Three-dimensional landform model reconstruction method and device based on geometry and texture combination

Abstract

The embodiment of the invention discloses a three-dimensional landform model reconstruction method and device based on geometry and texture combination. A specific embodiment of the method comprises the following steps: carrying out self-adaptive grid division on a landform three-dimensional point cloud and then carrying out geometric super-resolution to obtain a target three-dimensional landform sub-grid; performing color texture mapping and texture correction enhancement on the target three-dimensional landform sub-grids to obtain an enhanced landform texture map; performing collaborative bidirectional optimization on the enhanced landform texture map and the target three-dimensional landform sub-grids to obtain a super-resolution landform texture map and a super-resolution three-dimensional landform sub-grid; performing model recombination and fusion on the plurality of super-resolution landform texture maps and the plurality of super-resolution three-dimensional landform sub-grids to obtain a three-dimensional texture landform grid model; and performing density control reconstruction processing on the three-dimensional texture landform grid model to obtain a three-dimensional density texture landform model, and storing the three-dimensional density texture landform model. According to the embodiment, the quality of the three-dimensional density texture landform model can be improved, and waste of storage resources is reduced.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a method and apparatus for reconstructing three-dimensional terrain models based on a combination of geometry and texture. Background Technology

[0002] With the widespread adoption of 3D data acquisition technologies such as remote sensing, lidar, and UAV mapping, colored 3D terrain models are typically composed of a 3D mesh representing the terrain structure and texture attributes representing terrain features (e.g., color, physically based materials). These models are widely used in geographic information systems, digital earth, terrain visualization, and virtual simulation. However, limited by sampling hardware resources, the acquired colored 3D terrain models often suffer from spatial sparsity and lack of detail, affecting the accuracy of 3D terrain visualization. For 3D terrain model reconstruction based on a combination of geometry and texture, the common approach is as follows: acquire a 3D discrete point cloud of the target area. Then, input the 3D discrete point cloud into a cross-view interactive Transformer model to obtain terrain texture information. Next, perform super-resolution processing on the terrain texture information to obtain a super-resolution texture map. Finally, fuse the terrain texture information and the terrain point cloud to obtain a 3D textured terrain model and store it.

[0003] However, in practice, it has been found that when the above method is used to reconstruct 3D terrain models, the following technical problems often occur: the terrain texture information generated by the multi-view interactive Transformer model and the texture super-resolution are not taken into account, resulting in texture distortion problems such as distortion and breakage in the super-resolution texture map. The quality of the super-resolution texture map and the 3D texture terrain model is poor, and storage resources are wasted.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the present disclosure concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a method and apparatus for reconstructing three-dimensional terrain models based on the combination of geometry and texture, in order to solve one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a method for reconstructing a three-dimensional terrain model based on a combination of geometry and texture, comprising: acquiring a three-dimensional point cloud of terrain features in a target area; adaptively meshing the three-dimensional point cloud of terrain features to obtain a set of three-dimensional terrain sub-mesh sets; for each three-dimensional terrain sub-mesh in the set of three-dimensional terrain sub-mesh sets, performing the following collaborative bidirectional optimization steps: performing geometric super-resolution processing on the three-dimensional terrain sub-mesh sets to obtain a target three-dimensional terrain sub-mesh; performing color and texture mapping processing on the target three-dimensional terrain sub-mesh sets to obtain a terrain color and texture attribute map; and performing a color and texture mapping based on the target three-dimensional terrain sub-mesh sets. The above-mentioned landform color texture attribute map is subjected to texture correction and enhancement processing to obtain an enhanced landform texture map; the enhanced landform texture map and the above-mentioned target 3D landform sub-mesh are subjected to collaborative bidirectional optimization to obtain a super-resolution landform texture map and a super-resolution 3D landform sub-mesh; the multiple super-resolution landform texture maps and multiple super-resolution 3D landform sub-mesh obtained are model recombined and fused to obtain a 3D textured landform mesh model; the above-mentioned 3D textured landform mesh model is subjected to density-controlled reconstruction processing to obtain a 3D density textured landform model, and the above-mentioned 3D density textured landform model is stored as a landform model.

[0008] Secondly, some embodiments of this disclosure provide a three-dimensional terrain model reconstruction apparatus based on a combination of geometry and texture, comprising: an acquisition unit configured to acquire a three-dimensional point cloud of terrain features in a target area; an adaptive mesh generation unit configured to adaptively mesh the three-dimensional point cloud of terrain features to obtain a set of three-dimensional terrain sub-meshes; and an execution unit configured to perform the following collaborative bidirectional optimization steps for each three-dimensional terrain sub-mesh in the set of three-dimensional terrain sub-meshes: performing geometric super-resolution processing on the three-dimensional terrain sub-mesh to obtain a target three-dimensional terrain sub-mesh; performing color texture mapping processing on the target three-dimensional terrain sub-mesh to obtain a terrain color texture attribute map; and performing a color texture mapping based on the target three-dimensional terrain sub-mesh. The sub-mesh is used to perform texture correction and enhancement processing on the above-mentioned terrain color texture attribute map to obtain an enhanced terrain texture map; the enhanced terrain texture map and the above-mentioned target 3D terrain sub-mesh are jointly optimized in two directions to obtain a super-resolution terrain texture map and a super-resolution 3D terrain sub-mesh; the model recombination and fusion unit is configured to recombine and fuse the obtained multiple super-resolution terrain texture maps and multiple obtained super-resolution 3D terrain sub-meshes to obtain a 3D textured terrain mesh model; the density-controlled reconstruction unit is configured to perform density-controlled reconstruction processing on the above-mentioned 3D textured terrain mesh model to obtain a 3D density textured terrain model, and to store the above-mentioned 3D density textured terrain model.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0011] The various embodiments of this disclosure have the following beneficial effects: The 3D terrain model reconstruction method based on geometry and texture in some embodiments of this disclosure can improve the quality of 3D density texture terrain models and reduce the waste of display resources. Specifically, the reason why road markings in related maps are not accurate is that: when generating terrain texture information and performing texture super-resolution through a multi-view interactive Transformer model, the geometric structure information of the terrain in the color 3D terrain model is not considered, resulting in texture distortion problems such as distortion and breakage in the super-resolution texture map. The quality of the super-resolution texture map and the 3D texture terrain model is poor, and storage resources are wasted. Based on this, the 3D terrain model reconstruction method based on geometry and texture in some embodiments of this disclosure can first obtain the 3D point cloud of the terrain of the target area. Here, the 3D point cloud of terrain is used for subsequent adaptive meshing. Secondly, the above-mentioned 3D point cloud of terrain is adaptively meshed to obtain a 3D terrain sub-mesh set. Here, adaptive meshing based on the terrain corresponding to the 3D point cloud of terrain can improve the rationality of the 3D terrain sub-mesh and reduce the amount of data and the level of detail in subsequent processing. Next, for each 3D terrain sub-mesh in the aforementioned 3D terrain sub-mesh set, the following collaborative bidirectional optimization steps are performed: First, geometric super-resolution processing is performed on the aforementioned 3D terrain sub-mesh to obtain the target 3D terrain sub-mesh. This improves the representation of geometric details and topological relationships of the terrain in the target 3D terrain sub-mesh, making it closer to the real terrain. Second, color texture mapping processing is performed on the aforementioned target 3D terrain sub-mesh to obtain a terrain color texture attribute map. This reduces the mapping error from 3D to 2D and improves the texture quality of the terrain color texture attribute map. Third, based on the aforementioned target 3D terrain sub-mesh, texture correction and enhancement processing is performed on the aforementioned terrain color texture attribute map to obtain an enhanced terrain texture map. Further improving the quality of the enhanced terrain texture map effectively avoids texture distortion and improves the matching degree between texture and geometry. Fourth, collaborative bidirectional optimization is performed on the aforementioned enhanced terrain texture map and the aforementioned target 3D terrain sub-mesh to obtain a super-resolution terrain texture map and a super-resolution 3D terrain sub-mesh. Here, the coupling relationship between texture and geometry is considered, and a balanced co-optimization approach achieves a balance between spatial structure and texture fidelity. This effectively avoids misalignment issues between geometric textures and improves the quality of the super-resolution terrain texture map and the super-resolution 3D terrain sub-mesh. Then, the obtained multiple super-resolution terrain texture maps and multiple obtained super-resolution 3D terrain sub-meshes are modeled and fused to obtain a 3D textured terrain mesh model. This eliminates boundary seams and conflicts in the overlapping parts of the 3D terrain sub-mesh combination, improving the visual consistency of the 3D textured terrain mesh model.Finally, density-controlled reconstruction processing is performed on the aforementioned 3D textured terrain mesh model to obtain a 3D density-textured terrain model, which is then visualized. Here, density control improves the global uniformity of the 3D density-textured terrain model and maximizes the preservation of the macroscopic geometric structure and texture details of the target area's terrain, thereby improving the accuracy of 3D reconstruction, enhancing the quality of the 3D density-textured terrain model, and reducing storage resource waste. Therefore, this geometry-and-texture joint 3D terrain model reconstruction method can improve the quality of the 3D density-textured terrain model and reduce storage resource waste. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of the geometry-and-texture combined three-dimensional terrain model reconstruction method according to the present disclosure; Figure 2 These are schematic diagrams of some embodiments of the three-dimensional terrain model reconstruction apparatus based on the combination of geometry and texture according to this disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1 A flowchart 100 is shown, illustrating some embodiments of a geometry-and-texture combined-method 3D terrain model reconstruction method according to this disclosure. This geometry-and-texture combined-method 3D terrain model reconstruction method includes the following steps: Step 101: Obtain the topographic information of the target area.

