A method and apparatus for redundancy reduction of overlapping point clouds applicable to various scenarios

Abstract

This application provides a method and apparatus for redundancy removal from overlapping point clouds applicable to various scenarios. The method includes stitching and fusing 3D original point clouds from different perspectives to obtain a stitched point cloud; obtaining the overlapping region of the stitched point cloud; iteratively smoothing the overlapping region using a corresponding point cloud smoothing algorithm based on the scenario of the overlapping region to obtain a single-layer point cloud; and simplifying the single-layer point cloud using a corresponding point cloud redundancy removal algorithm based on the scenario of the single-layer point cloud to obtain a redundancy-removed point cloud. This application first employs an overlapping point judgment algorithm to quickly obtain the overlapping region, improving the processing efficiency of overlapping point cloud redundancy removal; secondly, it selects a more suitable point cloud smoothing algorithm to perform efficient and high-precision smoothing processing on the overlapping region, effectively eliminating the layering phenomenon of point cloud data in the overlapping region; and finally, it automatically selects a more suitable point cloud redundancy removal algorithm to simplify the overlapping points of the single-layer point cloud, effectively removing redundant data points.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and in particular to a method and apparatus for redundancy reduction of overlapping point clouds applicable to various scenarios. Background Technology

[0002] Currently, 3D machine vision inspection technology has been widely applied in the field of industrial intelligent manufacturing. In actual inspection projects, it is usually necessary to use high-resolution cameras to acquire 3D point cloud data from multiple perspectives and then stitch and fuse these 3D point cloud data from different perspectives to obtain a large-scale, high-precision 3D imaging result. However, when stitching and fusing 3D point cloud data from different perspectives using target calibration or point cloud registration, factors such as environmental noise, measurement errors, and registration and stitching errors can lead to data layering and redundancy in the overlapping areas of the stitched point cloud. Therefore, it is necessary to use overlapping point cloud deduplication algorithms to correct the stitched point cloud, eliminate point cloud data layering, remove redundant point cloud data, and simplify the stitched point cloud.

[0003] To improve the stitching accuracy of multi-view 3D point cloud data, existing technologies provide a voxel mesh-based method for redundancy reduction and simplification of the stitched point cloud. This method first directly encapsulates the stitched point cloud within a large voxel grid; then, it divides the large voxel grid into multiple uniformly sized small voxel grids; finally, it completes the redundancy reduction and simplification process by identifying or selecting feature points within the small voxel grids according to certain rules.

[0004] However, the aforementioned voxel grid-based point cloud redundancy reduction methods suffer from several drawbacks. The resulting stitched point clouds vary in size, and the subjective nature of manually setting the voxel grid makes it difficult to control the specific size of the voxel grid each time. Furthermore, the voxel grid size is ill-suited for the redundancy reduction and simplification requirements of higher-precision stitched point clouds. Additionally, methods that rely on rules to obtain or select feature points to represent the voxel grid cannot fully preserve the feature points of the stitched point cloud, nor can they resolve the point cloud data layering phenomenon, especially for stitched point clouds with uneven data distribution, where feature point loss is even more severe. Summary of the Invention

[0005] This application provides a method and apparatus for redundancy removal of overlapping point clouds applicable to various scenarios, in order to solve the problems in the prior art where the point cloud redundancy removal method based on voxel mesh cannot meet the requirements of redundancy removal and simplification of stitched point clouds with higher precision, cannot completely retain the feature points of the stitched point cloud, and cannot solve the problem of point cloud data layering in the stitched point cloud.

[0006] Firstly, this application provides a method for redundancy reduction of overlapping point clouds applicable to various scenarios, including:

[0007] The original 3D point clouds from different perspectives are stitched together to obtain the stitched point cloud;

[0008] Obtain the overlapping region of the stitched point cloud;

[0009] Based on the scenario of the overlapping region, the corresponding point cloud smoothing algorithm is used to iteratively smooth the overlapping region to obtain a single-layer point cloud.

[0010] Based on the scenario of the single-layer point cloud, a corresponding point cloud redundancy reduction algorithm is used to perform redundancy reduction and simplification processing on the single-layer point cloud to obtain the redundancy reduction result point cloud.

[0011] In a preferred embodiment of this application, obtaining the overlapping region of the stitched point cloud includes:

[0012] Construct the topological relationship of the stitched point cloud, wherein the stitched point cloud includes at least one target point;

[0013] Determine whether at least one of the target points is an overlapping point according to a preset determination algorithm;

[0014] All target points identified as overlapping points in the stitched point cloud are combined and output as the overlapping region.

