Private network park digital twin visualization method and apparatus, and storage medium

Abstract

The present invention relates to the technical field of digital twin, and in particular to a private network park digital twin visualization method, apparatus and device, and a computer storage medium. Disclosed in the present application is a private network park digital twin visualization method, comprising: using a clustering algorithm to classify ground objects in a remote sensing image of a park, dividing grids on the basis of formed pixel clusters, and calculating dilution ratios thereof; mapping grid data and point cloud data into a unified coordinate system, and using the clustering algorithm to cluster the point cloud data in each grid; sampling each point cloud cluster on the basis of the dilution ratio of the corresponding grid, and finding out non-edge points by means of a BoundED algorithm; on the basis of the proportion of the non-edge points in each point cloud cluster, calculating an initial dilution threshold of each point cloud cluster; and, after each point cloud cluster has been diluted, rendering and presenting same by means of a three-dimensional rendering engine and returning a rendering frame number and, if the rendering frame number meets a preset condition, ending the process.

Description

A digital twin visualization method, device and storage medium for a private network park

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Chinese Patent Application No. 202410123262.0 filed on January 29, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present invention relates to the field of digital twin technology, and in particular to a method, device, equipment and computer storage medium for visualizing a digital twin of a private network park. Background Art

[0004] The goal of 3D laser point cloud data thinning is to reduce redundant points, lower the number of point clouds, and improve the efficiency of subsequent processing. Traditional thinning methods generally use public point cloud processing platforms such as the Point Cloud Library (PCL). They generally use centroid bounding box clustering and uniform thinning methods to maximize the uniformity of the thinned point cloud and effectively prevent the generation of holes in the thinned point cloud data. This method is simple to calculate and very efficient. Most large-scale commercial software and point cloud data processing platforms use this method as a data thinning technology solution.

[0005] Existing solutions for point cloud data dilution in 5G private network digital twins are currently being developed. However, due to the significant differences in the features of 3D model elements in 5G private networks, a large number of 3D point cloud data scenes suffer from large data volumes, high rendering difficulty, and long loading times. To adapt to a variety of different client types for efficient, high-performance, and dynamic display of campus 3D scenes, the entire 3D scene model must address the data rendering issues of different clients. This paper proposes a dilution metric algorithm to address the issue of data model lightweighting and achieve lightweight rendering of 5G private network digital twin scenes, hoping to achieve a good balance between performance and rendering. Due to the inherent characteristics of the scene, the 3D scene element features of each 5G private network campus object are different. How to perform 3D reconstruction of point cloud data based on the unique borders and materials of the campus elements is an urgent problem to be solved in the automation of 5G private network digital twins. Existing technologies generally rely on manual setting of the thinning algorithm threshold, and use uniform sampling or chord deviation sampling to dilute the point cloud. This algorithm does not consider the role of different characteristic factors of various entities in 5G private network scenarios. It is too mechanical and single, which often makes three-dimensional reconstruction difficult after the point cloud data is diluted, hindering the automation of 5G digital twin three-dimensional scenes.

[0006] In recent years, the use of digital twins in 5G private networks has increased significantly. Simulations within private network campuses require high-precision 3D models. High-precision 3D device models are used for algorithm simulation, monitoring, and simulation within 5G private networks. Currently, the automated construction of detailed models for digital twin 3D scenes typically uses laser point cloud machines to scan and acquire point cloud data. However, point cloud data also suffers from large data volumes, difficult transmission, and high rendering resource requirements. Therefore, balancing the high-precision 3D model characteristics of 5G private network devices with current rendering computing power remains a pressing issue. Summary of the Invention

[0007] To this end, the technical problem to be solved by the present invention is to overcome the problem in the prior art that it is impossible to balance the high-precision three-dimensional model features of private network equipment and the current rendering computing power.

[0008] To solve the above technical problems, the present invention provides a method for visualizing a digital twin of a private network campus, comprising:

[0009] Use clustering algorithms to classify objects in remote sensing images within the park, perform grid division based on the formed pixel clusters, and calculate their dilution ratios;

[0010] Map the raster data and point cloud data into a unified coordinate system, and use a clustering algorithm to cluster the point cloud data within each raster;

[0011] Each point cloud cluster is sampled according to the dilution ratio of the corresponding grid, and the non-edge points are found through the BoundED algorithm;

[0012] Calculate the initial dilution threshold of each point cloud cluster according to the proportion of non-edge points in each point cloud cluster;

[0013] After each point cloud cluster is diluted, it is rendered through a three-dimensional rendering engine and the number of rendering frames is returned. If the number of rendering frames meets the preset conditions, the process ends.