[0021] In some embodiments, the execution entity (e.g., an electronic device) of the above-described geometry and texture-based 3D terrain model reconstruction method can acquire terrain point cloud information of the target area via a wired or wireless connection. The target area can be a large-scale region requiring digitization and visualization reconstruction of its spatial information, geometric features, and texture attributes. This large-scale sparse region can cover an area or a mountain range. The 3D terrain point cloud can be a point cloud describing the terrain of the target area. The terrain information can include: a large number of discrete 3D terrain points, the 3D coordinate information of these points, and the color information of the target area.

[0022] Step 102: Perform adaptive meshing on the 3D point cloud of terrain features to obtain a 3D terrain feature sub-mesh set.

[0023] In some embodiments, the execution entity may adaptively mesh the aforementioned three-dimensional point cloud of terrain features to obtain a set of three-dimensional terrain sub-mesh. The three-dimensional terrain sub-mesh in the aforementioned set of three-dimensional terrain sub-mesh can be a sub-mesh that includes three-dimensional point sets with a topological structure and overlapping regions between the three-dimensional terrain sub-mesh.

[0024] In some optional implementations of certain embodiments, the above-mentioned adaptive meshing of the 3D point cloud of terrain features to obtain a 3D terrain feature sub-mesh set may include the following steps: The first step is to perform filtering and denoising on the aforementioned 3D point cloud of terrain features to obtain a denoised 3D point cloud of terrain features. This denoised 3D point cloud can be obtained by removing outlier 3D points that deviate from the local regional trend, or by normalizing the 3D coordinates of the terrain feature 3D point cloud. The filtering and denoising process can be performed using a statistical filtering algorithm.

[0025] The second step involves normalizing the coordinates of the denoised 3D terrain point cloud to obtain a normalized 3D terrain point cloud. This coordinate normalization can be achieved by transforming the coordinate system of the denoised terrain point cloud to a local coordinate system, thereby reducing the impact of differences in absolute coordinate values. This local coordinate system can be a coordinate system with the smallest 3D point as its origin.

[0026] The third step involves determining the local density information of each 3D landform point in the normalized 3D point cloud, thus obtaining a local density information set. This local density information set characterizes the density of 3D points surrounding each landform point. In practice, the executing entity can determine the local density information by taking the reciprocal of the average distance between a predetermined number of adjacent 3D landform points and the points themselves. This predetermined number can be a pre-set number, such as 32.

[0027] The fourth step involves adaptively dividing the normalized 3D terrain point cloud into windows based on the aforementioned local density information set, resulting in a set of window-center terrain points. The window-center terrain points in this set can be the 3D terrain points located at the center of the window within the adaptive window. In practice, the executing entity can first determine an initial adaptive sliding window. This initial adaptive sliding window can be a window comprising a minimum window size, a maximum window size, and an adaptive step size. The minimum window size can be the length of a window centered on the 3D terrain point corresponding to the largest local density information, including at least 32 neighboring 3D terrain points. The maximum window size can be the length of a window centered on the 3D terrain point corresponding to the smallest local density information, including at most 512 neighboring 3D terrain points. The aforementioned adaptive step size can be determined based on the local density information of the point cloud of the 3D terrain points located at the initial position of the window included in the initial adaptive sliding window. If the local density information of the point cloud is greater than or equal to the density threshold, the step size is 1.2 times half the minimum window size; if it is less than the preset density threshold, the step size is 0.8 times half the minimum window size. The density threshold can be a pre-set threshold used to distinguish between high-density and low-density regions. The density threshold can be dynamically changed; if it is greater than the sum of the mean and standard deviation of the local density information set of the point cloud included in the previous adaptive sliding window, it is a high-density region; if it is less than the sum of the mean and standard deviation of the local density information set of the point cloud included in the previous adaptive sliding window, it is a low-density region. The aforementioned initial adaptive sliding window can be the minimum window. Next, starting from the 3D terrain point with the smallest coordinates included in the normalized terrain 3D point cloud, the normalized terrain 3D point cloud is traversed in row and column order to obtain an adaptive sliding window set. Among them, the adaptive sliding windows in the aforementioned adaptive sliding window set can be windows with half of their area overlapping between adjacent adaptive sliding windows. The next adaptive sliding window can be the sliding window determined by the previous adaptive sliding window according to the window size and adaptive step size. Then, the set of 3D terrain points located at the center position included in the above adaptive sliding window set is used as the initial window center terrain point set. Finally, the center point optimization processing is performed on the above initial window center terrain point set to obtain the window center terrain point set. The above center point optimization processing can include: deduplication and boundary supplementation. The deduplication can be a process of determining the distance between the initial window center terrain point sets, and retaining the points with higher local density if the distance is less than a distance threshold. The above distance threshold can be a threshold of one-quarter of the minimum window size. The above boundary supplementation can be a process of supplementing center points (e.g., uniform interpolation along the edge) for the parts not covered by the adaptive sliding window to avoid the loss of edge information.

[0028] The fifth step involves reconstructing a 3D mesh from the aforementioned set of terrain points at the center of the window, resulting in a 3D terrain surface mesh model. This 3D terrain surface mesh model can be data that connects discrete 3D terrain points to form a topological structure. Specifically, it can include a set of 3D terrain points (vertices), edges connecting adjacent 3D terrain points to form topological relationships, faces formed by these edges, and connection rules between vertices, edges, and faces. This 3D mesh reconstruction can be performed using the Poisson surface reconstruction algorithm.

[0029] The sixth step is to perform mesh overlap division on the above-mentioned three-dimensional terrain surface mesh model to obtain three-dimensional terrain surface mesh sub-models, which serve as the three-dimensional terrain sub-mesh set. The mesh overlap division can be based on a 50% boundary overlap between adjacent three-dimensional terrain surface mesh sub-models.

[0030] Step 103: For each 3D terrain subgrid in the above 3D terrain subgrid set, perform the following collaborative bidirectional optimization steps: Step 1031: Perform geometric super-resolution processing on the 3D terrain sub-mesh to obtain the target 3D terrain sub-mesh.

[0031] In some embodiments, the execution entity may perform geometric super-resolution processing on the aforementioned three-dimensional terrain sub-mesh to obtain a target three-dimensional terrain sub-mesh. The target three-dimensional terrain sub-mesh may be a sub-mesh that includes more vertices and faces than the standard three-dimensional terrain sub-mesh, and can represent the terrain of the target region in more detail.

[0032] In some optional implementations of certain embodiments, the above-mentioned geometric super-resolution processing of the three-dimensional terrain sub-mesh to obtain the target three-dimensional terrain sub-mesh may include the following steps: The first step involves performing graph convolution feature extraction on the aforementioned 3D terrain sub-grid to obtain the grid terrain geometric feature vector. This vector can represent geometric information of the target region, such as local curvature map, normal vector distribution, topological neighborhood features, and terrain trends. The graph convolution feature extraction can be performed using a graph convolutional network model.

[0033] The second step involves refining the three-dimensional terrain sub-grid based on the aforementioned geomorphic feature vectors to obtain an initial geomorphic geometric sub-grid. This initial geomorphic geometric sub-grid can be a sub-grid that increases the number of vertices (i.e., resolution) of the three-dimensional terrain sub-grid by enhancing vertex and face density.

[0034] As an example, the aforementioned execution entity can utilize a mesh subdivision algorithm to refine the three-dimensional geomorphic subgrid based on the aforementioned mesh geomorphic geometric feature vectors, thereby obtaining an initial geomorphic geometric subgrid.

[0035] The third step involves employing a mesh structure super-resolution model. Based on the aforementioned mesh geomorphic geometric feature vectors, the initial geomorphic geometric sub-mesh is structurally optimized to obtain a structurally optimized 3D geomorphic sub-mesh. This mesh structure super-resolution model can be a model that includes a forward process to progressively add noise to the initial geomorphic geometric sub-mesh and a reverse denoising process. Alternatively, the mesh structure super-resolution model can be a conditional diffusion model.

