A method and system for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds.

Abstract

This invention provides a method and system for extracting endpoints of multiple lines and filtering irrelevant line points based on line laser point clouds. First, a nearest neighbor set is recursively constructed from any point to extract the core flow direction and identify endpoints. Consistency judgment is used to cluster points in the same direction, accurately locating the initial structure of straight lines. Then, unclustered points are iteratively projected into the local streamline feature space of existing clusters. Clustering is adaptively expanded based on the statistical characteristics of the projection values ​​to solve the integrity problem in multi-line intersection scenarios. Next, projection uniformity analysis and radiative dispersion detection are performed on each cluster, using dual filtering to remove broken pseudo-straight lines and curved outliers, extracting the endpoints of valid straight line segments. Finally, a prediction model is constructed based on historical endpoint clustering sequences, matching newly added endpoints and updating clusters in real time. In dynamic time-series scenarios, unmatched noise points are filtered out, maintaining the stability of the endpoint set. This method possesses strong robustness, high accuracy, and versatility, and can meet diverse industrial needs.

Description

Technical Field

[0001] This invention belongs to the fields of industrial automation and point cloud processing technology, and in particular relates to a method for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds, which is applicable to feature extraction and robot guidance under complex bevel structures. Background Technology

[0002] In industrial intelligent manufacturing and welding tasks, robots often rely on point cloud information obtained from line laser sensors to extract structural edges or feature lines for path planning. However, industrial environments are often affected by various noise sources, resulting in a large amount of noise in the laser point cloud data. Common noises include curve noise and line noise, which not only degrade the quality of the point cloud data but also affect the accuracy of subsequent edge extraction and localization. Especially in complex industrial environments, effectively removing noise and accurately extracting the edge information of the target has always been a technical challenge.

[0003] Therefore, there is an urgent need for a line laser point cloud processing method that has multi-line recognition capabilities, can remove the influence of irrelevant structures, and can adapt to the dynamic changes in multi-frame temporal sequence, so as to achieve stable and reliable edge positioning. Summary of the Invention

[0004] To address the above technical problems, this invention provides a method and system for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A method for extracting multiple endpoints and filtering irrelevant line points based on line laser point clouds, the method comprising the following steps:

[0007] S100: Acquire line laser point cloud data, construct a set of nearest neighbor points starting from any point and extract the core flow direction of the nearest neighbor point set, calculate the projection value of the nearest neighbor point set on the core flow direction and identify the endpoints, recursively construct a new point set and the core flow direction of the new point set, and cluster point sets with the same core flow direction through consistency judgment;

[0008] S200: Iteratively project unclustered points onto the local streamline feature space of existing clusters, determine their affiliation based on the statistical characteristics of the projection values, until all points are labeled, and finally obtain the clustering results of multiple straight lines;

[0009] S300: Perform projection uniformity analysis and radiative dispersion detection on each straight line cluster, filter out non-uniform point sets and outliers, and extract the two endpoints of the effective straight line segments.

[0010] S400: For multiple laser point clouds arranged in chronological order, S100-S300 is performed to extract endpoints for the laser point cloud at each time point. The endpoint data of the previous preset number of time points are used as historical endpoints. A prediction model is built based on the clustering sequence of historical endpoints. New endpoints are matched with prediction data in real time and the clusters are updated. The endpoint set in the time series is iteratively maintained, and unmatched noise points are filtered out until the laser point cloud point optimization extraction for all time points is completed.

[0011] Preferably, S100 includes:

[0012] S110: For finite point cloud sets ,Depend on Composed of points, from the point set any point in the middle Start, search and nearest 1, 2, 3, 4, 5, 6, 7, 8, 9, 1, A set of points in a row of 3 columns The point set is calculated using the local streamline analysis method. Local streamlines, including core streamlines ;

[0013] S120: Calculate the point set In the core flow direction The projection values ​​are used to identify the two points with the smallest and largest projection values. and , respectively and Search with the center point and nearest A new point set is formed by 3 adjacent points, and the core flow direction of the new point set is calculated according to the method of S110. , and with A consistency check is performed to obtain clusters with the same core flow direction.

