A method for reconstructing a three-dimensional electromagnetic environment of a city based on point cloud data

Abstract

The application discloses a kind of city three-dimensional electromagnetic environment reconstruction methods based on point cloud data, it is related to digital twin online channel modeling technical field. Including: 1) three-dimensional point cloud data is obtained using unmanned aerial vehicle oblique photography;2) the ground and non-ground separation is carried out to point cloud using cloth simulation filtering algorithm;3) non-ground point cloud semantic classification is carried out, and reconstruction is carried out to building, vegetation, vehicle and street lamp etc.;4) ground point cloud semantic clustering is carried out, and semantic digital elevation model is generated, and ground surface reconstruction is completed;5) three-dimensional semantic electromagnetic environment model is constructed by fusing ground and non-ground reconstruction results, and material electromagnetic parameter matching is carried out;6) wireless channel simulation is carried out based on ray tracing method, and channel characteristics are obtained.The application can realize local area high-precision reconstruction while keeping global environment consistency, improve the simulation accuracy and calculation efficiency of digital twin online channel modeling.

Description

Technical Field

[0001] This invention relates to the field of wireless communication channel modeling technology, and in particular to a method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data. Background Technology

[0002] With the widespread deployment of 5G and the rapid development of 6G, wireless communication networks are evolving towards intelligence, digitalization, and full-scenario awareness. The Digital Twin Online Channel Model (DTOCM) constructs a digital twin that maps in real-time to the actual communication environment, enabling dynamic prediction and online updates of wireless channel characteristics. It is a crucial foundation for supporting 6G intelligent network design, network optimization, and resource management.

[0003] In the construction of DTOCM, 3D electromagnetic environment reconstruction is a crucial step in achieving high-precision channel modeling. The geometric accuracy and semantic integrity of the reconstructed environment directly determine the accuracy of subsequent ray tracing simulation results. Currently, commonly used 3D environment construction methods mainly include manual modeling, electronic map reconstruction, and photogrammetric reconstruction. Manual modeling methods rely on specialized software for scene construction, resulting in long modeling cycles and high labor costs. While electronic map-based methods can quickly construct building models, they typically lack environmental information that significantly impacts wireless propagation, such as vegetation, roads, water bodies, and street facilities. Point cloud data acquired through UAV photogrammetry can relatively completely reflect the 3D structure of the real scene, but its massive scale and complex categories present significant challenges in efficiently extracting environmental information relevant to electromagnetic propagation and achieving refined reconstruction.

[0004] Most existing point cloud reconstruction methods are geared towards visual scene modeling, focusing primarily on geometric appearance restoration. However, they neglect semantic recognition of key scatterers during wireless channel propagation, electromagnetic parameter assignment, and the adaptability to ray tracing simulations. Furthermore, traditional point cloud classification methods are susceptible to factors such as uneven class distribution, terrain undulations, and the similarity of building and ground features, leading to decreased classification accuracy. Additionally, applying a uniform reconstruction strategy to different types of terrain features, such as buildings, vegetation, and ground, can easily result in model redundancy, high computational complexity, and low simulation efficiency. Existing technologies suffer from insufficient accuracy in point cloud electromagnetic environment reconstruction, low classification efficiency, and difficulty in adapting the reconstructed models to ray tracing simulations.

[0005] Therefore, there is an urgent need for a fine-scale reconstruction method for the three-dimensional electromagnetic environment for wireless channel modeling applications, which can fully utilize the spatial and semantic information of point cloud data to achieve high-precision identification and differentiated reconstruction of key scatterers in complex scenes, and provide high-precision and high-efficiency electromagnetic environment support for digital twin online channel models. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for reconstructing the three-dimensional electromagnetic environment of a city based on point cloud data, so as to realize the automatic identification, fine reconstruction and electromagnetic environment modeling of key scatterers in complex communication scenarios, and improve the environment construction accuracy and channel simulation accuracy of digital twin online channel models.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data, proposed according to the present invention, includes:

[0009] Step S1: Obtain 3D point cloud data of the target communication scene;

[0010] Step S2: Pre-classify the ground types that affect wireless channel propagation in the target area to obtain semantic labels for each pre-classified ground type;

[0011] Step S3: Use the cloth simulation filtering algorithm to separate the ground point cloud from the non-ground point cloud in the 3D point cloud data to obtain the ground point cloud dataset and the non-ground point cloud dataset.

