Intersecting plane point cloud feature extraction method and system

Abstract

The invention discloses an intersection plane point cloud feature extraction method and system, and relates to the technical field of positioning, and the method comprises the steps: collecting a laser radar point cloud, and carrying out the preprocessing; calculating a normal vector of the preprocessed point cloud, and performing clustering analysis on the normal vector to obtain a cluster; performing plane fitting on the preprocessed point cloud to obtain a current plane equation; judging whether intersecting planes exist or not according to the clusters, if not, performing boundary smoothing processing on the planes corresponding to all the plane equations, outputting the processed planes, and ending the program; if so, extracting an inner point of the current plane equation; removing the inner point from the preprocessed point cloud to obtain a residual point cloud; and taking the residual point cloud as the pre-processing point cloud, and returning to the step of calculating the normal vector of the pre-processing point cloud. According to the method, through an iteration mode, the intersecting plane features in the point cloud can be effectively segmented, each independent plane is ensured to be extracted, and the accuracy and robustness of segmentation are improved while the calculation efficiency is ensured.

Description

Technical Field

[0001] This invention relates to the field of positioning technology, and more specifically, to a method and system for extracting features from intersecting plane point clouds. Background Technology

[0002] LiDAR point cloud feature extraction is a crucial step in improving localization accuracy in autonomous driving and robotic perception. Existing feature extraction methods typically extract planar features from the point cloud and use the point cloud map for matching within the robot's localization system. However, in complex scenarios, situations often arise where two or more planes intersect, such as building corners or corridor intersections. The presence of these intersecting planes often leads to the failure of planar feature extraction, resulting in reduced localization performance or even complete failure in such scenarios.

[0003] Currently, common point cloud planar feature extraction methods mainly rely on the Random Sample Consensus (RANSAC) algorithm for planar fitting. However, when faced with two perpendicularly intersecting planes, traditional RANSAC methods often fit a single plane, failing to effectively separate individual planes, leading to planar feature extraction failure. This, in turn, affects the accuracy of point cloud matching and reduces the robustness of localization. Some improved algorithms attempt to distinguish multiple planes by increasing computational load, resulting in excessively long computation times and impacting real-time performance.

[0004] Therefore, the present invention provides a method and system for extracting features from intersecting plane point clouds, which solves the problem that existing technologies cannot effectively segment multiple intersecting plane features in point clouds. Summary of the Invention

[0005] This invention provides a method and system for extracting features from intersecting plane point clouds, in order to overcome at least one technical problem existing in the prior art.

[0006] On one hand, embodiments of the present invention provide a method for extracting features from intersecting plane point clouds, including:

[0007] Acquire lidar point clouds;

[0008] The lidar point cloud is preprocessed to obtain a preprocessed point cloud;

[0009] Calculate the normal vectors of the preprocessed point cloud, and perform cluster analysis on the normal vectors to obtain clusters;

[0010] Perform plane fitting on the preprocessed point cloud to obtain the current plane equation;

[0011] Based on the clusters, determine whether there are intersecting planes. If not, perform boundary smoothing on all plane equations and output the processed planes, then the program ends. If they exist, extract the interior points of the current plane equation.

[0012] The interior points are removed from the preprocessed point cloud to obtain the remaining point cloud;

[0013] The remaining point cloud is used as a preprocessed point cloud, and the step of calculating the normal vector of the preprocessed point cloud is returned.

[0014] Optionally, the lidar point cloud is preprocessed, specifically as follows:

[0015] The radius filtering method is used to denoise the lidar point cloud and remove outliers.

[0016] Voxel filtering is used to downsample the point cloud after outlier removal.

[0017] Optionally, cluster analysis is performed on the normal vectors, specifically as follows:

[0018] Calculate the angles between all normal vectors;

[0019] Clustering algorithms are used to cluster the included angles to obtain clusters.

[0020] Optionally, cluster analysis is performed on the normal vectors, specifically as follows:

[0021] Calculate the angles between all normal vectors and construct a histogram;

[0022] Detect the peak value of the histogram;

[0023] Based on the peak values ​​of the histogram, clusters are obtained.