[0014] Preferably, in S110, the point set is calculated using the local streamline analysis method. Local streamlines, including:

[0015] S111: Calculate the point set covariance matrix :

[0016] ;

[0017] in Represents a point set The first in One point, Represent a column vectors, Represents the Kronecker product operation of matrices;

[0018] S112: For the covariance matrix Eigenvalue decomposition: , It is a matrix composed of eigenvectors. It is a diagonal matrix, with diagonal elements being eigenvalues. The eigenvector corresponding to the largest eigenvalue is the core flow direction, and the eigenvectors corresponding to other eigenvalues ​​are the radial flow directions.

[0019] Preferably, the consistency judgment method in S120 is as follows:

[0020] S121: For point neighborhood set and points neighborhood set Their core flow directions are respectively , If the following equation is satisfied, then the set is considered to be... and set Consistency:

[0021] ;

[0022] in It is a small positive number less than 1, used to define the allowable error range;

[0023] S122: If set and set Consistency, will and Clustering is performed to obtain clusters with the same core flow direction. If they are inconsistent, they are temporarily marked as noise.

[0024] Preferably, S200 includes:

[0025] S210: For all unclustered points, calculate their projection values ​​onto the local streamline feature space of existing clusters. If the projection value conforms to the statistical characteristics of the cluster distribution, then add the point to the corresponding cluster; otherwise, do not add it to the cluster.

[0026] S220: For For all non-noise and non-clustered points, repeat S110-S210 until... All points are marked as noise points or cluster points, and the clustering results of the entire point cloud are obtained. The clustering results are multiple straight line clusters.

[0027] Preferably, S210 specifically comprises:

[0028] S211: Assumption A set of cluster points , A set of unclustered points ,gather eigenvector matrix Calculate the matrix from all points to the eigenvectors. Projection of the central core flow direction :

[0029] ;

[0030] in , Represent a column vectors, For cluster point set The mean vector;

[0031] S212: Yes Calculate the mean for each flow direction. and standard deviation Specifically:

[0032] ;

[0033] ;

[0034] ;

[0035] in This indicates element-wise multiplication;

[0036] S213: For each unclustered point, determine whether it follows a normal distribution in each flow direction, i.e., for the th... If the deviation is less than a certain multiple of the standard value at each point... ,Right now Then the first If a point is added to a cluster, the mean and variance of the clusters are recalculated; otherwise, it is not added to the cluster and is marked as a non-clustered point. Represents points that are not clustered. This indicates the flow to the corresponding cluster center;

[0037] S214: Repeat S212-S213 until each cluster point is marked as a cluster point or a non-cluster point, completing all clustering.

[0038] Preferably, S300 includes:

[0039] S310: Clusters with fewer than a set number of clusters are marked as noise.

[0040] S320: For each remaining cluster, calculate the local streamlines of the cluster, project all points in this cluster onto the core streamline vector, and calculate the distance between each adjacent projected point. If there exists a distance greater than... If the distance to the average value is doubled, the clustering is considered uneven and the cluster is marked as noise; otherwise, execute S330.

[0041] S330: Project all points in this cluster onto the radiation flow vector, calculate the standard deviation of the projected points, and check if the standard deviation is less than a set threshold. If it is greater than the set threshold, the cluster is considered to be noise. If it is less than the set threshold, it is saved as the final cluster.

[0042] S340: Repeat S320-S330 until all clusters are labeled as final clusters or noise.

[0043] Preferably, S400 includes:

[0044] S410: For multiple laser point clouds arranged in chronological order, perform S100-S300 to extract endpoints for the laser point cloud at each time point, and then... The endpoint data at each time point is used as historical data. The endpoint data is Euclidean clustered, and the predicted data for each cluster is obtained by analyzing the cluster sequence.

[0045] S420: Yes At any given time, if the distance between each endpoint and the predicted data of a certain cluster is within a preset distance range, then the endpoint is considered a member of that cluster, and the cluster is updated. If the endpoint is not clustered, then it is marked as noise.

[0046] S430: Repeat S410-S420 until the line laser point clouds on all timelines are extracted to the correct endpoints.

