Radar-based adaptive target clustering method, device, equipment, and storage medium

Abstract

This application discloses a radar-based adaptive target clustering method, apparatus, device, and storage medium. The method includes: acquiring radar point cloud data and creating a four-dimensional grid of radar detection data; mapping each detection point of the radar point cloud data to the four-dimensional grid and recording the index information of each detection point in the grid; creating at least one grid-based three-dimensional clustering window; determining whether a processing point is a noise point and filtering out noise points; and performing adaptive clustering or fixed-parameter clustering on the detection points. This application considers the similarity of distance, elevation angle, horizontal angle, and velocity of radar detection points, and accelerates the target clustering process by replacing the traditional Euclidean distance with an adaptive clustering window based on grid index; this application applies a specific list structure of point indexes, recording only the index information of the detection points to minimize storage space.

Description

Technical Field

[0001] This application relates to the field of radar data processing technology, and in particular to a radar-based target adaptive clustering method, apparatus, device, and storage medium. Background Technology

[0002] Among various radar types, millimeter-wave radar is small in size, easy to integrate, and has high spatial resolution. However, due to the insufficient height measurement capability of traditional millimeter-wave radar, it has significant limitations in distinguishing and identifying low-lying targets, aerial targets, and static road targets. The overall output information volume is not on the same order of magnitude as that of visible light cameras and lidar, thus exhibiting certain limitations. With continuous breakthroughs in radar technology, 4D imaging radar sensors have overcome these limitations, but significant progress still needs to be made in data processing for 4D imaging radar.

[0003] Clustering has been widely used in processing image and radar data. Among various clustering algorithms, density-based spatial clustering of applications with noise (DBSCAN) has proven to be one of the most suitable methods. However, DBSCAN also has some drawbacks in radar data clustering, such as unequal dimensionality, high computational cost, and strong parameter dependence. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a radar-based target adaptive clustering method, apparatus, device, and storage medium. It utilizes a grid-indexed clustering window instead of Euclidean distance to accelerate the clustering process; simultaneously, it applies a specific list structure of point indexes to minimize storage space.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] The first aspect of this application provides a radar-based adaptive target clustering method, the method comprising:

[0007] Acquire radar point cloud data and create a four-dimensional grid of radar detection data;

[0008] Each detection point in the radar point cloud data is mapped to a four-dimensional grid, and the index information of each detection point in the grid is recorded;

[0009] Create at least one grid-based 3D clustering window containing detection points with similar velocities and similar distances;

[0010] Set a threshold for the number of adjacent points in the window, determine whether the processing point is a noise point, and filter out the noise points;

[0011] The target contour is determined based on preset information. If the target contour can be obtained, adaptive clustering is performed on the detection points; otherwise, fixed parameter clustering is performed.

[0012] Furthermore, the creation of the four-dimensional grid for radar detection data includes:

[0013] The Fast Fourier Transform is computed along the distance dimension, where the gate cell value of the distance and the FFT bin size are used as the grid size;

[0014] The grid size in the velocity dimension is determined by the FFT bin value over the velocity.

[0015] The angle is estimated using the DML algorithm, which generates an arithmetic sequence. The common difference of this estimated sequence is used to determine the grid size in the dimensional angle.

[0016] Furthermore, the index information is pre-set assignment information for each grid line. Index information with detection points is marked and stored, while index information without detection points is not recorded.

[0017] Furthermore, in the index information, the difference between the two indices in the distance dimension represents the physical distance of the point, and the difference between the indices in the angle dimension is converted into physical distance using the following formula:

[0018] Δd θ ≈r×sinΔθ.

[0019] For example, the physical distance between the pitch angle dimension and the horizontal angle dimension index can both be calculated using the formula above.

[0020] Furthermore, creating at least one mesh-based 3D clustering window includes:

[0021] Create a mesh-based 3D clustering window around the cells at the processing point. This window can be a fixed-parameter clustering window or an adaptive clustering window.

[0022] The size of the fixed-parameter clustering window is determined by two variables: the distance threshold d in either the distance dimension or the angular dimension. thr And velocity difference threshold dv thr ;

[0023] The three-dimensional clustering window is cylindrical, with an ellipse for the distance or angle dimension. Detection points with similar speeds and similar distances are located in this window.

[0024] Furthermore, the number of grid cells occupied by the three-dimensional clustering window decreases as the range increases, which ensures that the spatial size of the clustering window remains the same throughout the entire region.