[0024] Optionally, the preprocessed point cloud is fitted with the Random Sampling Consensus Algorithm (RANSAC) to obtain the plane equation Ax + By + Cz + D = 0.

[0025] Optionally, boundary smoothing is performed on the planes corresponding to all plane equations, specifically as follows:

[0026] Set boundary thresholds;

[0027] Boundary points are obtained based on the normal vector mutation method;

[0028] Calculate the standard deviation of the neighborhood normal vectors of the boundary points;

[0029] Remove boundary points whose standard deviation exceeds the boundary threshold.

[0030] Optionally, the normal vector can be calculated using principal component analysis (PCA) or covariance matrix decomposition.

[0031] On the other hand, the present invention also provides a feature extraction system for intersecting plane point clouds, comprising:

[0032] The acquisition module is used to acquire point clouds from the lidar.

[0033] The preprocessing module is used to preprocess the lidar point cloud to obtain a preprocessed point cloud;

[0034] The calculation module is used to calculate the normal vectors of the preprocessed point cloud and perform cluster analysis on the normal vectors to obtain clusters;

[0035] The fitting module is used to perform plane fitting on the preprocessed point cloud to obtain the current plane equation;

[0036] The system includes a judgment module and a smoothing module. The judgment module is used to determine whether there are intersecting planes based on the clusters. If there are no intersecting planes, the smoothing module is used to perform boundary smoothing on the planes corresponding to all plane equations and outputs the processed planes. The program then ends. If there are intersecting planes, the interior points of the current plane equation are extracted.

[0037] A removal module is used to remove the interior points from the preprocessed point cloud to obtain the remaining point cloud;

[0038] The return module is used to treat the remaining point cloud as a preprocessed point cloud and return the step of calculating the normal vector of the preprocessed point cloud.

[0039] Optionally, the preprocessing module is specifically used for:

[0040] The radius filtering method is used to denoise the lidar point cloud and remove outliers.

[0041] Voxel filtering is used to downsample the point cloud after outlier removal.

[0042] Optionally, the smoothing module is specifically used for:

[0043] Set boundary thresholds;

[0044] Boundary points are obtained based on the normal vector mutation method;

[0045] Calculate the standard deviation of the neighborhood normal vectors of the boundary points;

[0046] Remove boundary points whose standard deviation exceeds the boundary threshold.

[0047] The innovative aspects of this invention include:

[0048] In this embodiment, the intersecting plane features in the point cloud can be effectively segmented through iteration, ensuring the extraction of each independent plane. While ensuring computational efficiency, the accuracy and robustness of segmentation are also improved, which is one of the innovative points of this embodiment. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of a method for extracting features from intersecting plane point clouds provided in an embodiment of the present invention;

[0051] Figure 2 This invention provides a flowchart for cluster analysis of normal vectors.

[0052] Figure 3 This is another flowchart for cluster analysis of normal vectors provided by the present invention;

[0053] Figure 4 A flowchart of boundary smoothing processing provided in an embodiment of the present invention;

[0054] Figure 5 This is a schematic diagram of a feature extraction system for intersecting plane point clouds provided in an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that the terms "comprising" and "having," and any variations thereof, in the embodiments and drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0057] This invention discloses a method and system for extracting features from intersecting plane point clouds. These will be described in detail below.

[0058] Figure 1 This is a flowchart of an intersecting plane point cloud feature extraction method provided in an embodiment of the present invention. Please refer to it. Figure 1 The present invention provides a method for extracting features from intersecting plane point clouds, comprising:

[0059] Step 1: Acquire point clouds from lidar;

[0060] Step 2: Preprocess the lidar point cloud to obtain a preprocessed point cloud;

[0061] Step 3: Calculate the normal vectors of the preprocessed point cloud and perform cluster analysis on the normal vectors to obtain clusters;

[0062] Step 4: Perform plane fitting on the preprocessed point cloud to obtain the current plane equation;

[0063] Step 5: Determine whether there are intersecting planes based on the clusters. If not, perform boundary smoothing on all planes corresponding to the plane equations and output the processed planes. The program ends. If they exist, extract the interior points of the current plane equation.