[0015] In a preferred embodiment of this application, obtaining the overlapping region of the stitched point cloud includes:

[0016] The topological relationships of the stitched point cloud are constructed using common search trees;

[0017] Based on the aforementioned topological relationships, set the neighborhood search radius;

[0018] Search the sphere neighborhood corresponding to each target point in the stitched point cloud according to the neighborhood search radius;

[0019] Based on the ball's neighborhood, obtain the number of neighborhood points corresponding to the target point;

[0020] If the number of neighborhood points of the target point is greater than a preset threshold for the number of neighborhood points, then the target point is marked as an overlapping point;

[0021] All target points identified as overlapping points in the stitched point cloud are combined and output as the overlapping region.

[0022] In a preferred embodiment of this application, based on the scene of the overlapping region, a corresponding point cloud smoothing algorithm is used to iteratively smooth the overlapping region to obtain a single-layer point cloud, including:

[0023] The topological relationships of the overlapping regions are constructed using a common search tree.

[0024] Based on the aforementioned topological relationships, set the neighborhood search radius;

[0025] Search the neighborhood corresponding to each overlapping point in the overlapping region according to the neighborhood search radius;

[0026] Based on the neighborhood, obtain the neighborhood information of the overlapping points;

[0027] Based on the scene of the overlapping region and the neighborhood information of the overlapping points, the overlapping points are smoothed using the corresponding point cloud smoothing algorithm.

[0028] After smoothing all overlapping points for a preset number of iterations, a single-layer point cloud is obtained.

[0029] In a preferred embodiment of this application, the point cloud deduplication algorithm includes a voxel sampling-based point cloud deduplication method, a neighborhood mean-based point cloud deduplication method, and a mean shift-based point cloud deduplication method; wherein, the voxel sampling-based point cloud deduplication method includes the following process for simplifying a single-layer point cloud:

[0030] The single-layer point cloud is divided into at least one voxel grid;

[0031] Obtain the nearest center point or centroid of at least one overlapping point within each voxel grid;

[0032] The center nearest point or centroid point represents all overlapping points within the voxel grid;

[0033] The point cloud is output by combining all the nearest centers or centroids within the single-layer point cloud to form a deredundancy result point cloud.

[0034] In a preferred embodiment of this application, the steps of the voxel sampling-based point cloud redundancy reduction method for simplifying a single-layer point cloud are as follows:

[0035] Obtain the maximum and minimum boundary values ​​of a single-layer point cloud;

[0036] A voxel mesh of the corresponding size is obtained according to a pre-set voxel size, wherein the voxel size is set according to the maximum boundary value and the minimum boundary value;

[0037] Based on the voxel grid information, at least one overlapping point in the single-layer point cloud is assigned to the corresponding voxel grid to obtain the voxel grid index;

[0038] Based on the voxel grid index, the nearest center point or centroid point within the voxel grid is selected to represent all overlapping points within the voxel grid.

[0039] The point cloud is output by combining all the nearest centers or centroids within the single-layer point cloud to form a deredundancy result point cloud.

[0040] In a preferred embodiment of this application, if the single-layer point cloud is uniformly distributed, a voxel sampling-based point cloud redundancy reduction algorithm is used to simplify the single-layer point cloud.

[0041] If the single-layer point cloud is not uniformly distributed, a point cloud redundancy reduction algorithm based on neighborhood mean or a point cloud redundancy reduction algorithm based on mean drift is used to simplify the single-layer point cloud.

[0042] Secondly, this application provides an overlapping point cloud redundancy elimination device applicable to various scenarios, including an acquisition unit, an overlapping region judgment unit, an overlapping point cloud smoothing unit, and an overlapping point cloud redundancy elimination unit that are interconnected.

[0043] The acquisition unit is configured as follows:

[0044] The original 3D point clouds from different perspectives are stitched together to obtain the stitched point cloud;

[0045] The overlapping region determination unit is configured as follows:

[0046] Obtain the overlapping region of the stitched point cloud;

[0047] The overlapping point cloud smoothing unit is configured as follows:

[0048] Based on the scenario of the overlapping region, the corresponding point cloud smoothing algorithm is used to iteratively smooth the overlapping region to obtain a single-layer point cloud.

[0049] The overlapping point cloud redundancy reduction unit is configured as follows:

[0050] Based on the scenario of the single-layer point cloud, a corresponding point cloud redundancy reduction algorithm is used to perform redundancy reduction and simplification processing on the single-layer point cloud to obtain the redundancy reduction result point cloud.

[0051] Thirdly, this application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for eliminating redundant overlapping point clouds applicable to various scenarios.

[0052] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for redundancy reduction of overlapping point clouds applicable to various scenarios.