[0025] Further, the step of setting a threshold for the number of adjacent points in the window, determining whether a processing point is a noise point, and filtering out noise points includes:

[0026] Search for the index information of all cells in a window containing a processing point, and then compare it with the list of point indices; identify the adjacent points of the processing point in the window;

[0027] If the number of these points is less than a given threshold n thr If the condition is met, the processing point is marked as a noise point; otherwise, the point is the core point of a point cloud cluster, and all neighboring points are assigned to the cluster. Subsequently, these neighboring points are treated as new processing points and processed using the same process as the clustering window.

[0028] After processing all the detection points, multiple clusters with the selected points are identified, and noise points are eliminated.

[0029] Furthermore, the predicted target contour includes:

[0030] Using Support Vector Machine (SVM) and machine learning, potentially detectable targets are pre-classified into pedestrians, cyclists, and vehicles; the contour size of each target type is predefined.

[0031] The initial position and direction of motion of the cloud cluster are determined based on the average coordinates and average velocity of all detection points belonging to the same cloud cluster in the current frame; at the same time, the maximum and minimum velocities are stored for the clustering window of the next frame, that is, the upper and lower points of the velocity dimension in the cluster.

[0032] The estimated location of the point cloud cluster around the current frame is used to determine the contour based on the size of the point cloud cluster.

[0033] Furthermore, the adaptive clustering includes:

[0034] The clustering window size is determined based on the predefined target classification and the predicted target profile;

[0035] The clustering window is defined by the predicted target profile in the distance or angle dimension and the maximum and minimum velocities from the last cycle in the velocity dimension.

[0036] Furthermore, the four-dimensional grid of the radar detection data includes radial distance dimension, horizontal angle dimension, pitch angle dimension, and radial velocity dimension.

[0037] A second aspect of this application provides a radar-based target adaptive clustering device, comprising:

[0038] Four-dimensional mesh creation unit: Acquire radar point cloud data and create a four-dimensional mesh of radar detection data;

[0039] Point cloud indexing unit: maps each detection point of the radar point cloud data to a four-dimensional grid and records the index information of each detection point in the grid;

[0040] 3D clustering window creation unit: Creates at least one grid-based 3D clustering window, in which detection points with similar velocities and similar distances are located;

[0041] Noise point filtering unit: Sets a threshold for the number of adjacent points in the window, determines whether the processing point is a noise point, and filters out the noise points;

[0042] Adaptive clustering unit: Determines the target contour based on preset information. If the target contour can be obtained, adaptive clustering is performed on the detection points; otherwise, fixed parameter clustering is performed.

[0043] A third aspect of this application provides a radar-based target adaptive clustering device, comprising:

[0044] A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the radar-based target adaptive clustering method as described in the first aspect of this application.

[0045] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the radar-based target adaptive clustering method as described in the first aspect of this application.

[0046] Compared with the prior art, this application has the following advantages:

[0047] This application combines the traditional DBSCAN clustering algorithm with a target clustering method based on a four-dimensional grid, which considers the similarity of point distance, angle and velocity, and accelerates the target clustering process by replacing Euclidean distance with an adaptive clustering window based on grid index;

[0048] This application uses a specific list structure for the application point index, recording only the index information of the existence of detection points to minimize storage space. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1This application provides a schematic diagram of a radar-based target adaptive clustering method.

[0051] Figure 2 A schematic diagram illustrating the concept of a four-dimensional grid and clustering window for creating radar detection data, provided in an embodiment of this application.

[0052] Figure 3 This is a schematic diagram of the adaptive clustering process provided in an embodiment of this application;

[0053] Figure 4 A schematic diagram of a radar-based target adaptive clustering device provided in this application embodiment;

[0054] Figure 5 This is a schematic diagram of a radar-based target adaptive clustering device provided in an embodiment of this application. Detailed Implementation

[0055] As described earlier, clustering has been widely applied to process image and radar data. Among various clustering algorithms, density-based noisy applied spatial clustering (DBSCAN) has proven to be one of the most suitable methods because it is applicable to arbitrary object shapes and does not require prior information, such as the exact number of clusters in the dataset. However, DBSCAN also has some drawbacks in radar data clustering, such as unequal dimensionality, high computational cost, and strong parameter dependence.