[0064] Step 6: Remove the interior points from the preprocessed point cloud to obtain the remaining point cloud;

[0065] Step 7: Use the remaining point cloud as the preprocessed point cloud and return to the step of calculating the normal vector of the preprocessed point cloud.

[0066] For details, please refer to Figure 1 The intersecting plane point cloud feature extraction method provided in this embodiment of the invention improves upon the Random Sampling Consensus Algorithm (RANSAC) by iteratively segmenting independent planes. First, a lidar point cloud is acquired in step 1. Typically, the acquired lidar point cloud data contains outliers and noise, which can affect the calculation results. Therefore, this invention preprocesses the lidar point cloud in step 2 to obtain a preprocessed point cloud.

[0067] During preprocessing, noise reduction can be performed on the LiDAR point cloud to remove outliers. In this embodiment, a radius filtering method can be used to remove outliers. That is, a point is selected as the origin, the distance of each point from the origin is calculated, and only points within a predetermined radius are selected. For example, assuming the radius is set to 70m, only LiDAR point clouds within a distance of 70m are retained.

[0068] Furthermore, since large amounts of data can affect computation speed, after removing outliers, the remaining point cloud is downsampled to reduce subsequent computation. In this embodiment, speed-up filtering is used for downsampling. It should be noted that in other embodiments, uniform sampling can also be used for downsampling, and this invention does not specifically limit this method.

[0069] The normal vector of a point is also the normal vector of the plane containing that point. Therefore, points on the same plane have essentially the same normal vector; conversely, a group of points with similar normal vectors can be considered to be on the same plane. It is evident that by clustering the normal vectors of the lidar point cloud, the lidar point cloud can be classified. Based on this, in step 3 of this invention, the normal vectors of the preprocessed point cloud are first calculated. Principal Component Analysis (PCA) or covariance matrix decomposition can be used to calculate the normal vectors of the preprocessed point cloud; the specific method can be chosen according to the actual situation, and this invention does not impose specific limitations on this. Then, by performing cluster analysis on the normal vectors, the clusters of the lidar point cloud can be obtained.

[0070] In this invention, angle histograms or clustering algorithms can be used to perform cluster analysis on the normal vectors. Figure 2 Please refer to the flowchart provided by this invention for performing cluster analysis on normal vectors. Figure 2 In this embodiment, a clustering algorithm is used to perform cluster analysis on the normal vectors. First, in step 31, the angles between all normal vectors are counted. Here, the angles between normal vectors can be the angles between each normal vector and the reference direction, or they can be the angles between all normal vectors. After obtaining the angles between the normal vectors, in step 32, clustering algorithms such as K-Means and DBSCAN are used to cluster the angles, thus obtaining clusters.

[0071] Figure 3 For another flowchart of cluster analysis of normal vectors provided by this invention, please refer to [the original text]. Figure 3 In this embodiment, an angle histogram is used to perform cluster analysis on the normal vectors. First, in step 33, the angles between all normal vectors are counted. Here, the angles between normal vectors can be the angles between each normal vector and the reference direction, or they can be the angles between all normal vectors. After obtaining the angles, a histogram is constructed using these angles to represent the different categories of normal vectors. In step 34, the peak values ​​of the histogram are detected. In step 35, based on the histogram peak values, clusters are obtained. For example, assuming there are three peak values ​​in the histogram, the corresponding LiDAR point cloud is divided into three clusters.

[0072] In step 4, the Random Sampling Consensus (RANSAC) algorithm is used to fit the preprocessed point cloud to a plane, obtaining the current plane equation, which can be represented as Ax + By + Cz + D = 0. During plane fitting, three points are randomly selected from the preprocessed point cloud, and the parameters of the plane model Ax + By + Cz + D = 0 are calculated using these three points. Based on the calculated model parameters, a plane equation is obtained. After obtaining the plane equation, the distances from other preprocessed point clouds to this plane are calculated to identify interior points, which are points with a distance less than a threshold. By repeating the above process, multiple plane equations can be obtained. The model with the most interior points is used. Furthermore, to further optimize the plane, the plane refitted using all interior points corresponding to this model is used as the current plane equation.