[0014] Preferably, the method of using a clustering algorithm to classify the objects in the remote sensing image within the park, performing grid division based on the formed pixel clusters and calculating the dilution ratio thereof includes:

[0015] Randomly select multiple pixels from the remote sensing image as initial cluster centers, and calculate the color Euclidean distance between the remaining pixels and the initial cluster centers;

[0016] Assign each pixel to the cluster with the closest initial cluster center in Euclidean distance until the formed pixel cluster no longer changes or the maximum number of iterations is reached, and the classification is completed;

[0017] Based on the classification results, the region growing method is used to divide the region

[0018] Calculate the minimum and maximum coordinate pixel points of each area, construct a rectangular area bounding box, divide each area evenly into multiple grids, and calculate the image information entropy of each grid;

[0019] The dilution ratio of each region is calculated according to the image information entropy of each grid.

[0020] Preferably, mapping the grid data and the point cloud data into a unified coordinate system and clustering the point cloud data in each grid using a clustering algorithm comprises:

[0021] The raster data and point cloud data are mapped into a unified coordinate system through coordinate transformation, and the point cloud data projected on the corresponding classification grid of the remote sensing image is classified into the grid by combining the classification grid results of the remote sensing image and the three-dimensional coordinates of the point cloud data;

[0022] Traverse each grid and randomly select multiple points as the initial centroid;

[0023] Calculate the three-dimensional distance between each point in each grid and each centroid in the grid, and divide each point into the point cloud cluster to which the centroid with the closest three-dimensional distance belongs;

[0024] Update the centroid coordinates according to the formed point cloud cluster, and then update the point cloud cluster according to the updated centroid. Repeat this step until the centroid coordinates no longer change.

[0025] Preferably, sampling each point cloud cluster according to the dilution ratio of the corresponding grid and finding non-edge points by using the BoundED algorithm includes:

[0026] Each point cloud cluster is sampled according to the dilution ratio of the corresponding grid to obtain a point cloud set;

[0027] Set multiple different scale values ​​k and find the nearest k points to form a neighborhood for each point in the point cloud set;

[0028] Extract the multidimensional feature vector of each point at scale value k according to the neighborhood corresponding to each point at scale value k;

[0029] Traverse each point, remove the points contained in the neighborhood corresponding to scale value k from the neighborhood of the maximum scale value, and calculate the distance from the centroid of the neighborhood corresponding to scale value k to the centroid of the neighborhood of the maximum scale value after the removal, and decompose it into the first vertical component and the first tangential component;

[0030] According to the multidimensional feature vector, the first vertical component and the first tangential component corresponding to the scale value k, the multi-scale feature of each point is obtained;

[0031] The features of different scales of each point are spliced ​​in pairs, passed through the fully connected layer and input into the multi-layer perceptron to obtain the probability that the point is a non-edge point, a sharp edge point or a boundary point.

[0032] Preferably, extracting the multidimensional feature vector of each point at the scale value k according to the neighborhood corresponding to each point at the scale value k includes:

[0033] Calculate the covariance matrix of the neighborhood corresponding to each point under the scale value k, and perform singular value decomposition on the covariance matrix to obtain singular values;

[0034] Obtaining a least squares fitting plane according to the eigenvectors and neighborhood centroids corresponding to the singular values, and dividing the neighborhood corresponding to each point at the scale value k into two non-intersecting subsets according to the least squares fitting plane;

[0035] Performing singular value decomposition on the two disjoint subsets, and calculating the distance between the centroids of the two subsets, decomposing the distance into a second vertical component and a second tangential component;

[0036] Calculate the distance between each point and the corresponding neighborhood centroid at scale value k, decomposing it into the third vertical component and the third tangential component;

[0037] The multidimensional feature vector of each point is obtained according to the singular values ​​obtained by decomposing the two subsets after the neighborhood of each point is divided under the scale value k, as well as the second vertical component, second tangential component, third vertical component and third tangential component corresponding to each point.

[0038] Preferably, diluting each point cloud cluster includes:

[0039] The non-edge points in the point cloud cluster whose non-edge point probability is greater than the initial dilution threshold are discarded.

[0040] Preferably, if the number of rendering frames does not meet a preset condition, the thinning threshold of each point cloud cluster is updated by linear interpolation, and each point cloud cluster is thinned and rendered again.

[0041] The present invention also provides a private network park digital twin visualization device, comprising:

[0042] The object classification module is used to classify objects in remote sensing images within the park using a clustering algorithm, perform grid division based on the formed pixel clusters, and calculate their dilution ratios;

[0043] Point cloud data clustering module, used to map raster data and point cloud data into a unified coordinate system, and cluster the point cloud data within each grid using a clustering algorithm;

[0044] The non-edge point acquisition module is used to sample each point cloud cluster according to the dilution ratio of the corresponding grid and find non-edge points through the BoundED algorithm;

[0045] A dilution threshold calculation module is used to calculate the initial dilution threshold of each point cloud cluster according to the proportion of non-edge points in each point cloud cluster;

[0046] The rendering and visualization module is used to dilute each point cloud cluster, render it through a three-dimensional rendering engine, and return the number of rendering frames. If the number of rendering frames meets the preset conditions, the process ends.