[0053] This application provides a method and apparatus for redundancy reduction of overlapping point clouds applicable to various scenarios, which has the following advantages compared to the prior art:

[0054] This application first employs an overlapping point judgment algorithm, which can quickly obtain overlapping areas, improve the processing efficiency of redundancy removal in overlapping point clouds, and avoid excessive smoothing of all target points in the stitched point cloud. The algorithm process is simpler and more effective. Second, this application can automatically select a more suitable point cloud smoothing algorithm to perform efficient and high-precision smoothing processing on overlapping areas, which can effectively eliminate the point cloud data layering phenomenon in overlapping areas and obtain a single-layer point cloud. Finally, this application can automatically select a more suitable point cloud redundancy removal algorithm to perform redundancy removal and simplification processing on overlapping points of the single-layer point cloud, which can effectively remove redundant data points and facilitate subsequent rapid measurement and other processing of the point cloud. Attached Figure Description

[0055] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of an overlapping point cloud redundancy removal method applicable to multiple scenarios, as described in Embodiment 1 of this application;

[0057] Figure 2 This is a schematic diagram of the point cloud smoothing algorithm based on bilateral filtering, trilateral filtering, or linear interpolation in Embodiment 1 of this application.

[0058] Figure 3 This is a schematic diagram of the point cloud smoothing algorithm based on the MLS moving least squares method in Embodiment 1 of this application;

[0059] Figure 4a This is a schematic diagram of the first stitched point cloud of the application example of this application;

[0060] Figure 4b This is a schematic diagram of the point cloud of the first redundancy removal result in the application example of this application;

[0061] Figure 5a This is a schematic diagram of the second stitched point cloud of the application example of this application;

[0062] Figure 5b This is a schematic diagram of the point cloud of the second redundancy reduction result in the application example of this application. Detailed Implementation

[0063] To make the objectives, implementation methods and advantages of this application clearer, the exemplary implementation methods of this application will be clearly and completely described below with reference to the accompanying drawings of the exemplary embodiments of this application. Obviously, the described exemplary embodiments are only some embodiments of this application, and not all embodiments.

[0064] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0065] Based on the exemplary embodiments described in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the appended claims. Furthermore, although the disclosures in this application are presented by way of one or more exemplary examples, it should be understood that each aspect of these disclosures can also constitute a complete implementation on its own.

[0066] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0067] To facilitate the explanation of the technical solution of this application, some concepts involved in this application will be explained first below.

[0068] In this application, terms such as “first,” “second,” “third,” and “fourth” are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms “comprising,” “including,” or any other variations are intended to cover non-exclusive inclusion, thereby including not only the expressly listed elements but also other elements not expressly listed. Therefore, this does not cause ambiguity in the scheme of this application.

[0069] A point cloud is a collection of massive points that represent the spatial distribution and surface characteristics of a target within the same spatial reference frame.

[0070] Existing technologies also provide a point cloud redundancy reduction method based on clustering. This method uses a clustering algorithm to group similar points with the same name in the stitched point cloud into the same category, and then merges them according to certain rules, thereby achieving redundancy reduction in the stitched point cloud. However, this method has drawbacks such as difficulty in adjusting initial parameters, difficulty in handling the layering phenomenon of point cloud data, and uneven distribution of points in the fused result.

[0071] Existing technologies also provide a point cloud redundancy reduction method based on triangular meshes. This method first triangulates the obtained stitched point cloud, i.e., establishes the triangular mesh topological relationship; secondly, it calculates the normal vectors of the triangular meshes; and then, based on a weighted algorithm of the normal vectors, it judges the generated triangular meshes and deletes redundant meshes to achieve the requirements of redundancy reduction and simplification of the stitched point cloud. However, this method has disadvantages such as long processing time for triangulating the stitched point cloud, high complexity of the triangular mesh algorithm, large resource consumption, and susceptibility to cumulative errors.

[0072] Example 1

[0073] like Figure 1 As shown, this embodiment 1 provides a method for redundancy reduction of overlapping point clouds applicable to various scenarios. The method for redundancy reduction of overlapping point clouds includes:

[0074] S101, stitching and fusing the original 3D point clouds from different perspectives to obtain the stitched point cloud;

[0075] S102, Obtain the overlapping region of the stitched point cloud;

[0076] S103, based on the scene of the overlapping area, the overlapping area is iteratively smoothed using a corresponding point cloud smoothing algorithm to obtain a single-layer point cloud;

[0077] S104. Based on the scenario of the single-layer point cloud, a corresponding point cloud redundancy reduction algorithm is used to perform redundancy reduction and simplification processing on the single-layer point cloud to obtain the redundancy reduction result point cloud.

[0078] It should be noted that in step S101 of this embodiment, the original 3D point clouds from different perspectives are all captured by high-resolution cameras, and the stitched point cloud is obtained by stitching and fusing the original 3D point clouds from multiple perspectives. In step S102 of this embodiment, the overlapping area can be quickly obtained based on the overlapping point judgment algorithm, which can improve the processing efficiency of redundancy removal of overlapping point clouds, while avoiding over-smoothing of all target points in the stitched point cloud. Step S103 of this embodiment can automatically select a more suitable point cloud smoothing algorithm to perform efficient and high-precision smoothing processing on the overlapping area, which can effectively eliminate the point cloud data layering phenomenon in the overlapping area. Step S104 of this embodiment can automatically select a more suitable point cloud redundancy removal algorithm to perform redundancy removal and simplification processing on the overlapping points of a single-layer point cloud, which can effectively remove redundant data points.

