Point cloud clustering method, electronic equipment and storage medium
By using neighborhood range with a major axis and spatial orientation adjustment in point cloud clustering, combined with distance and orientation constraints, the problem of large clustering error in existing technologies is solved, and more accurate point cloud target segmentation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-03-10
AI Technical Summary
Existing point cloud clustering and segmentation algorithms use only one type of feature information, resulting in large clustering errors and difficulty in accurately distinguishing different targets.
Clustering is performed by using a neighborhood range with a major axis and adjusting the major axis direction of the neighborhood range based on the spatial orientation of the spatial points. By combining distance and orientation constraints, the accuracy of the clustering results is improved.
By combining spatial orientation and distance constraints of spatial points, the accuracy of clustering results is improved, solving the problems of over-segmentation and over-clustering in traditional algorithms, and improving the accuracy of point cloud clustering.
Smart Images
Figure CN121640112A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of radar technology, specifically to a point cloud clustering method, electronic device, and storage medium. Background Technology
[0002] With the rapid development of fields such as intelligent driving and smart transportation, various types of radar have become one of the core sensors in environmental perception systems. In radar applications, point cloud clustering and segmentation is a key preliminary step for achieving target recognition, tracking, and behavior prediction.
[0003] At present, although point cloud processing technology has made some progress and can initially divide the point cloud detected by radar, the traditional clustering and segmentation algorithms use only one type of point cloud feature information, which makes it difficult to accurately distinguish different targets, resulting in large clustering errors. Summary of the Invention
[0004] To address the aforementioned technical problems, this application provides a point cloud clustering method, electronic device, and storage medium. It uses a neighborhood range with a major axis, adjusts the major axis direction of the neighborhood range based on the spatial orientation of the spatial points, and combines distance and orientation constraints to improve the accuracy of the clustering results.
[0005] To address the aforementioned technical problems, this application provides a point cloud clustering method, comprising: Obtain the target point cloud to be clustered; Determine the spatial orientation of each spatial point in the target point cloud, wherein the spatial orientation is used to characterize the spatial orientation of the target to which the spatial point belongs; A first feature constraint is determined, which includes a first neighborhood range of a spatial point, and the first neighborhood range has a major axis. Based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint, density-based spatial clustering is performed on the target point cloud. During the clustering process, the direction of the major axis of the first neighborhood range changes with the spatial orientation of the spatial point. Obtain the clustering results of the target point cloud.
[0006] In some embodiments, obtaining the target point cloud to be clustered includes: Acquire the initial point cloud collected by the device; The initial point cloud is filtered based on a preset region of interest to obtain a first point cloud; The first point cloud is filtered based on the RCS values of each spatial point in the first point cloud to obtain the second point cloud; The second point cloud is filtered based on the existence probability, velocity ambiguity state, and effective state of each spatial point in the second point cloud to obtain the target point cloud.
[0007] In some embodiments, the first feature constraint further includes a first velocity range and a first RCS value range; the density-based spatial clustering of the target point cloud based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint includes: The target point cloud is divided based on the first velocity range and the first RCS value range to determine multiple first point cloud clusters; Based on the first neighborhood range, density-based spatial clustering is performed on the plurality of first point cloud clusters. During clustering, the direction of the major axis of the first neighborhood range follows the spatial orientation of the core spatial point among the spatial points.
[0008] In some embodiments, the method for determining the first neighborhood range includes: Determine the Euclidean distance of the circular neighbor gate; The Euclidean distance circular neighborhood gate is adjusted based on a preset weight value to obtain an Euclidean distance elliptical neighborhood gate with a major axis and a minor axis, which is used as the first neighborhood range.
[0009] In some embodiments, determining the spatial orientation of each spatial point in the target point cloud includes: The target point cloud is clustered based on the second feature constraint to determine multiple second point cloud clusters; Spatial attitude analysis is performed on the plurality of second point cloud clusters to determine their spatial orientation. Based on the spatial orientation of the plurality of second point cloud clusters, the spatial orientation of each spatial point in the target point cloud is obtained.
[0010] In some embodiments, the second feature constraint includes a second velocity range, a second RCS value range, and a second neighborhood range of the spatial point; the clustering of the target point cloud based on the second feature constraint to determine multiple second point cloud clusters includes: The target point cloud is divided based on the second velocity range and the second RCS value range to determine multiple third point cloud clusters; Based on the second neighborhood range, density-based spatial clustering is performed on the plurality of third point cloud clusters to obtain the plurality of second point cloud clusters.