[0036] The fourth step involves performing topological correction on the optimized 3D terrain sub-mesh to obtain a topological terrain sub-mesh, which serves as the target 3D terrain sub-mesh. This topological terrain sub-mesh can be obtained by performing topological conflict resolution on the optimized 3D terrain sub-mesh. This topological conflict resolution can be achieved by smoothing the vertex set of the optimized 3D terrain sub-mesh using the Laplacian operator, or by determining whether the vertices of each facet are collinear or self-intersecting based on the mesh's adjacency relationships, and if so, by delegating the vertices to other facets through topological constraints.

[0037] Step 1032: Perform color texture mapping on the target 3D terrain sub-mesh to obtain the terrain color texture attribute map.

[0038] In some embodiments, the execution entity may perform color texture mapping processing on the target three-dimensional terrain sub-mesh to obtain a terrain color texture attribute map. The terrain color texture attribute map may be a PBR (Physically Based Rendering) texture map, which includes albedo and roughness, obtained by mapping the color attributes of the target three-dimensional terrain sub-mesh to a two-dimensional plane.

[0039] In some optional implementations of certain embodiments, the above-mentioned color texture mapping processing of the target 3D terrain sub-mesh to obtain a terrain color texture attribute map may include the following steps: The first step is to determine the curvature weight of each grid edge within the aforementioned target 3D terrain sub-grid, thus obtaining a set of grid edge curvature weights. These grid edge curvature weights characterize the importance of the grid edge; that is, the larger the angle between the normal vectors of the two vertices of the grid edge, the lower the weight, and the smaller the angle, the higher the weight. In practice, the execution entity can first determine the angle between the normal vectors of the two vertices of each grid edge, obtaining the vertex angle. Secondly, through an objective weighting method, the weight value of each vertex angle is determined as the grid edge curvature weight.

[0040] The second step involves determining the set of surface extension curves for the target 3D terrain sub-mesh based on the aforementioned set of grid edge curvature weights. The surface extension curves in this set can be the cutting edges of the two-dimensional surface formed by extending the target 3D terrain sub-mesh. For example, the executing entity can first select a predetermined number of grid edge curvature weights from the set of grid edge curvature weights, sorted by numerical value, as the target grid edge curvature weight set. This predetermined number can be the minimum number of grid edges required to unfold the target 3D terrain sub-mesh into a two-dimensional surface. Then, the grid edges corresponding to the target grid edge curvature weight set are determined as the set of surface extension curves.

[0041] The third step involves splitting the target 3D terrain subgrid into a set of terrain grid blocks based on the aforementioned set of surface extension curves. The terrain grid blocks in this set can be obtained by further splitting the target 3D terrain subgrid so that the resulting grid blocks can be unfolded into a two-dimensional plane.

[0042] The fourth step involves performing a two-dimensional mapping process on the target three-dimensional terrain sub-grid based on the aforementioned terrain grid set, thereby obtaining a two-dimensional terrain texture map. This two-dimensional terrain texture map can be a two-dimensional image formed by mapping the terrain grid set from three-dimensional space to a two-dimensional UV plane and arranging the grid blocks within a UV space where the UV axis values ​​are all within the range of [0, 1].

[0043] As an example, the aforementioned execution entity can utilize a least-squares conformal mapping algorithm to perform two-dimensional mapping on the aforementioned terrain grid block set, obtaining a two-dimensional terrain plane set. Then, using a greedy algorithm, the aforementioned two-dimensional terrain plane set is arranged on a target two-dimensional plane to obtain a two-dimensional terrain texture map. The target two-dimensional plane can be a UV space where the UV axes all range from [0, 1]. The objective function of the aforementioned greedy algorithm can be a function that minimizes the area occupied by the UV plane.

[0044] The fifth step involves matching the aforementioned target 3D terrain sub-mesh and the aforementioned 2D terrain texture map to obtain a matching information set. This matching information set can record the mapping relationship between 3D terrain points and triangular facets. The matching process can involve searching the mesh for the nearest face (triangular facet) using a KD-tree to determine if the 3D terrain point is within the projection range of the triangular facet. If it is within the range, the correspondence between the 3D terrain point and the triangular facet is recorded; otherwise, it is marked as an invalid point and discarded.

[0045] Step 6: Based on the above matching information set, map the color attribute information included in the above three-dimensional point cloud of terrain to the above two-dimensional texture map of terrain to obtain the mapped two-dimensional texture map of terrain.

[0046] As an example, the aforementioned execution entity can first filter out the successfully matched 3D terrain point set and triangle facet set from the matching information set, as the matched 3D terrain point set and matched triangle facet set. Secondly, it determines the barycentric coordinates of each matched 3D terrain point in the matched 3D terrain point set on the corresponding matching triangle facet in the matched triangle facet set, obtaining the barycentric coordinate set. Then, it performs interpolation processing on the barycentric coordinate set to obtain a 2D coordinate set. Finally, it maps the color attribute information set corresponding to the matched 3D terrain point set to the pixel position set corresponding to the 2D coordinate set, obtaining the mapped 2D terrain texture map. If multiple matched 3D terrain points are mapped to the same 2D coordinate, the color attribute information corresponding to the multiple matched 3D terrain points is weighted and mapped. The weight can be the reciprocal of the distance from the matched 3D terrain point to the triangle facet; the closer the distance, the higher the weight.

[0047] Step 7: Fill in the blank areas of the mapped 2D terrain texture map to obtain the terrain color texture attribute map. This blank area filling process can be performed using a model that incorporates an attention mechanism within the U-Net network model.

[0048] Step 1033: Based on the target 3D terrain sub-mesh, perform texture correction and enhancement processing on the terrain color texture attribute map to obtain the enhanced terrain texture map.

[0049] In some embodiments, the execution entity may perform texture correction and enhancement processing on the terrain color texture attribute map based on the target three-dimensional terrain sub-mesh to obtain an enhanced terrain texture map. The enhanced terrain texture map may be a texture map obtained by performing texture offset correction on the terrain color texture attribute map.

[0050] In some optional implementations of certain embodiments, the above-mentioned texture correction and enhancement processing of the terrain color texture attribute map based on the target three-dimensional terrain sub-mesh to obtain an enhanced terrain texture map may include the following steps: The first step is to extract geometric features from the aforementioned target 3D terrain sub-mesh, obtaining a set of 3D geometric feature vectors for the mesh. These 3D geometric feature vectors can characterize the geometric features of the target region's terrain. This geometric feature information may include, but is not limited to, at least one of the following: normalized vertex position information, vertex normals, Gaussian curvature, and mean curvature.

[0051] The second step involves extracting two-dimensional texture from the aforementioned landform color texture attribute map to obtain a set of two-dimensional texture feature vectors. These two-dimensional texture feature vectors can be normalized pixel-level color values ​​representing the landform color texture attribute map.

[0052] The third step involves aligning and stitching the aforementioned 3D geometric feature vector set of the mesh and the aforementioned 2D texture feature vector set to obtain a joint geometric-texture feature map. This joint geometric-texture feature map can be obtained by mapping the aforementioned 3D geometric feature vector set of the mesh to the corresponding pixel positions in the corresponding terrain color texture attribute map, aligning them, and then stitching them along the channel dimension. In practice, the executing entity can determine the 2D coordinates of each pixel in the aforementioned terrain color texture attribute map, obtaining a set of pixel 2D coordinates. Then, the 2D coordinates of the pixel 2D coordinate set and the corresponding 2D coordinates of the aforementioned 3D geometric feature vector set of the mesh are matched to obtain a coordinate matching result set. Finally, using the coordinate matching result set, the aforementioned 3D geometric feature vector set of the mesh and the aforementioned 2D texture feature vector set are stitched together to obtain the joint geometric-texture feature map.

[0053] The fourth step involves inputting the aforementioned joint geometric texture feature map set into the color texture deviation prediction model to obtain a color texture offset set. The color texture offsets in this set can represent the color offset degree of each pixel in the aforementioned terrain color texture attribute map. The color texture deviation prediction model can be a deep neural network model that predicts the pixel-level color deviation of the input joint geometric texture feature map to obtain the color texture offset. For example, the color texture deviation prediction model could be RDB (Residual Dense Blocks).

[0054] The fifth step involves performing constraint correction processing on the aforementioned color texture offset set to obtain the corrected color texture offset set. This constraint correction processing can be performed using the Blinn-Phong model.

[0055] The sixth step involves residual fusion of the corrected color texture offset set and the aforementioned landform color texture attribute map to obtain a fused landform color texture attribute map. This residual fusion can be a process of adding the corrected color texture offset set and the aforementioned landform color texture attribute map together. It should be noted that residual fusion preserves the global color trend of the landform color texture attribute map (e.g., the overall hue of mountains), avoiding color shifts caused by de novo prediction in the color texture deviation prediction model. The corrected color texture offset set focuses only on high-frequency details (e.g., light and dark spots on rock surfaces, subtle color differences in vegetation), improving the learning efficiency of the color texture deviation prediction model.

