A lunar surface point cloud global registration method and device based on overlap classification

Abstract

This invention relates to a method and apparatus for global registration of lunar surface point clouds based on overlap classification. The method includes the following steps: acquiring multiple overlapping or partially overlapping point clouds; extracting the boundaries of each point cloud and calculating the overlap between point clouds; classifying the point clouds into overlapping point cloud sets and isolated point cloud sets based on the overlap and a preset threshold; acquiring a reference digital terrain model of the lunar surface; obtaining a reference point cloud through discretization; performing density consistency processing on each individual point cloud in the overlapping and isolated point cloud sets; performing iterative registration between the overlapping point cloud sets and the reference point cloud based on global generalized pluralistic analysis and the iterative nearest point algorithm; performing registration between the isolated point cloud sets and the reference point cloud based on the iterative nearest point algorithm; and completing the global point cloud registration to obtain a globally registered multi-view point cloud. Compared with existing technologies, this invention can generate higher quality fused terrain models.

Description

Technical Field

[0001] This invention relates to the field of point cloud registration and fusion, and in particular to a method and apparatus for global registration of lunar surface point clouds based on overlap classification. Background Technology

[0002] Due to the rapid advancements in photogrammetry and computer vision technologies, 3D point clouds generated from dense image matching have become an important data source for lunar topographic mapping due to their accuracy, reliability, and low cost. However, influenced by factors such as camera orientation accuracy and dense matching accuracy, layering and misalignment still exist between point clouds, posing significant challenges to subsequent multi-viewpoint cloud fusion. To meet the needs of large-scale lunar surface mapping, global registration of multi-viewpoint clouds is necessary to improve positional consistency between them, and appropriate weighting strategies should be employed during the fusion phase to obtain a higher-quality fused terrain model.

[0003] Global registration of point clouds is a more complex and challenging research problem than point cloud registration between two points, and it is mainly divided into sequential registration and global registration. Sequential registration methods involve aligning the two overlapping views sequentially each time, while global registration mainly addresses the cumulative registration error generated during sequential registration by evenly distributing the error across all overlapping views.

[0004] Chinese Patent Application Publication No. CN 110009667 A provides a global registration method for multi-view point clouds based on Rodrigues transform. First, based on obtaining multiple multi-view point clouds taken from different perspectives with partial overlap between the blocks, matching point pairs are found between any two overlapping point clouds. The corresponding rotation transformation matrix and translation transformation vector of each point cloud relative to itself are calculated. Then, the Rodrigues transform is used to convert all the calculated rotation transformation matrices into rotation transformation vectors. The rotation transformation vectors and translation transformation vectors are combined and substituted into the observation values ​​into the global optimization model of multi-view point clouds. Through multiple iterative adjustment calculations, the optimal transformation matrix of each multi-view scan surface point cloud is obtained, thereby completing the overall accurate registration of multi-view scan surface point clouds.

[0005] The aforementioned application can achieve global registration of multi-viewpoint clouds with a certain degree of overlap; however, it does not classify and process different degrees of overlap, resulting in unsatisfactory registration accuracy. Summary of the Invention

[0006] The purpose of this invention is to overcome the defects of the prior art by providing a global registration method and device for lunar surface point clouds based on overlap classification, so as to solve or partially solve the problem of unsatisfactory point cloud registration accuracy.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] One aspect of the present invention provides a global registration method for lunar surface point clouds based on overlap classification, comprising the following steps:

[0009] Acquire multiple overlapping or partially overlapping point clouds, extract the boundaries of each point cloud and calculate the overlap between point clouds, and divide the point clouds into overlapping point cloud sets and isolated point cloud sets based on the overlap and a preset threshold.

[0010] A reference digital terrain model of the lunar surface is obtained, and a reference point cloud is obtained by discretization. Density consistency processing is performed on each individual point cloud in the overlapping point cloud set and the isolated point cloud set.

[0011] For the overlapping point cloud, iterative registration is performed with the reference point cloud based on global generalized universal analysis and iterative nearest point algorithm;

[0012] For the isolated point cloud, registration is performed with the reference point cloud based on the iterative nearest point algorithm;

[0013] Complete global point cloud registration to obtain a globally registered multi-view point cloud.

[0014] As a preferred technical solution, the calculation process for the overlap between point clouds includes:

[0015] The point cloud is projected onto a two-dimensional plane, and the boundary point set of the point cloud is obtained based on the Alhpa Shapes algorithm. The boundary point set is simplified based on the convex hull algorithm, and the overlap between point clouds is calculated using the Geos library.