[0011] In some embodiments, performing spatial attitude analysis on the plurality of second point cloud clusters to determine the spatial orientation of the plurality of second point cloud clusters includes: Obtain the dynamic attributes of each of the plurality of second point cloud clusters; For a second point cloud cluster with a dynamic attribute of motion, the spatial orientation of the second point cloud cluster is determined based on the velocity vector of the spatial points in the second point cloud cluster. For a second point cloud cluster whose dynamic attribute is static, the spatial orientation of the second point cloud cluster is determined based on the spatial distribution characteristics of the spatial points in the second point cloud cluster.
[0012] In some embodiments, determining the spatial orientation of the second point cloud cluster based on the velocity vector of a spatial point in the second point cloud cluster includes: The velocity vector of all spatial points in the second point cloud cluster is obtained by weighted averaging the velocity vectors of all spatial points in the second point cloud cluster based on the existence probability of each spatial point in the second point cloud cluster. The directional components of the mean velocity vector in the radar coordinate system are analyzed to obtain the spatial orientation of the second point cloud cluster.
[0013] This application also provides an electronic device, including a storage medium and a controller, wherein a computer program is stored on the storage medium, and the computer program, when executed by the controller, implements the steps of the method described above.
[0014] This application also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described above.
[0015] This application discloses a point cloud clustering method, electronic device, and storage medium. The method includes: acquiring a target point cloud to be clustered; determining the spatial orientation of each spatial point in the target point cloud, whereby the spatial orientation characterizes the spatial orientation of the target to which the spatial point belongs; determining a first feature constraint, which includes a first neighborhood range of the spatial point, the first neighborhood range having a major axis; performing density-based spatial clustering of the target point cloud based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint, wherein the direction of the major axis of the first neighborhood range follows the spatial orientation of the spatial point; and acquiring the clustering result of the target point cloud. The technical solution of this application uses a neighborhood range with a major axis, adjusts the direction of the major axis of the neighborhood range based on the spatial orientation of the point cloud spatial points for clustering, and combines distance and orientation constraints to improve the accuracy of the clustering results. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a point cloud clustering method according to one embodiment.
[0017] Figure 2 This is another flowchart illustrating a point cloud clustering method according to one embodiment.
[0018] Figure 3 This is a schematic diagram of the structure of an electronic device according to one embodiment. Detailed Implementation
[0019] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.
[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. In this application, "each" includes one or more items.
[0021] Figure 1 This is a flowchart illustrating a point cloud clustering method according to one embodiment. For example... Figure 1 As shown, this application discloses a point cloud clustering method applied to electronic devices, comprising the following steps: S1, Obtain the target point cloud to be clustered; S2, determine the spatial orientation of each spatial point in the target point cloud. The spatial orientation is used to characterize the spatial orientation of the target to which the spatial point belongs. S3, determine the first feature constraint condition, which includes the first neighborhood range of the spatial point, and the first neighborhood range has a major axis; S4. Based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint, perform density-based spatial clustering on the target point cloud. During the clustering process, the direction of the major axis of the first neighborhood range changes with the spatial orientation of the spatial points. S5, obtain the clustering results of the target point cloud.
[0022] The target point cloud is a point cloud that needs to be clustered and segmented. The target point cloud consists of a large number of spatial points. The target to which the spatial points belong refers to the target obtained by clustering through a clustering algorithm. Compared with the target in the real environment, the target obtained by the clustering algorithm may correspond to a part of one or more targets in the real environment, or correspond to multiple targets in the real environment.
[0023] The spatial orientation of a target to which a spatial point belongs is the target's orientation in the radar coordinate system. Typically, different parts of the same target have the same orientation. Since clustering is the process of clustering spatial points to obtain targets, by assigning spatial points the characteristic of representing the target's spatial orientation, spatial orientation constraints can be imposed during the clustering process. This enhances the clustering of spatial points in terms of spatial orientation, allowing spatial points belonging to the same target to be more accurately classified into the same category.