[0056] Step 7: Post-processing optimization is performed on the fused terrain color texture attribute map to obtain an optimized terrain texture map, which serves as the enhanced terrain texture map. The post-processing optimization may include, but is not limited to, at least one of the following: color range mapping, smoothing filtering, and two-dimensional boundary consistency checking.

[0057] Step 1034: Perform collaborative bidirectional optimization on the enhanced terrain texture map and the target 3D terrain sub-mesh to obtain the super-resolution terrain texture map and the super-resolution 3D terrain sub-mesh.

[0058] In some embodiments, the execution entity may perform collaborative bidirectional optimization on the enhanced terrain texture map and the target 3D terrain sub-mesh to obtain a super-resolution terrain texture map and a super-resolution 3D terrain sub-mesh. The super-resolution terrain texture map may be a sub-mesh with higher resolution than the target 3D terrain sub-mesh, obtained through collaborative bidirectional optimization of geometry and texture. The super-resolution 3D terrain sub-mesh may be a texture map of higher quality than the enhanced terrain texture map, obtained through collaborative bidirectional optimization of geometry and texture.

[0059] In addressing the aforementioned technical challenges in the application of digital twins and smart cities, the following additional technical issues often arise: The construction of terrain texture maps and 3D terrain sub-mesh typically involves separate resolution enhancement and optimization. This separate enhancement and optimization leads to difficulties in recovering complex undulating structures in the 3D terrain sub-mesh, insufficient color detail extraction from the terrain texture map, or edge breakage. This makes it difficult to adapt to the 3D construction of large-scale terrain, prolonging the construction time and reducing the quality of both the terrain texture map and the 3D terrain sub-mesh, ultimately wasting significant storage resources. Based on the characteristics of this application scenario—large-scale urban and natural terrain coverage, the need for millimeter-level geometric accuracy, realistic texture representation, and multi-scale detail requirements—we have decided to adopt the following solution: In some optional implementations of certain embodiments, the above-mentioned collaborative bidirectional optimization of the enhanced terrain texture map and the target 3D terrain submesh to obtain the super-resolution terrain texture map and the super-resolution 3D terrain submesh may include the following steps: The first step, based on the enhanced terrain texture map and the target 3D terrain submesh, is to perform the following determination steps: Sub-step 1: Based on the target 3D terrain sub-mesh, perform the following reverse gradient update steps: The first sub-step involves performing differentiable rendering on the target 3D terrain sub-mesh based on the enhanced terrain texture map to obtain a 3D terrain rendering image. This 3D terrain rendering image can be a rendering color map generated by a differentiable renderer. As an example, the execution entity can use a differentiable renderer to perform differentiable rendering on the target 3D terrain sub-mesh based on the enhanced terrain texture map to obtain the 3D terrain rendering image.

[0060] The second sub-step involves determining the texture optimization geometric loss value set for the 3D terrain rendering image and the target terrain rendering image. The texture optimization geometric loss values ​​include color mapping loss values ​​and texture normal loss values. These loss values ​​can represent the color difference and texture normal difference between the rendered target 3D terrain sub-mesh (using the enhanced terrain texture image) and the actual terrain of the target region. The texture normal loss value can be the difference between the normal projection of the target 3D terrain sub-mesh onto the enhanced terrain texture image and the actual pixel normal of the enhanced terrain texture image. The target terrain rendering image can be a texture map of the actual terrain of the target region.

[0061] The third sub-step involves determining the set of grid vertex gradient values ​​for the grid vertex set included in the target 3D terrain sub-mesh, based on the texture optimization geometric loss set. The grid vertex gradient values ​​characterize the rate of change of function values ​​at grid vertices. As an example, the execution entity can first perform a weighted summation of the color mapping loss set and the texture normal loss set to obtain a weighted loss set. Then, it can determine the grid vertex gradient values ​​within the weighted loss set.

[0062] The fourth sub-step involves performing a reverse gradient update on the grid vertex set based on the grid vertex gradient value set, resulting in an updated 3D terrain sub-grid. As an example, the aforementioned execution entity can use the gradient descent algorithm to perform a reverse gradient update on the grid vertex set based on the grid vertex gradient value set, thus obtaining the updated 3D terrain sub-grid.

[0063] Sub-step 2: Based on the enhanced terrain texture map, perform the following texture map update steps: The first sub-step involves extracting the geometric attributes of the target 3D terrain sub-mesh to obtain the target terrain geometric attribute information. This target terrain geometric attribute information may include: Gaussian curvature and rate of change of normal.

[0064] The second sub-step generates a geometry-optimized texture regularization energy function value based on the target terrain geometry information and the enhanced terrain texture map. This geometry-optimized texture regularization energy function value can be an energy function value composed of a curvature texture detail energy term and a 0.5 times normal variation texture edge energy term. The curvature texture detail energy term can be the expectation function of the square of the difference between the magnitude of the texture gradient of the enhanced terrain texture map and the product of the absolute value of the Gaussian curvature included in the target terrain geometry information and a first scaling factor. The first scaling factor can be pre-set and has a value of 0.1. The curvature texture detail energy term can be a constraint function that penalizes overly smooth textures in high-curvature regions or overly cluttered textures in low-curvature regions. The normal variation texture edge energy term can be the expectation function of the square of the difference between the edge strength of the enhanced terrain texture map determined by the Sobel algorithm and the product of the magnitude of the normal variation rate included in the target terrain geometry information and a second scaling factor. The aforementioned normal variation texture edge energy term can constrain the function that penalizes situations such as no textured edges at geometrical corners or false textured edges in smooth areas. The aforementioned second scaling factor can be preset and has a value of 0.2.

[0065] The third sub-step involves determining the texture offset gradient value of the geometrically optimized texture regularization energy function value with respect to the modified color texture offset set corresponding to the enhanced terrain texture map. This texture offset gradient value characterizes the gradient of the geometrically optimized texture regularization energy function value with respect to the modified color texture offset set. In practice, the execution entity can first determine the sum of the geometrically optimized texture regularization energy function value and the square of the modulus of 0.1 times the modified color texture offset set, as the target energy function value. Then, it determines the gradient of the target energy function value with respect to the modified color texture offset set, thus obtaining the texture offset gradient value.

[0066] The fourth sub-step involves updating the enhanced terrain texture map based on the texture offset gradient value, resulting in an updated terrain texture map. As an example, the aforementioned execution entity can use a gradient descent algorithm to update the enhanced terrain texture map based on the texture offset gradient value, thus obtaining the updated terrain texture map.

[0067] Sub-step 3, in response to determining that the number of times the above determination step has been executed is greater than or equal to a preset execution threshold, determines the updated terrain texture map and the updated 3D terrain sub-mesh as a super-resolution terrain texture map and a super-resolution 3D terrain sub-mesh. The preset execution threshold can be a pre-set maximum number of executions. For example, the preset execution threshold can be 5.

[0068] The second step involves, in response to the determination that the number of times the above determination step has been executed is less than a preset execution threshold, identifying the updated terrain texture map and the updated 3D terrain sub-mesh as the enhanced terrain texture map and the target 3D terrain sub-mesh, and determining the sum of the number of executions and a preset value as the number of executions, to execute the above determination step again. The preset value can be a pre-defined value. For example, the preset value can be 1.