[0012] Step S4: Extract multi-scale geometric features, color features, and relative elevation features from the non-ground point cloud dataset. Construct a joint feature vector based on the multi-scale geometric features, color features, and relative elevation features. Input the constructed joint feature vector into the random forest model for training. Perform semantic classification on the non-ground point cloud to obtain the non-ground point cloud classification result.

[0013] Step S5: For the ground point cloud dataset, construct ground point cloud feature vectors, and use a clustering algorithm to perform unsupervised clustering of the ground point cloud feature vectors to obtain the land cover attributes corresponding to the clusters; complete the semantic classification of the ground point clouds based on the land cover attributes corresponding to the clusters obtained by clustering, and finally output the ground point cloud classification results with semantic labels.

[0014] Step S6: Based on the non-ground point cloud classification results, the building point cloud in the non-ground point cloud classification results is reconstructed in three dimensions using the planar fitting reconstruction method, and the vegetation, vehicle and rod point cloud in the non-ground point cloud classification results is reconstructed in three dimensions using the voxelization reconstruction method, so as to obtain the reconstructed building model, vegetation model, vehicle model and rod model.

[0015] Step S7: Construct a semantic ground elevation model based on the ground point cloud classification results, and complete the three-dimensional reconstruction of the ground area to obtain the reconstructed semantic ground elevation model.

[0016] Step S8: Integrate the reconstructed building model, vegetation model, vehicle model, pole model, and semantic ground elevation model to construct a three-dimensional environment model;

[0017] Step S9: Assign corresponding electromagnetic parameters to the 3D environment model based on the semantic tags of the pre-classified ground types to obtain the 3D electromagnetic model, and use the ray tracing algorithm to perform wireless channel simulation on the 3D electromagnetic model to obtain the channel characteristic parameters of the target communication scenario.

[0018] As a further optimization scheme of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, in step S2, the various types of ground that are pre-classified include ground, buildings, vegetation, traffic lights, street lights and street debris.

[0019] The ground includes asphalt roads, brick and cement roads, soil, grass, and water bodies;

[0020] The buildings include high-rise buildings and single-story temporary buildings;

[0021] The vegetation includes shrubs and tall vegetation.

[0022] As a further optimization scheme of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, in step S4, the non-ground point cloud classification results include the classification results of buildings, vegetation, vehicles and rod-shaped objects.

[0023] In step S5, the ground point cloud classification results include the classification results for roads, grasslands, bare soil, and water bodies.

[0024] As a further optimization of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, in step S1, the three-dimensional point cloud data is obtained by acquiring multi-view images through an UAV oblique photography system, followed by aerial triangulation, image matching, point cloud generation, and coordinate registration.

[0025] As a further optimization scheme of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, in step S3, the cloth simulation filtering algorithm constructs an inverted point cloud surface and simulates the free fall process of virtual cloth, so that the cloth surface conforms to the terrain contour, and realizes the separation of ground point cloud and non-ground point cloud according to the distance between cloth nodes and the original point cloud.

[0026] As a further optimization scheme for the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, the multi-scale geometric features in step S4 include linearity, flatness, sphericity, anisotropy, rate of curvature change, and verticality features; the color features include RGB color features, HSV color features, and greening index features; the relative elevation features are obtained by calculating the elevation difference between point cloud points and local ground reference surfaces.

[0027] In step S4, the random forest classification model uses a class-balanced sampling strategy to construct a training dataset and utilizes multiple decision tree classifiers for ensemble learning to achieve semantic classification of non-terrestrial point clouds.

[0028] As a further optimization scheme of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, the ground feature vector in step S5 includes color features and vertical roughness features, and the clustering algorithm is the K-Means clustering algorithm.

[0029] As a further optimization scheme of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, the building point cloud in step S6 is reconstructed using the PolyFit algorithm, including plane extraction, candidate polygon generation and global optimization steps, to generate a three-dimensional surface model of the building.

[0030] As a further optimization of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, the voxelization reconstruction method in step S6 includes: converting the point cloud into a voxel mesh; filling the voids in the voxel mesh; extracting the three-dimensional surface using the Marching Cubes algorithm to generate a triangular patch mesh model; and simplifying the triangular patch mesh model using the quadratic error metric QEM algorithm.