[0047] Preferably, the analysis of each cluster sequence in S410 includes:

[0048] S411: For the former Historical data Establish a vector error correction model:

[0049] ;

[0050] in, express Changes over time For known The value at time, For known The value at time, Let be the cointegration matrix, representing the long-run equilibrium relationship. This is a short-term dynamic adjustment coefficient matrix, representing the short-term impact of different lag periods. For error terms, The lag order;

[0051] S412: Parameters in the maximum likelihood or least squares estimation model, assuming the cointegration matrix... Decomposed into ,in It is an error correction term. It is a cointegrating vector, if the rank of the cointegrating matrix is... If it is greater than 0, then it exists. cointegration relationship;

[0052] S413: Predict based on error correction model Difference data at time points:

[0053] ;

[0054] in, It is known The value at time, It is known prior difference data. , This is the error term, which is assumed to be 0 during prediction, or approximated using historical errors;

[0055] S414: Pass The difference data at time points are obtained Predicted data for each moment :

[0056] .

[0057] A system for extracting multiple endpoints and filtering irrelevant line points based on line laser point clouds includes:

[0058] The consistency judgment module is used to acquire line laser point cloud data, build a set of nearest neighbors starting from any point and extract the core flow direction of the nearest neighbors, calculate the projection value of the nearest neighbors in the core flow direction and identify the endpoints, recursively build a new set of points and the core flow direction of the new set of points, and cluster the set of points with the same core flow direction through consistency judgment.

[0059] The point cloud clustering module iteratively projects unclustered points onto the local streamline feature space of existing clusters, determines their affiliation based on the statistical characteristics of the projection values, until all points are labeled, and finally obtains the clustering results of multiple straight lines.

[0060] The clustering filtering module is used to perform projection uniformity analysis and radiative dispersion detection on each straight line cluster, filter out non-uniform point sets and outliers, and extract the two endpoints of the effective straight line segments.

[0061] The temporal optimization module is used to extract endpoints from multiple laser point clouds arranged in chronological order by performing steps in the consistency judgment module, point cloud clustering module, and cluster filtering module for each time point of the laser point cloud. The endpoint data of the previous preset number of time points are used as historical endpoints. A prediction model is built based on the clustering sequence of historical endpoints. New endpoints are matched with prediction data in real time and the clusters are updated. The endpoint set in the temporal sequence is iteratively maintained, and unmatched noise points are filtered out until the point cloud optimization extraction of laser point clouds at all time points is completed.

[0062] The aforementioned method and system for multi-line endpoint extraction and irrelevant line point filtering based on line laser point clouds first recursively constructs a nearest neighbor point set from any point, extracts the core flow direction and identifies endpoints, and clusters the same-direction point set through consistency judgment to accurately locate the initial structure of the straight line. Then, iteratively projects unclustered points into the local streamline feature space of existing clusters, and adaptively expands the clusters based on the statistical characteristics of the projection values ​​to solve the integrity problem in multi-line intersection scenarios. Furthermore, projection uniformity analysis and radiative dispersion detection are performed on each cluster, with dual filtering ensuring the geometric accuracy of effective straight line segment endpoints. Finally, a prediction model is constructed based on historical endpoint clustering sequences, matching newly added endpoints and updating clusters in real time, filtering out unmatched noise points in dynamic time-series scenarios, and maintaining the stability of the endpoint set. This method significantly improves noise resistance, dynamic adaptability, and automation, and can meet diverse industrial needs. Attached Figure Description

[0063] Figure 1 This is a flowchart of a method for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds in one embodiment of the present invention. Detailed Implementation

[0064] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0065] In one embodiment, such as Figure 1 As shown, a method for extracting multiple endpoints and filtering irrelevant line points based on line laser point clouds includes the following steps:

[0066] S100: Acquire line laser point cloud data, construct a set of nearest neighbor points starting from any point and extract the core flow direction of the nearest neighbor point set, calculate the projection value of the nearest neighbor point set on the core flow direction and identify the endpoints, recursively construct a new point set and the core flow direction of the new point set, and cluster point sets with the same core flow direction through consistency judgment;

[0067] S200: Iteratively project unclustered points onto the local streamline feature space of existing clusters, determine their affiliation based on the statistical characteristics of the projection values, until all points are labeled, and finally obtain the clustering results of multiple straight lines;

[0068] S300: Perform projection uniformity analysis and radiative dispersion detection on each straight line cluster, filter out non-uniform point sets and outliers, and extract the two endpoints of the effective straight line segments.

