A point cloud data processing method, device and equipment of a laser radar and a medium

Abstract

The application provides a point cloud data processing method, device and equipment of a laser radar and a medium, comprising: adaptive local neighborhood construction is performed on each query point cloud to determine a local neighborhood point set of the query point cloud; a plurality of geometric features of each query point cloud are determined based on the local neighborhood point set; normalization processing and dynamic weight optimization are performed on the plurality of geometric features of each query point cloud to construct comprehensive features of each query point cloud, a plurality of original point cloud data are divided based on the plurality of comprehensive features to determine a plurality of point cloud categories; noise removal processing is performed on the original point cloud data in each point cloud category to determine updated point cloud data, and the local neighborhood point set determination and point cloud division processing are repeatedly performed on the updated point cloud data until a convergence condition is met, and the denoised point cloud data is output. The effectiveness and anti-interference ability of feature fusion are improved, and intelligent recognition and denoising of point clouds and accurate reservation of effective point clouds are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of point cloud data processing, in particular to a point cloud data processing method and device of a laser radar, equipment and a medium. BACKGROUND

[0002] The laser radar can efficiently obtain high-density and high-precision three-dimensional spatial coordinate data of objects or ground objects in a scene by actively emitting a laser beam and receiving reflection signals of a target surface to form a point cloud data set. However, in the actual collection process, a large number of noise points are inevitably mixed in the original point cloud data due to limitations of the sensor accuracy, measurement distance, target material reflection characteristics and complex environmental factors (such as rain, fog, dust, multi-path reflection, motion blur, etc.).

[0003] Traditional point cloud denoising methods mainly include filtering algorithms based on statistical characteristics and threshold judgment methods based on geometric rules. For example, statistical outlier removal (SOR) identifies and removes abnormal points by calculating the average distance and standard deviation of each point and points in its neighborhood; in addition, auxiliary means such as voxel grid downsampling and projection plane fitting are used for rough denoising. Although the above methods are simple to implement and have high computational efficiency, they have obvious limitations when facing large-scale, non-uniformly distributed and multi-scale structure coexisting actual point clouds: first, traditional methods usually rely on fixed neighborhood construction strategies (such as k-nearest neighbors or fixed radius sphere search), which are difficult to adapt to local feature changes in different density areas, and are prone to problems such as missing effective points in sparse areas and missing noise in dense areas; second, only relying on single geometric or statistical features (such as distance, density) for judgment lacks comprehensive understanding of the local topography of the point cloud (such as planes, curved surfaces and linear structures), resulting in poor performance in retaining small structures (such as railings, power lines and edges); third, most methods rely on manual experience for parameter setting and lack adaptive adjustment mechanisms, which have poor universality and robustness. In summary, there is an urgent need for a method for efficiently denoising point cloud data to solve the technical problem of severe noise interference in point cloud data in complex real scenes but complete retention of key structures. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a point cloud data processing method, device and equipment of a laser radar and a medium, which improves the effectiveness and anti-interference ability of feature fusion by cooperative analysis of multi-dimensional geometric features and introduction of a dynamic weight optimization mechanism, and realizes intelligent identification and denoising of point clouds and accurate retention of effective point clouds.

[0005] The present application provides a point cloud data processing method of a laser radar, which comprises:

[0006] constructing a local neighborhood point set of each query point cloud in the original point cloud data adaptively;

[0007] determining a plurality of geometric features of each query point cloud based on the local neighborhood point set of each query point cloud; wherein the geometric features at least include point density feature, normal consistency feature, local plane fitting residual feature, curvature change rate feature, principal component direction deviation degree feature, sphericity feature, planarity feature and straightness feature;

[0008] normalizing and dynamically weighting the plurality of geometric features of each query point cloud to construct a comprehensive feature of each query point cloud, and dividing the plurality of original point cloud data based on the comprehensive features of the plurality of query point clouds to determine a plurality of point cloud categories;

[0009] performing adaptive noise removal processing on the original point cloud data in each point cloud category to determine updated point cloud data, and repeatedly performing local neighborhood point set determination and point cloud division processing on the updated point cloud data until a convergence condition is met, and outputting denoised point cloud data.

[0010] In one possible implementation, the adaptive local neighborhood construction of each query point cloud in the original point cloud data to determine the local neighborhood point set of the query point cloud comprises:

[0011] dynamically selecting a spatial index method according to the local point cloud density of the region where the current query point cloud is located;

[0012] when the local point cloud density is higher than a first threshold, voxel index is used for neighborhood search;

[0013] when the local point cloud density is lower than the first threshold or located in an edge region, k-d tree index is used for neighborhood search, and the number of neighborhood points is adjusted according to the local point cloud density in the neighborhood search process to determine the local neighborhood point set of the query point cloud.

[0014] In one possible implementation, the determination of a plurality of geometric features of each query point cloud based on the local neighborhood point set of each query point cloud comprises:

[0015] constructing a covariance matrix based on the local neighborhood point set, and obtaining a first eigenvalue, a second eigenvalue and a third eigenvalue after singular value decomposition of the covariance matrix; wherein the first eigenvalue, the second eigenvalue and the third eigenvalue decrease in turn;

[0016] Based on the first characteristic value, the second characteristic value and the third characteristic value, a curvature change rate feature, a principal component direction deviation degree feature, a sphericity feature, a flatness feature and a straightness feature are determined.