[0069] The above technical solution and its related content, combined with "Step 104" and "Step 105" as an inventive point of this disclosure, solve the technical problem mentioned in the background art: "The 3D terrain sub-mesh is difficult to recover complex undulating structures, the terrain texture map extracts insufficient color details or has broken edges, making it difficult to adapt to the 3D construction of large-scale terrain, prolonging the construction time and quality of terrain texture maps and 3D terrain sub-mesh, and thus wasting a lot of storage resources." The factors that lead to prolonged construction time and quality of terrain texture maps and 3D terrain sub-mesh, and wasting a lot of storage resources, are often as follows: The construction of terrain texture maps and 3D terrain sub-mesh is usually performed by separate resolution enhancement optimization. This separate enhancement optimization leads to the 3D terrain sub-mesh being difficult to recover complex undulating structures, the terrain texture map extracting insufficient color details or having broken edges, making it difficult to adapt to the 3D construction of large-scale terrain, prolonging the construction time and quality of terrain texture maps and 3D terrain sub-mesh, and thus wasting a lot of storage resources. If the above factors are addressed, it is possible to achieve 3D construction adaptable to a wide range of terrain features, shorten the construction time and improve the quality of terrain texture maps and 3D terrain sub-mesh, and reduce the waste of storage resources. To achieve this effect, this disclosure firstly optimizes the target 3D terrain sub-mesh iteratively using an enhanced terrain texture map, based on a texture-optimized geometric loss set. The enhanced terrain texture map guides the optimization of the target 3D terrain sub-mesh, ensuring that the rendering result of the target 3D terrain sub-mesh is consistent with the target rendering image in terms of structure and normals. The texture-optimized geometric loss set jointly transforms the geometry-texture inconsistency into an optimizable numerical loss, making the abstract "realism" quantifiable and optimizable. Inverse gradient updates allow the undulations of the geometric surface to actively adapt to the brightness changes of the texture, directly correcting the contradiction between geometry and texture, making the color mapping and texture normals of the rendering image and the target image more accurate and realistic, and improving the quality and resolution of the target 3D terrain sub-mesh. Then, using the target 3D terrain sub-mesh, the enhanced terrain texture map is optimized by geometrically optimizing the texture regularization energy function value. The extracted target terrain geometric attribute information can provide structural priors for texture enhancement, avoiding artifacts caused by indiscriminate enhancement. Geometrically optimizing the texture regularization energy function value can ensure that the distribution of texture details is in sync with the geometric structure, so that high-frequency details of the texture (such as spots and edges) are concentrated in geometrically complex areas, reducing false textures in flat areas and improving the structural matching degree of texture-geometry. Texture map updates can make the brightness and color of the texture transition naturally with the geometric curvature (such as enhanced texture contrast at edges and softer texture gradients in flat areas), correcting the problems of over-sharpening or uneven blurring in texture enhancement, and improving the visual naturalness of the texture.Subsequently, a bidirectional iterative process based on texture and geometry is performed on the above two steps. This solves the local optima problem of unidirectional optimization, improves the quality of the super-resolution terrain texture map and the super-resolution 3D terrain sub-mesh, enhances realism, and models the coupling relationship between the mesh geometry and texture attributes of the terrain, achieving a balance between spatial structure and texture fidelity, and avoiding the geometry-texture misalignment problem of unidirectional optimization. Finally, 3D reconstruction is performed using the super-resolution terrain texture map and the super-resolution 3D terrain sub-mesh to obtain a 3D density texture terrain model, which is then stored, reducing the waste of storage resources.

[0070] Step 104: The obtained multiple super-resolution terrain texture maps and multiple obtained super-resolution 3D terrain sub-mesh are recombined and fused to obtain a 3D textured terrain mesh model.

[0071] In some embodiments, the execution entity may perform model recombination and fusion on the obtained multiple super-resolution terrain texture maps and multiple obtained super-resolution 3D terrain sub-mesh to obtain a 3D textured terrain mesh model. The 3D textured terrain mesh model may be a 3D digital model representing the terrain of a target area, obtained by mapping the texture map obtained from the fusion of multiple super-resolution terrain texture maps to the mesh obtained from the fusion of multiple super-resolution 3D terrain sub-mesh.

[0072] In some optional implementations of certain embodiments, the above-described model recombination and fusion of multiple obtained super-resolution terrain texture maps and multiple obtained super-resolution 3D terrain sub-mesh to obtain a 3D textured terrain mesh model may include the following steps: The first step involves performing global relocation processing on the aforementioned multiple super-resolution terrain texture maps and multiple super-resolution 3D terrain sub-mesh sets, respectively, to obtain a relocated terrain texture map set and a relocated 3D terrain sub-mesh set. The relocated terrain texture maps in the relocated texture map set can be obtained by transforming the local coordinate system to the global coordinate system of the entire region corresponding to the 3D terrain point cloud. Similarly, the relocated 3D terrain sub-mesh sets can be obtained by transforming the local coordinate system to the global coordinate system of the entire region corresponding to the 3D terrain point cloud.

[0073] The second step involves determining the overlapping terrain texture atlas and the overlapping 3D terrain sub-mesh set of the aforementioned relocated terrain texture atlas and relocated 3D terrain sub-mesh set. Specifically, the overlapping terrain texture atlas in the aforementioned overlapping terrain texture atlas can be the texture map of the overlapping region between two adjacent relocated terrain texture maps in the relocated terrain texture atlas. Similarly, the overlapping 3D terrain sub-mesh in the aforementioned overlapping 3D terrain sub-mesh set can be the mesh of the overlapping region between two adjacent overlapping 3D terrain sub-mesh sets.

[0074] The third step involves performing vertex distance-weighted integration on the aforementioned overlapping terrain texture maps to obtain an overlapping integrated texture map set. This overlapping integrated texture map can be a mesh texture map obtained by weighting and integrating two adjacent overlapping terrain texture maps according to the pixel-level distance to the corresponding relocated terrain texture map.

[0075] The fourth step involves weighted merging of the texture attributes of the aforementioned overlapping 3D terrain sub-mesh sets to obtain an overlapping merged 3D terrain mesh set. The overlapping merged 3D terrain meshes in this set can be obtained by weighted merging of two adjacent overlapping 3D terrain sub-mesh sets according to their distances to the corresponding relocated 3D terrain sub-mesh sets. In practice, the execution entity can perform the following merging process for each overlapping 3D terrain sub-mesh in the overlapping 3D terrain sub-mesh set: First, determine the distance from each overlapping 3D terrain point included in the overlapping 3D terrain sub-mesh set to the center 3D terrain point of the corresponding overlapping 3D terrain sub-mesh, obtaining an overlapping distance value set. Then, determine the negative of the ratio of the square of each overlapping distance value in the overlapping distance value set to a preset weight attenuation factor, using this as the weight exponent, obtaining a weight exponent set. Next, normalize the exponential function values ​​with base e and the weight exponent set as the exponent, obtaining a mesh overlapping weight value set. Finally, based on the aforementioned mesh overlapping weight value set, perform weighted merging of the aforementioned overlapping 3D terrain points to obtain the merged 3D terrain mesh set.

[0076] The fifth step involves integrating the non-overlapping terrain texture maps in the aforementioned relocated terrain texture map set to obtain a non-overlapping integrated texture map. This texture map integration can be performed using a texture packing algorithm.

[0077] The sixth step involves merging the non-overlapping 3D terrain sub-grids in the aforementioned relocated 3D terrain sub-grid set to obtain a non-overlapping merged 3D terrain mesh. This non-overlapping merged 3D terrain mesh can be obtained by integrating the vertices, faces, and topological relationships of the non-overlapping 3D terrain sub-grids.

[0078] Step 7: Globally integrate the above-mentioned overlapping and integrated texture map set and the above-mentioned non-overlapping and integrated texture map to obtain a global texture map, and globally integrate the above-mentioned overlapping and merged 3D terrain mesh set and the non-overlapping and merged 3D terrain mesh set to obtain a global 3D terrain mesh.

[0079] Step 8: Map the global texture map to the global 3D terrain mesh to obtain the 3D texture terrain mesh model.

[0080] Step 105: Perform density-controlled reconstruction processing on the three-dimensional textured terrain mesh model to obtain a three-dimensional density-textured terrain model.

[0081] In some embodiments, the execution entity may perform density-controlled reconstruction processing on the three-dimensional textured terrain mesh model to obtain a three-dimensional density-textured terrain model. The three-dimensional density-textured terrain model may be a three-dimensional digital model that meets user requirements, obtained by upsampling the three-dimensional textured terrain mesh model.

[0082] In addressing the aforementioned technical problems in the application scenario of restoring cultural relics and their surrounding terrain, the following technical challenges often arise: Due to the presence of small objects and the precise reproduction of details within the cultural relics and their surrounding terrain, accurate extraction of the 3D point cloud of the relics and their surrounding terrain becomes difficult, resulting in a large amount of redundant and erroneous texture information. This leads to numerous missing and incomplete parts in the 3D density texture terrain model, resulting in low accuracy and wasted storage resources. Based on the following characteristics of this application scenario: the existence of clear linear terrain features, high requirements for terrain detail accuracy, balancing large-scale coverage with local refinement, and the need to preserve the true texture and physical properties of the surface, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the above-described density-controlled reconstruction processing of the three-dimensional textured terrain mesh model to obtain a three-dimensional density-textured terrain model, and the storage of the three-dimensional density-textured terrain model, may include the following steps: The first step involves determining the set of grid geometric vertex curvatures for each grid vertex in the global 3D terrain mesh corresponding to the aforementioned 3D textured terrain mesh model, thus obtaining a set of grid geometric vertex curvature sets. The second step involves determining the texture complexity of each texture pixel in the global texture map corresponding to the aforementioned 3D textured terrain mesh model, thus obtaining a texture complexity set. The aforementioned set of grid geometric vertex curvatures characterizes the degree of curvature of the surface at its vertices. This set of grid geometric vertex curvatures may include principal curvatures and Gaussian curvatures. The texture complexity in the texture complexity set characterizes the richness of detail in the surface texture. This texture complexity can be obtained through the following steps: First, using the Sobel operator, determine the texture gradient value of the global texture map at each pixel, thus obtaining a set of texture gradient values. These texture gradient values ​​characterize the degree of local variation in the global texture map near the pixel. Then, using LBP (Local Binary Pattern), determine the texture attribute value of the global texture map at each pixel, thus obtaining a set of texture attribute values. These texture attribute values ​​characterize the irregularity of the surface texture. Finally, the texture gradient value set and the texture degree attribute value set are averaged to obtain the texture mean set, which is used as the texture complexity set.