[0047] The present invention also provides a private network park digital twin visualization device, including:

[0048] memory for storing computer programs;

[0049] A processor is used to implement the above-mentioned steps of a private network park digital twin visualization method when executing the computer program.

[0050] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for visualizing a digital twin of a private network park are implemented.

[0051] The above technical solution of the present invention has the following advantages over the prior art:

[0052] The digital twin visualization method for a private network campus described in the present invention extracts features from private network campus entities, thins out point cloud three-dimensional data based on the BoundED algorithm, and adjusts the dilution threshold of the algorithm based on feedback from the server rendering capability, so that it can be dynamically adapted according to the client's capabilities, ensuring the client's access performance and efficiency, thereby ensuring the stability of the overall service. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0054] FIG1 is a flowchart of a method for visualizing a digital twin of a private network park provided by the present invention;

[0055] Figure 2 is a schematic diagram of point cloud cluster update;

[0056] Figure 3 is a schematic diagram of the BoundED model architecture;

[0057] Figure 4 is a schematic diagram of the classification results of the BoundED model;

[0058] Figure 5 is an overall implementation flow chart of a private network campus digital twin visualization method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The core of the present invention is to provide a private network campus digital twin visualization method, device, equipment and computer storage medium, which effectively adapts to the current rendering computing power while ensuring rendering performance.

[0060] In order to enable those skilled in the art to better understand the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0061] Please refer to Figure 1, which is a flowchart of a method for visualizing a private network campus digital twin provided by the present invention. The specific steps are as follows:

[0062] S101: using a clustering algorithm to classify the objects in the remote sensing image of the park, performing grid division based on the formed pixel clusters and calculating their dilution ratios;

[0063] S102: Mapping the grid data and the point cloud data into a unified coordinate system, and clustering the point cloud data in each grid using a clustering algorithm;

[0064] S103: Sampling each point cloud cluster according to the dilution ratio of the corresponding grid, and finding non-edge points using the BoundED algorithm;

[0065] S104: Calculating an initial dilution threshold for each point cloud cluster based on the proportion of non-edge points in each point cloud cluster;

[0066] S105: After each point cloud cluster is diluted, it is rendered and presented through a three-dimensional rendering engine, and the number of rendering frames is returned. If the number of rendering frames meets the preset conditions, the process ends.

[0067] Based on the above embodiment, this embodiment describes step S101 in detail:

[0068] The method of using a clustering algorithm to classify the objects in the remote sensing image of the park, performing grid division based on the formed pixel clusters and calculating the dilution ratio thereof includes:

[0069] Randomly select multiple pixels from the remote sensing image as initial cluster centers, and calculate the color Euclidean distance between the remaining pixels and the initial cluster centers;

[0070] 5G private network park subdivisions can generally be summarized into the following categories: smart factory, smart campus, smart port, smart park, smart agriculture, etc. The reference values ​​of parameters such as park element type K are shown in the following table:

[0071] This paper uses a smart park as an example. The landforms of a smart park generally include the geographical environment (green space, rivers and lakes), buildings, roads, and park equipment (visible by satellites or aerial photography). First, based on the number of landform types in the park, the upper limit of the number of typical divisible areas within the park is determined to be K, and the upper limit of the park landform area is no more than J.

[0072] Randomly select K pixels from the remote sensing image I as the initial cluster center C i (1≤i≤K), calculate the color Euclidean distance between the remaining data pixels X and the cluster center Ci, as shown in the following formula:

[0073] Where d is the Euclidean distance and m is the number of image channels.

[0074] Assign each pixel to the cluster with the closest initial cluster center in Euclidean distance until the formed pixel cluster no longer changes or the maximum number of iterations is reached, and the classification is completed;

[0075] Find the cluster center C closest to each X i , and assign the data objects to the cluster center C i Then calculate the color average of the pixels in each cluster as the new cluster center and perform the next iteration until the cluster center no longer changes or the maximum number of iterations is reached. The final remote sensing image can be clustered into I g {G1,G2,G3,...,G K}, where G is the pixel cluster of each category.

[0076] Based on the classification results, the region growing method is used to perform region division;

[0077] Based on the clustering results I g , region growing method is used for region segmentation.

