Method for multi-source GIS data fusion and three-dimensional terrain construction based on flow mode

Abstract

The application provides a multi-source GIS data fusion and three-dimensional terrain construction method based on a stream mode, relates to the technical field of terrain construction, and comprises the following steps: collecting multi-source GIS data, fusing the multi-source GIS data according to a stream data processing framework to obtain terrain basic data, and generating three-dimensional terrain data. A three-dimensional terrain model is constructed according to a top layer reconstruction technology and the three-dimensional terrain data, and an optimized terrain model is obtained by repairing geometric holes based on Poisson surface reconstruction and an improved ICP algorithm. The optimized terrain model is published to a GIS platform and is adapted to a mobile terminal, and is subjected to lightweight processing through edge computing and cross-platform deployment. The application solves the defects of traditional methods in efficiency, accuracy and real-time performance, improves the data fusion efficiency, and the constructed three-dimensional terrain model has high integrity, high realism and high efficient deployment capability.

Description

Technical Field

[0001] This invention relates to the field of terrain construction technology, and in particular to a method for multi-source GIS data fusion and three-dimensional terrain construction based on a streaming approach. Background Technology

[0002] In Geographic Information System (GIS) applications, the construction of 3D terrain models relies on the effective fusion of multi-source data. Traditional GIS data sources are wide-ranging, covering satellite remote sensing images, UAV aerial survey data, lidar point clouds, BIM models, etc. These data have significant differences in accuracy, coordinate system, and resolution, resulting in a serious data "island" phenomenon and making fusion difficult.

[0003] The demand for streaming data processing is becoming increasingly urgent. In dynamic scenarios such as disaster early warning and construction monitoring, real-time fusion of sensor data streams is required, but traditional batch processing methods cannot meet real-time requirements due to their excessive time consumption. Meanwhile, the spatiotemporal alignment and standardization issues of heterogeneous data streams are prominent, such as mismatched timestamps and inconsistent spatial references among multi-source data, leading to deviations in the fusion results. The computational load of large-scale point cloud and image data has also become a bottleneck, necessitating algorithm optimization to reduce processing pressure.

[0004] In multi-scale modeling, the challenge of smoothly transitioning data at different resolutions has not been fully resolved, easily leading to visual defects such as "terrain jumps." Dynamic update requirements (such as those for water conservancy and urban construction) necessitate models that support incremental fusion, while traditional global recalculation methods are inefficient. Furthermore, the interactive analysis needs of industries such as power and geology place higher demands on model accuracy, realism, and real-time visualization, requiring a balance between geometric integrity and rendering efficiency.

[0005] Cross-domain applications are further driving technological innovation. Disaster early warning requires dynamic simulation support, engineering construction relies on real-time earthwork calculation, and smart cities require digital twin modeling. These scenarios all demand efficient fusion of multi-source data and accurate construction of 3D models, which traditional methods can no longer meet. Therefore, this paper proposes a streaming-based method for multi-source GIS data fusion and 3D terrain construction to address these shortcomings. Summary of the Invention

[0006] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, which solves the problems of traditional GIS data fusion being difficult, computational efficiency being poor, and terrain construction accuracy and realism being poor in the existing technology.

[0007] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, including:

[0008] Collect multi-source GIS data, which includes topographic data, tilt data, image data, and elevation data.

[0009] Based on the streaming data processing architecture, multi-source GIS data are fused to obtain basic terrain data, and three-dimensional terrain data with influence texture and elevation relief information are generated based on the basic terrain data.

[0010] A three-dimensional terrain model is constructed based on top-level reconstruction technology and three-dimensional terrain data, and an optimized terrain model is obtained by repairing geometric voids based on Poisson surface reconstruction and improved ICP algorithm.

[0011] Optimize the terrain model, publish it to the GIS platform and adapt it to mobile devices, and use edge computing to perform lightweight processing and cross-platform deployment.

[0012] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, the steps of which include:

[0013] The system performs format conversion, coordinate system 1 and resolution adaptation on multi-source GIS data, and sets a sliding time window for data whose temporal characteristics exceed the threshold to extract related data within the same spatiotemporal range.

[0014] An R-tree index is used to build an index structure for related data, quickly matching spatially overlapping areas of different data sources and aligning them in the spatiotemporal dimension.

[0015] For high-density laser point clouds and image data exceeding the preset size, layered sampling is performed according to spatial hierarchy to obtain the detail level of different data blocks, and hole repair is performed using the neighborhood difference method.