[0017] In one possible implementation, based on the first characteristic value, the second characteristic value and the third characteristic value, the curvature change rate feature, the principal component direction deviation degree feature, the sphericity feature, the flatness feature and the straightness feature are determined, including:

[0018] The sum value between the first characteristic value, the second characteristic value and the third characteristic value is calculated, and based on the ratio between the third characteristic value and the sum value, the curvature change rate feature is determined;

[0019] Based on the ratio between the first characteristic value and the sum value, the principal component direction deviation degree feature is determined;

[0020] Based on the ratio between the third characteristic value and the first characteristic value, the sphericity feature is determined;

[0021] The difference value between the second characteristic value and the third characteristic value is calculated, and based on the ratio between the difference value and the first characteristic value, the flatness feature is determined;

[0022] The difference value between the first characteristic value and the second characteristic value is calculated, and based on the ratio between the difference value and the first characteristic value, the straightness feature is determined.

[0023] In one possible implementation, the normalization processing and dynamic weight optimization are performed on the plurality of geometric features of each query point cloud to construct a comprehensive feature of each query point cloud, including:

[0024] The normalization processing is performed on the plurality of geometric features of each query point cloud to determine a plurality of normalized geometric features of each query point cloud;

[0025] Based on an optimization objective function, dynamic optimization processing is performed on initial weights of the plurality of normalized geometric features to determine an optimal weight vector;

[0026] Based on the optimal weight vector, the plurality of normalized geometric features are weighted and fused to generate the comprehensive feature.

[0027] In one possible implementation, based on the optimization objective function, dynamic optimization processing is performed on initial weights of the plurality of normalized geometric features to determine an optimal weight vector, including:

[0028] The discriminative ability of multiple normalized geometric features on a point cloud dataset is quantitatively evaluated to obtain a representation ability score for each normalized geometric feature; wherein the representation ability score is calculated based on at least one of the feature's information gain, variance response, or attention weight.

[0029] An initial feature weight allocation scheme is generated based on the representation ability score. Based on the initial feature weight allocation scheme, the normalized geometric features of the query point cloud are weighted and fused to determine the initial comprehensive features.

[0030] One-dimensional clustering is performed using multiple initial integrated features to determine the clustering results;

[0031] Based on the clustering results, an optimization objective function is constructed, and an optimal weight vector that maximizes the optimization objective function is searched in the probabilistic simplex space using a gradient-free optimization algorithm.

[0032] In one possible implementation, the optimization objective function is:

[0033]

[0034] in, To optimize the objective function, α and β are balance coefficients. For the first Clustered point cloud dataset of classes, This represents the number of clusters after clustering. For querying point clouds The initial comprehensive characteristics, For the first The category center of the class.

[0035] This application embodiment also provides a point cloud data processing device for lidar, the point cloud data processing device comprising:

[0036] The neighborhood construction module is used to adaptively construct a local neighborhood for each query point cloud in the original point cloud data, and determine the local neighborhood point set of the query point cloud.

[0037] The geometric feature determination module is used to determine multiple geometric features of each query point cloud based on the local neighborhood point set of each query point cloud; wherein, the geometric features include at least point density features, normal vector consistency features, local plane fitting residual features, curvature change rate features, principal component direction deviation features, sphericity features, flatness features, and straightness features.

[0038] The point cloud division module is configured to normalize and optimize the dynamic weight of the multiple geometric features of each query point cloud, construct comprehensive features of each query point cloud, divide the multiple original point cloud data based on the comprehensive features of the multiple query point clouds, and determine multiple point cloud categories.

[0039] The noise removal module is configured to perform adaptive noise removal processing on the original point cloud data in each point cloud category, determine updated point cloud data, repeatedly perform local neighborhood point set determination and point cloud division processing on the updated point cloud data, and output denoised point cloud data when a convergence condition is met.

[0040] The embodiments of the present application also provide an electronic device, which comprises a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus. The machine-readable instructions are executed by the processor to perform the steps of the point cloud data processing method of the laser radar as described above.

[0041] The embodiments of the present application also provide a computer-readable storage medium, which stores a computer program. When the computer program is run by a processor, the steps of the point cloud data processing method of the laser radar as described above are performed.

[0042] The embodiments of the present application provide a point cloud data processing method, device, equipment, and medium of a laser radar. The point cloud data processing method comprises: performing adaptive local neighborhood construction on each query point cloud in original point cloud data to determine a local neighborhood point set of the query point cloud; determining multiple geometric features of each query point cloud based on the local neighborhood point set of each query point cloud. The geometric features at least include point density features, normal vector consistency features, local plane fitting residual features, curvature change rate features, principal component direction deviation degree features, sphericity features, planarity features, and straightness features. The multiple geometric features of each query point cloud are normalized and dynamically weighted to construct comprehensive features of each query point cloud. The multiple original point cloud data are divided based on the comprehensive features of the multiple query point clouds to determine multiple point cloud categories. The original point cloud data in each point cloud category are adaptively removed from noise to determine updated point cloud data. The local neighborhood point set determination and point cloud division processing are repeatedly performed on the updated point cloud data until a convergence condition is met, and denoised point cloud data is output. Through the collaborative analysis of multiple-dimensional geometric features and the introduction of a dynamic weight optimization mechanism, the effectiveness and anti-interference ability of feature fusion are improved, and intelligent identification and denoising of point clouds and accurate preservation of effective point clouds are realized.