[0016] As a preferred technical solution, the density consistency processing includes:

[0017] By using voxel filtering, the density of each individual point cloud in the overlapping point cloud set and the isolated point cloud set is made consistent with the density of the reference point cloud.

[0018] As a preferred technical solution, the registration process for the aforementioned overlapping point cloud includes:

[0019] For each point in each point cloud, the nearest neighbor matching relationship is searched in the remaining point clouds. Based on the nearest neighbor matching relationship and the preset objective function, the rotation matrix and translation matrix of each point cloud are calculated, and the point cloud is updated. This step is repeated until the preset global convergence condition is met, and the registration is completed.

[0020] As a preferred technical solution, the global convergence condition is:

[0021] If the number of iterations exceeds a preset value, or if the global registration error before and after global registration is less than a preset value, wherein the global registration error is based on the average distance between corresponding points in the point cloud and the centroid cloud, and the centroid cloud is calculated based on an independent subset of each nearest neighbor matching relationship.

[0022] As a preferred technical solution, the objective function is:

[0023]

[0024] Where m is the number of overlapping point clouds, A1, A2, ..., A m Let m be a set of points, and p be the number of points in the point cloud set. T is the rotation matrix, j is a p×1 unit vector, t is the translation matrix, and M is a p×p matrix. i This indicates the visibility of a point set, with 1 indicating visibility and 0 indicating absence.

[0025] As a preferred technical solution, the following are also included:

[0026] Based on the multi-view point cloud after global registration, a spatial raster fusion strategy based on triangulation error and distance weighting is adopted to obtain the fused point cloud, and interpolation is used to obtain the fused digital terrain model.

[0027] The fused digital terrain model is subjected to slope-based outlier detection and contour-based hole filling to complete the fusion of point cloud and digital terrain model.

[0028] As a preferred technical solution, the process of acquiring the fused digital terrain model includes:

[0029] Based on the extreme values ​​on the three coordinate axes of the three-dimensional point set of the multi-view point cloud after global registration, calculate the size parameters of the minimum bounding box of the point cloud;

[0030] The minimum bounding box is divided into multiple spatial grids and numbered based on the set voxel grid side length;

[0031] The spatial raster fusion value is calculated by using a fusion strategy that weights triangulation error and distance, and the fused digital terrain model is obtained.

[0032] As a preferred technical solution, the calculation of the spatial grid fusion value is achieved using the following formula:

[0033]

[0034]

[0035] Among them, w i,t Indicates the i-th th Point to tth Weights in the grid, The value represented by d indicates the spatial raster fusion value. i,t and e i,t Let represent the Euclidean distance and triangulation error between the i-th point and the t-th grid center, respectively, and let α1 and α2 be the weight values ​​of the relative error of triangulation and the distance term, respectively.

[0036] In another aspect, an electronic device is provided, comprising: one or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for performing the above-described method for global registration of lunar surface point clouds based on overlap classification.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] (1) High point cloud registration accuracy: This invention overcomes or partially overcomes the shortcomings of low registration accuracy of multi-view point clouds in complex point cloud coverage. It combines global GPA+ICP adjustment with ICP registration based on reference point cloud and uses a global registration method for lunar surface point clouds that uses the above two methods according to the overlap degree classification. This achieves global registration of lunar surface multi-view point clouds and improves the positional consistency and absolute accuracy between point clouds.

[0039] (2) High registration efficiency: Existing global registration methods do not fully consider the low overlap rate of real photogrammetric point clouds. They cannot guarantee reliable registration accuracy for complex point cloud distributions. This invention classifies based on overlap and adopts two strategies for different overlap situations: global GPA+ICP adjustment and ICP registration based on reference point clouds. These strategies guarantee registration accuracy in both cases. At the same time, the point cloud with consistent density is used as input data to improve the time efficiency of multi-view point cloud registration.

[0040] (3) High accuracy of fused digital terrain model: This invention adopts a weighting strategy that comprehensively considers the measurement error and distance of three-dimensional point triangulation for multi-view cloud fusion, which effectively eliminates the subtle geometric inconsistencies that still exist in multi-view cloud, and generates a smooth terrain model. At the same time, this paper performs outlier detection and hole filling on the fused terrain model to ensure the integrity and accuracy of the terrain model. Attached Figure Description

[0041] Figure 1 This is a flowchart of the lunar surface point cloud global registration method based on overlap classification in the embodiment.