[0079] Furthermore, in one specific embodiment of this Example 1, step S102, obtaining the overlapping region of the stitched point cloud, includes:

[0080] Construct the topological relationship of the stitched point cloud, wherein the stitched point cloud includes at least one target point;

[0081] Determine whether at least one of the target points is an overlapping point according to a preset determination algorithm;

[0082] All target points identified as overlapping points in the stitched point cloud are combined and output as the overlapping region.

[0083] Specifically, in this embodiment 1, the specific steps for obtaining the overlapping region of the stitched point cloud are as follows:

[0084] The topological relationships of the stitched point cloud are constructed using common search trees such as kd-tree, oc-tree, and quadtree.

[0085] Based on the aforementioned topological relationships, set the neighborhood search radius;

[0086] Search the sphere neighborhood corresponding to each target point in the stitched point cloud according to the neighborhood search radius;

[0087] Based on the ball's neighborhood, obtain the number of neighborhood points corresponding to the target point;

[0088] If the number of neighborhood points of the target point is greater than a preset threshold for the number of neighborhood points, then the target point is marked as an overlapping point;

[0089] Traverse each target point in the stitched point cloud and repeat the above steps to determine whether the target points are overlapping points;

[0090] All target points identified as overlapping points in the stitched point cloud are combined and output as the overlapping region.

[0091] Specifically, in this embodiment 1, the steps for obtaining the overlapping region of the stitched point cloud further include:

[0092] The topological relationships of the stitched point cloud are constructed using common search trees such as kd-tree, oc-tree, and quadtree.

[0093] Based on the aforementioned topological relationships, set the neighborhood search radius;

[0094] Search the sphere neighborhood corresponding to each target point in the stitched point cloud according to the neighborhood search radius;

[0095] Based on the sphere neighborhood, obtain the distance between the nearest neighbor of the target point and the target point;

[0096] If the distance between the nearest neighbor point and the target point is greater than a preset distance threshold, and the angle between the normal of the nearest neighbor point and the target point is less than a preset angle threshold, then the target point is marked as an overlapping point.

[0097] Traverse each target point in the stitched point cloud and repeat the above steps to determine whether the target points are overlapping points;

[0098] All target points identified as overlapping points in the stitched point cloud are combined and output as the overlapping region.

[0099] Furthermore, in one specific embodiment of this Example 1, S103, based on the scene of the overlapping region, an appropriate point cloud smoothing algorithm is used to iteratively smooth the overlapping region to obtain a single-layer point cloud, including:

[0100] Construct the topological relationships of overlapping regions using common search trees such as kd-tree, oc-tree, and quadtree;

[0101] Based on the aforementioned topological relationship, set the neighborhood search radius or the number of nearest neighbors K;

[0102] Search the neighborhood corresponding to each overlapping point in the overlapping region according to the neighborhood search radius or the nearest neighbor number K;

[0103] Based on the neighborhood, obtain the neighborhood information of the overlapping point, which includes the normal information of the neighborhood;

[0104] The overlapping point normals are weighted and corrected based on the normal information of the neighborhood, and the neighborhood is constrained based on the preset neighborhood normal angle threshold and the preset cylindrical neighborhood size.

[0105] Based on the scene and neighborhood information of the overlapping region, the overlapping points are smoothed using a corresponding point cloud smoothing algorithm. The point cloud smoothing algorithm includes bilateral filtering, trilateral filtering, linear interpolation, and MLS moving least squares.

[0106] After smoothing all overlapping points for a preset number of iterations, a single-layer point cloud is obtained.

[0107] It's important to note that bilateral filtering essentially performs weighted interpolation on overlapping points based on their neighborhood information. The weights include the normal projection distance between the overlapping point and a neighboring point, and the distance between the overlapping point and the axis of the cylinder's neighborhood. The larger the distance, the smaller the corresponding weight. Trilateral filtering also performs weighted interpolation on overlapping points based on their neighborhood information. The weights include the normal projection distance between the overlapping point and a neighboring point, the distance between the overlapping point and the axis of the cylinder's neighborhood, and the angle between the overlapping point and the normals of its neighbors. Linear interpolation divides the layered overlapping point cloud into three layers, applies bilateral / trilateral filtering to each layer to obtain a smoothing result, and then uses linear interpolation to obtain the final smoothing result. Moving Least Squares (MLS) fits a quadratic surface based on the overlapping point neighborhood information and projects the overlapping points onto this surface to obtain a smoothing result.