[0083] The second step involves determining the vertex sampling probability of each mesh vertex based on the aforementioned set of mesh geometric vertex curvatures and texture complexity sets, thus obtaining a vertex sampling probability set. The vertex sampling probabilities in this set can describe the probability of a mesh vertex being sampled. For example, the execution entity can perform a weighted summation of the aforementioned set of mesh geometric vertex curvatures and texture complexity sets to obtain an initial vertex sampling probability set. Then, the sum of each initial vertex sampling probability and density offset factor in the initial vertex sampling probability set is determined as the vertex sampling probability, resulting in the vertex sampling probability set. The density offset factor can be a pre-defined adjustment term, a constant value used to shift or offset the overall sampling probability. The density offset factor ensures that even when the mesh geometric vertex curvature and texture complexity are both zero, there is still a basic value for the sampling density, helping to ensure a certain sampling density in flat and uniformly textured regions, avoiding information loss due to excessively low sampling density.

[0084] The third step involves performing global-local collaborative sampling on the global 3D terrain mesh based on the aforementioned vertex sampling probability set, to obtain an initial sampled vertex set. This initial sampled vertex set can be obtained by upsampling the global 3D terrain mesh.

[0085] As an example, the aforementioned execution entity can first input the global 3D terrain mesh into a global vertex structure evaluation model to obtain a set of vertex structure importance values. This global vertex structure evaluation model can be a deep neural network model that evaluates the geometric importance of each vertex in the input global 3D terrain mesh. For example, the global vertex structure evaluation model could be a PointNet model. Next, the aforementioned vertex sampling probability set and the aforementioned vertex structure importance value set are multiplied accordingly to obtain a set of sampling target weight values. Then, the ratio of each sampling target weight in the sampling target weight set to the sum of the sampling target weight sets is determined as the target sampling probability set. Next, using a farthest point sampling algorithm, the global 3D terrain mesh is initially sampled according to the aforementioned target sampling probability set to obtain a sampled terrain vertex set. Subsequently, the sampled terrain vertex set is input into a hierarchical attention network to obtain local feature information of each sampled terrain vertex in its local neighborhood. This local neighborhood can be a region determined by the k-nearest neighbor algorithm. Finally, the obtained sets of multiple local feature information are input into a sampling classifier to obtain a vertex detail criticality value set. Finally, in response to the determination that at least one vertex detail critical value in the aforementioned vertex detail critical value set is greater than or equal to a preset detail criticality threshold, the sampled terrain vertices are supplemented by sampling to obtain a supplemented sampled vertex set. The preset detail criticality threshold can be a pre-defined threshold for determining whether resampling is necessary, and will not be elaborated further here. Finally, the supplemented sampled vertex set and the sampled terrain vertex set are used as the initial sampled vertex set.

[0086] The fourth step involves performing multi-resolution hybrid sampling optimization on the initial sampled vertex set to obtain a terrain sampled vertex set. This terrain sampled vertex set can be obtained by performing layered sampling on the initial sampled vertex set. In practice, the execution entity can input the initial sampled vertex set into a multi-resolution pyramid for layered sampling to obtain a layered sampling optimized vertex set.

[0087] The fifth step involves generating a set of terrain constraint segments based on the aforementioned set of terrain sampling vertices. These terrain constraint segments can be segments that play a crucial role in shaping the topography of the target area. For example, the set of terrain constraint segments may include, but is not limited to, at least one of the following: ridgelines, river centerlines, and road boundary lines. As an example, the executing entity can first determine the elevation gradient set and vertex curvature set of the aforementioned set of terrain sampling vertices. Then, it can select at least one terrain sampling vertex from the elevation gradient set and vertex curvature set that is greater than or equal to a preset elevation gradient threshold and vertex curvature threshold. Afterward, it can perform line segment fitting on at least one terrain sampling vertex to obtain the set of terrain constraint segments.

[0088] Step 6: Based on the above set of terrain constraint segments, perform the following tetrahedral surface construction steps: Sub-step 1: Generate a set of target constraint segments based on the set of terrain constraint segments. The target constraint segments in the target constraint segment set can be segments obtained by breaking at the midpoint of at least one terrain constraint segment whose length is less than or equal to the radius confidence value and greater than the radius confidence value. The radius confidence value can be the product of the radius of the terrain constraint segment and a preset threshold. The preset threshold can be a pre-defined value. For example, the preset threshold can be 0.5.

[0089] As an example, the aforementioned executing entity may first, in response to determining that there are geomorphic constraint segments in the set of geomorphic constraint segments whose lengths are greater than the radius confidence value, divide each geomorphic constraint segment in the set of geomorphic constraint segments corresponding to at least one constraint segment length greater than the radius confidence value by midpoint division, thereby obtaining a midpoint constraint segment set. Then, the midpoint constraint segment set and each geomorphic constraint segment with a radius confidence value less than or equal to the radius confidence value are determined as the target constraint segment set.

[0090] Sub-step 2 involves constructing tetrahedrons from the target constraint line segment set to obtain an initial terrain tetrahedron set. This initial terrain tetrahedron set can be a set of non-overlapping tetrahedrons formed from the three-dimensional endpoints included in the target constraint line segment set. The tetrahedron construction can be performed using the Delaunay tetrahedronization algorithm.

[0091] Sub-step 3: Based on the initial set of landform tetrahedrons and the aforementioned set of target constraint segments, generate a target landform tetrahedron set. This target landform tetrahedron set can be a tetrahedron obtained by optimizing the initial set of landform tetrahedrons. As an example, the executing entity can first, in response to determining that at least one target constraint segment exists in the target constraint segment set that is not included in the initial landform tetrahedron set, add a 3D point cloud to the midpoint of each target constraint segment in the at least one target constraint segment, obtaining a first newly added 3D point cloud set. Secondly, for each initial landform tetrahedron in the initial landform tetrahedron set, compare the radius-to-side-length ratio of the initial landform tetrahedron with a preset ratio threshold to obtain a comparison result. This radius-to-side-length ratio can be the ratio of the radius of the circumscribed sphere corresponding to the initial landform tetrahedron to the target side length. The target side length can be the minimum side length of the initial landform tetrahedron. The preset ratio threshold can be a pre-set maximum value of the ratio. For example, the preset ratio threshold can be 2. Then, in response to determining that at least one comparison result in the obtained comparison result set has a radius-to-side-length ratio greater than the aforementioned preset ratio threshold, point cloud data is added to the circumcenter of the sphere of at least one initial terrain tetrahedron corresponding to at least one comparison result, resulting in a second newly added 3D point cloud set. Finally, tetrahedrons are constructed from the target newly added point cloud dataset and the 3D endpoint set to obtain the target point cloud tetrahedron set. The aforementioned target newly added point cloud dataset can be a first newly added 3D point cloud set, a second newly added 3D point cloud set, or a point cloud set composed of the first and second newly added 3D point cloud sets.

[0092] Sub-step 4: In response to determining that the target constraint segment set and the target terrain tetrahedron set satisfy a preset constraint condition set, the target terrain tetrahedron set is subjected to mesh surface segmentation processing to obtain a terrain triangular mesh. The preset constraint condition set may include: each target constraint segment in the target constraint segment set is located within the target terrain tetrahedron set; the radius-to-side-length ratio of the target terrain tetrahedron is less than a preset ratio threshold; and the length of the target constraint segment is less than the radius confidence value. In practice, the execution entity can use the maximum flow minimum cut algorithm to perform mesh surface segmentation processing on the target terrain tetrahedron set to obtain a terrain triangular mesh.

[0093] Step 7: In response to the determination that the target constraint line segment set and the target point cloud tetrahedron set do not meet the preset constraint condition set, the target constraint line segment set is determined as the terrain constraint line segment set, so as to execute the above tetrahedron surface construction step again.

[0094] Step 8: Perform texture mapping on the above-mentioned terrain triangular mesh and the above-mentioned global texture map to obtain a three-dimensional density texture terrain model, and store the above-mentioned three-dimensional density texture terrain model.