[0078] 1) First, scan the image column by column (usually, scan the image starting from column 0) and find the first pixel that has not been visited. Let this pixel be I(x0, y0);

[0079] 2) Taking I(x0,y0) as the center, select its 8 neighboring pixels I(x,y n ), if the neighboring pixel and I(x0,y0) belong to the same category (the current common 8-neighborhood algorithm can be used to determine whether they are of the same category, which will not be explained here), I(x n ,y n ) and I(x0,y0) into the same area, and I(x n ,y n ) is pushed onto the stack;

[0080] 3) Randomly pick a pixel from the stack, treat it as I(x0, y0), and return it to step 2);

[0081] 4) When the stack is empty, complete the segmentation of an independent region.

[0082] 5) Repeat steps 1)-4) until all pixels in the image are traversed.

[0083] Finally, the image can be segmented into I p {P1, P2, P3,..., P L}

[0084] Perform Mahalanobis distance calculation and merge the small regions (area less than S) after the above segmentation:

[0085] Set the minimum area of the region as S. If P l < S, then calculate the Mahalanobis distance of its neighborhood region P n as shown in the following formula:

[0086] where l is the segmentation region number, 1 ≤ l ≤ L

[0087] where P n is resampled by interpolation to ensure the same dimension as P l , and Σ -1 is the covariance calculation operator. Merge P l into the category with the smallest Mahalanobis distance. According to the above algorithm, perform regional segmentation on the 5G park ground objects, and the final segmentation result is: I P {P1, P2, P3,..., P T}, where T ≤ L.

[0088] Calculate the minimum coordinate pixel point and the maximum coordinate pixel point of each region, construct the rectangular region bounding box, evenly divide each region into multiple grids, and calculate the image information entropy of each grid;

[0089] According to the above segmentation result, calculate the pixel point (x min , y min ) of the minimum coordinate and the pixel point (x max , y max ) of the maximum coordinate of each region, construct the rectangular region bounding box with this, evenly divide the region into N 2 grids, and discard the grids that do not contain clustering data. Calculate the image information entropy of each grid:

[0090] where p i is the frequency of occurrence of the i-th gray pixel.

[0091] Calculate the dilution ratio of each region according to the image information entropy of each grid.

[0092] The larger the information entropy, the richer the information contained in the image, and the lower the sparse sampling multiple. The relationship between it and the sparse sampling multiple Th can be expressed as: Th = floor(aH + Th max )

[0093] where Th max is the maximum sparse sampling multiple, Th, Th max ≥ 0 and is an integer; a is the attenuation coefficient (-1 ≤ a ≤ 0), obtained by manual parameter adjustment; floor() means rounding down the value in the parentheses.

[0094] Based on the above embodiments, this embodiment elaborates on step S102 in detail:

[0095] The mapping of the grid data and the point cloud data into a unified coordinate system and the use of the clustering algorithm to cluster the point cloud data within each grid include:

[0096] Map the grid data and the point cloud data into a unified coordinate system through coordinate transformation, and combine the classification grid results of the remote sensing image and the three-dimensional coordinates of the point cloud data to assign the point cloud data projected on the corresponding classification grid of the remote sensing image to that grid;

[0097] Use a laser point cloud device to scan the park and obtain the point cloud data set P{p1, p2, p3,..., p t}, map the point cloud data and the sub-grid data of the remote sensing image into a unified coordinate system through coordinate transformation, and combine the above classification grid results of the remote sensing image and the three-dimensional coordinates of the point cloud to assign the point cloud data projected on the corresponding classification grid of the remote sensing image to that grid.

[0098] Traverse each grid and randomly select multiple points as the initial centroids;

[0099] For the point cloud data cluster set P, traverse by grid and then cluster. The formula for the number of initial centroids is as follows: F = bN / Area

[0100] F is the number of initial centroids, b is the centroid reduction coefficient (0 < b < 1), N is the number of point clouds, and Area is the area of this grid.

[0101] Calculate the three-dimensional space distance from each point in each grid to each centroid in that grid, and divide each point into the point cloud cluster to which the centroid with the closest three-dimensional space distance belongs;

[0102] Calculate the three-dimensional space distance from each point p i in the grid to each centroid, find the centroid m i closest to this point, and mark this point pi Belong to m i Point cloud clusters.

[0103] Update the centroid coordinates according to the formed point cloud cluster, and then update the point cloud cluster according to the updated centroid. Repeat this step until the centroid coordinates no longer change.

[0104] For all point cloud data of each point cloud cluster, its single point cloud data can be expressed as, p i ={x i ,y i ,z i}Use the Euclidean distance formula to calculate the new centroid coordinates of each cluster.

[0105] Calculate the distance from each point to the new centroid m i ', and then update the point cloud cluster to which each point belongs according to the minimum distance principle, as shown in Figure 2.

[0106] Repeat the above until the centroids of all point cloud clusters no longer change, and generate the point cloud cluster set of the grid: Y{C1,C2,C3,...,C F Finally, the point cloud cluster set Y of all grids is output all {Y1,Y2,Y2,...,Y h}.