[0108] Specifically, such as Figure 2 The diagram shown illustrates the point cloud smoothing principles of bilateral filtering, trilateral filtering, and linear interpolation. Figure 2 The scan data 1 and scan data 2 in the image are the original 3D point clouds captured from different viewpoints. Figure 2 The dashed box in the figure represents the cylindrical neighborhood of the overlapping points. Represents the normal, d i r represents the distance between overlapping points and their neighboring points. i The radius of the neighborhood search is indicated by the arrow, which points to a single-layer point cloud after the overlapping point clouds of scan 1 and scan 2 data have been smoothed and corrected using a point cloud smoothing algorithm that employs bilateral filtering, trilateral filtering, or linear interpolation.

[0109] like Figure 3 As shown, the MLS moving least squares point cloud smoothing algorithm is used to smooth the overlapping point clouds. The two dashed circles on the locally fitted quadratic surface are the projections of scan 1 data and scan 2 data onto the locally fitted quadratic surface, respectively. Figure 3 The middle arrow points to a single-layer point cloud after the overlapping point clouds of scan 1 and scan 2 data have been smoothed and corrected using the MLS moving least squares point cloud smoothing algorithm.

[0110] It should be noted that the above only lists a few algorithms for smoothing and correcting overlapping point clouds. The specific processing procedures of these algorithms are not disclosed. Those skilled in the art can derive the specific processing procedures of each point cloud smoothing algorithm based on the scheme of this application and conventional technical means in the field, which will not be elaborated here. In practical applications, those skilled in the art can also use a series of improved neighborhood weighted average point cloud smoothing algorithms to smooth and correct overlapping point clouds according to the actual situation, and this application does not impose any restrictions on them.

[0111] Furthermore, in this embodiment 1, if the overlapping point cloud has only two layers of overlap, a point cloud smoothing algorithm using bilateral filtering or trilateral filtering is preferred for smoothing correction; if the overlapping point cloud has three or more layers of overlap, a point cloud smoothing algorithm using linear interpolation is preferred for smoothing correction; if the overlapping point cloud has high requirements for redundancy reduction accuracy but low requirements for efficiency, the MLS moving least squares method is preferred for smoothing correction. The scenarios for which each point cloud smoothing algorithm listed in this application is applicable are the best applicable scenarios derived from the experience of those skilled in the art. However, in practical applications, the specific point cloud smoothing algorithm used can be automatically determined by a computer program based on the obtained overlapping area information, thereby achieving a better smoothing effect. This application does not impose specific limitations on it.

[0112] After the smoothing correction described above, the overlapping points in the overlapping region are distributed in a single layer, i.e., a single-layer point cloud. At this point, the density of overlapping points in the single-layer point cloud varies, resulting in uneven density and a large amount of redundant data. To facilitate subsequent measurement processing, it is necessary to merge two redundant overlapping points with the same name (the two overlapping points are close enough to each other) in the single-layer point cloud, thereby making the single-layer point cloud distribution more uniform. Point cloud redundancy reduction algorithms such as voxel sampling, neighborhood averaging, or mean shift are used to simplify and merge the single-layer point cloud. Among them, the voxel sampling point cloud redundancy elimination algorithm divides a single-layer point cloud into voxel grids and obtains a point within the voxel grid to represent all overlapping points within the grid according to a certain point selection strategy; the neighborhood mean point cloud redundancy elimination algorithm uses a sufficiently small spherical neighborhood for searching, and if it meets a certain principle of being close enough, it replaces the overlapping points in the neighborhood with the centroid point; the mean drift algorithm point cloud redundancy elimination algorithm obtains the cluster center of each overlapping point in a single-layer point cloud, and then performs clustering through the Euclidean clustering algorithm to aggregate overlapping points whose cluster centers are close enough, and then uses the centroid point of each cluster set to represent all overlapping points in this cluster.

[0113] Furthermore, in one specific embodiment of this Example 1, the point cloud redundancy reduction algorithm in step S104 includes point cloud redundancy reduction methods based on voxel sampling, point cloud redundancy reduction methods based on neighborhood mean, and point cloud redundancy reduction methods based on mean drift.

[0114] Furthermore, in one specific embodiment of this Example 1, the process of simplifying a single-layer point cloud using the voxel sampling-based point cloud redundancy reduction method includes:

[0115] The single-layer point cloud is divided into at least one voxel grid;

[0116] Obtain the nearest center point or centroid of at least one overlapping point within each voxel grid;

[0117] The center nearest point or centroid point represents all overlapping points within the voxel grid;

[0118] The point cloud is output by combining all the nearest centers or centroids within the single-layer point cloud to form a deredundancy result point cloud.