[0024] The first feature constraint is a condition imposed on the features of spatial points when clustering a target point cloud. This first feature constraint can include one or more constraints. The first neighborhood range of a spatial point refers to a certain distance range centered on that spatial point. It constrains the distance characteristics of the spatial point; spatial points within this distance range are considered "neighbors" of the centered point. In density-based spatial clustering, when the number of spatial points within the first neighborhood range reaches or exceeds a preset minimum number of points (MinPts), this centered point is considered the core spatial point. The core spatial point is the basis for forming clusters. The clustering algorithm expands and connects density-reachable clusters through the core spatial point and the neighborhood range, thus forming the final clustering result. Density-based spatial clustering algorithms, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), are well-known to those skilled in the art and will not be elaborated upon further.
[0025] Existing density-based spatial clustering algorithms typically use circular neighborhood ranges, meaning that the neighborhood ranges do not have a distinction between major and minor axes. Therefore, during the clustering process, the neighborhood ranges lack directionality and only perform distance constraints, which may result in the point cloud of a real target being clustered into multiple targets or the point clouds of multiple targets being clustered into a single target, leading to low accuracy in the clustering results.
[0026] To address this, based on the spatial distribution characteristics of targets relative to radar equipment—"extending along the orientation and contracting vertically"—this application employs a first neighborhood range with a major axis. During clustering, the direction of the major axis of the first neighborhood range follows the spatial orientation of the spatial points, enabling dynamic matching between neighborhood parameters and target characteristics. By combining distance and spatial orientation constraints, core spatial points can be accurately determined. Furthermore, when expanding and connecting densely accessible clusters through core spatial points and neighborhood ranges, on the one hand, spatial points with the same spatial orientation and meeting distance requirements can be grouped into the same target, improving the over-segmentation phenomenon caused by distance-based clustering. On the other hand, spatial points with the same distance but different spatial orientations will not be grouped into the same target, thus accurately distinguishing adjacent targets with different spatial orientations and improving the over-clustering problem caused by merging adjacent targets with opposite orientations in distance-based clustering. Therefore, the accuracy of the clustering results is effectively improved.
[0027] In some embodiments, step S1, obtaining the target point cloud to be clustered, includes: Acquire the initial point cloud collected by the device; The initial point cloud is filtered based on a preset region of interest to obtain the first point cloud; The first point cloud is filtered based on the RCS values of each spatial point in the first point cloud to obtain the second point cloud; The second point cloud is filtered based on the existence probability, velocity fuzziness state, and effective state of each spatial point in the second point cloud to obtain the target point cloud.
[0028] In this process, the initial point cloud collected by the device is preprocessed to obtain the target point cloud, thereby improving the purity and reliability of the point cloud data and laying a solid foundation for the accurate execution of subsequent target clustering and segmentation tasks.
[0029] First, the initial point cloud is filtered based on a preset region of interest (ROI) to initially identify potential targets within a specific spatial range, achieving initial focusing on valid data. For example, the preset ROI can be set to within 70m, then the point cloud within the radar detection range of 70m is the first point cloud. Next, the first point cloud is filtered based on the RCS (radar cross-section) values of each spatial point to further filter out targets of interest that meet the requirements, eliminating redundant data that does not meet the scene requirements, and obtaining the second point cloud. Specifically, by setting the selection range of RCS values, targets with different reflection intensities, such as roads, rocks, and objects with different levels of transparency, are distinguished, and unnecessary point clouds such as roads and rocks are eliminated. Afterward, the second point cloud is filtered based on the existence probability, velocity ambiguity state, and valid state of each spatial point, performing a final verification of the point cloud data, and selecting the point clouds corresponding to reasonably existing targets to obtain the target point cloud. Specifically, by setting the selection range of existence probability, velocity ambiguity state, and valid state, reasonably existing target point clouds are selected. When the point cloud is a millimeter-wave radar point cloud, it can be 2D point cloud data projected onto a set plane. Typically, the point cloud data covers three core dimensions of information: target state information (e.g., number of points, point cloud ID), target general information (e.g., velocity, RCS), and target quality information (e.g., existence probability, velocity ambiguity, and effective state).
[0030] In some embodiments, when preprocessing the initial point cloud, a temporal processing mechanism can be introduced to filter dynamic point clouds with large positional offsets between consecutive frames by combining multiple frames of historical point cloud data, thereby preserving a stable target point cloud in the current frame.
[0031] In some embodiments, the first feature constraint further includes a first velocity range and a first RCS value range; density-based spatial clustering of the target point cloud based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint includes: The target point cloud is divided based on the first velocity range and the first RCS value range to determine multiple first point cloud clusters; Based on the first neighborhood range, density-based spatial clustering is performed on multiple first point cloud clusters. During clustering, the direction of the major axis of the first neighborhood range follows the spatial orientation of the core spatial point in the spatial points.