[0107] Based on the above embodiment, this embodiment describes step S103 in detail:

[0108] As shown in FIG3 and FIG4, the sampling of each point cloud cluster according to the dilution ratio of the corresponding grid and finding the non-edge points by the BoundED algorithm include:

[0109] Each point cloud cluster is sampled according to the dilution ratio of the corresponding grid to obtain a point cloud set;

[0110] According to the output point cloud cluster Y all And the sparse sampling multiple Th of the corresponding grid, for Y all Each point cloud cluster C in i , sample the point cloud data according to the sparse sampling multiple (for example, the sparse sampling multiple is 4, and one point cloud data is sampled every 3 points from the minimum x value to the maximum x value according to its coordinates) to form a new point cloud set

[0111] Set multiple different scale values ​​k and find the nearest k points to form a neighborhood for each point in the point cloud set;

[0112] Based on the k-nearest neighbor algorithm, find the distance from p i The nearest k points form the neighborhood N i,k；

[0113] Extract the multidimensional feature vector of each point at scale value k according to the neighborhood corresponding to each point at scale value k:

[0114] Calculate the covariance matrix of the neighborhood corresponding to each point under the scale value k, and perform singular value decomposition on the covariance matrix to obtain singular values;

[0115] Calculate the covariance matrix K i =cov(N i,k ), for K i Perform SVD (singular value decomposition) to obtain the singular value σ i,1 ≥σ i,2 ≥σ i,3 .

[0116] Obtaining a least squares fitting plane according to the eigenvectors and neighborhood centroids corresponding to the singular values, and dividing the neighborhood corresponding to each point at the scale value k into two non-intersecting subsets according to the least squares fitting plane;

[0117] Set N i,k The points in are normalized with the origin as the center according to the following formula:

[0118] Among them, the neighborhood centroid

[0119] With the singular value σ i,3 The corresponding eigenvector n i and the centroid of the neighborhood A least squares fitting plane π is formed i,k , the plane will be the neighbor set N i,k Split into two disjoint subsets N i,k,upper and N i,k,lower .

[0120] Performing singular value decomposition on the two disjoint subsets, and calculating the distance between the centroids of the two subsets, decomposing the distance into a second vertical component and a second tangential component;

[0121] Extract statistical information of two subsets respectively: i,k,upper and N i,k,lower The covariance matrix of the singular value decomposition is performed to obtain the singular value σ i,k,upper =(σi,k,upper,1,σi,k,upper,2,σi,k,upper,3) T and σ i,k,lower =(σi,k,lower,1,σi,k,lower,2,σi,k,lower,3) T , find the distance between the centroids of the two subsets and decompose it into the vertical component and the tangential component d i,k =(di,k,⊥ ,d i,k,∥ ) T .

[0122] Calculate the distance between each point and the corresponding neighborhood centroid at scale value k, decomposing it into the third vertical component and the third tangential component;

[0123] Ask for point p i and the centroid of the neighborhood The distance between them is decomposed into vertical and tangential components s i,k =(s i,k,⊥ ,s i,k,∥ ) T .

[0124] The multidimensional feature vector of each point is obtained according to the singular values ​​obtained by decomposing the two subsets after the neighborhood of each point is divided under the scale value k, as well as the second vertical component, second tangential component, third vertical component and third tangential component corresponding to each point.

[0125] Finally, for every point p i , a ten-dimensional feature vector can be generated:

[0126] Traverse each point, remove the points contained in the neighborhood corresponding to scale value k from the neighborhood of the maximum scale value, and calculate the distance from the centroid of the neighborhood corresponding to scale value k to the centroid of the neighborhood of the maximum scale value after the removal, and decompose it into the first vertical component and the first tangential component;

[0127] The largest scale neighborhood is denoted as N i,k0 , from N i,k0 Eliminate N i,k The point cloud set after the midpoint is recorded as <N i,k0 -N i,k >, find the neighborhood N i,k Centroid to <N i,k0 -N i,k >The distance to the center of mass, decomposed into vertical and tangential components c i,k =(c i,k,⊥ ,c i,k,∥ ) T , then the point p at scale k i Features are:

[0128] According to the multidimensional feature vector, the first vertical component and the first tangential component corresponding to the scale value k, the multi-scale feature of each point is obtained;

[0129] After feature extraction at different scales (the scale here refers to the k value), for point p i The following complete multi-scale features are formed:

[0130] The features of different scales of each point are spliced ​​in pairs, passed through the fully connected layer and input into the multi-layer perceptron to obtain the probability that the point is a non-edge point, a sharp edge point or a boundary point.