[0069] S400: For multiple laser point clouds arranged in chronological order, S100-S300 is performed to extract endpoints for the laser point cloud at each time point. The endpoint data of the previous preset number of time points are used as historical endpoints. A prediction model is built based on the clustering sequence of historical endpoints. New endpoints are matched with prediction data in real time and the clusters are updated. The endpoint set in the time series is iteratively maintained, and unmatched noise points are filtered out until the laser point cloud point optimization extraction for all time points is completed.

[0070] The aforementioned method for extracting multiple endpoints and filtering irrelevant line points based on line laser point clouds achieves accurate localization of line endpoints in line laser point clouds by recursively constructing nearest neighbor point sets, core flow direction clustering, and consistency judgment. It iteratively labels unclustered points using the statistical characteristics of projection values, addressing the problems of multi-line intersections and noise interference in complex scenarios and improving the integrity of line clustering. Combined with projection uniformity analysis and radiation dispersion detection, it effectively eliminates non-uniform point sets and outliers, retaining valid line segment endpoints. A prediction model is constructed based on historical endpoint clustering sequences, enabling real-time endpoint matching and noise filtering in dynamic scenarios, thus improving the robustness of temporal endpoint extraction.

[0071] In one embodiment, S100 includes:

[0072] S110: For finite point cloud sets ,Depend on Composed of points, from the point set any point in the middle Start, search and nearest 1, 2, 3, 4, 5, 6, 7, 8, 9, 1, A set of points in a row of 3 columns The point set is calculated using the local streamline analysis method. Local streamlines, including core streamlines ;

[0073] S120: Calculate the point set In the core flow direction The projection values ​​are used to identify the two points with the smallest and largest projection values. and , respectively and Search with the center point and nearest A new point set is formed by 3 adjacent points, and the core flow direction of the new point set is calculated according to the method of S110. , and with A consistency check is performed to obtain clusters with the same core flow direction.

[0074] Specifically, a neighborhood point set is recursively constructed starting from any point to avoid the influence of initial point selection on the clustering results; by judging the consistency of core flow direction, the clustering results are ensured to conform to the local linear structure characteristics, thereby improving the accuracy of line recognition.

[0075] In one embodiment, the point set is calculated in S110 using a local streamline analysis method. Local streamlines, including:

[0076] S111: Calculate the point set covariance matrix :

[0077] ;

[0078] in Represents a point set The first in One point, Represent a column vectors, Represents the Kronecker product operation of matrices;

[0079] S112: For the covariance matrix Eigenvalue decomposition: , It is a matrix composed of eigenvectors. It is a diagonal matrix, with diagonal elements being eigenvalues. The eigenvector corresponding to the largest eigenvalue is the core flow direction, and the eigenvectors corresponding to other eigenvalues ​​are the radial flow directions.

[0080] Specifically, based on the eigenvalue decomposition of the covariance matrix, the core flow direction and radiative flow direction of the point set are quickly calculated, providing a mathematical basis for projection analysis. The eigenvalue decomposition algorithm is mature and efficient, and suitable for real-time point cloud processing scenarios.

[0081] In one embodiment, the consistency determination method in S120 is as follows:

[0082] S121: For point neighborhood set and points neighborhood set Their core flow directions are respectively , If the following equation is satisfied, then the set is considered to be... and set Consistency:

[0083] ;

[0084] in It is a small positive number less than 1, used to define the allowable error range;

[0085] S122: If set and set Consistency, will and Clustering is performed to obtain clusters with the same core flow direction. If they are inconsistent, they are temporarily marked as noise.

[0086] Specifically, through the error threshold The system controls the consistency of flow direction judgment, adapts to local deformation or noise interference in point clouds, and marks inconsistent point sets as noise points in real time to avoid invalid clustering from interfering with subsequent processes.

[0087] In one embodiment, S200 includes:

[0088] S210: For all unclustered points, calculate their projection values ​​onto the local streamline feature space of existing clusters. If the projection value conforms to the statistical characteristics of the cluster distribution, then add the point to the corresponding cluster; otherwise, do not add it to the cluster.