[0032] Before performing density-based spatial clustering on multiple first point cloud clusters based on their first neighborhood range, constraints are first applied using velocity and RCS values. On one hand, a velocity range is set to consider the consistency and continuity of target motion, effectively distinguishing adjacent targets with different speeds (such as stationary guardrails and moving vehicles), reducing mis-clustering in dynamic scenes. On the other hand, a range for RCS values is set, further optimizing the merging logic of targets within point cloud clusters obtained from velocity constraints, based on the characteristic that the reflection intensity of the same target is usually similar. This allows point clouds of the same target to be more accurately classified into one category, thus identifying multiple first point cloud clusters. For each of the multiple first point cloud clusters, density-based spatial clustering is performed. During clustering, a neighborhood range with a major axis is used. The direction of the major axis of the neighborhood range follows the spatial orientation of the core spatial point. By combining distance and spatial orientation, multiple temporary point cloud clusters centered on the core spatial point are identified. Temporary point cloud clusters with consistent or similar spatial orientations are clustered into point clouds of the same target, while temporary point cloud clusters with significantly different spatial orientations are clustered into point clouds of different targets. In actual implementation, to further optimize the oversegmentation problem, the range of the first RCS value can be selected to be relatively wide, so that the various parts of the large target are divided into the same point cloud cluster as much as possible. Then, by combining spatial clustering based on distance and spatial orientation, the point cloud clusters are further divided.
[0033] Thus, this application integrates geometric and motion features to construct a density-based spatial clustering framework with multi-dimensional constraints, which can solve the target confusion problem that is prone to occur in single feature clustering, such as the mis-merging of targets with different speeds at close range and the mis-splitting of targets with the same speed but different distances, thereby improving the accuracy of clustering results.
[0034] In some embodiments, the method for determining the first neighborhood range in step S2 includes: Determine the Euclidean distance of the circular neighbor gate; The Euclidean distance circular neighborhood gate is adjusted based on the preset weight value to obtain the Euclidean distance elliptical neighborhood gate with major and minor axes, which is used as the first neighborhood range.
[0035] The preset weight values can be the system default weight values or the weight values set by the user. They can also be dynamically adjusted during the clustering process based on the distance between the spatial points and the radar, further improving the accuracy of the clustering results. For example, the weights of the major axis can be positively correlated with the distance, while the weights of the minor axis can be negatively correlated with the distance.
[0036] like Figure 2 As shown, to achieve automatic and accurate acquisition of the spatial orientation of spatial points, step S2, determining the spatial orientation of each spatial point in the target point cloud, includes the following steps: S21, cluster the target point cloud based on the second feature constraint to determine multiple second point cloud clusters; S22, Perform spatial attitude analysis on multiple second-point cloud clusters to determine the spatial orientation of multiple second-point cloud clusters; S23, based on the spatial orientation of multiple second point cloud clusters, obtain the spatial orientation of each spatial point in the target point cloud.
[0037] The second feature constraint is a condition imposed on the features of spatial points when clustering the target point cloud. The second feature constraint can include one or more constraints. First, a first-stage clustering is performed, which clusters the target point cloud based on the second feature constraint. The purpose of the first-stage clustering is to obtain the spatial orientation of the spatial points based on the clustering results. Then, a second-stage clustering is performed, which clusters the target point cloud based on the first feature constraint. In the second-stage clustering process, the spatial orientation is combined with the first feature constraint to obtain more accurate clustering results.
[0038] After clustering the target point cloud based on the second feature constraint, multiple second point cloud clusters are determined. At this time, each point cloud cluster is equivalent to a target. Spatial attitude analysis is performed on multiple second point cloud clusters to determine the spatial orientation of multiple second point cloud clusters, which is also the spatial orientation of the target. Then, the spatial orientation of the target is assigned to the corresponding spatial point. At this time, the spatial point in the target point cloud has a spatial orientation, and the spatial orientation of the spatial point can represent the spatial orientation of the target to which it belongs.