[0131] The features of different scales are concatenated in pairs and passed through a fully connected layer respectively, and then input into the MLP to output the probability that the point is a non-edge point, a sharp edge point or a boundary point [p ne ,p se ,p b ]. Compare the sizes of the three probabilities. If p se The maximum, then point p i It is judged as a sharp edge point; if p b The maximum, then point p i is judged as a boundary point; and for p ne The largest point, p i It is judged as a non-edge point, and the subsequent dilution threshold is used to determine whether the point needs to be discarded.

[0132] Based on the above embodiment, this embodiment describes step S104 in detail:

[0133] The initial dilution threshold for each point cloud cluster is set according to the following formula:

[0134] Where: t j represents the initial dilution threshold of the j-th point cloud cluster, n ne Indicates the number of points classified as non-edge points in the j-th point cloud cluster, n j Represents the total number of point clouds in the jth point cloud cluster.

[0135] Based on the above embodiment, this embodiment describes step S105 in detail:

[0136] The diluting of each point cloud cluster includes:

[0137] Discard non-edge points in the point cloud cluster whose probability of non-edge points is greater than the initial dilution threshold:

[0138] Initial dilution threshold set: [t1,t2,...,t j ,...], for the non-edge points in the j-th point cloud cluster, when p ne >t j , the point cloud will be discarded, forming a point cloud set based on the dilution of the original point cloud cluster.

[0139] If the number of rendered frames does not meet the preset conditions, the thinning threshold of each point cloud cluster is updated by linear interpolation, and each point cloud cluster is thinned and rendered again:

[0140] The diluted point cloud cluster set is obtained through the above steps, and the point cloud data is rendered and presented in 3D through a 3D rendering engine. The present invention sets the 3D scene rendering frame number requirements for the 5G private network park as follows:

[0141] By using the OpenGL graphics API interface of the target computer, the point cloud data is rendered and the display frame number of the three-dimensional scene is obtained.

[0142] The obtained number of rendered frames is returned to the dilution algorithm. If the number of rendered frames meets the requirement of smooth browsing of the scene, the point cloud data is diluted according to the current dilution threshold. If the number of rendered frames does not meet the requirement of smooth browsing of the scene, the dilution threshold of each point cloud cluster is reset through linear interpolation to replace the initial dilution threshold. The calculation formula of the dilution threshold is as follows:

[0143] Where: t j represents the new dilution threshold of the j-th point cloud cluster, t j represents the initial thinning threshold of the j-th point cloud cluster, and x represents the current rendering frame number.

[0144] After calculating the new dilution threshold, the point cloud data is diluted and rendered again, and the number of rendered frames is re-judged to see whether it meets the requirements. If not, the steps are repeated until the number of rendered frames is in the range [30,60].

[0145] If the number of rendered frames fed back after dilution according to the initial dilution threshold is greater than 60 frames in the rendering process, the dilution threshold t j According to the formula, if the number of rendered frames is still > 60, then adjust it according to the step size Ω a Increase or decrease the centroid coefficient b value. If the rendering frame number is still > 60 frames, then increase the value according to the step size Ω. a Lower the attenuation coefficient a, recalculate the sparse sampling multiple Th, and continue the algorithm process.

[0146] If the number of rendered frames fed back after dilution according to the initial dilution threshold is less than 30 frames in the rendering process, the dilution threshold will be lowered according to the formula. If the number of rendered frames is still less than 30 frames, the number of rendered frames will be increased according to the step size Ω. a Lower the centroid coefficient b value. If the rendering frame number is still less than 30 frames, then increase it according to the step size Ω. a Increase the value of the attenuation coefficient a, recalculate the sparse sampling multiple Th, and continue the algorithm process.

[0147] For example, suppose the initial threshold for a point cloud cluster is set to 0.4 based on the calculation in stage 3. After diluting the point cloud cluster data, the point cloud is rendered 3D using the OpenGL API, and the real-time rendering frame rate is 20. The updated threshold is calculated to be 0.3 based on the dilution threshold calculation formula. This means that more point cloud data is diluted, and the real-time rendering shows that the 3D rendering frame rate for this point cloud cluster will increase. The final diluted point cloud cluster data is output.