[0031] As a further optimization of the urban three-dimensional electromagnetic environment reconstruction method based on point cloud data described in this invention, the step S7 of constructing a semantic ground elevation model includes: projecting the ground point cloud onto a regular grid; generating corresponding grid cells according to semantic labels; filling in missing areas using a hybrid interpolation method; removing isolated noise areas using a connected component filtering method; and generating a continuous closed three-dimensional ground model.

[0032] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0033] (1) The present invention uses the CSF algorithm to achieve pre-separation of ground point cloud and non-ground point cloud, which effectively reduces the problem of class imbalance and confusion between ground and buildings, and improves the accuracy and computational efficiency of subsequent semantic segmentation;

[0034] (2) The present invention constructs a joint feature system that integrates multi-scale geometric features, color features and relative elevation features to achieve high-precision identification of key scatterers such as buildings, vegetation, vehicles and poles in complex communication scenarios;

[0035] (3) The present invention designs a differentiated reconstruction strategy for different types of land features. PolyFit is used to reconstruct buildings, voxelization is used to reconstruct irregular scattering bodies such as vegetation, and semantic digital elevation model is used to reconstruct the ground. This significantly reduces the complexity of the model while ensuring geometric accuracy.

[0036] (4) The three-dimensional electromagnetic environment model constructed by this invention contains rich semantic information and electromagnetic attribute information, which can accurately characterize the propagation environment in real communication scenarios and improve the accuracy of ray tracing channel simulation.

[0037] (5) This invention can be used as a local fine environment update module in the online channel model of digital twins, and realizes high-precision reconstruction of key areas on the basis of large-scale scene reconstruction, providing efficient and accurate electromagnetic environment support for 6G digital twin networks. Attached Figure Description

[0038] Figure 1 This is a block diagram of the urban three-dimensional electromagnetic environment fine reconstruction method based on point cloud data provided in Embodiment 1 of the present invention;

[0039] Figure 2a This is a diagram showing the non-ground point cloud separation result in Embodiment 1 of the present invention;

[0040] Figure 2b This is a ground point cloud classification result diagram from Embodiment 1 of the present invention;

[0041] Figure 3 This is a schematic diagram of the non-ground point cloud classification results in Embodiment 1 of the present invention;

[0042] Figure 4 This is a schematic diagram of the ground point cloud classification results in Embodiment 1 of the present invention;

[0043] Figure 5 This is a diagram showing the point cloud reconstruction result of a building in Embodiment 1 of the present invention;

[0044] Figure 6 This is a schematic diagram of the irregular target reconstruction result in Embodiment 1 of the present invention; wherein, (a) is vegetation point cloud, (b) is vegetation voxelization, (c) is vegetation reconstruction result, (d) is vehicle point cloud, (e) is vehicle voxelization, and (f) is vehicle reconstruction structure.

[0045] Figure 7 This is a schematic diagram of the ground three-dimensional reconstruction result in Embodiment 1 of the present invention;

[0046] Figure 8 This is a schematic diagram of the complete three-dimensional reconstruction result in Embodiment 1 of the present invention;

[0047] Figure 9a This is a schematic diagram comparing the path loss results of the line-of-sight path in Embodiment 1 of the present invention;

[0048] Figure 9b This is a schematic diagram comparing the path loss results of non-line-of-sight path 1 in Embodiment 1 of the present invention;

[0049] Figure 9c This is a schematic diagram comparing the path loss results of non-line-of-sight path 2 in Embodiment 1 of the present invention;

[0050] Figure 10a This is a schematic diagram comparing the time delay spread results of the line-of-sight path in Embodiment 1 of the present invention;

[0051] Figure 10b This is a schematic diagram comparing the time delay spread results of non-line-of-sight path 1 in Embodiment 1 of the present invention;

[0052] Figure 10c This is a schematic diagram comparing the time delay spread results of non-line-of-sight path 2 in Embodiment 1 of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] It can be used for communication scenario construction, offline channel map generation, and ray tracing channel simulation in Digital Twin Online Channel Model (DTOCM).