[0039] In some embodiments, the second feature constraint includes a second velocity range, a second RCS value range, and a second neighborhood range of the spatial point; step S21, clustering the target point cloud based on the second feature constraint to determine multiple second point cloud clusters, including: The target point cloud is divided based on the second velocity range and the second RCS value range to determine multiple third point cloud clusters; Based on the second neighborhood range, multiple third point cloud clusters are spatially clustered according to density to obtain multiple second point cloud clusters.
[0040] The second neighborhood of a spatial point refers to a certain distance range centered on a spatial point. It is a constraint on the distance characteristics of the spatial point. Spatial points within this distance range are the "neighbors" of the central point. When performing density-based spatial clustering, if the number of spatial points contained in the second neighborhood reaches or exceeds the preset minimum number of points, this central point is called the core spatial point. The core spatial point is the basis for forming clusters. The clustering algorithm expands and connects the density-reachable clusters through the core spatial point and the neighborhood range, thereby forming the final clustering result.
[0041] Since the purpose of one-stage clustering is to obtain point cloud clusters for analyzing spatial orientation, it is necessary to cover all spatial points of the target point cloud as much as possible and perform finer segmentation to improve the accuracy of spatial orientation. When selecting the second velocity range, the second RCS value range, and the second neighborhood range of the spatial points, the second feature constraint conditions can be different. The second RCS value range can be a smaller numerical range to achieve finer target segmentation, and the second neighborhood range can be a traditional circular neighborhood range to cover all spatial points as much as possible without spatial orientation.
[0042] After the first-stage clustering, some spatial points in the target point cloud may be identified as noise points. Since noise points are usually not clustered into point cloud clusters, their spatial orientation cannot be obtained through spatial pose analysis. Therefore, the spatial orientation of noise points is defined as zero, and the spatial orientation of their target is also defined as zero. In this case, when performing the second-stage clustering, the direction of the major axis of the first neighborhood range follows the spatial orientation of the spatial points, including at least one of the following: When the spatial orientation of a spatial point is not zero, the direction of the major axis of the first neighborhood range is parallel to the spatial orientation of the spatial point. When the spatial orientation of a point is zero, the direction of the major axis of the first neighborhood range is parallel to any coordinate axis of the radar coordinate system or the default direction is used.
[0043] In some embodiments, step S22, performing spatial attitude analysis on multiple second point cloud clusters to determine the spatial orientation of the multiple second point cloud clusters, includes: Obtain the dynamic attributes of each second point cloud cluster in multiple second point cloud clusters; For a second point cloud cluster with a dynamic attribute of motion, the spatial orientation of the second point cloud cluster is determined based on the velocity vector of the spatial points in the second point cloud cluster. For a second point cloud cluster whose dynamic attribute is static, the spatial orientation of the second point cloud cluster is determined based on the spatial distribution characteristics of the spatial points in the second point cloud cluster.
[0044] Among them, the second feature constraint includes a velocity constraint, which makes each second point cloud cluster obtained by clustering have corresponding dynamic attributes. The target (second point cloud cluster) can be divided into moving targets and stationary targets. Different target orientation estimation methods are adopted in a targeted manner, which can reduce the computational complexity and achieve full coverage of dynamic and static scenes.
[0045] For second point cloud clusters with a dynamic attribute of motion, the spatial orientation of the second point cloud cluster is determined based on the velocity vectors of the spatial points within it. This directly utilizes the inherent dynamic characteristics of the moving target, eliminating the need for extracting complex spatial structure information and resulting in high computational efficiency. For second point cloud clusters with a dynamic attribute of stationary motion, the spatial orientation of the second point cloud cluster is determined based on the spatial distribution characteristics of the spatial points within it. The orientation of a stationary target can be derived solely from the spatial distribution characteristics, making the computation simple and demonstrating robustness even under conditions of low point cloud density and uneven distribution.
[0046] In some embodiments, determining the spatial orientation of the second point cloud cluster based on the velocity vector of a spatial point in the second point cloud cluster includes: The velocity vector of the second point cloud cluster is obtained by weighted averaging the velocity vectors of all spatial points in the second point cloud cluster based on the existence probability of each spatial point in the second point cloud cluster. The directional components of the mean velocity vector in the radar coordinate system are analyzed to obtain the spatial orientation of the second point cloud cluster.