[0043] In order to make the above objectives, characteristics and advantages of the present application more apparent, more comprehensible, the following preferred embodiments are specifically described in detail below, together with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0045] Figure 1 A flow chart of a point cloud data processing method of a laser radar provided by the embodiments of the present application;

[0046] Figure 2 A schematic diagram of a point cloud data processing method of a laser radar provided by the embodiments of the present application;

[0047] Figure 3 A structural schematic diagram of a point cloud data processing device of a laser radar provided by the embodiments of the present application;

[0048] Figure 4 A structural schematic diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0049] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more apparent, the following will combine the drawings in the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by those skilled in the art without creative labor belongs to the scope of protection of the present application.

[0050] Firstly, the application scenarios applicable to the present application are introduced. The present application can be applied to the technical field of point cloud data processing.

[0051] It is found through research that the traditional method usually relies on fixed neighborhood construction strategies (such as k-nearest neighbors or fixed radius sphere search), which is difficult to adapt to local feature changes in different density areas, and is prone to cause problems such as missing effective points in sparse areas and missing noise in dense areas; secondly, only relying on single geometric or statistical features (such as distance, density) for judgment lacks comprehensive understanding of the local topography of point cloud (such as plane, curved surface, linear structure), resulting in poor performance in retaining small structures (such as railings, wires, edges); thirdly, most methods rely on manual experience for parameter setting, lack adaptive adjustment mechanism, and have poor universality and robustness. In summary, there is an urgent need for an efficient point cloud data denoising method to solve the technical problem of serious noise interference in complex real scene point cloud data but key structure needs to be completely retained.

[0052] Based on this, the embodiment of the application provides a point cloud data processing method of a laser radar, which improves the effectiveness and anti-interference ability of feature fusion through cooperative analysis of multi-dimensional geometric features and introduction of a dynamic weight optimization mechanism, and realizes intelligent identification and denoising of point cloud and accurate reservation of effective point cloud.

[0053] Please refer to Figure 1 , Figure 1 The flowchart of the point cloud data processing method of the laser radar provided by the embodiment of the application is shown in FIG. 1. Figure 1 The point cloud data processing method provided by the embodiment of the application includes the following steps.

[0054] S101: adaptively constructing a local neighborhood for each query point cloud in the original point cloud data to determine a local neighborhood point set of the query point cloud.

[0055] In this step, the local neighborhood of each query point cloud in the original point cloud data is adaptively constructed to determine the local neighborhood point set of the query point cloud.

[0056] It should be noted that the local neighborhood point set is composed of a plurality of adjacent point cloud data of the query point cloud.

[0057] It should be noted that assuming that there is a point cloud representing a point on a railing. The point cloud data is taken as a "query point cloud", and the nearest k points are searched around it to form a small area.

[0058] In one possible implementation, the adaptively constructing a local neighborhood for each query point cloud in the original point cloud data to determine a local neighborhood point set of the query point cloud includes:

[0059] A: dynamically selecting a spatial index method according to the local point cloud density of the area where the current query point cloud is located.

[0060] Here, the local point cloud density of the region where the current query point cloud is located is determined, and a spatial indexing method is dynamically selected according to the local point cloud density.

[0061] It should be noted that the local point cloud density can be determined in the following manner: setting an initial search radius centered on the query point cloud, and counting the number of neighboring points within the radius; or obtaining the average number of points in the grid based on the pre-built spatial division structure (such as a coarse-grained voxel grid) as a density reference value.

[0062] B: When the local point cloud density is higher than the first threshold, voxel indexing is used for neighborhood search.

[0063] Here, when the local point cloud density is higher than the first threshold, it indicates that the region has a dense point distribution and a high data redundancy, and voxel indexing (Voxel Grid Indexing) is used for neighborhood search. Voxel indexing divides the three-dimensional space regularly and assigns the point cloud to the corresponding voxel unit, which can quickly locate the neighboring points in high-density scenarios and effectively suppress the calculation noise caused by local oversampling, thereby improving the search efficiency and stability.

[0064] C: When the local point cloud density is lower than the first threshold or located in the edge region, k-d tree indexing is used for neighborhood search, and the number of neighborhood points is adjusted according to the local point cloud density during the neighborhood search process to determine the local neighborhood point set of the query point cloud.

[0065] Here, when the local point cloud density is lower than the first threshold or the query point cloud is located in the edge, contour or sparse transition region of the target object, k-d tree indexing (k-dimensional Tree Indexing) is used for neighborhood search. K-d tree can efficiently support nearest neighbor search of irregularly distributed point clouds, and can accurately capture limited but key neighboring points in low-density or complex structure regions, avoiding the problem of missing neighborhoods caused by voxel voids.

[0066] After the selection of the index structure is completed, the application also introduces a dynamic adjustment mechanism for the number of neighborhood points: the number of required neighborhood points is adaptively determined according to the current local density level. For example, a larger number of neighborhood points is set in high-density regions to fully represent the local planar or curved structure; while in low-density regions, a smaller number of neighborhood points is used to prevent the introduction of irrelevant points at a long distance from causing feature distortion. Finally, in combination with the selected index structure and the dynamic number of neighborhood points, the neighborhood search operation is performed to obtain a set of nearest neighbor points around the query point, which constitutes the local neighborhood point set of the query point, and is used for subsequent calculation and analysis of multi-dimensional geometric features.

[0067] S102: Determine a plurality of geometric features of each of the query point clouds based on a local neighborhood point set of each of the query point clouds; wherein the geometric features at least include a point density feature, a normal consistency feature, a local plane fitting residual feature, a curvature variation rate feature, a principal component direction deviation degree feature, a sphericity feature, a planarity feature, and a straightness feature.

[0068] In this step, a plurality of geometric features of each of the query point clouds are determined based on a local neighborhood point set of each of the query point clouds.