[0042] Figure 2 This is a schematic diagram illustrating the point cloud coverage and the distribution of point cloud sets with an overlap greater than 50% in the embodiment.

[0043] Figure 3This is a schematic diagram of the reference DEM and reference point cloud in the embodiment;

[0044] Figure 4 This is a schematic diagram showing the point cloud density before and after unification in the embodiment;

[0045] Figure 5 This is a schematic diagram of the projection point cloud distribution before and after GPA+ICP global registration in the embodiment.

[0046] Figure 6 This is a schematic diagram of slope-based anomaly grid detection in the embodiment.

[0047] Figure 7 This is a schematic diagram of hole filling based on contour detection in the embodiment;

[0048] Figure 8 The example uses a fused DEM and a mountain shadow map. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0050] Example 1

[0051] To provide a global registration and fusion method for lunar surface point clouds based on overlap classification, overcome the uncertainty of registration errors caused by complex point cloud coverage, improve the positional consistency after global registration of multi-view point clouds, and thus generate higher-quality fused terrain models, see [link to relevant documentation]. Figure 1 This embodiment provides a global registration method for lunar surface point clouds based on overlap classification, including the following steps.

[0052] S1, Obtain the boundaries of each point cloud and calculate the overlap between point clouds. This step specifically includes the following sub-steps.

[0053] S11, Project the point onto a two-dimensional plane;

[0054] S12, use the alpha shapes algorithm to obtain the point cloud boundary point set;

[0055] S13, using the convex hull algorithm to simplify the point cloud boundary point set;

[0056] S14, calculate the overlap between point clouds using the geos library.

[0057] Among them, the point cloud is the photogrammetric point cloud generated by the lunar orbiter's multi-view LROC NAC image, and the total number of point clouds is ≥3.

[0058] See Figure 2 (a) shows the coverage of point clouds in the Apollo 17 landing area of ​​this paper, totaling 12 points. The overlap between the point clouds was calculated for each point cloud. The resulting point cloud set a{4,8,11,12} has an overlap greater than 50%, indicating sufficient overlap (e.g., ...). Figure 2 (b)); For the remaining isolated point clouds, ICP registration based on the reference point cloud is used, taking point clouds 2, 3, and 5 as examples.

[0059] S2, the reference digital terrain model (DEM) is discretized to obtain the reference point cloud.

[0060] Specifically, a three-dimensional point is generated for each pixel unit, and its geographic coordinates and elevation information are assigned. The geographic coordinates are calculated from the starting coordinate information in the DEM header file and the ground resolution in the x and y directions, and the elevation information is the pixel value of each pixel.

[0061] This application references the SLDEM2015 data (download address: http: / / imbrium.mit.edu / EXTRAS / SLDEM2015 / ), with a resolution of approximately 60m / pixel. The data is preprocessed, including elevation value reduction, projection, and clipping steps, and a three-dimensional point cloud is obtained according to the discretization method described above. Figure 3 (a) and (b) are the reference DEM and reference point cloud, respectively.

[0062] S3 processes the cloud map after the overlap calculation in S1 to obtain a downsampled point cloud with the same density as the reference point cloud.

[0063] Specifically, voxel filtering is used to ensure that the density of a single-point cloud after downsampling is consistent with that of the reference point cloud. The formula is as follows:

[0064]

[0065] Among them, the minimum value of the point cloud point set is X. min Y min Z min The voxel grid has a side length of cell, and the t-th voxel grid is numbered (i,j,k), where t = i·l x +j·l y +k.

[0066] By using voxel filtering, the downsampling parameter was obtained by parameter tuning: 60. Figure 4 (a) and (b) show the point cloud before and after downsampling. The red dots are SLDEM point clouds, and the blue dots are single-point point clouds. It can be seen intuitively that the point cloud densities of the two tend to be consistent.

[0067] S4. For point clouds with high overlap, global generalized Protodyakonov analysis and iterative nearest point (GPA+ICP) algorithm are used for registration to evaluate the global adjustment accuracy. This step includes the following sub-steps.

[0068] S41, search for the "nearest neighbor match" for all points within each point cloud in the remaining point cloud views.