[0047] For moving targets, their orientation is usually strongly correlated with their direction of motion. Therefore, for the second cloud cluster with the dynamic attribute of motion, a weighted velocity component estimation method is used to determine the spatial orientation of the second cloud cluster. The steps are as follows: The velocity vector of the second point cloud cluster is obtained by weighted averaging of the velocity vectors of all spatial points within the second point cloud cluster. The weights used in the weighting are combined with the existence probability in the point cloud quality information, and the magnitude of the weights is positively correlated with the existence probability. The directional component of the average velocity vector in the radar coordinate system is analyzed, and this direction is used as the spatial orientation of the target.
[0048] Thus, by combining the existence probability of spatial points to calculate the mean velocity vector, the accuracy of spatial orientation can be improved.
[0049] For stationary targets, since they have no obvious motion trend, their orientation cannot be derived from velocity characteristics. Therefore, Principal Component Analysis (PCA) is used to determine the spatial orientation of the second point cloud cluster based on the spatial distribution characteristics of the spatial points within the second point cloud cluster. The specific steps are as follows: Extract the spatial coordinates of all spatial points within the second cloud cluster; Calculate the covariance of the coordinate matrix, and analyze the degree of dispersion of the second point cloud cluster in different spatial directions through the covariance matrix, so as to obtain the spatial distribution characteristics of the spatial points in the second point cloud cluster. The covariance matrix is decomposed into eigenvalues. The direction corresponding to the eigenvector with the largest eigenvalue is the principal component direction of the second point cloud cluster. This direction is taken as the spatial orientation of the second point cloud cluster.
[0050] The technical solution of this application first inputs an initial point cloud collected by the device and preprocesses it to obtain a target point cloud. Next, a neighborhood range is defined, and multi-dimensional features of the point cloud are fused to perform a first-stage clustering of the target point cloud, yielding a clustering result. Then, based on the dynamic attributes of the clustered point cloud clusters, the clustering result is divided into stationary and moving targets, and their spatial orientation is estimated using mean velocity components and principal component analysis, respectively, to obtain the spatial orientation of the spatial points in the target point cloud. Finally, for the target point cloud, based on the spatial orientation of the spatial points and combined with multi-feature constraints, a second-stage clustering is performed using a neighborhood range with a major axis combined with the direction of rotation of the neighborhood range to obtain the final clustering result. Thus, the clustering method can flexibly adapt to point cloud data of targets of different sizes, and can adaptively adjust the neighborhood range of the clustering according to the point cloud orientation. Especially for large targets, it can effectively improve the over-segmentation problem, effectively overcome the limitations of traditional algorithms, significantly improve clustering accuracy, and is computationally simple, saving computational resources.
[0051] Based on the same inventive concept as the foregoing embodiments, this invention provides an electronic device, such as... Figure 3 As shown, the electronic device includes: a processor 310 and a memory 311 storing a computer program; wherein, Figure 3 The processor 310 shown in the diagram does not indicate that there is only one processor 310, but only indicates the positional relationship of the processor 310 relative to other devices. In practical applications, there can be one or more processors 310; similarly, Figure 3 The memory 311 shown in the diagram has the same meaning, that is, it is only used to indicate the positional relationship of memory 311 relative to other devices. In practical applications, there can be one or more memories 311. When the processor 310 runs the computer program, it implements the point cloud clustering method described above.
[0052] The electronic device may also include at least one network interface 312. The various components of the electronic device are coupled together via a bus system 313. It is understood that the bus system 313 is used to implement communication between these components. In addition to a data bus, the bus system 313 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 3 The general designated all buses as Bus System 313.
[0053] The memory 311 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Sync Link Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 311 described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.
[0054] The memory 311 in this embodiment of the invention is used to store various types of data to support the operation of the electronic device. Examples of this data include: any computer programs used to operate on the electronic device, such as operating systems and applications; contact data; phonebook data; messages; pictures; videos, etc. The operating system includes various system programs, such as the framework layer, core library layer, driver layer, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications, such as media players, browsers, etc., used to implement various application services. Here, the program implementing the method of this embodiment of the invention can be included in the application.
[0055] Based on the same inventive concept as the foregoing embodiments, this embodiment also provides a computer-readable storage medium storing a computer program. The computer-readable storage medium can be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM), etc.; it can also be various devices including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc. When the computer program stored in the computer-readable storage medium is executed by a processor, it implements the point cloud clustering method described above. For the specific steps implemented when the computer program is executed by the processor, please refer to [link to relevant documentation]. Figure 1 The description of the illustrated embodiments will not be repeated here.