[0055] Example 1:

[0056] A three-dimensional electromagnetic environment reconstruction system for wireless channels based on multi-source data, comprising:

[0057] The point cloud acquisition module is used to acquire 3D point cloud data of the target communication scene;

[0058] The point cloud separation module is used to separate ground point clouds from non-ground point clouds;

[0059] The semantic classification module is used to implement semantic classification of ground point clouds and non-ground point clouds;

[0060] The 3D reconstruction module is used to complete the 3D reconstruction of ground features based on the classification results;

[0061] The model fusion module is used to construct a three-dimensional electromagnetic environment model;

[0062] The channel simulation module is used for assigning electromagnetic parameters and performing ray tracing channel simulation.

[0063] See Figure 1 This embodiment provides a method for fine reconstruction of urban three-dimensional electromagnetic environment based on point cloud data, including the following steps:

[0064] Step S1: Acquisition of point cloud data for the target area. First, acquire the 3D point cloud data of the target communication scene.

[0065] In this embodiment, an unmanned aerial vehicle (UAV) oblique photography system is used for data acquisition. Multiple high-resolution area array digital cameras are mounted on the same flight platform to simultaneously acquire image data of the target area from the vertical direction and multiple oblique directions. Subsequently, through aerial triangulation, multi-view stereo matching, point cloud generation, and coordinate registration processing, three-dimensional point cloud data of the target area is obtained.

[0066] The acquired point cloud data not only contains spatial coordinate information, but also corresponding color information, providing a data foundation for subsequent semantic classification.

[0067] Step S2: Classification of ground environment information.

[0068] The ground types affecting wireless channel propagation in the target area are classified. In this example, the pre-classified ground types include ground, buildings, vegetation, traffic lights, street lights, and street debris; ground includes asphalt roads, masonry and cement roads, soil, grassland, and water bodies; buildings include high-rise buildings and single-story temporary buildings; vegetation includes shrubs and tall vegetation.

[0069] Step S3: Separate the ground point cloud from the non-ground point cloud.

[0070] Since ground point clouds typically account for more than 50% of the total point cloud data in urban environments, direct overall classification can easily cause the classifier to shift towards the majority category, thereby reducing the classification accuracy of targets such as buildings, vehicles, and poles. Therefore, this embodiment uses the Cloth Simulation Filter (CSF) algorithm to preprocess the original point cloud.

[0071] The specific steps are as follows:

[0072] Step S31: Invert the original point cloud;

[0073] Step S32: Construct a regular cloth mesh above the inverted point cloud;

[0074] Step S33: Simulate the free fall of fabric under the influence of gravity, with the mass point... Position updates under gravity depend on a preset step size:

[0075]

[0076] in, yes At the temporary height before being constrained by internal forces, It is the downward displacement driven by gravity. (Point mass) Subject to adjacent points Constrained displacement The calculation is as follows:

[0077]

[0078] in This is the tension coefficient, set to 0.15. If node j is already a fixed point, i.e., in contact with the ground, then the coefficient becomes 1.

[0079] In each fall iteration, the height of the particle must satisfy the following constraint:

[0080]

[0081] It is a point The height of the lowest point at the corresponding grid location.

[0082] Step S34: The fabric reaches a stable state after contacting the point cloud surface;

[0083] Step S35: Classify points based on the distance threshold between cloth nodes and point cloud points. Points with a distance less than the threshold are classified as ground point clouds, and points with a distance greater than the threshold are classified as non-ground point clouds. After processing, a set of ground point clouds and a set of non-ground point clouds are obtained. Figure 2a This is a diagram showing the non-ground point cloud separation result in Embodiment 1 of the present invention; Figure 2b This is a diagram showing the ground point cloud classification results in Embodiment 1 of the present invention.

[0084] Step S4: Semantic classification of non-terrestrial point clouds.

[0085] Non-terrestrial point clouds mainly include important scatterers in the wireless propagation process, such as buildings, vegetation, vehicles, and pole-shaped objects. To improve classification accuracy, this embodiment constructs a joint feature vector that integrates multi-scale geometric features, color features, and relative elevation features.

[0086] in:

[0087] (1) Multiscale geometric features

[0088] The point cloud covariance matrix was calculated at neighborhood scales of 0.5m, 1m, and 2m, and the results were extracted.