[0089] S220: For For all non-noise and non-clustered points, repeat S110-S210 until... All points are marked as noise points or cluster points, and the clustering results of the entire point cloud are obtained. The clustering results are multiple straight line clusters.

[0090] In one embodiment, S210 specifically includes:

[0091] S211: Assumption A set of cluster points , A set of unclustered points ,gather eigenvector matrix Calculate the matrix from all points to the eigenvectors. Projection of the central core flow direction :

[0092] ;

[0093] in , Represent a column vectors, For cluster point set The mean vector;

[0094] S212: Yes Calculate the mean for each flow direction. and standard deviation Specifically:

[0095] ;

[0096] ;

[0097] ;

[0098] in This indicates element-wise multiplication;

[0099] S213: For each unclustered point, determine whether it follows a normal distribution in each flow direction, i.e., for the th... If the deviation is less than a certain multiple of the standard value at each point... ,Right now Then the first If a point is added to a cluster, the mean and variance of the clusters are recalculated; otherwise, it is not added to the cluster and is marked as a non-clustered point. Represents points that are not clustered. This indicates the flow to the corresponding cluster center;

[0100] S214: Repeat S212-S213 until each cluster point is marked as a cluster point or a non-cluster point, completing all clustering.

[0101] Specifically, based on the normal distribution test of the projected values, non-clustered points that meet the statistical characteristics are dynamically absorbed to enhance the integrity of clustering.

[0102] In one embodiment, S300 includes:

[0103] S310: Clusters with fewer than a set number of clusters are marked as noise.

[0104] S320: For each remaining cluster, calculate the local streamlines of the cluster, project all points in this cluster onto the core streamline vector, and calculate the distance between each adjacent projected point. If there exists a distance greater than... If the distance to the average value is doubled, the clustering is considered uneven and the cluster is marked as noise; otherwise, execute S330.

[0105] S330: Project all points in this cluster onto the radiation flow vector, calculate the standard deviation of the projected points, and check if the standard deviation is less than a set threshold. If it is greater than the set threshold, the cluster is considered to be noise. If it is less than the set threshold, it is saved as the final cluster.

[0106] S340: Repeat S320-S330 until all clusters are labeled as final clusters or noise.

[0107] Specifically, projection uniformity detection eliminates broken or discontinuous pseudo-lines. Radiation dispersion detection removes invalid clusters caused by curvature or outliers, using distance multiples. The filter sensitivity can be flexibly adjusted according to the set standard deviation threshold to adapt to different scenario requirements.

[0108] In one embodiment, S400 includes:

[0109] S410: For multiple laser point clouds arranged in chronological order, perform S100-S300 to extract endpoints for the laser point cloud at each time point, and then... The endpoint data at each time point is used as historical data. The endpoint data is Euclidean clustered, and the predicted data for each cluster is obtained by analyzing the cluster sequence.

[0110] S420: Yes At any given time, if the distance between each endpoint and the predicted data of a certain cluster is within a preset distance range, then the endpoint is considered a member of that cluster, and the cluster is updated. If the endpoint is not clustered, it is marked as noise. In this embodiment, the preset distance range is set to 0-0.1mm.

[0111] S430: Repeat S410-S420 until the line laser point clouds on all timelines are extracted to the correct endpoints.

[0112] In one embodiment, the analysis of each cluster sequence in S410 includes:

[0113] S411: For the former Historical data Establish a vector error correction model:

[0114] ;

[0115] in, express Changes over time For known The value at time, For known The value at time, Let be the cointegration matrix, representing the long-run equilibrium relationship. This is a short-term dynamic adjustment coefficient matrix, representing the short-term impact of different lag periods. For error terms, The lag order;

[0116] S412: Parameters in the maximum likelihood or least squares estimation model, assuming the cointegration matrix... Decomposed into ,in It is an error correction term. It is a cointegrating vector, if the rank of the cointegrating matrix is... If it is greater than 0, then it exists. cointegration relationship;

[0117] S413: Predict based on error correction model Difference data at time points:

[0118] ;

[0119] in, It is known The value at time, It is known prior difference data. , This is the error term, which is assumed to be 0 during prediction, or approximated using historical errors;

[0120] S414: Pass The difference data at time points are obtained Predicted data for each moment :

[0121] .