[0069] The geometric feature calculation includes a plurality of geometric features such as point density, normal consistency, local plane fitting residual, curvature variation rate, deviation degree from principal component direction, sphericity, planarity, and straightness, and the purpose is to classify the point clouds into different topographic point cloud regions, such as sparse and dense point clouds, planar point clouds, spherical point clouds, etc.

[0070] In one possible implementation, the determining of the plurality of geometric features of each of the query point clouds based on the local neighborhood point set of each of the query point clouds includes:

[0071] a: constructing a covariance matrix based on the local neighborhood point set, and obtaining a first eigenvalue, a second eigenvalue, and a third eigenvalue after singular value decomposition of the covariance matrix; wherein the first eigenvalue, the second eigenvalue, and the third eigenvalue decrease in turn.

[0072] Here, the covariance calculation formula is:

[0073]

[0074] wherein, is the covariance of the local neighborhood point set, , , denotes the coordinates of the neighborhood point cloud in the local neighborhood point set, , , is the neighborhood mean point, and N is the number of point clouds in the local neighborhood point set. The first eigenvalue , the second eigenvalue , and the third eigenvalue of the covariance matrix can be obtained by svd decomposition.

[0075] b: determining the curvature variation rate feature, the principal component direction deviation degree feature, the sphericity feature, the planarity feature, and the straightness feature based on the first eigenvalue, the second eigenvalue, and the third eigenvalue.

[0076] Here, according to the first eigenvalue, the second eigenvalue and the third eigenvalue, a curvature change rate feature, a principal component direction deviation degree feature, a sphericity feature, a planeness feature and a straightness feature are determined.

[0077] In one possible implementation, the determination of the curvature change rate feature, the principal component direction deviation degree feature, the sphericity feature, the planeness feature and the straightness feature based on the first eigenvalue, the second eigenvalue and the third eigenvalue comprises:

[0078] The sum value between the first eigenvalue, the second eigenvalue and the third eigenvalue is calculated, the curvature change rate feature is determined based on the ratio between the third eigenvalue and the sum value, the principal component direction deviation degree feature is determined based on the ratio between the first eigenvalue and the sum value, the sphericity feature is determined based on the ratio between the third eigenvalue and the first eigenvalue, the difference value between the second eigenvalue and the third eigenvalue is calculated, and the planeness feature is determined based on the ratio between the difference value and the first eigenvalue, and the difference value between the first eigenvalue and the second eigenvalue is calculated, and the straightness feature is determined based on the ratio between the difference value and the first eigenvalue.

[0079] Here, the curvature change rate feature, the principal component direction deviation degree feature, the sphericity feature, the planeness feature and the straightness feature are determined by the following formulas:

[0080] Curvature change rate feature = (third eigenvalue - first eigenvalue) / first eigenvalue ;

[0081] Principal component direction deviation degree feature = (first eigenvalue - second eigenvalue) / second eigenvalue ;

[0082] Sphericity feature = (third eigenvalue - second eigenvalue) / second eigenvalue ;

[0083] Planeness feature = (first eigenvalue - second eigenvalue) / second eigenvalue ;

[0084] Straightness feature = (first eigenvalue - third eigenvalue) / third eigenvalue ;

[0085] In the present application, multiple geometric features including point density, normal vector consistency, local plane fitting residual, curvature change rate, principal component direction deviation degree, sphericity, planeness and straightness are comprehensively extracted, and a comprehensive feature vector is constructed through a normalization and dynamic weighting fusion strategy, so as to realize fine classification of local topography (such as plane, curved surface, linear structure and isolated point) of the point cloud. Compared with the traditional method which only relies on a single feature, the present scheme can more accurately distinguish noise points and effective structure points, and significantly improves the pertinence and reliability of denoising.

[0086] S103: Normalizing and dynamically optimizing the plurality of geometric features of each of the query point clouds to construct a comprehensive feature of each of the query point clouds, and dividing the plurality of original point cloud data based on the comprehensive features of the plurality of query point clouds to determine a plurality of point cloud categories.

[0087] In this step, the plurality of geometric features of each query point cloud is normalized and dynamically optimized to construct a comprehensive feature of each query point cloud, and the plurality of original point cloud data is divided based on the comprehensive features of the plurality of query point clouds to determine a plurality of point cloud categories.

[0088] It should be noted that the point cloud categories at least include a high-density planar region, a low-density linear structure, an isolated outlier point, a simple planar region, and a complex curved surface region.

[0089] In one possible implementation, the normalizing and dynamically optimizing the plurality of geometric features of each of the query point clouds to construct a comprehensive feature of each of the query point clouds includes:

[0090] (1) Normalizing the plurality of geometric features of each of the query point clouds to determine a plurality of normalized geometric features of each of the query point clouds.

[0091] Here, the plurality of geometric features of each query point cloud is normalized by using a quartile range (IQR) truncation normalization method to determine a plurality of normalized geometric features of each query point cloud.

[0092] (2) Dynamically optimizing the initial weights of the plurality of normalized geometric features based on an optimization objective function to determine an optimal weight vector.

[0093] Here, the initial weights of the plurality of normalized geometric features are dynamically optimized based on an optimization objective function to determine an optimal weight vector.

[0094] In one possible implementation, the dynamically optimizing the initial weights of the plurality of normalized geometric features based on an optimization objective function to determine an optimal weight vector includes:

[0095] I: Quantitatively evaluating the discriminative ability of the plurality of normalized geometric features on the point cloud data set to obtain a representation ability score of each normalized geometric feature; wherein the representation ability score is calculated based on at least one of information gain, variance response, or attention weight.