[0119] Specifically, in this embodiment 1, the specific steps of the voxel sampling-based point cloud redundancy reduction method for simplifying a single-layer point cloud are as follows:

[0120] Obtain the maximum and minimum boundary values ​​of a single-layer point cloud;

[0121] A voxel mesh of the corresponding size is obtained according to a pre-set voxel size, wherein the voxel size is set according to the maximum boundary value and the minimum boundary value;

[0122] Based on the voxel grid information, at least one overlapping point in the single-layer point cloud is assigned to the corresponding voxel grid to obtain the voxel grid index;

[0123] Based on the voxel grid index, the nearest center point or centroid point within the voxel grid is selected to represent all overlapping points within the voxel grid.

[0124] The point cloud is output by combining all the nearest centers or centroids within the single-layer point cloud to form a deredundancy result point cloud.

[0125] Furthermore, in one specific embodiment of this Example 1, the point cloud redundancy reduction method based on neighborhood mean performs a simplification process for a single-layer point cloud by searching for a sufficiently small spherical neighborhood for each overlapping point in the single-layer point cloud. If the centroid of the spherical neighborhood coincides with the overlapping point, then the centroid replaces all overlapping points in the spherical neighborhood, thereby achieving the effect of redundancy reduction and simplification of overlapping point clouds.

[0126] Specifically, in this embodiment 1, the specific steps of the point cloud redundancy reduction method based on neighborhood mean to simplify a single-layer point cloud are as follows:

[0127] The topological relationships of the single-layer point cloud are constructed using common search trees such as kd-tree, oc-tree, and quadtree.

[0128] Based on the topological relationship, search the sphere neighborhood corresponding to each overlapping point with a sufficiently small radius;

[0129] Calculate the centroid of the neighborhood of the sphere;

[0130] If the centroid coincides with the overlapping point, the iteration stops and all overlapping points in the neighborhood of the sphere, except for the centroid, are deleted.

[0131] If the centroid does not coincide with the overlapping point, then search the sphere neighborhood of the centroid again and repeat the above steps.

[0132] Once all overlapping points with the same name have been merged, the centroids that coincide with the overlapping points will be combined and output as a deredundant point cloud.

[0133] Furthermore, in one specific embodiment of this Example 1, the point cloud redundancy reduction method based on mean drift simplifies the single-layer point cloud by repeatedly searching to gradually approach and converge to the center point of the optimal position of the overlapping points in the single-layer point cloud. Then, the overlapping points that are close enough are aggregated by Euclidean clustering algorithm, and the centroid of each cluster set represents all overlapping points in the cluster set.

[0134] Specifically, in this embodiment 1, the specific steps of the point cloud redundancy reduction method based on mean shift to simplify a single-layer point cloud are as follows:

[0135] The topological relationships of the single-layer point cloud are constructed using common search trees such as kd-tree, oc-tree, and quadtree.

[0136] Based on the aforementioned topological relationship, any overlapping point within the single-layer point cloud is selected to search its neighborhood;

[0137] The cluster center of the overlapping points is calculated by performing a Gaussian weighted average on all neighboring points within the neighborhood.

[0138] Using the cluster center point as the new target point, search its neighborhood again and calculate the new cluster center point;

[0139] Until the cluster centers converge, that is, the cluster centers no longer change;

[0140] Repeat the above steps to obtain the cluster center point corresponding to each overlapping point, and output the set of all cluster center points as a cluster center point set.

[0141] In addition, after obtaining the cluster centers corresponding to each overlapping point according to the above steps, the set of similar points is obtained by Euclidean distance clustering. The steps of Euclidean distance clustering are as follows:

[0142] Search for each cluster center in the set of cluster center points;

[0143] If the cluster center point is not labeled, it is placed in the seed point set C;

[0144] Search the seed point set C, and use the distance between two sufficiently close cluster centers that are considered to be of the same category as the neighborhood search radius to obtain the spherical neighborhood;

[0145] Each cluster center point within the neighborhood of the sphere is determined to be a point of the same category, and the classified cluster center points are marked as classified points and pushed into the seed point set C. This process continues until all cluster center points in the seed point set C have been processed, thus obtaining the current cluster set.

[0146] Repeat the above steps until all cluster centers in the cluster center set have been classified.

[0147] The centroids of the cluster centers in each current cluster set are calculated, and all centroids are combined and output as a deredundant point cloud.

[0148] It should be particularly noted that the specific values ​​of the pre-set parameters not given in Embodiment 1 can be set by those skilled in the art based on actual application conditions, and this application does not impose any restrictions on them. Furthermore, the above only lists a few algorithms for deduplication and simplification of overlapping point clouds. In practical applications, those skilled in the art can also employ a series of improved point cloud deduplication algorithms to deduplication and simplify overlapping point clouds according to actual conditions, such as point cloud deduplication algorithms based on triangulation, curvature sampling, index sampling, or random sampling; this application does not impose any restrictions on them.