[0069] Specifically, for the corresponding point obtained by the nearest neighbor search of the Iterative Closest Point (ICP) algorithm, we define when point p i ∈A i With p j ∈A j The two points are the nearest neighbors (i.e., p) i In A j The nearest neighbor in is p j p j In A i The nearest neighbor in is p i ) represents the "nearest neighbor matching relationship".

[0070] S42, calculate the centroid K for each independent subset of the nearest neighbor matching relation.

[0071] S43, Solve the rotation matrix T1, T2, ..., T for each point cloud by minimizing the function. m Translation matrices t1, t2, ..., t m .

[0072] The objective function to be minimized is:

[0073]

[0074] Where m is the number of point clouds (point sets A1, A2, ..., A m Let p be the number of points in the set. T is a 3×3 rotation matrix, j is a p×1 unit vector, t is a 3×1 translation matrix, and a p×p matrix M is introduced. i Let represent the visibility of a point set, where 1 indicates visibility and 0 indicates absence.

[0075] S44, the obtained transformation matrix is ​​applied to each scenic point cloud to obtain the updated point cloud.

[0076] S45, return to step S41, iterate until global convergence.

[0077] Specifically, Point-to-Point ICP is used to register the two, and the maximum number of iterations, the difference between the two transformation matrices, and the mean square error are set as conditions for ICP iteration convergence.

[0078] After the iterative calculation is completed, the accuracy of this method is evaluated by calculating the global registration error and absolute registration error of GPA+ICP.

[0079] The point clouds of individual viewpoints were downsampled to 150 meters, and then used as input data for global adjustment, resulting in 16346, 12710, 9995, and 28206 3D points, respectively. The iteration termination condition set in this paper was (errAfter-errPrev) / errAfter < 1e-10 or the number of iterations reached 100. errAfter and errPrev represent the global registration error before and after global adjustment (the average distance from each viewpoint cloud to the corresponding point in the centroid cloud). The global adjustment process iterated 31 times until convergence. Table 1 shows the RT transformation matrix obtained from the global ICP adjustment of the four viewpoint clouds. Figure 5 (a) and (b) show the point cloud distribution before and after global adjustment, respectively. It can be seen that there is obvious elevation inconsistency between point clouds before registration, and point cloud 11 is significantly higher than the other three scenes. After global registration, it can be seen that the four scene clouds maintain good spatial self-consistency.

[0080] Table 1. Point Cloud RT Matrix

[0081]

[0082] S5 uses ICP to register the globally adjusted point cloud group to the reference point cloud, and directly registers isolated point clouds with low overlap to the reference point cloud.

[0083] The four point cloud sets (point cloud set a) that have undergone global GPA+ICP registration are registered as a whole to the reference point cloud. The specific configuration parameters of ICP are: maximum number of iterations: 500; difference between the two transformation matrices: 1e-10; mean square error: 1. The obtained RT transformation matrix is ​​shown in Table 2. For isolated point clouds with very low overlap with neighboring point clouds, they are directly registered to the reference point cloud. The ICP registration parameters are the same as above. The single-scene registration effect of point clouds 2, 3, and 5 is listed below, and their RT matrices are shown in Table 2.

[0084] Table 2 RT Matrix of Point Cloud Aggregates and Isolated Point Clouds

[0085]

[0086]

[0087] S6 uses multiple verification methods to evaluate the global point cloud registration accuracy.

[0088] First, the global GPA+ICP registration accuracy was evaluated. The global GPA+ICP registration error (the average distance between corresponding points in each point cloud and the centroid cloud) decreased from 14.0388 meters to 7.04875 meters. Second, the overall registration accuracy was reflected by evaluating the error between each pair of point clouds before and after global registration, as shown in Table 3. Point clouds with no overlap or very small overlap are indicated by " / ". The table shows that after global registration, the average registration error between point clouds decreased significantly. For example, the average registration error between point cloud 4 and point cloud 12 decreased from 43 meters to 5.2 meters, and the standard deviation also decreased to varying degrees, indicating that the registration error was more concentrated.