[0056] In view of this, this application provides a radar-based adaptive target clustering method, comprising: acquiring radar point cloud data and creating a four-dimensional grid of radar detection data; mapping each detection point of the radar point cloud data to the four-dimensional grid and recording the index information of each detection point in the grid; creating at least one grid-based three-dimensional clustering window, in which detection points with similar speeds and similar distances are located; setting a threshold for the number of adjacent points in the window, determining whether the processing point is a noise point, and filtering out noise points; determining the target contour according to preset information, and if the target contour can be obtained, performing adaptive clustering on the detection points; otherwise, performing fixed parameter clustering. This application combines the traditional DBSCAN clustering algorithm to provide a target clustering method based on a four-dimensional grid, considering the similarity of distance, angle, and speed of points, and accelerating the target clustering process by replacing Euclidean distance with an adaptive clustering window based on grid index; this application also uses a specific list structure of point indexes, recording only the index information of the existence of detection points to minimize storage space.

[0057] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0058] See Figure 1 The figure is a schematic diagram of a radar-based target adaptive clustering method provided in an embodiment of this application, as shown below. Figure 1 As shown, the radar-based target adaptive clustering method includes:

[0059] S110: Acquire radar point cloud data and create a four-dimensional grid of radar detection data;

[0060] S110 includes:

[0061] S1101. Create a four-dimensional grid for radar detection data:

[0062] For example, see Figure 2 The Fast Fourier Transform is computed in the distance dimension, where the gate cell of the distance, i.e. the size of the FFT bin value, is used as the grid size;

[0063] As one possible implementation, see Δr in the figure, where the FFT bin value is 0.0633m.

[0064] The mesh size in the velocity dimension is determined by the gate element in the velocity dimension, i.e., the FFT bin value; as one possible implementation, this FFT bin value is 0.0397 m / s.

[0065] The DML algorithm, or Deterministic Maximum Likelihood Algorithm, is used to estimate the angle. This algorithm generates an arithmetic sequence, and the common difference of the estimated sequence is used to determine the grid size in the dimensional angle.

[0066] As one possible implementation, the common difference is 0.25 degrees.

[0067] S120: Map each detection point of the radar point cloud data to a four-dimensional grid and record the index information of each detection point in the grid;

[0068] S120 includes:

[0069] S1201. Determine index information:

[0070] The index information is the pre-set value information for each grid line. Index information with detection points is marked and stored, while index information without detection points is not recorded.

[0071] S1202. In the index information, the difference between the two indices in the distance dimension represents the physical distance of the point, and the difference between the indices in the angle dimension is converted into physical distance using the following formula:

[0072] Δd θ ≈r×sinΔθ.

[0073] S130: Create at least one grid-based 3D clustering window containing detection points with similar velocities and similar distances;

[0074] S130 includes:

[0075] S1301. Creating at least one mesh-based 3D clustering window includes:

[0076] Create a mesh-based 3D clustering window around the cells at the processing point; this window can be a fixed-parameter clustering window or an adaptive clustering window.

[0077] The size of the fixed-parameter clustering window is determined by two variables: the distance threshold d in either the distance dimension or the angular dimension. thr And velocity difference threshold dv thr ;

[0078] The three-dimensional clustering window is cylindrical, with an ellipse for the distance or angle dimension. Detection points with similar speeds and similar distances are located in this window.

[0079] S1302, The number of grid cells occupied by the three-dimensional clustering window decreases as the range increases, which ensures that the spatial size of the clustering window remains the same throughout the entire area.

[0080] S140: Set the threshold for the number of adjacent points in this window, determine whether the processing point is a noise point, and filter out the noise points;

[0081] S1401. Set a threshold for the number of adjacent points in the window, determine whether the processing point is a noise point, and filter out the noise points:

[0082] Search the index information of all cells in a window for a processing point, and then compare it with the list of point indices; identify the adjacent points of the processing point in the window;

[0083] If the number of these points is less than a given threshold n thrIf the condition is met, the processing point is marked as a noise point; otherwise, the point is the core point of a point cloud cluster, and all neighboring points are assigned to the cluster. Subsequently, these neighboring points are treated as new processing points and processed using the same process as the clustering window.

[0084] After processing all the detection points, multiple clusters with the selected points are identified, and noise points are eliminated.

[0085] S150: Determine the target contour based on preset information. If the target contour can be obtained, perform adaptive clustering on the detection points; otherwise, perform fixed parameter clustering.

[0086] S1501, Predict the target outline:

[0087] Using Support Vector Machine (SVM) and machine learning, potentially detectable targets are pre-classified into pedestrians, cyclists, and vehicles; the contour size of each target type is predefined.

[0088] The initial position and direction of motion of the cloud cluster are determined based on the average coordinates and average velocity of all detection points belonging to the same cloud cluster in the current frame; at the same time, the maximum and minimum velocities are stored for the clustering window of the next frame, that is, the upper and lower points of the velocity dimension in the cluster.