[0149] Furthermore, in this embodiment 1, if the single-layer point cloud distribution is uniform and the point cloud redundancy reduction efficiency requirement is high, then the voxel sampling point cloud redundancy reduction algorithm is preferred; if the single-layer point cloud distribution is uneven and the point cloud redundancy reduction accuracy requirement is low but the efficiency requirement is high, then the neighborhood mean point cloud redundancy reduction algorithm is preferred; if the single-layer point cloud distribution is uneven and the point cloud redundancy reduction accuracy requirement is high but the efficiency requirement is low, then the mean shift point cloud redundancy reduction algorithm is preferred. The scenarios for which each point cloud redundancy reduction algorithm listed in this application is applicable are the best applicable scenarios derived from the experience of those skilled in the art. However, in actual applications, the specific point cloud redundancy reduction algorithm used can be automatically determined by the computer program based on the obtained overlapping area information, thereby achieving better point cloud redundancy reduction and simplification effects. This application does not impose specific limitations on it.

[0150] Example 2

[0151] Corresponding to Embodiment 1 of the above-described method for eliminating redundant overlapping point clouds applicable to multiple scenarios, this application also provides Embodiment 2 of an apparatus for eliminating redundant overlapping point clouds applicable to multiple scenarios. The apparatus includes an acquisition unit, an overlapping region determination unit, an overlapping point cloud smoothing unit, and an overlapping point cloud redundancy elimination unit connected to each other.

[0152] The acquisition unit is configured as follows:

[0153] The original 3D point clouds from different perspectives are stitched together to obtain the stitched point cloud;

[0154] The overlapping region determination unit is configured as follows:

[0155] Obtain the overlapping region of the stitched point cloud;

[0156] The overlapping point cloud smoothing unit is configured as follows:

[0157] Based on the scenario of the overlapping region, the corresponding point cloud smoothing algorithm is used to iteratively smooth the overlapping region to obtain a single-layer point cloud.

[0158] The overlapping point cloud redundancy reduction unit is configured as follows:

[0159] Based on the scenario of the single-layer point cloud, a corresponding point cloud redundancy reduction algorithm is used to perform redundancy reduction and simplification processing on the single-layer point cloud to obtain the redundancy reduction result point cloud.

[0160] Example 3

[0161] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the overlapping point cloud redundancy reduction method applicable to multiple scenarios in Embodiment 1.

[0162] Example 4

[0163] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the overlapping point cloud redundancy reduction method applicable to multiple scenarios in Embodiment 1.

[0164] Application examples

[0165] The solution in this application is illustrated by using two overlapping point layers with a uniform point cloud distribution:

[0166] like Figure 4a As shown, this is a schematic diagram of the first stitched point cloud. After processing using the point cloud redundancy reduction method of this application, specifically, a bilateral filtering point cloud smoothing algorithm is used to eliminate the layering phenomenon of overlapping point cloud data, and a voxel sampling point cloud redundancy reduction algorithm is used to perform redundancy reduction and simplification processing on the single-layer point cloud after iterative smoothing correction, resulting in the following... Figure 4b A schematic diagram of the first deredundancy result point cloud.

[0167] like Figure 5a As shown, this is a schematic diagram of the point cloud obtained from the second stitching result. After processing using the point cloud redundancy reduction method of this application, specifically, a bilateral filtering point cloud smoothing algorithm is used to eliminate the layering phenomenon of overlapping point cloud data, and a voxel sampling point cloud redundancy reduction algorithm is used to perform redundancy reduction and simplification processing on the single-layer point cloud after iterative smoothing correction, resulting in the following... Figure 5b The second deredundancy result is shown in the point cloud diagram.

[0168] from Figure 4b and Figure 5bIt can be clearly seen that the point cloud redundancy reduction method of this application can effectively eliminate the layering phenomenon of point cloud data in overlapping areas, and has a better effect on redundancy reduction and simplification of the stitched point cloud, resulting in a clearer point cloud.

[0169] Similar parts between the embodiments provided in this application can be referred to mutually. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. A method for removing redundancy of overlapping point clouds suitable for various scenarios, characterized in that, The method for removing redundancy of the overlapping point cloud comprises: splicing and fusing 3D original point clouds under different visual angles to obtain a spliced result point cloud; acquiring an overlapping region of the spliced result point cloud; automatically adopting a corresponding point cloud smoothing algorithm to iteratively smooth the overlapping region according to a scene of the overlapping region to obtain a single-layer point cloud, the point cloud smoothing algorithm comprising bilateral filtering, tri-linear filtering, linear interpolation and MLS (Moving Least Squares) method, if the overlapping region is two-layer overlapping, adopting bilateral filtering or tri-linear filtering to iteratively smooth, if the overlapping region is at least three-layer overlapping, adopting linear interpolation to iteratively smooth, if the redundancy removal accuracy of the overlapping region is higher than a first preset threshold and the efficiency is lower than a second preset threshold, adopting the MLS method to iteratively smooth; automatically adopting a corresponding point cloud redundancy removal algorithm to remove redundancy of the single-layer point cloud according to a scene of the single-layer point cloud to obtain a redundancy removal result point cloud, the point cloud redundancy removal algorithm comprising a point cloud redundancy removal method based on voxel sampling, a point cloud redundancy removal method based on neighborhood mean and a point cloud redundancy removal method based on mean shift; if the single-layer point cloud is uniformly distributed, adopting the point cloud redundancy removal method based on voxel sampling to simplify the single-layer point cloud; if the single-layer point cloud is not uniformly distributed, adopting the point cloud redundancy removal method based on neighborhood mean or the point cloud redundancy removal method based on mean shift to simplify the single-layer point cloud.