[0096] Here, the discriminative ability of each normalized geometric feature is evaluated according to the information gain, variance response and / or attention weight of the feature, and a capability score is generated.

[0097] wherein the information gain is used to measure the amount of information provided by a certain feature in dividing the "noise points" and "effective structure points" into two categories of samples. The variance response reflects the sensitivity of the numerical fluctuation of a certain feature in different structure type regions, and is used to evaluate its response ability to local topographic changes. For each type of typical structure (such as plane, edge, line, and isolated point), the variance of the feature in each category is calculated, and the inter-class dispersion is calculated. The feature attention network based on MLP automatically evaluates the importance of each feature. The network takes the normalized feature vector as input, and outputs the attention weight of each dimension. The above indicators are weighted and fused to obtain the comprehensive representation capability score of the feature.

[0098] II: An initial feature weight distribution scheme is generated according to the representation capability score, and each normalized geometric feature of the query point cloud is weighted and fused based on the initial feature weight distribution scheme to determine an initial comprehensive feature.

[0099] Here, the initial feature weight distribution scheme is generated according to the representation capability score, and each normalized geometric feature of the query point cloud is weighted and fused according to the initial feature weight distribution scheme to determine an initial comprehensive feature.

[0100] It should be noted that the representation capability score of each feature is normalized to form a weight set in the form of a probability distribution. The weight distribution principle embodies the design idea of "giving higher weight to features with higher discriminative ability", ensuring that in the subsequent comprehensive feature value calculation, features with strong noise and effective structure distinguishing ability (such as the performance of planarity in the plane region and straightness in the line structure) can play a dominant role.

[0101] III: One-dimensional clustering is performed using a plurality of initial comprehensive features to determine a clustering result.

[0102] It should be noted that the clustering result refers to dividing the original point cloud data into a plurality of point sets (i.e. categories or clusters) with similar local geometric characteristics, obtaining category division , ,..., }.

[0103] IV: An optimization objective function is constructed according to the clustering result, and a gradient-free optimization algorithm is used in the probability simplex space to search for an optimal weight vector that maximizes the optimization objective function.

[0104] Here, an optimization objective function is constructed based on the clustering results, and an optimal weight vector that maximizes the optimization objective function is searched in the probabilistic simplex space using a gradient-free optimization algorithm.

[0105] The optimization objective function is:

[0106]

[0107] in, To optimize the objective function, α and β are balance coefficients. For the first Clustered point cloud dataset of classes, This represents the number of clusters after clustering. For querying point clouds The initial comprehensive characteristics, For the first The category center of the class.

[0108] Here, the category center is determined using the following formula:

[0109]

[0110] (3): Based on the optimal weight vector, multiple normalized geometric features are weighted and fused to generate the comprehensive feature.

[0111] Here, the optimal weight vector is used to weight and fuse multiple normalized geometric features to generate a comprehensive feature.

[0112] It should be noted that the optimal weight vector refers to the set of optimal weight coefficients obtained in the process of weighted fusion of multidimensional normalized geometric features to make the classification result most discriminative, where each weight value corresponds to a normalized geometric feature.

[0113] In this application, to address the issue of uneven point cloud density, an adaptive neighborhood search mechanism is employed. This mechanism dynamically selects between kd-trees, octrees, or voxel indexes based on different region densities, and combines this with a self-adjusting strategy for the number of neighboring points to avoid insufficient or redundant neighbors, thus ensuring the stability of feature computation. Simultaneously, a dynamic weight optimization mechanism based on clustering results is introduced, utilizing gradient-free optimization to solve for the optimal feature weight combination, further enhancing the effectiveness and anti-interference capability of feature fusion.

[0114] S104: Perform adaptive noise removal processing on the original point cloud data in each point cloud category to determine the updated point cloud data. Repeat the local neighborhood point set determination and point cloud partitioning processing on the updated point cloud data until the convergence condition is met, and then output the denoised point cloud data.

[0115] In this step, adaptive noise removal processing is performed on the original point cloud data in each point cloud category to determine the updated point cloud data. The above steps are repeated on the updated point cloud data until the convergence condition is met, and then the denoised point cloud data is output.

[0116] It should be noted that the convergence condition is to calculate the mean value of features for each type of point cloud after each iteration and calculate its standard deviation. If the standard deviation is less than the threshold, the processing stops and the denoised point cloud data is output.

[0117] Here, based on the structural characteristics and noise sensitivity of different categories, the denoising algorithm and its parameter configuration are dynamically selected: For point sets identified as high-density planar regions, the Statistical Outlier Removal (SOR) method is adopted, and a relatively strict statistical threshold (such as the average distance multiple) is set to efficiently remove small noise points that deviate from the main distribution; For low-density linear structures or small edge regions, Radius Outlier Removal (ROR) is preferred, or the sensitivity of SOR is reduced to avoid the false deletion of valid points due to sparse neighborhoods; At the same time, topological connectivity constraints can be introduced to ensure that the continuity of linear structures is not destroyed; For points identified as isolated points or low-neighborhood density floating points, they are directly marked as noise and removed; For complex curved surface regions, a flexible threshold is set by combining the local curvature change rate and the plane fitting residual to filter out obvious outliers while preserving surface details.

[0118] In this application, a closed-loop iterative denoising process is designed. After each iteration, features are recalculated, classifications are updated, and denoising intensity is adjusted to gradually approach the optimal state. By introducing a fusion convergence criterion—including a joint judgment mechanism of multiple indicators such as the standard deviation of point cloud feature values ​​of each category, structural similarity, and geometric detail retention rate—premature termination or over-denoising is effectively avoided. At the same time, the denoising intensity adaptively decays with the iteration process, enhancing the stability of the algorithm.