[0095] The above-described technical solution and its related content, as an inventive point of this disclosure, solve the technical problem mentioned in the background art: "The 3D density texture terrain model contains a large number of missing and incomplete parts, resulting in low accuracy and wasting a large amount of storage resources." The factors leading to low accuracy and wasted storage resources in the 3D density texture terrain model are often as follows: Due to the difficulty in accurately reproducing the details of small objects and artifacts in the restoration of cultural relics and their surrounding terrain, it is difficult to accurately extract the 3D point cloud of cultural relics and their surrounding terrain, resulting in a large amount of redundant and erroneous texture information. This leads to a large number of missing and incomplete parts in the 3D density texture terrain model, resulting in low accuracy and wasted storage resources. Solving these factors can reduce the number of missing and incomplete parts in the 3D density texture terrain model, improve its accuracy, and reduce the waste of storage resources. To achieve this effect, this disclosure first optimizes global-local collaborative sampling and multi-resolution hybrid sampling by using a set of mesh geometric vertex curvatures and a set of texture complexity. This improves sampling accuracy, allowing the collected vertices to accurately represent the topographic features of the target region with a minimum number of vertices, effectively reducing the occurrence of missing or incomplete areas. Second, it generates a set of terrain constraint segments and performs midpoint partitioning on these segments. This set accurately represents the topographic information of the target region, and midpoint partitioning ensures that all subsequently generated target constraint segments are included within the generated tetrahedron. Next, tetrahedron construction is performed on the target constraint segments to better describe the spatial distribution and geometric features of the point cloud data, facilitating subsequent 3D surface reconstruction. Finally, determining whether all target constraint segments are included within the initial terrain tetrahedron improves the integrity of the generated tetrahedron, reduces gaps and breaks in the generated tetrahedron set, and enhances the overall integrity and stability of the tetrahedron. Then, when the target constraint line segment set is not included in the initial terrain tetrahedron set and the radius-to-side-length ratio is greater than a preset threshold, a new 3D point cloud set is added to optimize the tetrahedron shape and quality of the target terrain tetrahedron. Finally, the loop ends, and the target terrain tetrahedron set is meshed to obtain a terrain triangular mesh. Texture mapping is then performed on the terrain triangular mesh and the global texture map to obtain a 3D density-textured terrain model, which improves the shape and quality of the generated tetrahedrons and the accuracy of the generated 3D density-textured terrain model. Finally, the 3D density-textured terrain model is stored to reduce storage resource waste.

[0096] The various embodiments of this disclosure have the following beneficial effects: The 3D terrain model reconstruction method based on geometry and texture in some embodiments of this disclosure can improve the quality of 3D density texture terrain models and reduce the waste of display resources. Specifically, the reason why road markings in related maps are not accurate is that: when generating terrain texture information and performing texture super-resolution through a multi-view interactive Transformer model, the geometric structure information of the terrain in the color 3D terrain model is not considered, resulting in texture distortion problems such as distortion and breakage in the super-resolution texture map. The quality of the super-resolution texture map and the 3D texture terrain model is poor, and storage resources are wasted. Based on this, the 3D terrain model reconstruction method based on geometry and texture in some embodiments of this disclosure can first obtain the 3D point cloud of the terrain of the target area. Here, the 3D point cloud of terrain is used for subsequent adaptive meshing. Secondly, the above-mentioned 3D point cloud of terrain is adaptively meshed to obtain a 3D terrain sub-mesh set. Here, adaptive meshing based on the terrain corresponding to the 3D point cloud of terrain can improve the rationality of the 3D terrain sub-mesh and reduce the amount of data and the level of detail in subsequent processing. Next, for each 3D terrain sub-mesh in the aforementioned 3D terrain sub-mesh set, the following collaborative bidirectional optimization steps are performed: First, geometric super-resolution processing is performed on the aforementioned 3D terrain sub-mesh to obtain the target 3D terrain sub-mesh. This improves the representation of geometric details and topological relationships of the terrain in the target 3D terrain sub-mesh, making it closer to the real terrain. Second, color texture mapping processing is performed on the aforementioned target 3D terrain sub-mesh to obtain a terrain color texture attribute map. This reduces the mapping error from 3D to 2D and improves the texture quality of the terrain color texture attribute map. Third, based on the aforementioned target 3D terrain sub-mesh, texture correction and enhancement processing is performed on the aforementioned terrain color texture attribute map to obtain an enhanced terrain texture map. Further improving the quality of the enhanced terrain texture map effectively avoids texture distortion and improves the matching degree between texture and geometry. Fourth, collaborative bidirectional optimization is performed on the aforementioned enhanced terrain texture map and the aforementioned target 3D terrain sub-mesh to obtain a super-resolution terrain texture map and a super-resolution 3D terrain sub-mesh. Here, the coupling relationship between texture and geometry is considered, and a balanced co-optimization approach achieves a balance between spatial structure and texture fidelity. This effectively avoids misalignment issues between geometric textures and improves the quality of the super-resolution terrain texture map and the super-resolution 3D terrain sub-mesh. Then, the obtained multiple super-resolution terrain texture maps and multiple obtained super-resolution 3D terrain sub-meshes are modeled and fused to obtain a 3D textured terrain mesh model. This eliminates boundary seams and conflicts in the overlapping parts of the 3D terrain sub-mesh combination, improving the visual consistency of the 3D textured terrain mesh model.Finally, density-controlled reconstruction processing is performed on the aforementioned 3D textured terrain mesh model to obtain a 3D density-textured terrain model, which is then visualized. Here, density control improves the global uniformity of the 3D density-textured terrain model and maximizes the preservation of the macroscopic geometric structure and texture details of the target area's terrain, thereby improving the accuracy of 3D reconstruction, enhancing the quality of the 3D density-textured terrain model, and reducing storage resource waste. Therefore, this geometry-and-texture joint 3D terrain model reconstruction method can improve the quality of the 3D density-textured terrain model and reduce storage resource waste.

[0097] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a three-dimensional terrain model reconstruction device based on a combination of geometry and texture. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this geometry-and-texture combined 3D terrain model reconstruction device can be specifically applied to various electronic devices.

[0098] like Figure 2 As shown, a 3D terrain model reconstruction device 200 based on a combination of geometry and texture includes: an acquisition unit 201, an adaptive mesh generation unit 202, an execution unit 203, a model reconstruction and fusion unit 204, and a density-controlled reconstruction unit 205. The acquisition unit 201 is configured to acquire a 3D point cloud of terrain features in the target area. The adaptive mesh generation unit 202 is configured to adaptively mesh the aforementioned 3D point cloud of terrain features to obtain a 3D terrain sub-mesh set. Execution unit 203 is configured to perform the following collaborative bidirectional optimization steps for each of the three-dimensional terrain sub-mesh sets: perform geometric super-resolution processing on the three-dimensional terrain sub-mesh to obtain a target three-dimensional terrain sub-mesh; perform color texture mapping processing on the target three-dimensional terrain sub-mesh to obtain a terrain color texture attribute map; perform texture correction and enhancement processing on the terrain color texture attribute map based on the target three-dimensional terrain sub-mesh to obtain an enhanced terrain texture map; and perform collaborative bidirectional optimization on the enhanced terrain texture map and the target three-dimensional terrain sub-mesh to obtain a super-resolution terrain texture map and a super-resolution three-dimensional terrain sub-mesh. Model recombination and fusion unit 204 is configured to perform model recombination and fusion on the obtained multiple super-resolution terrain texture maps and multiple obtained super-resolution three-dimensional terrain sub-mesh to obtain a three-dimensional textured terrain mesh model. Density control reconstruction unit 205 is configured to perform density control reconstruction processing on the above three-dimensional textured terrain mesh model to obtain a three-dimensional density textured terrain model, and store the above three-dimensional density textured terrain model as a terrain model.

[0099] It is understandable that the units and references recorded in the three-dimensional terrain model reconstruction device 200 based on the combination of geometry and texture are... Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the geometry-texture combined 3D terrain model reconstruction device 200 and the units contained therein, and will not be repeated here.

[0100] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device (e.g., an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0101] like Figure 3 As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0102] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0103] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0104] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0105] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0106] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: acquire a 3D point cloud of terrain features corresponding to the terrain features of the target area; adaptively mesh the aforementioned 3D point cloud of terrain features to obtain a set of 3D terrain sub-mesh; for each 3D terrain sub-mesh in the aforementioned 3D terrain sub-mesh set, perform the following cooperative bidirectional optimization steps: perform geometric super-resolution processing on the aforementioned 3D terrain sub-mesh to obtain a target 3D terrain sub-mesh; perform color texture mapping processing on the aforementioned target 3D terrain sub-mesh to obtain a terrain color texture attribute map; and, according to the aforementioned target... A 3D terrain submesh is defined, and the terrain color texture attribute map is subjected to texture correction and enhancement processing to obtain an enhanced terrain texture map. The enhanced terrain texture map and the target 3D terrain submesh are then subjected to collaborative bidirectional optimization to obtain a super-resolution terrain texture map and a super-resolution 3D terrain submesh. The multiple super-resolution terrain texture maps and multiple super-resolution 3D terrain submeshes are then combined and fused to obtain a 3D textured terrain mesh model. The 3D textured terrain mesh model is then subjected to density-controlled reconstruction processing to obtain a 3D density textured terrain model, and the 3D density textured terrain model is then stored.