[0016] Based on terrain relief and line-of-sight parameters, a dynamic LOD strategy is obtained by adjusting each level of detail in real time through a dynamic LOD optimization algorithm. Multiple data blocks are aligned and feature sets are fused to obtain a basic terrain database.

[0017] An incremental computing model is used to perform real-time fusion processing on newly added and changed data from different dynamic scenarios, and then upload the data to the terrain database to obtain basic terrain data.

[0018] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, and the steps to obtain a dynamic LOD strategy include:

[0019] The terrain data is divided into spatial grids to generate multi-level LOD models, and the terrain undulation of each spatial grid is calculated.

[0020] The line-of-sight distance is calculated based on the camera position and the coordinates of the center of the spatial grid.

[0021] The screen error is obtained by determining the impact of terrain relief on the screen at the current viewing distance and introducing a relief correction coefficient.

[0022] The error threshold for each LOD model is set according to the level, and the corresponding level threshold is obtained. The LOD model that satisfies the screen error not being greater than the level threshold is selected as the optimal LOD level.

[0023] When the absolute value of the change in LOD level between adjacent frames is not less than two, linear interpolation is used for transition.

[0024] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, and the steps for real-time fusion processing of newly added and changed data include:

[0025] Spatial change detection and attribute change detection are performed on newly added and changed data to obtain the bounding box of the changed region and the changed data type.

[0026] The newly added and changed data undergoes format conversion and coordinate correction, and is then aligned in time and space.

[0027] The corresponding fusion algorithm is used to perform fusion based on the data type of the change, and the historical fusion results of the changed area are updated based on the bounding box.

[0028] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, wherein the steps for generating 3D terrain data include:

[0029] Elevation data blocks and image texture data are extracted from the terrain baseline data and spatial reference consistency processing is performed.

[0030] Using the spatial range of elevation data blocks, crop the image texture data to the same range to remove redundant parts.

[0031] Elevation data blocks are converted into corresponding triangular meshes based on different structured data, and then optimized accordingly based on the hole type.

[0032] Texture coordinates are calculated using affine transformation formulas and assigned to each vertex of the triangular mesh.

[0033] Image texture data is attached to the surface of a triangular mesh, and the texture coordinates are spatially aligned with the image pixels through graphics rendering to obtain three-dimensional terrain data.

[0034] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, wherein the steps for constructing a 3D terrain model include:

[0035] Invalid value removal and data cropping are performed on the elevation data blocks and image texture data that have undergone spatial reference consistency processing in the software tool, and the overlapping effects of multiple images are stitched together according to the color equalization algorithm to obtain preprocessed data.

[0036] Using the geographic coordinate system as a reference, a three-dimensional coordinate system is created in the software tool, and the preprocessed data is mapped according to the coordinate relationship to form a basic terrain model.

[0037] The terrain files are created using software tools and the terrain data is stored at the optimal LOD level. The number of rendering requests is reduced by merging adjacent low-precision data blocks using top-level reconstruction technology.

[0038] Image data is applied to the surface of the base terrain model using texture mapping technology, and combined with a dynamic LOD strategy, the matching texture resolution is automatically switched under different viewpoints.

[0039] Edge detection is used to identify the boundaries and attributes of typical land features from image data to obtain land feature characteristics, which are then fused with the basic terrain model to obtain a three-dimensional terrain model.

[0040] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, the steps for obtaining ground feature characteristics include:

[0041] The impact data is subjected to contrast stretching and image noise removal processing, and appropriate band combinations are selected for different types of impact.

[0042] For different features of land cover, select the corresponding edge detection algorithm to obtain the detection results, and perform threshold segmentation to binarize edge pixels and non-edge pixels.

[0043] The Hough transform is used to detect straight lines and curves at discrete edge points, connecting broken edge segments, and then eliminating and filling them through morphological operations to obtain the boundary contours of ground features.

[0044] Extract image features within the boundary contour area of ​​land features to distinguish land feature types.

[0045] The feature of a land feature is obtained by associating its boundary outline with its corresponding attribute information.

[0046] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, and the steps for Poisson surface reconstruction include:

[0047] High curvature vertices are extracted from the cavity boundary as feature points, and a local point cloud dataset is constructed by combining it with the point cloud data of the surrounding complete area. Then, noise reduction and downsampling are performed.

[0048] The implicit surface function is obtained by solving the Poisson equation using the normal vector of the local point cloud dataset as a constraint.