[0119] For further details, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating a point cloud data processing method for a lidar system provided in an embodiment of this application. Figure 2 As shown, adaptive local neighborhood construction is performed on each query point cloud in the original point cloud data to determine the local neighborhood point set of the query point cloud and determine multiple geometric features of each query point cloud; feature normalization processing, weighted fusion, and point cloud classification processing are performed; adaptive noise removal processing is performed on the original point cloud data in each point cloud category; it is checked whether the convergence condition is met; if yes, the denoised point cloud data is output; otherwise, the above steps are repeated.

[0120] The point cloud data processing method of the laser radar provided in the embodiments of the present application comprises: performing adaptive local neighborhood construction on each query point cloud in original point cloud data to determine a local neighborhood point set of the query point cloud; determining a plurality of geometric features of each query point cloud based on the local neighborhood point set of the query point cloud; wherein the geometric features at least include point density features, normal vector consistency features, local plane fitting residual features, curvature change rate features, principal component direction deviation degree features, sphericity features, planeness features and straightness features; performing normalization processing and dynamic weight optimization on the plurality of geometric features of each query point cloud to construct comprehensive features of each query point cloud, dividing the plurality of original point cloud data based on the comprehensive features of the plurality of query point clouds to determine a plurality of point cloud categories; performing adaptive noise removal processing on the original point cloud data in each point cloud category to determine updated point cloud data, repeatedly performing local neighborhood point set determination and point cloud division processing on the updated point cloud data until a convergence condition is met, and outputting denoised point cloud data. Through the collaborative analysis of multi-dimensional geometric features and the introduction of a dynamic weight optimization mechanism, the effectiveness of feature fusion and the anti-interference ability are improved, and intelligent recognition and denoising of point clouds and accurate reservation of effective point clouds are realized.

[0121] Please refer to Figure 3 , Figure 3 The structure diagram of the point cloud data processing device of the laser radar provided in the embodiments of the present application is shown in FIG. 3. Figure 3 The point cloud data processing device 300 of the laser radar comprises:

[0122] The neighborhood construction module 310 is configured to perform adaptive local neighborhood construction on each query point cloud in original point cloud data to determine a local neighborhood point set of the query point cloud.

[0123] The geometric feature determination module 320 is configured to determine a plurality of geometric features of each query point cloud based on the local neighborhood point set of the query point cloud; wherein the geometric features at least include point density features, normal vector consistency features, local plane fitting residual features, curvature change rate features, principal component direction deviation degree features, sphericity features, planeness features and straightness features.

[0124] The point cloud division module 330 is configured to perform normalization processing and dynamic weight optimization on the plurality of geometric features of each query point cloud to construct comprehensive features of each query point cloud, divide the plurality of original point cloud data based on the comprehensive features of the plurality of query point clouds, and determine a plurality of point cloud categories.

[0125] The noise removal module 340 is configured to perform adaptive noise removal processing on the original point cloud data in each point cloud category to determine updated point cloud data, repeatedly perform local neighborhood point set determination and point cloud division processing on the updated point cloud data, and output denoised point cloud data when a convergence condition is met.

[0126] Further, the neighborhood construction module 310 is configured to perform adaptive local neighborhood construction on each query point cloud in the original point cloud data to determine a local neighborhood point set of the query point cloud.

[0127] According to the local point cloud density of the region where the current query point cloud is located, a spatial index mode is dynamically selected.

[0128] When the local point cloud density is higher than a first threshold, voxel indexing is used for neighborhood search.

[0129] When the local point cloud density is lower than the first threshold or is located in an edge region, k-d tree indexing is used for neighborhood search, and the number of neighborhood points is adjusted according to the local point cloud density during the neighborhood search process to determine the local neighborhood point set of the query point cloud.

[0130] Further, the geometric feature determination module 320 is configured to determine a plurality of geometric features of each query point cloud based on the local neighborhood point set of the query point cloud.

[0131] A covariance matrix is constructed based on the local neighborhood point set, and first, second, and third eigenvalues are obtained by singular value decomposition of the covariance matrix; wherein the first, second, and third eigenvalues decrease in turn.

[0132] Based on the first, second, and third eigenvalues, a curvature change rate feature, a principal component direction deviation degree feature, a sphericity feature, a planeness feature, and a straightness feature are determined.

[0133] Further, the geometric feature determination module 320 is configured to determine a curvature change rate feature, a principal component direction deviation degree feature, a sphericity feature, a planeness feature, and a straightness feature based on the first, second, and third eigenvalues.

[0134] The sum of the first, second, and third eigenvalues is calculated, and the curvature change rate feature is determined based on the ratio between the third eigenvalue and the sum.

[0135] The principal component direction deviation degree feature is determined based on the ratio between the first eigenvalue and the sum.

[0136] determine the sphericity feature based on a ratio between the third eigenvalue and the first eigenvalue;

[0137] calculate a difference between the second eigenvalue and the third eigenvalue, and determine the flatness feature based on a ratio between the difference and the first eigenvalue;

[0138] calculate a difference between the first eigenvalue and the second eigenvalue, and determine the straightness feature based on a ratio between the difference and the first eigenvalue.