[0089] Linearity:

[0090]

[0091] Flatness:

[0092]

[0093] Sphericity:

[0094]

[0095] Anisotropy:

[0096]

[0097] Verticality:

[0098]

[0099] in, are the eigenvalues ​​of the covariance matrix of the point cloud's neighborhood points, and , This is the vertical component.

[0100] (2) Color characteristics

[0101] RGB color features, HSV color features, and Excess Green Index (ExG) features are extracted using the image information corresponding to the point cloud. The Excess Green Index is defined as:

[0102]

[0103] in These represent the values ​​for the red, green, and blue channels, respectively.

[0104] (3) Relative elevation characteristics

[0105] A local reference ground is constructed using the ground point cloud obtained in step S2.

[0106] The relative elevation of each point is defined as:

[0107] in, Indicates the elevation of the target point. This indicates the corresponding reference ground elevation.

[0108] The aforementioned features are then input into a random forest model. The random forest model consists of multiple decision trees, and the final classification is achieved through a majority voting mechanism. The final classification results are as follows: building category, vegetation category, vehicle category, and pole category. Figure 3 As shown.

[0109] Step S5: Semantic classification of ground point clouds

[0110] Ground point clouds mainly include roads, grasslands, bare soil, and water bodies. Because these categories exhibit strong color differences, this embodiment employs unsupervised clustering for classification.

[0111] First, construct the four-dimensional feature vector:

[0112]

[0113] in, The saturation level is extracted in HSV space and used to distinguish between low-saturation gray artificial pavements and high-saturation natural features. It is a vertical roughness feature. Redness is defined as It is used to capture the reddish-brown color response characteristic of soil.

[0114] The K-Means clustering algorithm was then used for classification.

[0115] The clustering objective function is:

[0116]

[0117] The number of clusters in the clustering is 4. For the centroid feature of the j-th cluster, This represents the Euclidean distance. The standardized feature vectors are then input into the K-Means algorithm. After clustering, different ground material categories are obtained.

[0118] Classification results as follows Figure 4 As shown.

[0119] Step S6: 3D reconstruction of non-terrestrial point clouds

[0120] Different reconstruction strategies are adopted for different types of scatterers.

[0121] Step S61: Building Reconstruction

[0122] The PolyFit algorithm is used to reconstruct the point cloud of buildings.

[0123] Specifically, this includes: plane detection, candidate surface generation, patch optimization, and global combination optimization. The final result is a model of the building's exterior surface.

[0124] 1) In the planar detection stage, the region growth method is used to extract planar primitives from the original building point cloud. By calculating the curvature and normal direction of the points, adjacent points with similar geometric properties are clustered and grown into planar regions.

[0125] 2) In the candidate plane generation and patch optimization stage, the algorithm extends the extracted finite planes and calculates the pairwise intersection relationships between the planes. The intersection lines of these planes constitute the candidate boundaries in three-dimensional space, further enclosing and forming a large number of candidate patches.

[0126] 3) In the global optimization generation stage, PolyFit constructs an energy function to globally optimize candidate faces, comprehensively considering data fit, model complexity, and manifold closure constraints. It then selects suitable combinations from the candidate faces by solving a binary integer programming problem, ultimately reconstructing a topologically consistent 3D building model. The building reconstruction results are shown below. Figure 5 As shown.

[0127] Step S62: Vegetation and Vehicle Reconstruction

[0128] For irregular targets such as vegetation, vehicles, and poles, a voxelization reconstruction method is used.

[0129] The specific steps are as follows:

[0130] 1) Discretize the point cloud into a voxel mesh. For any point in the point cloud... Its corresponding voxel index coordinates This is obtained by mapping using the floor function:

[0131]

[0132] By traversing the point cloud set Construct a non-empty voxel set This set represents the discrete occupancy state of vegetation in three-dimensional space:

[0133]

[0134] in, Each element in the array represents a side with length . voxel units.

[0135] 2) Morphological operations are used to fill internal voids. Due to the uneven density of point clouds, local voids may appear in the voxel structure. Therefore, neighborhood filling is performed on the voxels to enhance the connectivity of the volumetric structure.