[0122] Specifically, the vector error correction model is used to analyze historical endpoint sequences, capture the long-term equilibrium and short-term dynamic characteristics of endpoint motion, and match new endpoints with predicted data to effectively distinguish between real endpoints and instantaneous noise, thereby improving the endpoint tracking stability in dynamic scenes.

[0123] Compared with the prior art, the above-described technical solutions conceived in this invention have the following beneficial effects:

[0124] 1. High robustness. This invention effectively removes curve noise from laser point clouds through local streamline analysis and projection denoising methods, thereby improving the quality of point cloud data. Furthermore, combining sequence analysis methods to remove endpoints of straight-line noise effectively enhances the reliability and accuracy of the results.

[0125] 2. High precision. This invention designs a local streamline analysis method, which effectively extracts multiple straight-line structures from line laser point clouds, solving the problems of noise interference and low extraction accuracy in traditional methods, and ensuring high precision of the extraction results.

[0126] 3. Versatility. Due to the widespread application of line laser sensors in intelligent industrial manufacturing, the method for extracting edge points and feature points based on line laser point clouds has strong versatility. The method of this invention is not only applicable to real-time processing scenarios such as skin milling and bevel positioning, but also capable of handling the positioning of feature points for various bevel shapes, including conventional bevels, V-shaped bevels, U-shaped bevels, and butt joints, demonstrating extremely high adaptability and meeting diverse industrial needs.

[0127] In one embodiment, a system for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds is also provided, comprising:

[0128] The consistency judgment module is used to acquire line laser point cloud data, build a set of nearest neighbors starting from any point and extract the core flow direction of the nearest neighbors, calculate the projection value of the nearest neighbors in the core flow direction and identify the endpoints, recursively build a new set of points and the core flow direction of the new set of points, and cluster the set of points with the same core flow direction through consistency judgment.

[0129] The point cloud clustering module iteratively projects unclustered points onto the local streamline feature space of existing clusters, determines their affiliation based on the statistical characteristics of the projection values, until all points are labeled, and finally obtains the clustering results of multiple straight lines.

[0130] The clustering filtering module is used to perform projection uniformity analysis and radiative dispersion detection on each straight line cluster, filter out non-uniform point sets and outliers, and extract the two endpoints of the effective straight line segments.

[0131] The temporal optimization module is used to extract endpoints from multiple laser point clouds arranged in chronological order by performing steps in the consistency judgment module, point cloud clustering module, and cluster filtering module for each time point of the laser point cloud. The endpoint data of the previous preset number of time points are used as historical endpoints. A prediction model is built based on the clustering sequence of historical endpoints. New endpoints are matched with prediction data in real time and the clusters are updated. The endpoint set in the temporal sequence is iteratively maintained, and unmatched noise points are filtered out until the point cloud optimization extraction of laser point clouds at all time points is completed.

[0132] For specific limitations regarding the multi-line endpoint extraction and irrelevant line point filtering system based on line laser point clouds, please refer to the limitations of the multi-line endpoint extraction and irrelevant line point filtering method based on line laser point clouds mentioned above, which will not be repeated here. Each module in the aforementioned multi-line endpoint extraction and irrelevant line point filtering system based on line laser point clouds can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independent of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0133] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a method for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds.

[0134] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for extracting multi-line endpoints and filtering irrelevant line points based on line laser point clouds.

[0135] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0136] The present invention has provided a detailed description of a method and system for multi-line endpoint extraction and irrelevant line point filtering based on line laser point clouds. Specific examples have been used to illustrate the principles and implementation methods of the invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the core ideas of the invention. It should be noted that those skilled in the art can make various improvements and modifications to the invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A multi-line endpoint extraction and irrelevant line point filtering method based on line laser point cloud, characterized in that, The method comprises the following steps: S100: acquiring linear laser point cloud data, constructing a near neighbor point set from an arbitrary point and extracting a core flow direction of the near neighbor point set, calculating a projection value of the near neighbor point set on the core flow direction and identifying an end point, recursively constructing a new point set and a core flow direction of the new point set, and clustering point sets with the same core flow direction through consistency judgment; S200: iteratively projecting unclustered points to a local flow line feature space of an existing cluster, judging the attribution of the unclustered points according to statistical characteristics of the projection values, until all points are marked, and finally obtaining a plurality of straight line clustering results; S300: performing projection uniformity analysis and radiation dispersion degree detection on each straight line cluster, filtering non-uniform point sets and outlier noise points, and extracting two side end points of an effective straight line segment; S400: performing S100-S300 on each time point of a plurality of time-sequentially arranged laser point clouds to extract end points, taking end point data of a preset number of previous time points as historical end points, constructing a prediction model based on a historical end point clustering sequence, real-time matching prediction data for new end points and updating a clustering cluster, iteratively maintaining an end point set over time, filtering out unmatched noise points, and until the optimization and extraction of end points of the laser point clouds of all time points are completed.