[0056] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A clustering method of point clouds, characterized in that, The method comprises the following steps: acquiring a target point cloud to be clustered; determining a spatial orientation of each spatial point in the target point cloud, the spatial orientation being used to represent a spatial orientation of a target to which the spatial point belongs; determining a first feature constraint condition, the first feature constraint condition comprising a first neighborhood range of a spatial point, the first neighborhood range having a long axis; performing density-based spatial clustering on the target point cloud based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint condition, during the clustering process, the direction of the long axis of the first neighborhood range changing along with the spatial orientation of the spatial point; acquiring a clustering result of the target point cloud.
2. The method of claim 1, wherein, The method comprises the following steps: acquiring an initial point cloud collected by a device; performing filtering on the initial point cloud based on a preset region of interest to acquire a first point cloud; performing filtering on the first point cloud based on an RCS value of each spatial point in the first point cloud to acquire a second point cloud; performing filtering on the second point cloud based on an existence probability, a velocity ambiguity state and an effective state of each spatial point in the second point cloud to acquire the target point cloud.
3. The method of claim 1, wherein, The first feature constraint condition further comprises a first velocity range and a first RCS value range; and the method of performing density-based spatial clustering on the target point cloud based on the spatial orientation of each spatial point in the target point cloud and the first feature constraint condition comprises the following steps: dividing the target point cloud based on the first velocity range and the first RCS value range to determine a plurality of first point cloud clusters; performing density-based spatial clustering on the plurality of first point cloud clusters based on the first neighborhood range, during the clustering process, the direction of the long axis of the first neighborhood range changing along with the spatial orientation of a core spatial point in the spatial point.
4. The method of claim 1, wherein, The method of determining the first neighborhood range comprises the following steps: determining a Euclidean distance circular neighborhood gate; adjusting the Euclidean distance circular neighborhood gate based on a preset weight value to obtain a Euclidean distance elliptical neighborhood gate having a long axis and a short axis, so as to serve as the first neighborhood range.
5. The method according to any one of claims 1 to 4, characterized in that, The method of determining the spatial orientation of each spatial point in the target point cloud comprises the following steps: performing clustering on the target point cloud based on a second feature constraint condition to determine a plurality of second point cloud clusters; performing spatial posture analysis on the plurality of second point cloud clusters to determine a spatial orientation of the plurality of second point cloud clusters; acquiring the spatial orientation of each spatial point in the target point cloud based on the spatial orientation of the plurality of second point cloud clusters.
6. The method of claim 5, wherein, The second feature constraint condition comprises a second velocity range, a second RCS value range and a second neighborhood range of a spatial point; and the method of performing clustering on the target point cloud based on the second feature constraint condition to determine a plurality of second point cloud clusters comprises the following steps: dividing the target point cloud based on the second velocity range and the second RCS value range to determine a plurality of third point cloud clusters; performing density-based spatial clustering on the plurality of third point cloud clusters based on the second neighborhood range to obtain the plurality of second point cloud clusters.
7. The method of claim 5, wherein, The method of performing spatial posture analysis on the plurality of second point cloud clusters to determine a spatial orientation of the plurality of second point cloud clusters comprises the following steps: acquiring a dynamic attribute of each second point cloud cluster in the plurality of second point cloud clusters; for a second point cloud cluster with a dynamic attribute of motion, determining a spatial orientation of the second point cloud cluster according to velocity vectors of spatial points in the second point cloud cluster; for a second point cloud cluster with a dynamic attribute of stillness, determining a spatial orientation of the second point cloud cluster according to spatial distribution characteristics of spatial points in the second point cloud cluster.
8. The method of claim 7, wherein, The determining of the spatial orientation of the second point cloud cluster according to the velocity vectors of the spatial points in the second point cloud cluster comprises: weighting and averaging the velocity vectors of all spatial points in the second point cloud cluster according to the existence probabilities of the spatial points in the second point cloud cluster to obtain a mean velocity vector of the second point cloud cluster; analyzing a direction component of the mean velocity vector in a radar coordinate system to obtain the spatial orientation of the second point cloud cluster.
9. An electronic device, comprising: A storage medium and a controller are included, and the storage medium has a computer program stored thereon, and the computer program, when executed by the controller, implements the steps of the method according to any one of claims 1 to 8.
10. A storage medium, characterized by The storage medium has a computer program stored thereon, and the computer program, when executed by the processor, implements the steps of the method according to any one of claims 1 to 8.