[0073] After fitting the equation of the current plane, step 5 determines whether fitting all planes has been completed. The number of point cloud clusters represents the number of planes. Therefore, the existence of intersecting planes can be determined based on the number of clusters. For example, when the number of clusters is 2, it means that there are two planes. After fitting one plane in step 4, there is still another plane. Therefore, it is necessary to fit the other plane.

[0074] To fit another plane, we first need to obtain the point cloud corresponding to that plane. At this point, we can count the interior points corresponding to the current plane equation. In step 6, the interior points of the current plane equation are removed from the preprocessed point cloud, and the remaining point cloud is the point cloud corresponding to the other plane. Therefore, in step 7, the remaining point cloud is used as the preprocessed point cloud, and we return to step 3 to execute the subsequent steps again, thus obtaining all the plane equations sequentially.

[0075] When the number of clusters is 1, it indicates that there is only one plane, and at this point, fitting has been completed for all planes. The planes obtained by point cloud segmentation usually have edge jaggedness, resulting in uneven edges. Therefore, after segmenting all planes, it is necessary to perform boundary smoothing processing on the planes corresponding to the plane equations before outputting them.

[0076] Figure 4 A flowchart illustrating boundary smoothing processing provided in an embodiment of the present invention is provided below. Figure 4 When performing boundary smoothing on a plane, a boundary threshold is first set in step 51. In step 52, the boundary points of the plane are identified. Since the normal vectors of point clouds located on the same plane are consistent, the boundary points can be identified using the normal vector mutation method. In step 53, the standard deviation of the neighborhood normal vectors of each identified boundary point is calculated, and in step 54, it is compared with the boundary threshold. Boundary points with a standard deviation exceeding the boundary threshold are removed, thus completing the boundary smoothing process for the plane.

[0077] The intersecting plane point cloud feature extraction method provided by this invention can effectively segment intersecting plane features in point clouds through an iterative approach, ensuring the extraction of each independent plane. While maintaining computational efficiency, it also improves the accuracy and robustness of segmentation.

[0078] Based on the same inventive concept, this invention also provides a system for extracting features from intersecting plane point clouds. Figure 5 This is a schematic diagram of a structural system for extracting intersecting plane point cloud features according to an embodiment of the present invention. Please refer to [the diagram]. Figure 5 The intersecting plane point cloud feature extraction system 100 provided by the present invention includes:

[0079] The acquisition module is used to acquire point clouds from the lidar.

[0080] The preprocessing module is used to preprocess the lidar point cloud to obtain a preprocessed point cloud;

[0081] The calculation module is used to calculate the normal vectors of the preprocessed point cloud and perform cluster analysis on the normal vectors to obtain clusters;

[0082] The fitting module is used to perform plane fitting on the preprocessed point cloud to obtain the equation of the current plane.

[0083] The program consists of a judgment module and a smoothing module. The judgment module is used to determine whether there are intersecting planes based on the clusters. If there are no intersecting planes, the smoothing module is used to smooth the boundaries of all plane equations and output the processed planes. The program then ends. If there are intersecting planes, the interior points of the current plane equation are extracted.

[0084] The removal module is used to remove interior points from the preprocessed point cloud to obtain the remaining point cloud;

[0085] The return module is used to treat the remaining point cloud as a preprocessed point cloud and returns the steps for calculating the normal vectors of the preprocessed point cloud.

[0086] For details, please refer to Figure 5 The intersecting plane point cloud feature extraction system 100 provided in this embodiment of the invention improves upon the Random Sample Consensus Algorithm (RANSAC) by iteratively segmenting independent planes. First, a lidar point cloud is acquired using an acquisition module. Typically, the acquired lidar point cloud data contains outliers and noise, which can affect the calculation results. Therefore, this invention uses a preprocessing module to preprocess the lidar point cloud, obtaining a preprocessed point cloud.

[0087] During preprocessing, noise reduction can be performed on the LiDAR point cloud to remove outliers. In this embodiment, a radius filtering method can be used to remove outliers. That is, a point is selected as the origin, the distance of each point from the origin is calculated, and only points within a predetermined radius are selected. For example, assuming the radius is set to 70m, only LiDAR point clouds within a distance of 70m are retained.