[0139] Further, the point cloud division module 330 is configured to normalize the multiple geometric features of each query point cloud and dynamically optimize the weights to construct a comprehensive feature of each query point cloud:

[0140] normalize the multiple geometric features of each query point cloud to determine multiple normalized geometric features of each query point cloud;

[0141] dynamically optimize the initial weights of the multiple normalized geometric features based on an optimization objective function to determine an optimal weight vector;

[0142] weight and fuse the multiple normalized geometric features based on the optimal weight vector to generate the comprehensive feature.

[0143] Further, the point cloud division module 330 is configured to dynamically optimize the initial weights of the multiple normalized geometric features based on an optimization objective function to determine an optimal weight vector:

[0144] quantitatively evaluate the discriminative ability of the multiple normalized geometric features on the point cloud dataset to obtain a representation ability score of each normalized geometric feature; wherein the representation ability score is calculated based on at least one of information gain, variance response, or attention weight;

[0145] generate an initial feature weight distribution scheme according to the representation ability score, and weight and fuse the normalized geometric features of the query point cloud based on the initial feature weight distribution scheme to determine an initial comprehensive feature;

[0146] perform one-dimensional clustering using the multiple initial comprehensive features to determine a clustering result;

[0147] construct an optimization objective function according to the clustering result, and search for an optimal weight vector that maximizes the optimization objective function in a probability simplex space using a gradient-free optimization algorithm.

[0148] The point cloud data processing device of the laser radar provided in the embodiment of the application comprises: a neighborhood construction module, configured to perform adaptive local neighborhood construction on each query point cloud in original point cloud data, and determine a local neighborhood point set of the query point cloud; a geometric feature determination module, configured to determine a plurality of geometric features of each query point cloud based on the local neighborhood point set of the query point cloud; wherein the geometric features at least include point density features, normal vector consistency features, local plane fitting residual features, curvature change rate features, principal component direction deviation degree features, sphericity features, planeness features and straightness features; a point cloud division module, configured to perform normalization processing and dynamic weight optimization on the plurality of geometric features of each query point cloud, construct comprehensive features of each query point cloud, divide the plurality of original point cloud data based on the comprehensive features of the plurality of query point clouds, and determine a plurality of point cloud categories; and a noise removal module, configured to perform adaptive noise removal processing on the original point cloud data in each point cloud category, determine updated point cloud data, repeatedly perform local neighborhood point set determination and point cloud division processing on the updated point cloud data, and output denoised point cloud data when a convergence condition is met. Through the cooperative analysis of multi-dimensional geometric features and the introduction of a dynamic weight optimization mechanism, the effectiveness of feature fusion and the anti-interference ability are improved, and intelligent recognition and denoising of point clouds and accurate reservation of effective point clouds are realized.

[0149] Please refer to Figure 4 , Figure 4 A structural schematic diagram of an electronic device provided by the embodiment of the application is shown in FIG. 4. Figure 4 As shown in FIG. 4, the electronic device 400 comprises a processor 410, a memory 420 and a bus 430.

[0150] The memory 420 stores machine readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 and the memory 420 communicate through the bus 430. When the machine readable instructions are executed by the processor 410, the steps of the point cloud data processing method of the laser radar in the method embodiments shown in FIG. 1 and FIG. 2 can be performed. For specific implementation manners, please refer to the method embodiments, which will not be described here. Figure 1 and Figure 2 The specific implementation manners can be referred to the method embodiments, which will not be described here.

[0151] The embodiment of the application further provides a computer readable storage medium, which stores a computer program. When the computer program is run by a processor, the steps of the point cloud data processing method of the laser radar in the method embodiments shown in FIG. 1 and FIG. 2 can be performed. For specific implementation manners, please refer to the method embodiments, which will not be described here. Figure 1 and Figure 2 The specific implementation manners can be referred to the method embodiments, which will not be described here.

[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0153] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some communication interfaces, devices or units, and can be electrical, mechanical or other forms.

[0154] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0155] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0156] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a non-volatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0157] Finally, it should be noted that the above-described embodiments are merely specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit the same. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that any skilled person in the art can still modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some of the technical features, within the technical scope disclosed by the present application. The modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for processing point cloud data of a laser radar, characterized in that, The point cloud data processing method comprises: Adaptive local neighborhood construction is performed on each query point cloud in the original point cloud data to determine a local neighborhood point set of the query point cloud; A plurality of geometric features of each query point cloud are determined based on the local neighborhood point set of the query point cloud; wherein the geometric features at least include a point density feature, a normal vector consistency feature, a local plane fitting residual feature, a curvature change rate feature, a principal component direction deviation degree feature, a sphericity feature, a planeness feature and a straightness feature; Normalization processing and dynamic weight optimization are performed on the plurality of geometric features of each query point cloud to construct a comprehensive feature of each query point cloud, and the original point cloud data is divided based on the comprehensive features of the plurality of query point clouds to determine a plurality of point cloud categories; Adaptive noise removal processing is performed on the original point cloud data in each point cloud category to determine updated point cloud data, and the local neighborhood point set determination and point cloud division processing are repeatedly performed on the updated point cloud data until a convergence condition is met, and the denoised point cloud data is output; The plurality of geometric features of each query point cloud are determined based on the local neighborhood point set of the query point cloud, comprising: A covariance matrix is constructed based on the local neighborhood point set, and first, second and third eigenvalues are obtained after singular value decomposition of the covariance matrix; wherein the first, second and third eigenvalues decrease in turn; Based on the first, second and third eigenvalues, the curvature change rate feature, the principal component direction deviation degree feature, the sphericity feature, the planeness feature and the straightness feature are determined, comprising: The sum of the first, second and third eigenvalues is calculated, and the ratio between the third eigenvalue and the sum is used to determine the curvature change rate feature; The ratio between the first eigenvalue and the sum is used to determine the principal component direction deviation degree feature; The ratio between the third eigenvalue and the first eigenvalue is used to determine the sphericity feature; The difference between the second eigenvalue and the third eigenvalue is calculated, and the ratio between the difference and the first eigenvalue is used to determine the planeness feature; The difference between the first eigenvalue and the second eigenvalue is calculated, and the ratio between the difference and the first eigenvalue is used to determine the straightness feature.

