moving body measurement point cloud and the second moving body measurement point cloud that can be associated by a simple matching process (first matching process) and the detected moving body data. This allows the matching process with the detected moving body data to be performed with high accuracy while further reducing the processing speed and processing load required for the moving body detection process.

[0065] In this embodiment, the predetermined length threshold in the filtering process is set to the maximum length of an expected moving object, and the object detection processing unit 63 labels and records measurement point clouds whose lengths in any direction are equal to or greater than the predetermined length threshold as road surface irregularities or road installations. This makes it possible to easily identify measurement point clouds that exceed the size of an expected moving object as road surface irregularities or road installations.

[0066] The above-described effects of the object detection device can also be achieved by an object detection method and a computer program that executes object detection processing.

[0067] It should be noted that in the matching process, the first matching process may be omitted and only the second matching process may be performed, and it is understood that an object detection device that executes this processing method also falls within the technical scope of the present disclosure.

[0068] Although the preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to these examples. It is clear that a person skilled in the art to which the present invention pertains can conceive of various modifications or alterations within the scope of the technical ideas set forth in the claims, and it is understood that these also naturally fall within the technical scope of the present invention.

[0069] For example, in the above embodiment, LiDAR 11 calculates the position and speed of the measurement point and transmits the calculated position and speed information to the object detection device 50, but LiDAR 11 may calculate the position of the measurement point and transmit the calculated position information to the object detection device 50, and the object detection device 50 may calculate the speed information of the measurement point. [Explanation of symbols]

[0070] 1: vehicle, 11·11LF: radar sensor, 17: position detection sensor, 40: vehicle control device, 50: object detection device, 51: communication unit, 53: processing unit, 55: storage unit, 61: acquisition unit, 63: object detection processing unit, C R : sensor coordinate system, C V : Vehicle coordinate system, C W :Fixed coordinate system, Pf: Erroneous measurement point, Pa_i・Pb_i: Measurement point for judgment, αr・αv: Azimuth angle

Claims

1. In an object detection device (100) that detects an object based on data of measurement points acquired by a LiDAR (11), A filtering process is performed to extract a group of measurement points whose adjacent measurement points are within a predetermined distance from the primary measurement point data, which is data of the measurement points acquired by the LiDAR (11), and to exclude the non-moving body measurement point group from the primary measurement point data by defining the group of measurement points whose length in any direction is equal to or greater than a predetermined length threshold as a non-moving body measurement point group. a dynamic measurement point extraction process for extracting, as dynamic measurement points, measurement points from secondary measurement point data, which is data on measurement points obtained by excluding the non-moving body measurement point group from the primary measurement point data, excluding static measurement points that overlap with the primary measurement point data used in the previous processing cycle; a clustering process for extracting measurement point clouds from the secondary measurement point data where the distance between adjacent measurement points is within the predetermined distance, and identifying a first moving object measurement point cloud made up of the dynamic measurement points and a second moving object measurement point cloud made up of the dynamic measurement points and the static measurement points; a matching process for associating the first moving object measurement point cloud and the second moving object measurement point cloud identified by the clustering process with moving object data that has been detected up to a previous processing cycle; An object detection device that performs the above.

2. In the matching process, execute a first matching process to determine whether or not a moving body among the detected moving body data whose speed is equal to or greater than a predetermined speed threshold can be associated with any of the first moving body measurement point clouds, and to determine whether or not a moving body among the detected moving body data whose speed is less than the predetermined speed threshold can be associated with any of the first moving body measurement point cloud and the second moving body measurement point cloud; The object detection device according to claim 1 .

3. The object detection device Equipped with a machine learning model that has learned the movement patterns of various moving objects, a second matching process is executed in which the data of the first moving body measurement point cloud and the second moving body measurement point cloud that could not be associated by the first matching process, and the moving body data that has been detected up to the previous processing cycle, are input to the machine learning model, and it is determined whether or not the data of the first moving body measurement point cloud and the second moving body measurement point cloud can be associated with the moving body data that has been detected up to the previous processing cycle. The object detection device according to claim 2 .

4. The object detection device Equipped with a machine learning model that has learned the movement patterns of various moving objects, In the matching process, the data of the first moving body measurement point cloud and the second moving body measurement point cloud, and moving body data detected up to the previous processing cycle are input to the machine learning model, and it is determined whether the data of the first moving body measurement point cloud and the second moving body measurement point cloud can be associated with the moving body data detected up to the previous processing cycle. The object detection device according to claim 1 .

5. The predetermined length threshold in the filtering process is set to a value of the maximum length of an assumed moving object, In the filtering process, the measurement point cloud whose length in any one of the directions is equal to or greater than the predetermined length threshold is labeled as a road surface irregularity or a road installation and recorded. The object detection device according to claim 1 .

6. An object detection method for detecting an object based on measurement point data acquired by LiDAR (11), From the primary measurement point data, which is data of the measurement points acquired by the LiDAR (11), a measurement point group in which the distance between adjacent measurement points is within a predetermined distance is extracted, and among the extracted measurement point groups, a measurement point group whose length in any direction is equal to or greater than a predetermined length threshold is defined as a non-moving body measurement point group, and the non-moving body measurement point group is excluded from the primary measurement point data (step S11); extracting, as dynamic measurement points, measurement points from the secondary measurement point data, which is data of measurement points obtained by excluding the non-moving body measurement point group from the primary measurement point data, excluding static measurement points, which are measurement points that overlap with the primary measurement point data used in the previous processing cycle (step S15); a clustering process (step S17) for extracting measurement point clouds in which the distance between adjacent measurement points is within the predetermined distance from the secondary measurement point data, and identifying a first moving object measurement point cloud constituted by the dynamic measurement points and a second moving object measurement point cloud constituted by the dynamic measurement points and the static measurement points; a matching process (steps S19 to S29) for associating the first moving object measurement point cloud and the second moving object measurement point cloud identified by the clustering process with moving object data that has been detected up to the previous processing cycle; 1. An object detection method comprising:

7. A computer program that causes a computer to execute a process of detecting an object based on data of measurement points acquired by a LiDAR (11), From the primary measurement point data, which is data of the measurement points acquired by the LiDAR (11), a measurement point group in which the distance between adjacent measurement points is within a predetermined distance is extracted, and among the extracted measurement point groups, a measurement point group whose length in any direction is equal to or greater than a predetermined length threshold is defined as a non-moving body measurement point group, and the non-moving body measurement point group is excluded from the primary measurement point data (step S11); extracting, as dynamic measurement points, measurement points from the secondary measurement point data, which is data of measurement points obtained by excluding the non-moving body measurement point group from the primary measurement point data, excluding static measurement points, which are measurement points that overlap with the primary measurement point data used in the previous processing cycle (step S15); a clustering process (step S17) for extracting measurement point clouds in which the distance between adjacent measurement points is within the predetermined distance from the secondary measurement point data, and identifying a first moving object measurement point cloud constituted by the dynamic measurement points and a second moving object measurement point cloud constituted by the dynamic measurement points and the static measurement points; a matching process (steps S19 to S29) for associating the first moving object measurement point cloud and the second moving object measurement point cloud identified by the clustering process with moving object data that has been detected up to the previous processing cycle; A computer program that causes a process including the steps of:

Images