[0107] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0108] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0109] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition unit, an adaptive mesh generation unit, an execution unit, a model reorganization and fusion unit, and a density-controlled reconstruction unit. The names of these units do not necessarily limit the specific unit; for example, an acquisition unit may also be described as "a unit that acquires topographic information of a target area."

[0110] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0111] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for reconstructing a three-dimensional terrain model based on joint geometry and texture, comprising: obtaining a three-dimensional point cloud of a target region; performing adaptive grid division on the three-dimensional point cloud to obtain a set of three-dimensional terrain sub-grids; for each three-dimensional terrain sub-grid in the set of three-dimensional terrain sub-grids, performing the following steps of cooperative bidirectional optimization: performing geometric super-resolution processing on the three-dimensional terrain sub-grid to obtain a target three-dimensional terrain sub-grid; performing color texture mapping processing on the target three-dimensional terrain sub-grid to obtain a terrain color texture attribute map; performing texture correction and enhancement processing on the terrain color texture attribute map based on the target three-dimensional terrain sub-grid to obtain an enhanced terrain texture map; performing cooperative bidirectional optimization on the enhanced terrain texture map and the target three-dimensional terrain sub-grid to obtain a super-resolution terrain texture map and a super-resolution three-dimensional terrain sub-grid; performing model recombination and fusion on the obtained super-resolution terrain texture maps and the obtained super-resolution three-dimensional terrain sub-grids to obtain a three-dimensional texture terrain grid model; performing density control reconstruction processing on the three-dimensional texture terrain grid model to obtain a three-dimensional density texture terrain model, and storing the three-dimensional density texture terrain model.

2. The method of claim 1, wherein, The adaptive grid division on the three-dimensional point cloud to obtain a set of three-dimensional terrain sub-grids comprises: performing filtering and denoising processing on the three-dimensional point cloud to obtain a denoised three-dimensional point cloud; performing point cloud coordinate normalization processing on the denoised three-dimensional point cloud to obtain a normalized three-dimensional point cloud; determining local density information of each three-dimensional terrain point in the normalized three-dimensional point cloud to obtain a set of point cloud local density information; performing adaptive window division on the normalized three-dimensional point cloud based on the set of point cloud local density information to obtain a set of window center terrain points; performing three-dimensional grid reconstruction on the set of window center terrain points to obtain a three-dimensional terrain surface grid model; performing grid overlap division on the three-dimensional terrain surface grid model to obtain a three-dimensional terrain surface grid sub-model as a set of three-dimensional terrain sub-grids.

3. The method of claim 1, wherein, The geometric super-resolution processing on the three-dimensional terrain sub-grid to obtain a target three-dimensional terrain sub-grid comprises: performing graph convolution feature extraction on the three-dimensional terrain sub-grid to obtain a grid terrain geometry feature vector; performing grid geometry refinement on the three-dimensional terrain sub-grid based on the grid terrain geometry feature vector to obtain an initial terrain geometry sub-grid; performing structure optimization processing on the initial terrain geometry sub-grid based on the grid terrain geometry feature vector using a grid structure super-resolution model to obtain a structure-optimized three-dimensional terrain sub-grid; performing topology correction processing on the structure-optimized three-dimensional terrain sub-grid to obtain a topology terrain sub-grid as a target three-dimensional terrain sub-grid.

4. The method of claim 1, wherein, The color texture mapping processing on the target three-dimensional terrain sub-grid to obtain a terrain color texture attribute map comprises: determining a grid edge curvature weight of each grid edge included in the target three-dimensional terrain sub-grid to obtain a set of grid edge curvature weights; determine a set of surface stretch curves for the target three-dimensional terrain sub-grid according to the set of mesh edge curvature weights; perform mesh splitting on the target three-dimensional terrain sub-grid according to the set of surface stretch curves to obtain a set of terrain mesh blocks; perform two-dimensional mapping processing on the target three-dimensional terrain sub-grid according to the set of terrain mesh blocks to obtain a terrain two-dimensional texture map; perform matching processing on the target three-dimensional terrain sub-grid and the terrain two-dimensional texture map to obtain a set of matching information; map color attribute information included in the terrain three-dimensional point cloud to the terrain two-dimensional texture map according to the set of matching information to obtain a mapped terrain two-dimensional texture map; perform blank area filling processing on the mapped terrain two-dimensional texture map to obtain a terrain color texture attribute map.

5. The method of claim 1, wherein, the texture correction enhancement processing on the terrain color texture attribute map according to the target three-dimensional terrain sub-grid to obtain an enhanced terrain texture map, comprising: perform geometric feature extraction on the target three-dimensional terrain sub-grid to obtain a set of mesh three-dimensional geometric feature vectors; perform two-dimensional texture extraction on the terrain color texture attribute map to obtain a set of two-dimensional texture feature vectors; perform feature alignment splicing on the set of mesh three-dimensional geometric feature vectors and the set of two-dimensional texture feature vectors to obtain a set of geometric texture joint splicing feature maps; input the set of geometric texture joint splicing feature maps into a color texture deviation prediction model to obtain a set of color texture offset amounts; perform constraint correction processing on the set of color texture offset amounts to obtain a set of corrected color texture offset amounts; perform residual fusion on the set of corrected color texture offset amounts and the terrain color texture attribute map to obtain a fused terrain color texture attribute map; perform post-processing optimization on the fused terrain color texture attribute map to obtain an optimized terrain texture map as the enhanced terrain texture map.

6. The method of claim 1, wherein, the model reorganization fusion on the obtained multiple super-resolution terrain texture maps and the obtained multiple super-resolution three-dimensional terrain sub-grids to obtain a three-dimensional textured terrain mesh model, comprising: perform global position repositioning processing on the multiple super-resolution terrain texture maps and the multiple super-resolution three-dimensional terrain sub-grids respectively to obtain a set of repositioned terrain texture maps and a set of repositioned three-dimensional terrain sub-grids; determine a set of overlapping terrain texture maps and a set of overlapping three-dimensional terrain sub-grids of the set of repositioned terrain texture maps and the set of repositioned three-dimensional terrain sub-grids; perform vertex distance weighted integration on the set of overlapping terrain texture maps to obtain a set of overlapping integrated texture maps; perform texture attribute weighted merging on the set of overlapping three-dimensional terrain sub-grids to obtain a set of overlapping merged three-dimensional terrain meshes; perform texture map integration on a set of non-overlapping terrain texture maps in the set of repositioned terrain texture maps to obtain non-overlapping integrated texture maps; perform mesh merging on a set of non-overlapping three-dimensional terrain sub-grids in the set of repositioned three-dimensional terrain sub-grids to obtain non-overlapping merged three-dimensional terrain meshes; performing global integration on the overlapped integrated texture atlas and the non-overlapped integrated texture atlas to obtain a global texture atlas, and performing global integration on the overlapped merged three-dimensional terrain mesh set and the non-overlapped merged three-dimensional terrain mesh to obtain a global three-dimensional terrain mesh; mapping the global texture atlas to the global three-dimensional terrain mesh to obtain a three-dimensional texture terrain mesh model.

7. A three-dimensional terrain model reconstruction device based on joint geometry and texture, comprising: an acquisition unit configured to acquire a three-dimensional point cloud of a terrain of a target region; an adaptive mesh division unit configured to perform adaptive mesh division on the three-dimensional point cloud of the terrain to obtain a three-dimensional terrain sub-mesh set; a performing unit configured to, for each three-dimensional terrain sub-mesh in the three-dimensional terrain sub-mesh set, perform the following steps of cooperative bidirectional optimization: performing geometry super-resolution processing on the three-dimensional terrain sub-mesh to obtain a target three-dimensional terrain sub-mesh; performing color texture mapping processing on the target three-dimensional terrain sub-mesh to obtain a terrain color texture attribute map; performing texture correction enhancement processing on the terrain color texture attribute map according to the target three-dimensional terrain sub-mesh to obtain an enhanced terrain texture map; performing cooperative bidirectional optimization on the enhanced terrain texture map and the target three-dimensional terrain sub-mesh to obtain a super-resolution terrain texture map and a super-resolution three-dimensional terrain sub-mesh; a model reorganization fusion unit configured to perform model reorganization fusion on the obtained plurality of super-resolution terrain texture maps and the obtained plurality of super-resolution three-dimensional terrain sub-meshes to obtain a three-dimensional texture terrain mesh model; a density control reconstruction unit configured to perform density control reconstruction processing on the three-dimensional texture terrain mesh model to obtain a three-dimensional density texture terrain model, and to store the three-dimensional density texture terrain model.

8. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-6.

9. A computer readable medium having stored thereon a computer program, wherein, The computer program is executed by the processor to implement the method of any one of claims 1-6. The computer program is executed by the processor to implement the method of any one of claims 1-6.

Images