[0136] 3) Extract isosurfaces using the Marching Cubes algorithm. First, assign voxel occupancy sets... Convert to a three-dimensional binary occupied field :

[0137]

[0138] Based on this, the isosurface extraction algorithm using moving cubes is used to extract... Extract the continuous triangular mesh surface. Assume that an edge of a voxel starts from an endpoint. and Composition, its in The corresponding scalar values ​​are respectively and And satisfy That is, the isosurface passes through this edge. The exact intersection point of the isosurface and the edge of this voxel. The coordinates are calculated using a linear interpolation formula:

[0139]

[0140] in, The target threshold for isosurface extraction is set to 0.5, so that the intersection points... Always located at the geometric midpoint of the voxel mesh edge, the initial mesh is guaranteed to be spatially uniform. After calculating all intersection points, the algorithm connects the intersection points within the same voxel into one or more triangular patches in sequence according to a pre-established topology lookup table, outputting a closed and continuous external surface mesh.

[0141] 4) Mesh simplification using the QEM algorithm. To further reduce model complexity, while maintaining the overall geometric contour and vegetation scattering volume characteristics, the generated triangular mesh is simplified using a quadratic error metric (QEM). The core idea of ​​the QEM algorithm is to simplify the mesh by iteratively performing an "edge folding" operation, while constraining the geometric error introduced by each fold to be minimized. For any vertex in the mesh... Its local geometric topology is compressed into a quadratic error matrix. The error cost of a vertex is defined as:

[0142]

[0143] The algorithm uses a greedy strategy to prioritize cost folding. The smallest edge. After simplification, a reconstructed model that balances geometric shape and computational efficiency can be generated.

[0144] In this embodiment: the voxel size is set to 0.3m, the number of model facets is reduced by more than 80%, while maintaining the main electromagnetic scattering contours unchanged. The reconstruction result is as follows. Figure 6 As shown, Figure 6 This is a schematic diagram of the irregular target reconstruction result in Embodiment 1 of the present invention; wherein, Figure 6 (a) in the image represents vegetation point cloud. Figure 6 (b) in the text represents vegetation voxelization. Figure 6(c) in the figure represents the vegetation reconstruction result. Figure 6 (d) in the figure represents the vehicle point cloud. Figure 6 (e) in the text represents vehicle voxelization. Figure 6 (f) in the diagram represents the vehicle reconfiguration structure.

[0145] Step S7: Ground 3D Reconstruction

[0146] To ensure accurate calculation of ground reflection paths, a continuously closed ground model needs to be constructed. This includes:

[0147] Step S71:

[0148] Project the ground point cloud onto a regular two-dimensional plane and construct a regular mesh;

[0149] Step S72:

[0150] Based on the ground semantic classification results obtained in step S4, the categories such as roads, grasslands, bare soil, and water bodies are mapped to the corresponding grid cells;

[0151] Step S73:

[0152] A digital elevation model is constructed using the average elevation values ​​in the semantic grid. For grid cells with valid elevation values, the corresponding elevation is directly assigned; for areas without point cloud coverage, they are marked as empty areas, thus obtaining the initial semantic digital elevation model.

[0153] Step S74:

[0154] Due to factors such as occlusion, shadows, and uneven point cloud density during UAV photogrammetry, missing regions may exist in the digital elevation model. To obtain a continuous surface, this embodiment employs a hybrid interpolation strategy for cavity repair. When sufficient neighboring samples exist around the cavity region, bilinear interpolation is used to calculate the missing elevation:

[0155]

[0156] in: Indicates the known elevation of the neighborhood. This indicates the corresponding weight.

[0157] For edge regions or regions with few samples, nearest neighbor interpolation is used for completion. Using a hybrid interpolation method can effectively improve terrain continuity while avoiding terrain distortion caused by excessive smoothing.

[0158] Step S74:

[0159] To eliminate semantic classification errors and small noise regions introduced during interpolation, a connected component analysis is performed on the generated semantic digital elevation model. When the area of ​​a connected component is less than a preset threshold, it is considered a noise region and is deleted or merged. This step effectively eliminates isolated patches and anomalous regions, improving the integrity and stability of the surface model.