[0148] As shown in Figure 5, an embodiment of the present invention provides an overall implementation flow chart of a private network park digital twin visualization method; the present invention protects the process of rapid dilution and rendering of the entire 5G park digital twin. First, the scene is segmented for the 5G private network park, and then the 5G park remote sensing image is extracted through a classification algorithm. The extracted elements are then grid-divided according to an adaptive grid division algorithm. Next, the image complexity of the grid is marked using an information entropy algorithm, and a data basis is provided for the subsequent BoundED algorithm to thin out the point cloud clusters, accelerate the processing of massive point cloud data, improve the speed of the dilution algorithm, and perform fine classification and thinning of different object shapes and three-dimensional feature types to ensure the accuracy of subsequent point cloud data rendering after three-dimensional dilution. This invention introduces remote sensing images to measure regional information of 5G private network parks, thereby guiding the optimization of multi-sampling rates of park point cloud data, and rendering the diluted point cloud data. The frame number is obtained through OpenGL, and combined with the basic standards of the three-dimensional digital twin platform, feedback is given to the dilution algorithm. In this way, the feasibility of point cloud data dilution is evaluated according to the rendering capability standard while ensuring the characteristics of the point cloud data. This proposes an engineering application solution for the dilution, transmission, and rendering of the 5G private network park digital twin visualization three-dimensional data model.

[0149] The existing scheme of dilution of three-dimensional scene data refers to the static sampling algorithm to dilute the data; the present invention introduces remote sensing images to measure the regional information of the 5G private network park, thereby guiding the optimization of the multi-sampling rate of the park point cloud data, and uses k-means for clustering. Then, the point cloud clusters are classified based on edge features by using the BoundED algorithm, and the initial threshold is set in layers according to the classification results. Subsequently, the number of three-dimensional scene rendering frames of the diluted point cloud data is obtained through the WebGL API. By adjusting the dilution threshold or the sparse sampling multiple, the appropriate threshold is fed back to the algorithm, and the point cloud data is adaptively diluted, thereby improving the problem of point cloud data feature value loss caused by the feature value of the point cloud data being diluted by the general thinning algorithm.

[0150] The embodiment of the present invention further provides a private network campus digital twin visualization device; the specific device may include:

[0151] The object classification module is used to classify objects in remote sensing images within the park using a clustering algorithm, perform grid division based on the formed pixel clusters, and calculate their dilution ratios;

[0152] Point cloud data clustering module, used to map raster data and point cloud data into a unified coordinate system, and cluster the point cloud data within each grid using a clustering algorithm;

[0153] The non-edge point acquisition module is used to sample each point cloud cluster according to the dilution ratio of the corresponding grid and find non-edge points through the BoundED algorithm;

[0154] A dilution threshold calculation module is used to calculate the initial dilution threshold of each point cloud cluster according to the proportion of non-edge points in each point cloud cluster;

[0155] The rendering and visualization module is used to dilute each point cloud cluster, render it through a three-dimensional rendering engine, and return the number of rendering frames. If the number of rendering frames meets the preset conditions, the process ends.

[0156] The private network campus digital twin visualization device of this embodiment is used to implement the aforementioned private network campus digital twin visualization method. Therefore, the specific implementation methods of the private network campus digital twin visualization device can be seen in the embodiment part of the private network campus digital twin visualization method mentioned above. For example, the ground feature classification module, the point cloud data clustering module, the non-edge point acquisition module, the dilution threshold calculation module, and the rendering visualization module are respectively used to implement steps S101, S102, S103, S104 and S105 in the above-mentioned private network campus digital twin visualization method. Therefore, its specific implementation methods can refer to the descriptions of the corresponding embodiments of each part and will not be repeated here.

[0157] A specific embodiment of the present invention also provides a private network campus digital twin visualization device, including: a memory for storing a computer program; and a processor for implementing the steps of the above-mentioned private network campus digital twin visualization method when executing the computer program.

[0158] A specific embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned private network campus digital twin visualization method are implemented.

[0159] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0160] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0161] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0163] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A digital twin visualization method for a private network campus, comprising: Use clustering algorithms to classify objects in remote sensing images within the park, perform grid division based on the formed pixel clusters, and calculate their dilution ratios; Map the raster data and point cloud data into a unified coordinate system, and use a clustering algorithm to cluster the point cloud data within each raster; Each point cloud cluster is sampled according to the dilution ratio of the corresponding grid, and the non-edge points are found through the BoundED algorithm; Calculate the initial dilution threshold of each point cloud cluster according to the proportion of non-edge points in each point cloud cluster; After each point cloud cluster is diluted, it is rendered through a three-dimensional rendering engine and the number of rendering frames is returned. If the number of rendering frames meets the preset conditions, the process ends.

2. The private network park digital twin visualization method according to claim 1, wherein: The method of using a clustering algorithm to classify the objects in the remote sensing image of the park, performing grid division based on the formed pixel clusters and calculating the dilution ratio thereof includes: Randomly select multiple pixels from the remote sensing image as initial cluster centers, and calculate the color Euclidean distance between the remaining pixels and the initial cluster centers; Assign each pixel to the cluster with the closest initial cluster center in Euclidean distance until the formed pixel cluster no longer changes or the maximum number of iterations is reached, and the classification is completed; Based on the classification results, the region growing method is used to perform region division; Calculate the minimum and maximum coordinate pixel points of each area, construct a rectangular area bounding box, divide each area evenly into multiple grids, and calculate the image information entropy of each grid; The dilution ratio of each region is calculated according to the image information entropy of each grid.