2. The method of claim 1, wherein, S100 comprises: S110: For finite point cloud sets ,Depend on Composed of points, from the point set any point in the middle Start, search and nearest 1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 1, 2, 3, 1, 2, 3, 4, 5, 1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 1, 2, 3 ... A set of points in a row of 3 columns The point set is calculated using the local streamline analysis method. Local streamlines, including core streamlines ; S120: calculate the point set In the core flow direction The projection value of the two points with the minimum and maximum projection values is identified And , respectively, with And As the center point, search And The nearest Adjacent point set consisting of new point set, and calculate the core flow direction of new point set according to the method of S110 , , and consistency judgment with Get the same core flow direction clustering.

3. The method of claim 2, wherein, In S110, the point set is calculated using the local streamline analysis method. Local streamlines, including: S111: Calculate the point set Covariance matrix of : ； in Represents a point set The first in One point, Represent a column vectors, Represents the Kronecker product operation of matrices; S112: Covariance matrix Eigenvalue decomposition: , is a matrix composed of eigenvectors, is a diagonal matrix, the diagonal elements are eigenvalues, the eigenvector corresponding to the largest eigenvalue is the core flow direction, and the eigenvectors corresponding to other eigenvalues are the radiation flow directions.

4. The method of claim 2, wherein, The consistency judgment method in S120 is: S121: For point neighborhood set and points neighborhood set Their core flow directions are respectively , If the following equation is satisfied, then the set is considered to be... and set Consistency: ； wherein is a small positive number smaller than 1, defining the allowed error range; S122: If the set and the set are consistent, then and are clustered to get the cluster of the same core flow direction, if not, then temporarily mark as noise.

5. The method of claim 4, wherein, S200 comprises: S210: calculating a projection value of all unclustered points to a local flow line feature space of an existing cluster, and if the projection value meets statistical characteristics of a cluster distribution, the point is added to the corresponding cluster, otherwise, the point is not added to the cluster; S220: For Repeat S110-S210 for all non-noise and non-cluster points in All points in are marked as noise or cluster points, and the clustering result of the entire point cloud is obtained, which is a plurality of straight line clusters.

6. The method of claim 5, wherein, S210 specifically comprises: S211: Assume a set of cluster points , a set of unclustered points , a set of feature vectors a matrix of feature vectors compute the projection of all points to the core flow in the matrix of feature vectors :​ ； wherein , denotes a column vector, is the mean vector of the cluster point set ; S212: Calculate the mean value of each flow direction Calculate the mean value of each flow direction and standard deviation , in particular: ； ； ； wherein denotes the multiplication of the elements one by one; S213: For each unclustered point, determine whether it follows a normal distribution in each flow direction, i.e., for the th... If the deviation is less than a certain multiple of the standard value at each point... ,Right now Then the first If a point is added to a cluster, the mean and variance of the clusters are recalculated; otherwise, it is not added to the cluster and is marked as a non-clustered point. Represents points that are not clustered. This indicates the flow to the corresponding cluster center; S214: repeating S212-S213 until each cluster point is marked as a cluster point or a non-cluster point, and all clusters are completed.