[0088] Furthermore, since large amounts of data can affect computation speed, after removing outliers, the remaining point cloud is downsampled to reduce subsequent computation. In this embodiment, speed-up filtering is used for downsampling. It should be noted that in other embodiments, uniform sampling can also be used for downsampling, and this invention does not specifically limit this method.

[0089] The normal vector of a point is also the normal vector of the plane containing that point. Therefore, points on the same plane have essentially the same normal vector; conversely, a group of points with similar normal vectors can be considered to be on the same plane. It is evident that by clustering the normal vectors of the lidar point cloud, the lidar point cloud can be classified. Based on this, this invention calculates the normal vectors of the preprocessed point cloud using a computation module. When calculating the normal vectors of the preprocessed point cloud, principal component analysis (PCA) or covariance matrix decomposition can be used, and the choice can be made according to the actual situation; this invention does not impose specific limitations on this. Then, by performing cluster analysis on the normal vectors, the clusters of the lidar point cloud can be obtained.

[0090] In this invention, angle histograms or clustering algorithms can be used to perform cluster analysis on the normal vectors. When using a clustering algorithm to perform cluster analysis on the normal vectors, the angles between all normal vectors are first counted. Here, the angles between normal vectors can be the angles between each normal vector and the reference direction, or they can be the angles between all normal vectors. After obtaining the angles between the normal vectors, clustering algorithms such as K-Means and DBSCAN are used to cluster the angles, thus obtaining clusters.

[0091] When using angle histograms to cluster normal vectors, the first step is to calculate the angles between all normal vectors. These angles can be the angles between each normal vector and the reference direction, or the angles between all normal vectors themselves. After obtaining the angles, a histogram is constructed using these angles to represent the different categories of normal vectors. The peak values ​​in the histograms are then detected, and based on these peak values, clusters are obtained. For example, if there are three peak values ​​in the histogram, the corresponding LiDAR point cloud is divided into three clusters.

[0092] The fitting module uses the Random Sampling Consensus (RANSAC) algorithm to perform plane fitting on the preprocessed point cloud, obtaining the current plane equation, which can be represented as Ax + By + Cz + D = 0. During plane fitting, three points are randomly selected from the preprocessed point cloud. These three points are then used to calculate the parameters of the plane model Ax + By + Cz + D = 0. Based on the calculated model parameters, a plane equation is obtained. After obtaining the plane equation, the distances from other preprocessed point clouds to this plane are calculated to identify interior points (points with a distance less than a threshold). By repeating this process, multiple plane equations are obtained. The model with the most interior points is used. Furthermore, to further optimize the plane, the plane refitted using all interior points corresponding to this model is used as the current plane equation.

[0093] After fitting the equation of the current plane, the judgment module determines whether fitting all planes has been completed. The number of point cloud clusters represents the number of planes. Therefore, the existence of intersecting planes can be determined based on the number of clusters. For example, when the number of clusters is 2, it means that there are two planes. After fitting one plane, there is still another plane. Therefore, it is necessary to fit the other plane.

[0094] To fit another plane, we first need to obtain the point cloud corresponding to that plane. At this point, we can count the interior points corresponding to the current plane's equation. Then, the removal module removes the interior points of the current plane's equation from the preprocessed point cloud, resulting in the remaining point cloud corresponding to the other plane. Therefore, the return module uses the remaining point cloud as the preprocessed point cloud and returns to the normal vector calculation step, re-executing subsequent steps to sequentially derive all the plane equations.

[0095] When the number of clusters is 1, it indicates that there is only one plane, and at this point, fitting has been completed for all planes. The planes obtained by point cloud segmentation usually have edge jaggedness, resulting in uneven edges. Therefore, after segmenting all planes, a smoothing module is needed to smooth the boundaries of the planes corresponding to the plane equations before outputting the results.

[0096] When smoothing the boundaries of a plane, a boundary threshold is first set, and then the boundary points of the plane are identified. Since the normal vectors of point clouds located on the same plane are consistent, the boundary points can be identified using the normal vector mutation method. The standard deviation of the neighborhood normal vectors of each identified boundary point is calculated and compared with the boundary threshold. Boundary points with a standard deviation exceeding the boundary threshold are removed, thus completing the boundary smoothing process of the plane.