[0089] Table 3. Statistics of Point Cloud Deviations Before and After Global GPA+ICP Registration

[0090]

[0091] Secondly, the absolute accuracy of the point cloud before and after registration with the reference point cloud was evaluated using LOLA laser checkpoints. A total of 710 laser checkpoints were selected. The correlation statistics of the absolute deviation between the point cloud and the LOLA laser checkpoints before and after registration are shown in Table 4. The absolute deviation RMSE decreased from 30.13m to 3.86m, and the standard deviation also decreased significantly, indicating that the overall deviation was greatly reduced and more concentrated. For isolated point cloud registration, the correlation statistics of the absolute deviation between the isolated point cloud and the LOLA laser checkpoints before and after registration are shown in Table 5. It can be seen that: before registration, the errors of individual point clouds with the reference point cloud varied. For example, point cloud 2 had a smaller error, while points clouds 3 and 5 had larger errors. This situation was mainly affected by the accuracy of the foreign side. After registration, the average absolute deviation of individual point clouds with LOLA laser points was about 3 meters, and the absolute deviation was between 3 and 4 meters. The errors decreased significantly, indicating that the registration effect of individual point clouds was good and they were all aligned with the reference point cloud. All point clouds are globally adjusted using the aforementioned global registration strategy to ensure the consistency and absolute accuracy of their spatial positions.

[0092] Table 4. Absolute Deviation Between the Registered Point Cloud DEM and the LOLA Laser Check Points

[0093]

[0094] Table 5. Absolute Deviations Between Single-Scene DEM and LOLA Laser Checkpoints Before and After Registration Table 5.

[0095]

[0096] S7. For the multi-view point cloud after global registration, a spatial raster fusion strategy that weights triangulation error and distance is used to obtain the fused point cloud, and interpolation is used to obtain the fused DEM. This step includes the following sub-steps.

[0097] S71, Calculate the side length l of the minimum bounding box of the point cloud based on the maximum and minimum values ​​on the X, Y, and Z coordinate axes of the multi-view point cloud 3D point set. x l y l z 。;

[0098] S72, set the cell side length of the voxel grid, and divide the minimum bounding box into M*N*L spatial grids;

[0099] S73, number each spatial grid to determine the spatial grid to which each 3D point belongs;

[0100] S74 employs a fusion strategy that considers a weighted average of triangulation error and distance to calculate the spatial raster fusion value.

[0101] The formulas for determining the weight of spatial raster values ​​and calculating raster values ​​are as follows:

[0102]

[0103]

[0104] Among them, w i,t Indicates the i-th th Point to t th The weights in the grid, d i,t and e i,t Let represent the Euclidean distance and triangulation error between the i-th point and the t-th grid center, respectively. α1 and α2 are the weight values ​​for the relative error of triangulation and the distance term, respectively. t To merge spatial raster values.

[0105] The weighting formula of this invention sets α1 = 0.5 and α2 = 0.5 respectively. A fused terrain model is obtained through fusion and interpolation.

[0106] S8 performs slope-based outlier detection and contour-based hole filling on the fused DEM.

[0107] Specifically, the formula for slope-based anomaly raster detection is as follows:

[0108] Thresholdx = k·RMSEx DSC

[0109] Thresholdy = k·RMSEy DSC

[0110] DSC (Differences in Slope Change) represents the difference in slope change at a single point. The RMSE (RMSEx) value of this sum for each grid cell is calculated by summing the DSC values ​​for each data point in the same direction. DSC The threshold is set to k times the RMSE in each direction.

[0111] First, on the anomaly detection side, with k=3, Thresholdx=0.196135 and Thresholdy=0.197602. 865 pixels are retrieved as coarse errors from the 18,605,325 grids of the fused DEM. Figure 6 This shows the anomaly detection status in the magnified area.

[0112] Secondly, regarding hole filling, based on the binary image of the fused DEM, all contours are detected and saved to a linked list. The contours are output in the form of Freeman chain code, and linear interpolation is performed to fill the holes in the x and y directions. Figure 7 This is a schematic diagram illustrating the filling of holes in the magnified area. After the above post-processing steps, the final fused DEM and mountain shadow image are shown below. Figure 8 As shown.

[0113] Example 2

[0114] Based on Embodiment 1, this embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the above-described global registration method for lunar surface point clouds based on overlap classification.

[0115] This application employs a global registration method to adjust multi-viewpoint clouds without requiring prior knowledge of view order. Furthermore, considering the complex distribution of multi-viewpoint clouds, it categorizes them based on overlap and uses a global registration method combining global GPA+ICP adjustment with ICP registration based on reference point clouds. On the point cloud fusion side, a fusion strategy considering triangulation error and distance weighting is adopted to achieve the fusion of multi-viewpoint clouds, effectively eliminating subtle inconsistencies between point clouds and removing data redundancy. Post-processing of the fused terrain model ensures its integrity and accuracy.