3. The private network park digital twin visualization method according to claim 1, wherein: Mapping the grid data and the point cloud data into a unified coordinate system and clustering the point cloud data in each grid using a clustering algorithm includes: The raster data and point cloud data are mapped into a unified coordinate system through coordinate transformation, and the point cloud data projected on the corresponding classification grid of the remote sensing image is classified into the grid by combining the classification grid results of the remote sensing image and the three-dimensional coordinates of the point cloud data; Traverse each grid and randomly select multiple points as the initial centroid; Calculate the three-dimensional distance between each point in each grid and each centroid in the grid, and divide each point into the point cloud cluster to which the centroid with the closest three-dimensional distance belongs; Update the centroid coordinates according to the formed point cloud cluster, and then update the point cloud cluster according to the updated centroid. Repeat this step until the centroid coordinates no longer change.

4. The private network park digital twin visualization method according to claim 1, wherein: The method of sampling each point cloud cluster according to the dilution ratio of the corresponding grid and finding non-edge points through the BoundED algorithm includes: Each point cloud cluster is sampled according to the dilution ratio of the corresponding grid to obtain a point cloud set; Set multiple different scale values k and find the nearest k points to form a neighborhood for each point in the point cloud set; Extract the multidimensional feature vector of each point at scale value k according to the neighborhood corresponding to each point at scale value k; Traverse each point, remove the points contained in the neighborhood corresponding to scale value k from the neighborhood of the maximum scale value, and calculate the distance from the centroid of the neighborhood corresponding to scale value k to the centroid of the neighborhood of the maximum scale value after the removal, and decompose it into the first vertical component and the first tangential component; According to the multidimensional feature vector, the first vertical component and the first tangential component corresponding to the scale value k, the multi-scale feature of each point is obtained; The features of different scales of each point are spliced in pairs, passed through the fully connected layer and input into the multi-layer perceptron to obtain the probability that the point is a non-edge point, a sharp edge point or a boundary point.

5. The private network park digital twin visualization method according to claim 4, wherein: Extracting the multidimensional feature vector of each point at scale value k according to the neighborhood corresponding to each point at scale value k includes: Calculate the covariance matrix of the neighborhood corresponding to each point under the scale value k, and perform singular value decomposition on the covariance matrix to obtain singular values; Obtaining a least squares fitting plane according to the eigenvectors and neighborhood centroids corresponding to the singular values, and dividing the neighborhood corresponding to each point at the scale value k into two non-intersecting subsets according to the least squares fitting plane; Performing singular value decomposition on the two disjoint subsets, and calculating the distance between the centroids of the two subsets, decomposing the distance into a second vertical component and a second tangential component; Calculate the distance between each point and the corresponding neighborhood centroid at scale value k, decomposing it into the third vertical component and the third tangential component; The multidimensional feature vector of each point is obtained according to the singular values obtained by decomposing the two subsets after the neighborhood of each point is divided under the scale value k, as well as the second vertical component, second tangential component, third vertical component and third tangential component corresponding to each point.

6. The private network park digital twin visualization method according to claim 1, wherein: The diluting of each point cloud cluster includes: The non-edge points in the point cloud cluster whose non-edge point probability is greater than the initial dilution threshold are discarded.

7. The private network campus digital twin visualization method according to claim 1, wherein: If the number of rendering frames does not meet the preset conditions, the thinning threshold of each point cloud cluster is updated by linear interpolation, and each point cloud cluster is thinned and rendered again.

8. A digital twin visualization device for a private network campus, comprising: The object classification module is used to classify objects in remote sensing images within the park using a clustering algorithm, perform grid division based on the formed pixel clusters, and calculate their dilution ratios; Point cloud data clustering module, used to map raster data and point cloud data into a unified coordinate system, and cluster the point cloud data within each grid using a clustering algorithm; The non-edge point acquisition module is used to sample each point cloud cluster according to the dilution ratio of the corresponding grid and find non-edge points through the BoundED algorithm; A dilution threshold calculation module is used to calculate the initial dilution threshold of each point cloud cluster according to the proportion of non-edge points in each point cloud cluster; The rendering and visualization module is used to dilute each point cloud cluster, render it through a three-dimensional rendering engine, and return the number of rendering frames. If the number of rendering frames meets the preset conditions, the process ends.

9. A digital twin visualization device for a private network campus, comprising: memory for storing computer programs; A processor is used to implement the steps of a private network campus digital twin visualization method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of a method for visualizing a digital twin of a private network campus as described in any one of claims 1 to 7 are implemented.

Images