2. The method of claim 1, wherein, The method for acquiring the overlapping region of the spliced result point cloud comprises: constructing a topological relationship of the spliced result point cloud, the spliced result point cloud comprising at least one target point; judging whether at least one target point is an overlapping point according to a preset judgment algorithm; combining all target points determined as overlapping points in the spliced result point cloud and outputting as an overlapping region.

3. The method of claim 2, wherein, The method for acquiring the overlapping region of the spliced result point cloud comprises: constructing a topological relationship of the spliced result point cloud by a common search tree; setting a neighborhood search radius according to the topological relationship; searching a spherical neighborhood corresponding to each target point in the spliced result point cloud according to the neighborhood search radius; acquiring a target point neighborhood point number corresponding to the target point according to the spherical neighborhood; if the target point neighborhood point number is greater than a preset neighborhood point number threshold, marking the target point as an overlapping point; combining all target points determined as overlapping points in the spliced result point cloud and outputting as an overlapping region.

4. The method of claim 1, wherein, According to the scene of the overlapping region, adopting a corresponding point cloud smoothing algorithm to iteratively smooth the overlapping region to obtain a single-layer point cloud, comprising: constructing a topological relationship of the overlapping region by a common search tree; setting a neighborhood search radius according to the topological relationship; searching a neighborhood corresponding to each overlapping point in the overlapping region according to the neighborhood search radius; acquiring overlapping point neighborhood information according to the neighborhood; adopting a corresponding point cloud smoothing algorithm to smooth the overlapping point according to the scene of the overlapping region and the overlapping point neighborhood information; after completing the smoothing processing of all overlapping points for a preset iteration number, obtaining a single-layer point cloud.

5. The method of claim 1-4, wherein, The processing procedure of the point cloud redundancy removal method based on voxel sampling for simplifying the single-layer point cloud comprises: Dividing the single-layer point cloud into at least one voxel grid; Obtaining the center nearest point or barycenter point of at least one overlapping point in each voxel grid; Representing all overlapping points in the voxel grid by the center nearest point or barycenter point; Combining all center nearest points or barycenter points in the single-layer point cloud to output a de-duplication result point cloud.

6. The method of claim 5, wherein, The steps of the point cloud de-duplication method based on voxel sampling for simplifying a single-layer point cloud are as follows: Obtaining the maximum boundary value and the minimum boundary value of the single-layer point cloud; Obtaining a voxel grid of a corresponding size according to a preset voxel size, wherein the voxel size is set according to the maximum boundary value and the minimum boundary value; Dividing at least one overlapping point in the single-layer point cloud into the corresponding voxel grid according to the voxel grid information to obtain a voxel grid index; Selecting the center nearest point or barycenter point in the voxel grid to represent all overlapping points in the voxel grid according to the voxel grid index; Combining all center nearest points or barycenter points in the single-layer point cloud to output a de-duplication result point cloud.

7. An apparatus for removing redundancy of overlapping point clouds suitable for various scenarios, comprising: The overlapping point cloud de-duplication device according to any one of claims 1-6 is applied to multiple scenes, and comprises an acquisition unit, an overlapping region judgment unit, an overlapping point cloud smoothing unit, and an overlapping point cloud de-duplication unit connected with each other. The acquisition unit is configured to: Splice and fuse 3D original point clouds under different viewing angles to obtain a spliced result point cloud. The overlapping region judgment unit is configured to: Obtain an overlapping region of the spliced result point cloud. The overlapping point cloud smoothing unit is configured to: According to the scene of the overlapping region, perform iterative smoothing processing on the overlapping region by using a corresponding point cloud smoothing algorithm to obtain a single-layer point cloud. The overlapping point cloud de-duplication unit is configured to: According to the scene of the single-layer point cloud, perform de-duplication and simplification processing on the single-layer point cloud by using a corresponding point cloud de-duplication algorithm to obtain a de-duplication result point cloud.

8. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the overlapping point cloud de-duplication method according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the steps of the overlapping point cloud de-duplication method according to any one of claims 1-6.

Images