[0089] The estimated location of the point cloud cluster around the current frame is used to determine the contour based on the size of the point cloud cluster.

[0090] S1502, Adaptive Clustering:

[0091] The clustering window size is determined based on the predefined target classification and the predicted target profile;

[0092] The clustering window is defined by the predicted target profile in the distance or angle dimension and the maximum and minimum velocities from the last cycle in the velocity dimension.

[0093] Furthermore, the four-dimensional grid of the radar detection data includes radial distance dimension, horizontal angle dimension, pitch angle dimension, and radial velocity dimension.

[0094] See Figure 3 Let P be the radar point cloud dataset. Target contour information is determined based on a predefined target classification. If initial contour information is available, a clustering window of the predefined contour size is selected to perform target clustering on the radar point cloud dataset P, obtaining the first clustering result P. m If target contour information cannot be obtained due to sparse point cloud or other reasons, a clustering window with fixed parameters is used to perform target clustering on the radar point cloud dataset P to obtain a second clustering result P. g The first clustering result P m Compared with the second clustering result Pg The clustering information is merged to obtain the final target clustering result P. c .

[0095] See Figure 4 The figure is a schematic diagram of a radar-based target adaptive clustering device provided in an embodiment of this application, as shown below. Figure 4 As shown, a radar-based target adaptive clustering device includes:

[0096] Four-dimensional mesh creation unit: Acquire radar point cloud data and create a four-dimensional mesh of radar detection data;

[0097] Point cloud indexing unit: maps each detection point of the radar point cloud data to a four-dimensional grid and records the index information of each detection point in the grid;

[0098] 3D clustering window creation unit: Creates at least one grid-based 3D clustering window, in which detection points with similar velocities and similar distances are located;

[0099] Noise point filtering unit: Sets a threshold for the number of adjacent points in the window, determines whether the processing point is a noise point, and filters out the noise points;

[0100] Adaptive clustering unit: Determines the target contour based on preset information. If the target contour can be obtained, adaptive clustering is performed on the detection points; otherwise, fixed parameter clustering is performed.

[0101] The technical terms used in this application are explained as follows:

[0102] "4D" refers to the addition of elevation angle information perception capability compared to traditional millimeter-wave radar, also known as 3D radar, which only provides information in three dimensions: distance, velocity, and azimuth. This allows for high-resolution identification of longitudinal targets. "Imaging" is similar to the point cloud imaging effect of lidar. Compared to traditional millimeter-wave radar, 4D imaging radar has more than ten times more radio frequency transceiver channels. With the significant improvement in elevation angle resolution, it can present rich point cloud images of targets and the environment, as well as distance, velocity, and angle information.

[0103] Detection point: The detection point mentioned in the embodiments of this application refers to the direct microwave reflection point detected by the 4D radar from the sensor. One detection point does not necessarily directly reflect a target.

[0104] Point cloud: In microwave radar, especially millimeter-wave radar, the collection of point data reflecting back information from the sensor beam is called a point cloud. Points with larger spacing between them are called sparse point clouds, while points with smaller spacing form denser point cloud clusters.

[0105] Processing point: The detection point that the algorithm is currently calculating and processing when processing radar data is the processing point.

[0106] See Figure 5 It shows a schematic diagram of the structure of an electronic device 700 suitable for implementing embodiments of the present disclosure, the electronic device being used to implement, as Figure 5 The functions corresponding to the radar-based overflow detection device shown are as follows. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0107] like Figure 5 As shown, the electronic device 700 may include a processing unit (e.g., a central processing unit, a graphics processor, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 into a random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device 700. The processing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0108] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic device 700 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 700 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0109] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 709, or installed from storage device 708, or installed from ROM 702. When the computer program is executed by processing device 701, it performs the functions defined in the methods of embodiments of this disclosure.

[0110] It should be noted that the computer-readable medium of this application may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0111] A computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0112] In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, capable of transmitting, propagating, or transmitting a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0113] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0114] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A radar-based adaptive target clustering method, characterized in that, The method includes: Acquire radar point cloud data and create a four-dimensional grid of radar detection data; Each detection point in the radar point cloud data is mapped to a four-dimensional grid, and the index information of each detection point in the grid is recorded; Create at least one grid-based 3D clustering window containing detection points with similar velocities and similar distances; Set a threshold for the number of adjacent points in the window, determine whether the processing point is a noise point, and filter out the noise points; The target contour is determined based on preset information. If the target contour can be obtained, the detection points are adaptively clustered; otherwise, fixed parameter clustering is performed. The four-dimensional grid for creating radar detection data includes: The Fast Fourier Transform is computed along the distance dimension, where the gate cell value of the distance is used as the grid size; The mesh size in the velocity dimension is determined by the velocity gate element value; The angle is estimated using the DML algorithm, which generates an arithmetic sequence. The common difference of this sequence is used to determine the grid size in the dimensional angle. The four-dimensional grid of the radar detection data includes radial distance dimension, horizontal angle dimension, pitch angle dimension, and radial velocity dimension.