[0160] Step S75:

[0161] A 3D ground mesh model is generated based on the optimized semantic digital elevation model. For adjacent mesh nodes, a triangular mesh connection method is used to construct a continuous surface. Each rectangular mesh cell is divided into two triangular faces. The 3D coordinates of the vertices are determined based on the mesh elevation values, and corresponding land feature category attributes are assigned according to semantic tags. The final result is a 3D ground model containing information on land features such as roads, grasslands, bare soil, and water bodies. The reconstruction result is shown below. Figure 7 As shown;

[0162] Step S8: Construction of a 3D Electromagnetic Environment Model

[0163] The building models, vegetation models, vehicle models, pole models, and ground models are registered to a unified coordinate system. Then, the models are merged to obtain a complete 3D electromagnetic environment model.

[0164] The reconstruction result is as follows Figure 8 As shown.

[0165] Step S9: Electromagnetic parameter assignment and ray tracing simulation

[0166] Step S91:

[0167] Import the 3D electromagnetic environment model generated in step S8 into the Wireless InSite ray tracing platform;

[0168] Step S92:

[0169] In Wireless InSite, electromagnetic parameters are assigned to environmental information such as buildings, roads, bare soil, forests, water bodies, vehicles, and poles in the 3D environment model based on the optimized environmental perception classification results.

[0170] Step S93:

[0171] Configure the location and antenna parameters of the transceiver in Wireless InSite, including antenna height, radiation pattern and transmit power, to complete the construction and initialization of the simulation environment, and perform ray tracing (RT) simulation to obtain the required channel characteristics;

[0172] Step S94:

[0173] The channel characteristics were evaluated and compared with real channel measurement data to ensure the accuracy of the three-dimensional electromagnetic environment and to verify the model's improvement in channel characteristic simulation. In Example 1, the channel measurement environment was a suburban scene near a building. The transmitter was placed at a fixed location near the building, and the receiver's location changed continuously. The measurement frequency was 7 GHz, including line-of-sight (LoS) and non-line-of-sight (NLoS) scenarios. The path loss comparison results in this example are shown in the figure. Figure 9a , Figure 9b , Figure 9c , Figure 9a This is a schematic diagram comparing the path loss results of the line-of-sight path in Embodiment 1 of the present invention. Figure 9b This is a schematic diagram comparing the path loss results of non-line-of-sight path 1 in Embodiment 1 of the present invention. Figure 9c This is a schematic diagram comparing the path loss results of non-line-of-sight path 2 in Embodiment 1 of the present invention;

[0174] The results show that the electromagnetic environment model reconstructed from point cloud data can accurately describe the propagation structure characteristics in real-world scenarios, thus effectively capturing the signal attenuation characteristics under both Loss and Non-LoS propagation conditions. The RT simulation results based on the point cloud-reconstructed environment show high consistency with the actual measurement results in both the scatter distribution and overall trend of path loss. Figure 10a This is a schematic diagram comparing the time delay spread results of the line-of-sight path in Embodiment 1 of the present invention; Figure 10b This is a schematic diagram comparing the time delay spread results of non-line-of-sight path 1 in Embodiment 1 of the present invention; Figure 10c This is a schematic diagram comparing the time delay spread results of non-line-of-sight path 2 in Embodiment 1 of the present invention. The RT simulation results based on the reconstructed electromagnetic environment model and the actual measured data show good consistency in the shape and distribution trend of the CDF curve, indicating that the constructed environment model can accurately reflect the multipath propagation characteristics in the LoS scenario. Under NLoS propagation conditions, factors such as buildings, vegetation obstruction, and other scatterers in the scene increase the complexity of the propagation path, resulting in greater dispersion of the time delay spread, a larger overall time delay spread, and more pronounced curve fluctuations. Overall, the DS simulation results obtained from the electromagnetic environment model based on point cloud reconstruction under the NLoS scenario still maintain a good degree of agreement with the measured channel data, and the two show high similarity in the shape and distribution trend of the CDF curve.