2. The point cloud data processing method of claim 1, wherein, The adaptive local neighborhood construction on each query point cloud in the original point cloud data to determine the local neighborhood point set of the query point cloud comprises: The spatial index mode is dynamically selected according to the local point cloud density of the region where the current query point cloud is located; When the local point cloud density is higher than a first threshold, voxel index is used for neighborhood search; When the local point cloud density is lower than the first threshold or located in an edge region, k-d tree index is used for neighborhood search, and the number of neighborhood points is adjusted according to the local point cloud density in the neighborhood search process to determine the local neighborhood point set of the query point cloud.

3. The point cloud data processing method of claim 1, wherein, The normalization processing and dynamic weight optimization of the plurality of geometric features of each of the query point clouds are performed to construct a comprehensive feature of each of the query point clouds, including: The plurality of geometric features of each of the query point clouds are normalized to determine a plurality of normalized geometric features of each of the query point clouds; The initial weights of the plurality of normalized geometric features are dynamically optimized based on an optimization objective function to determine an optimal weight vector; The plurality of normalized geometric features are weighted and fused based on the optimal weight vector to generate the comprehensive feature.

4. The point cloud data processing method of claim 3, wherein, The initial weights of the plurality of normalized geometric features are dynamically optimized based on an optimization objective function to determine an optimal weight vector, including: The discrimination abilities of the plurality of normalized geometric features on a point cloud dataset are quantitatively evaluated to obtain a representation ability score of each normalized geometric feature; wherein the representation ability score is calculated based on at least one of information gain, variance response, or attention weight; An initial feature weight distribution scheme is generated according to the representation ability score, and the initial comprehensive feature is determined by weighted fusion of each normalized geometric feature of the query point cloud based on the initial feature weight distribution scheme; One-dimensional clustering is performed using a plurality of the initial comprehensive features to determine a clustering result; An optimization objective function is constructed according to the clustering result, and a gradient-free optimization algorithm is used in a probability simplex space to search for an optimal weight vector that maximizes the optimization objective function.

5. The point cloud data processing method of claim 4, wherein, The optimization objective function is: wherein, are balance coefficients for optimizing the objective function, is the first class of clustered point cloud data sets, is the number of classes after clustering, is the initial comprehensive feature of the query point cloud is the initial comprehensive feature of the query point cloud is the class center of the first class. 6.A point cloud data processing apparatus of a laser radar, characterized by comprising: The point cloud data processing apparatus includes: A neighborhood construction module configured to perform adaptive local neighborhood construction on each query point cloud in original point cloud data to determine a local neighborhood point set of the query point cloud; A geometric feature determination module configured to determine a plurality of geometric features of each query point cloud based on the local neighborhood point set of the query point cloud; wherein when the geometric features include at least point density feature, normal vector consistency feature, local plane fitting residual feature, curvature change rate feature, principal component direction deviation degree feature, sphericity feature, planeness feature, and straightness feature, the geometric feature determination module is specifically configured to: Construct a covariance matrix based on the local neighborhood point set, and obtain a first eigenvalue, a second eigenvalue, and a third eigenvalue after singular value decomposition of the covariance matrix; wherein the first eigenvalue, the second eigenvalue, and the third eigenvalue decrease in turn; Determine the curvature change rate feature, the principal component direction deviation degree feature, the sphericity feature, the planeness feature, and the straightness feature based on the first eigenvalue, the second eigenvalue, and the third eigenvalue, including: Calculate the sum of the first eigenvalue, the second eigenvalue, and the third eigenvalue, and determine the curvature change rate feature based on the ratio between the third eigenvalue and the sum; Determine the principal component direction deviation degree feature based on the ratio between the first eigenvalue and the sum; Determine the sphericity feature based on the ratio between the third eigenvalue and the first eigenvalue; a difference between the second eigenvalue and the third eigenvalue is calculated, and the flatness feature is determined based on a ratio between the difference and the first eigenvalue; a difference between the first eigenvalue and the second eigenvalue is calculated, and the straightness feature is determined based on a ratio between the difference and the first eigenvalue; a point cloud division module is configured to normalize and dynamically optimize a plurality of geometric features of each of the query point clouds, construct comprehensive features of each of the query point clouds, and divide the plurality of original point cloud data based on the comprehensive features of the plurality of query point clouds to determine a plurality of point cloud categories; a noise removal module is configured to perform adaptive noise removal processing on the original point cloud data in each point cloud category to determine updated point cloud data, repeatedly perform local neighborhood point set determination and point cloud division processing on the updated point cloud data until a convergence condition is met, and output denoised point cloud data.

7. An electronic device, comprising: comprising: a processor, a memory and a bus, the memory storing machine readable instructions executable by the processor, when the electronic device is running, the processor and the memory communicate through the bus, the machine readable instructions are executed by the processor to perform the steps of the lidar point cloud data processing method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, the computer readable storage medium stores a computer program, the computer program is executed by the processor to perform the steps of the lidar point cloud data processing method of any one of claims 1 to 5.

Images