7. The method of claim 6, wherein, S300 comprises: S310: marking a cluster with a number less than a set value as noise; S320: For each of the remaining clusters, compute the local streamline of the cluster, and project all points in this cluster to the core streamline vector, and compute the distance of each neighboring projected point, if there is some distance greater than 1.5 times the average distance, then consider the cluster non-uniform, mark this cluster as a noise point, otherwise perform S330; S320: For each of the remaining clusters, compute the local streamline of the cluster, and project all points in this cluster to the core streamline vector, and compute the distance of each neighboring projected point, if there is some distance greater than 1.5 times the average distance, then consider the cluster non-uniform, mark this cluster as a noise point, otherwise perform S330; S320: For each of the remaining clusters, compute the local streamline S330: projecting all points in the cluster to a radiation flow direction vector, calculating a standard deviation of the projected points, checking whether the standard deviation is less than a set threshold, if the standard deviation is greater than the set threshold, the cluster is considered as noise, and if the standard deviation is less than the set threshold, the cluster is saved as a final cluster; S340: repeating S320-S330 until all clusters are marked as final clusters or noise.

8. The method of claim 7, wherein, S400 comprises: S410: for a plurality of time-sequentially arranged laser point clouds, performing S100-S300 extraction of end points on each time point laser point cloud, and taking the end point data of the previous time point as historical data, performing Euclidean clustering on the end point data, performing sequence analysis on each cluster to obtain prediction data of each cluster; S410: for a plurality of time-sequentially arranged laser point clouds, performing S100-S300 extraction of end points on each time point laser point cloud, and taking the end point data of the previous time point as historical data, performing Euclidean clustering on the end point data, performing sequence analysis on each cluster to obtain prediction data of each cluster; S420: For each endpoint at the time instant, if its distance to the prediction data of a cluster is within a preset distance range, the endpoint is considered as a member of the cluster, the cluster is updated, and if the endpoint is not clustered, it is marked as a noise point. S420: For each endpoint at the time instant, if its distance to the prediction data of a cluster is within a preset distance range, the endpoint is considered as a member of the cluster, the cluster is updated, and if the endpoint is not clustered, it is marked as a noise point. S430: repeating S410-S420 until all linear laser point clouds on a time line are extracted to correct end points.

9. The method of claim 8, wherein, The analysis of each cluster sequence in S410 comprises: S411: For the previous historical data , a vector error correction model is established: ； wherein, represents the change in the time, is known the value of the time, is known the value of the time, is a cointegration matrix, representing a long-term equilibrium relationship, is a short-term dynamic adjustment coefficient matrix, representing the short-term influence of different lags, is an error term, is the lag order; S412: Estimate parameters in the maximum likelihood or least squares estimation model, assuming that the cointegration matrix is decomposed into where is the error correction term, is a cointegrating vector, and if the rank of the cointegration matrix is greater than 0, then there are cointegrating relationships. S413: predict according to error correction model Difference data of time instant: ； wherein is known the value of the time instant, is known previous difference data, , is an error term, assumed to be 0 at prediction time, or approximated using historical errors; S414: by differential data of the time instant prediction data of the time instant : 。 10. A multi-line endpoint extraction and irrelevant line point filtering system based on line laser point cloud, characterized in that, It comprises: A consistency judgment module is configured to acquire linear laser point cloud data, construct a near neighbor point set from an arbitrary point and extract a core flow direction of the near neighbor point set, calculate a projection value of the near neighbor point set on the core flow direction and identify an end point, recursively construct a new point set and a core flow direction of the new point set, and cluster point sets with the same core flow direction through consistency judgment; A point cloud clustering module is configured to iteratively project unclustered points to a local flow line feature space of an existing cluster, judge the attribution of the unclustered points according to statistical characteristics of the projection values, until all points are marked, and finally obtain a plurality of straight line clustering results; A cluster filtering module is configured to perform projection uniformity analysis and radiation dispersion degree detection on each straight line cluster, filter non-uniform point sets and outlier noise points, and extract two side end points of an effective straight line segment. The time sequence optimization module is used for a plurality of laser point clouds arranged in time sequence, and the steps of the consistency judgment module, the point cloud clustering module and the clustering filtering module are used for extracting end points for each time point laser point cloud, and the end point data of the previous preset number of time points is used as historical end points, a prediction model is constructed based on the historical end point clustering sequence, the prediction data is matched in real time for the new end point and the clustering cluster is updated, the end point set in time sequence is iteratively maintained, the unmatched noise is filtered out, and the extraction of the end points of the laser point cloud of all time points is completed.

Images