[0097] The intersecting plane point cloud feature extraction system provided by this invention can effectively segment intersecting plane features in point clouds through an iterative approach, ensuring the extraction of each independent plane. While maintaining computational efficiency, it also improves the accuracy and robustness of segmentation.

[0098] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0099] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting features from intersecting plane point clouds, characterized in that, include: Acquire lidar point clouds; The lidar point cloud is preprocessed to obtain a preprocessed point cloud; Calculate the normal vectors of the preprocessed point cloud, and perform cluster analysis on the normal vectors to obtain clusters; Perform plane fitting on the preprocessed point cloud to obtain the current plane equation; Based on the clusters, determine whether there are intersecting planes. If not, perform boundary smoothing on all plane equations and output the processed planes, then the program ends. If they exist, extract the interior points of the current plane equation. The interior points are removed from the preprocessed point cloud to obtain the remaining point cloud; The remaining point cloud is used as a preprocessed point cloud, and the step of calculating the normal vector of the preprocessed point cloud is returned.

2. The method for extracting features from intersecting plane point clouds according to claim 1, characterized in that, The lidar point cloud is preprocessed, specifically as follows: The radius filtering method is used to denoise the lidar point cloud and remove outliers. Voxel filtering is used to downsample the point cloud after outlier removal.

3. The method for extracting features from intersecting plane point clouds according to claim 1, characterized in that, Cluster analysis is performed on the normal vectors, specifically as follows: Calculate the angles between all normal vectors; Clustering algorithms are used to cluster the included angles to obtain clusters.

4. The method for extracting features from intersecting plane point clouds according to claim 1, characterized in that, Cluster analysis is performed on the normal vectors, specifically as follows: Calculate the angles between all normal vectors and construct a histogram; Detect the peak value of the histogram; Based on the peak values ​​of the histogram, clusters are obtained.

5. The method for extracting features from intersecting plane point clouds according to claim 1, characterized in that, The preprocessed point cloud was fitted to a plane using the Random Sampling Consensus Algorithm (RANSAC) to obtain the plane equation Ax + By + Cz + D = 0.

6. The method for extracting features from intersecting plane point clouds according to claim 1, characterized in that, Boundary smoothing is performed on the planes corresponding to all plane equations, specifically as follows: Set boundary thresholds; Boundary points are obtained based on the normal vector mutation method; Calculate the standard deviation of the neighborhood normal vectors of the boundary points; Remove boundary points whose standard deviation exceeds the boundary threshold.

7. The method for extracting features from intersecting plane point clouds according to claim 1, characterized in that, The normal vector is calculated using principal component analysis (PCA) or covariance matrix decomposition.

8. A feature extraction system for intersecting plane point clouds, characterized in that, include: The acquisition module is used to acquire point clouds from the lidar. The preprocessing module is used to preprocess the lidar point cloud to obtain a preprocessed point cloud; The calculation module is used to calculate the normal vectors of the preprocessed point cloud and perform cluster analysis on the normal vectors to obtain clusters; The fitting module is used to perform plane fitting on the preprocessed point cloud to obtain the current plane equation; The system includes a judgment module and a smoothing module. The judgment module is used to determine whether there are intersecting planes based on the clusters. If there are no intersecting planes, the smoothing module is used to perform boundary smoothing on the planes corresponding to all plane equations and outputs the processed planes. The program then ends. If there are intersecting planes, the interior points of the current plane equation are extracted. A removal module is used to remove the interior points from the preprocessed point cloud to obtain the remaining point cloud; The return module is used to treat the remaining point cloud as a preprocessed point cloud and return the step of calculating the normal vector of the preprocessed point cloud.

9. The intersecting plane point cloud feature extraction system according to claim 8, characterized in that, The preprocessing module is specifically used for: The radius filtering method is used to denoise the lidar point cloud and remove outliers. Voxel filtering is used to downsample the point cloud after outlier removal.

10. The intersecting plane point cloud feature extraction system according to claim 8, characterized in that, The smoothing module is specifically used for: Set boundary thresholds; Boundary points are obtained based on the normal vector mutation method; Calculate the standard deviation of the neighborhood normal vectors of the boundary points; Remove boundary points whose standard deviation exceeds the boundary threshold.

Images