[0175] 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 variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data, characterized in that, include: Step S1: Obtain 3D point cloud data of the target communication scene; Step S2: Pre-classify the ground types that affect wireless channel propagation in the target area to obtain semantic labels for each pre-classified ground type; Step S3: Use the cloth simulation filtering algorithm to separate the ground point cloud from the non-ground point cloud in the 3D point cloud data to obtain the ground point cloud dataset and the non-ground point cloud dataset. Step S4: Extract multi-scale geometric features, color features, and relative elevation features from the non-ground point cloud dataset. Construct a joint feature vector based on the multi-scale geometric features, color features, and relative elevation features. Input the constructed joint feature vector into the random forest model for training. Perform semantic classification on the non-ground point cloud to obtain the non-ground point cloud classification result. Step S5: For the ground point cloud dataset, construct ground point cloud feature vectors, and use a clustering algorithm to perform unsupervised clustering of the ground point cloud feature vectors to obtain the land cover attributes corresponding to the clusters; complete the semantic classification of the ground point clouds based on the land cover attributes corresponding to the clusters obtained by clustering, and finally output the ground point cloud classification results with semantic labels. Step S6: Based on the non-ground point cloud classification results, the building point cloud in the non-ground point cloud classification results is reconstructed in three dimensions using the planar fitting reconstruction method, and the vegetation, vehicle and rod point cloud in the non-ground point cloud classification results is reconstructed in three dimensions using the voxelization reconstruction method, so as to obtain the reconstructed building model, vegetation model, vehicle model and rod model. Step S7: Construct a semantic ground elevation model based on the ground point cloud classification results, and complete the three-dimensional reconstruction of the ground area to obtain the reconstructed semantic ground elevation model. Step S8: Integrate the reconstructed building model, vegetation model, vehicle model, pole model, and semantic ground elevation model to construct a three-dimensional environment model; Step S9: Assign corresponding electromagnetic parameters to the 3D environment model based on the semantic tags of the pre-classified ground types to obtain the 3D electromagnetic model, and use the ray tracing algorithm to perform wireless channel simulation on the 3D electromagnetic model to obtain the channel characteristic parameters of the target communication scenario.

2. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, In step S2, the various ground types pre-classified include ground, buildings, vegetation, traffic lights, streetlights, and street debris; The ground includes asphalt roads, brick and cement roads, soil, grass, and water bodies; The buildings include high-rise buildings and single-story temporary buildings; The vegetation includes shrubs and tall vegetation.

3. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, In step S4, the non-ground point cloud classification results include the classification results for buildings, vegetation, vehicles, and pole-shaped objects; In step S5, the ground point cloud classification results include the classification results for roads, grasslands, bare soil, and water bodies.

4. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, In step S1, the 3D point cloud data is obtained by acquiring multi-view images through an UAV oblique photography system, followed by aerial triangulation, image matching, point cloud generation, and coordinate registration.

5. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, In step S3, the cloth simulation filtering algorithm constructs an inverted point cloud surface and simulates the free fall of virtual cloth, so that the cloth surface conforms to the terrain contour and the ground point cloud and non-ground point cloud are separated according to the distance between the cloth nodes and the original point cloud.

6. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, The multi-scale geometric features in step S4 include linearity, flatness, sphericity, anisotropy, rate of curvature change, and verticality features; the color features include RGB color features, HSV color features, and greenness index features; the relative elevation features are calculated by the elevation difference between point cloud points and the local ground reference surface. In step S4, the random forest classification model uses a class-balanced sampling strategy to construct a training dataset and utilizes multiple decision tree classifiers for ensemble learning to achieve semantic classification of non-terrestrial point clouds.

7. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, The ground feature vector in step S5 includes color features and vertical roughness features, and the clustering algorithm is the K-Means clustering algorithm.

8. The method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, In step S6, the building point cloud is reconstructed using the PolyFit algorithm, which includes plane extraction, candidate polygon generation, and global optimization steps to generate a three-dimensional surface model of the building.

9. A method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, The voxelization reconstruction method in step S6 includes: converting the point cloud into a voxel mesh; filling holes in the voxel mesh; extracting the three-dimensional surface using the Marching Cubes algorithm to generate a triangular mesh model; and simplifying the triangular mesh model using the quadratic error metric QEM algorithm.

10. A method for reconstructing a city's three-dimensional electromagnetic environment based on point cloud data according to claim 1, characterized in that, Step S7, which involves constructing a semantic ground elevation model, includes: projecting the ground point cloud onto a regular grid; generating corresponding grid cells based on semantic labels; filling in missing regions using a hybrid interpolation method; removing isolated noise regions using a connected component filtering method; and generating a continuous, closed 3D ground model.

Images