[0116] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A global registration method for lunar surface point clouds based on overlap classification, characterized in that, Includes the following steps: Acquire multiple overlapping or partially overlapping point clouds, extract the boundaries of each point cloud and calculate the overlap between point clouds, and divide the point clouds into overlapping point cloud sets and isolated point cloud sets based on the overlap and a preset threshold. A reference digital terrain model of the lunar surface is obtained, and a reference point cloud is obtained by discretization. Density consistency processing is performed on each individual point cloud in the overlapping point cloud set and the isolated point cloud set. For the overlapping point cloud, iterative registration is performed with the reference point cloud based on global generalized universal analysis and iterative nearest point algorithm; For the isolated point cloud, registration is performed with the reference point cloud based on the iterative nearest point algorithm; Complete global point cloud registration to obtain the globally registered multi-view point cloud; Based on the multi-view point cloud after global registration, a spatial raster fusion strategy based on triangulation error and distance weighting is adopted to obtain the fused point cloud, and interpolation is used to obtain the fused digital terrain model. The fused digital terrain model is subjected to slope-based outlier detection and contour-based hole filling to complete the fusion of point cloud and digital terrain model.

2. The global registration method for lunar surface point clouds based on overlap classification according to claim 1, characterized in that, The calculation process for the overlap between point clouds includes: The point cloud is projected onto a two-dimensional plane, and the boundary point set of the point cloud is obtained based on the Alhpa Shapes algorithm. The boundary point set is simplified based on the convex hull algorithm, and the overlap between point clouds is calculated using the Geos library.

3. The global registration method for lunar surface point clouds based on overlap classification according to claim 1, characterized in that, The density uniformity processing includes: By using voxel filtering, the density of each individual point cloud in the overlapping point cloud set and the isolated point cloud set is made consistent with the density of the reference point cloud.

4. The global registration method for lunar surface point clouds based on overlap classification according to claim 1, characterized in that, The registration process for the aforementioned overlapping point cloud includes: For each point in each point cloud, the nearest neighbor matching relationship is searched in the remaining point clouds. Based on the nearest neighbor matching relationship and the preset objective function, the rotation matrix and translation matrix of each point cloud are calculated, and the point cloud is updated. This step is repeated until the preset global convergence condition is met, and the registration is completed.

5. The global registration method for lunar surface point clouds based on overlap classification according to claim 4, characterized in that, The global convergence condition is as follows: If the number of iterations exceeds a preset value, or if the global registration error before and after global registration is less than a preset value, wherein the global registration error is based on the average distance between corresponding points in the point cloud and the centroid cloud, and the centroid cloud is calculated based on an independent subset of each nearest neighbor matching relationship.

6. The global registration method for lunar surface point clouds based on overlap classification according to claim 4, characterized in that, The objective function is: in, m The number of overlapping points in the cloud. A 1 ,A 2 ,…,A m They represent m A set of points, p To represent the number of point clouds in a point cloud cluster, let , T Let be a rotation matrix. j for p A unit vector of size 1. t It is a translation matrix. p × p matrix M i This indicates the visibility of a point set, with 1 indicating visibility and 0 indicating absence. .

7. The global registration method for lunar surface point clouds based on overlap classification according to claim 1, characterized in that, The process of acquiring the fused digital terrain model includes: Based on the extreme values ​​on the three coordinate axes of the three-dimensional point set of the multi-view point cloud after global registration, calculate the size parameters of the minimum bounding box of the point cloud; The minimum bounding box is divided into multiple spatial grids and numbered based on the set voxel grid side length; The spatial raster fusion value is calculated by using a fusion strategy that weights triangulation error and distance, and the fused digital terrain model is obtained.

8. The global registration method for lunar surface point clouds based on overlap classification according to claim 7, characterized in that, The spatial raster fusion value is calculated using the following formula: in, w i,t Indicates the i th Point to point t th Weights in the grid, This represents the spatial raster fusion value. d i,t and e i,t They represent the first i The point and the first t Euclidean distance between grid centers and triangulation error a 1 and a 2 These are the relative error of triangulation and the weight values ​​of the distance term, respectively.

9. An electronic device, characterized in that, include: One or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for executing the lunar surface point cloud global registration method based on overlap classification as described in any one of claims 1-8.

Images