[0049] Extract isosurfaces that match the curvature of the surrounding mesh from the implicit surface function, generate meshes to fill holes, and then apply Laplacian smoothing to the reconstructed surface.

[0050] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on streaming, and the formula for Laplace smoothing is expressed as follows:

[0051]

[0052] In the formula, It is the vertex The set of neighboring vertices, It is the number of vertices in the domain. It is a smoothing factor. It is the first vertex in the neighborhood vertex set. vertex coordinates It is the smoothed first The new coordinates of each vertex.

[0053] This invention provides a method for multi-source GIS data fusion and 3D terrain construction based on a streaming approach. The steps for repairing geometric holes using an improved ICP algorithm include:

[0054] The point cloud surrounding the cavity repair area is selected as the source point cloud, and the vertices of the repaired triangular mesh surface are selected as the target points. Initial alignment is performed through coarse registration.

[0055] Point weights are introduced for key feature points in the source point cloud to resolve outlier interference.

[0056] The convergence conditions are dynamically adjusted based on the current registration error, and normal vector constraints are introduced to ensure that the direction of the vertex normal vector is consistent with the surrounding terrain trend during the registration process.

[0057] The process involves iteratively calculating the optimal transformation matrix between the source point cloud and the target point until the deviation between the vertex coordinates of the repaired area and the original point cloud is within a preset threshold.

[0058] This invention provides a streaming-based method for multi-source GIS data fusion and 3D terrain construction. It processes multi-source GIS data and sets a sliding time window for data whose temporal characteristics exceed a threshold, extracting related data within the same spatiotemporal range. This solves the inconsistencies in format, coordinate system, resolution, and time reference of multi-source data, eliminating data heterogeneity and achieving seamless fusion of multi-source data. This improves data usability and consistency, providing high-quality foundational data for subsequent 3D terrain construction. An R-tree index is used to establish an index structure for related data, quickly matching spatially overlapping areas from different data sources and aligning them in the spatiotemporal dimension. This solves the problem of misalignment in the spatiotemporal dimension of multi-source data, especially in dynamic scenes where inconsistent spatiotemporal references can lead to model misalignment or accuracy loss. This improves the accuracy and efficiency of data fusion, ensuring the accuracy and reliability of the 3D terrain model.

[0059] This invention provides a streaming-based multi-source GIS data fusion and 3D terrain construction method. For high-density laser point clouds and image data exceeding a preset size, it performs layered sampling according to spatial hierarchy to obtain detailed levels of different data blocks. It then employs a neighborhood interpolation method for hole repair, solving the problem of excessive computational load when processing large-scale point clouds or image data. This avoids processing delays caused by large data volumes, improves data processing efficiency, and reduces computational resource consumption. Furthermore, it repairs geometric holes through Poisson surface reconstruction and an improved ICP algorithm, enhancing the geometric integrity of complex features. This addresses issues in traditional methods such as unnatural terrain and feature fusion, difficulty in multi-scale representation, and interpolation errors due to missing data. It improves the accuracy and realism of the 3D terrain model, making it more lifelike and suitable for high-precision applications. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 This is one of the flowcharts illustrating the multi-source GIS data fusion and 3D terrain construction method based on streaming provided in this embodiment of the invention;

[0062] Figure 2 This is a data processing diagram of the multi-source GIS data fusion and 3D terrain construction method based on streaming provided in this embodiment of the invention;

[0063] Figure 3 This is one of the principle diagrams for creating three-dimensional terrain in the multi-source GIS data fusion and three-dimensional terrain construction method based on streaming provided in this embodiment of the invention;

[0064] Figure 4 This is the second schematic diagram of the principle of three-dimensional terrain creation in the multi-source GIS data fusion and three-dimensional terrain construction method based on the flow method provided in the embodiments of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0066] The following is combined Figures 1-4 This invention describes a method for multi-source GIS data fusion and 3D terrain construction based on a streaming approach.

[0067] like Figure 1 As shown, the multi-source GIS data fusion and 3D terrain construction method based on streaming provided by this embodiment of the invention includes:

[0068] Collect multi-source GIS data, which includes topographic data, tilt data, image data, and elevation data.

[0069] The data processing task of this invention is enormous and involves a large amount of three-dimensional data, thus placing certain demands on the computing and display performance of the computer.

[0070] Hardware environment:

[0071] CPU: Xeon MP2 200MHz or higher.

[0072] Processors: 8 or more.

[0073] Memory: DDR16GB or higher.

[0074] Hard drive: Requires at least 5TB of space.