2. The radar-based target adaptive clustering method according to claim 1, characterized in that, The index information is the pre-set value information for each grid line. Index information with detection points is marked and stored, while index information without detection points is not recorded.

3. The radar-based target adaptive clustering method according to claim 1, characterized in that, In the index information, the difference between the two indices in the distance dimension represents the physical distance of the point, and the difference between the indices in the angle dimension is converted into physical distance using the following formula: 。 4. The radar-based target adaptive clustering method according to claim 1, characterized in that, Creating at least one mesh-based 3D clustering window includes: Create a mesh-based 3D clustering window around the cells at the processing point. This window can be a fixed-parameter clustering window or an adaptive clustering window. The size of the fixed-parameter clustering window is determined by two variables: a distance threshold in either the distance dimension or the angular dimension. and speed difference threshold ; The three-dimensional clustering window is a cylinder, and detection points with similar speeds and similar distances are located in this window.

5. The radar-based target adaptive clustering method according to claim 1, characterized in that, The number of grid cells occupied by the three-dimensional clustering window decreases as the range increases.

6. The radar-based target adaptive clustering method according to claim 1, characterized in that, The step of setting a threshold for the number of adjacent points in the window, determining whether a processing point is a noise point, and filtering out noise points includes: Search for the index information of all cells in a window containing a processing point, and then compare it with the list of point indices; identify the adjacent points of the processing point in the window; If the number of these points is less than the preset threshold If the condition is met, the processing point is marked as a noise point; otherwise, the point is the core point of a point cloud cluster, and all neighboring points are assigned to the cluster. Subsequently, these neighboring points are treated as new processing points and processed using the same process as the clustering window. After processing all the detection points, multiple clusters with the selected points are identified, and noise points are eliminated.

7. The radar-based target adaptive clustering method according to claim 1, characterized in that, The target contour includes: Using vector machines and machine learning, detected targets are pre-classified into pedestrians, cyclists, and vehicles; the contour size of each target type is predefined. The initial position and direction of motion of the cloud cluster are determined based on the average coordinates and average velocity of all detection points belonging to the same cloud cluster in the current frame; at the same time, the maximum and minimum velocities are stored for the clustering window of the next frame, that is, the upper and lower points of the velocity dimension in the cluster. The estimated location of the point cloud cluster around the current frame is used to determine the contour based on the size of the point cloud cluster.

8. The radar-based target adaptive clustering method according to claim 1, characterized in that, The adaptive clustering includes: The clustering window size is determined based on the predefined target classification and the predicted target profile; The clustering window is defined by the predicted target profile in the distance or angle dimension and the maximum and minimum velocities from the last cycle in the velocity dimension.

9. A radar-based target adaptive clustering device, characterized in that, include: Four-dimensional mesh creation unit: Acquire radar point cloud data and create a four-dimensional mesh of radar detection data; Point cloud indexing unit: maps each detection point of the radar point cloud data to a four-dimensional grid and records the index information of each detection point in the grid; 3D clustering window creation unit: Creates at least one grid-based 3D clustering window, in which detection points with similar velocities and similar distances are located; Noise point filtering unit: Sets a threshold for the number of adjacent points in the window, determines whether the processing point is a noise point, and filters out the noise points; Adaptive clustering unit: Determines the target contour based on preset information. If the target contour can be obtained, adaptive clustering is performed on the detection points; otherwise, fixed parameter clustering is performed. The four-dimensional grid for creating radar detection data includes: The Fast Fourier Transform is computed along the distance dimension, where the gate cell value of the distance is used as the grid size; The mesh size in the velocity dimension is determined by the velocity gate element value; The angle is estimated using the DML algorithm, which generates an arithmetic sequence. The common difference of this sequence is used to determine the grid size in the dimensional angle. The four-dimensional grid of the radar detection data includes radial distance dimension, horizontal angle dimension, pitch angle dimension, and radial velocity dimension.

10. A radar-based target adaptive clustering device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the radar-based target adaptive clustering method as described in any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the radar-based target adaptive clustering method as described in any one of claims 1-8.

Images