[0075] Graphics card: It is recommended that the monitor have 2GB or more of video memory.

[0076] Software environment:

[0077] Operating system: Windows, Windows Server, Win10 recommended.

[0078] GIS platform software: SkylineGlobe products (TerraExplorerpro, TerraBuilder, SkylineGlobeServer), ArcGIS products (ArcGISdesktop), cesiumlab.

[0079] Web application software: Tomcat 7.0.14 (or higher), IIS 7.0 (for SkylineGlobeServer), JDK 7.0.

[0080] Based on the streaming data processing architecture, multi-source GIS data are fused to obtain basic terrain data, and three-dimensional terrain data with influence texture and elevation relief information are generated based on the basic terrain data.

[0081] The steps involved in fusing the terrain baseline data include:

[0082] The system performs format conversion, coordinate system 1 and resolution adaptation on multi-source GIS data, and sets a sliding time window for data whose temporal characteristics exceed the threshold to extract related data within the same spatiotemporal range.

[0083] An R-tree index is used to build an index structure for related data, quickly matching spatially overlapping areas of different data sources and aligning them in the spatiotemporal dimension.

[0084] A global R-tree index is constructed by recording the smallest bounding rectangle containing the longitude, latitude, and elevation range from the associated data. The spatial range of multi-source GIS data is used as R-tree nodes, and fast retrieval is performed through hierarchical division (the root node covers the entire area, and the child nodes are subdivided to the region and plot level).

[0085] The global R-tree index is further subdivided into spatial blocks within each terrain, tilt, image, and elevation data.

[0086] The time window size is dynamically set according to the application scenario, and the time range is defined based on the target time point.

[0087] Based on the global R-tree index nodes, data sources whose timestamps fall within the sliding window are selected to form a spatiotemporal candidate dataset, excluding data with mismatched time dimensions (such as old images from many years ago and real-time monitoring data).

[0088] For candidate data in the spatiotemporal candidate dataset, using the minimum bounding rectangle of the target region as the retrieval condition, the range query function of the global R-tree index is used to quickly locate all data sources that spatially overlap with the target region.

[0089] Calculate the overlap between candidate data and target region (overlap area / total data area), retain candidate data with overlap exceeding a threshold (e.g., 30%), and filter out redundant data with low spatial correlation to obtain matching data.

[0090] The spatial offset is calculated by comparing the coordinates of the vertices of the minimum bounding rectangle for the matching data. Affine transformation or thin plate spline interpolation is used to perform geometric correction on low-precision data (such as remote sensing images) so that the coordinate deviation of the overlapping area is controlled within a preset threshold.

[0091] For multi-temporal data (such as images taken at different times) within the same spatial area, weighted fusion is used in combination with timestamps (new data has higher weight) to eliminate conflicts in terrain changes caused by time differences (such as terrain differences before and after construction).

[0092] For conflicting areas (such as elevation discrepancies between the BIM model and the laser point cloud), corrections are made based on data accuracy priorities (such as laser point cloud > image > vector) to ensure consistency of the aligned data.

[0093] For high-density laser point clouds and image data exceeding the preset size, layered sampling is performed according to spatial hierarchy to obtain the detail level of different data blocks, and hole repair is performed using the neighborhood difference method.

[0094] Define multi-level spatial resolutions based on application scenario requirements, set point cloud density thresholds and influence size thresholds, and formulate hole determination criteria.

[0095] Using the global coordinate system as a reference, high-density laser point cloud data exceeding the point cloud density threshold is divided into regular grid blocks according to the grid size of the preset level.

[0096] The specific steps for calculating the point cloud density for each regular grid block include:

[0097] If the density exceeds the current level point cloud density threshold, random sampling or uniform grid sampling is used to retain the target points and discard redundant points.

[0098] If the density is below the threshold but above the point cloud density threshold of the next level, it will automatically be downgraded to the next level and sampled according to the corresponding threshold.

[0099] For key areas (such as abrupt terrain changes and building edges), high curvature points are preserved through feature point detection (such as based on the rate of change of normal vectors) to ensure that critical details are not lost.

[0100] Assign a unique ID to the sampled regular grid block, record the spatial range, hierarchical information and storage path, and support dynamic loading by hierarchy.

[0101] For image data exceeding a preset size, Gaussian pyramid downsampling is used to generate multi-resolution images:

[0102] The top layer (Level 0) is the original resolution image.

[0103] The next level (Level 1) is generated by 2×2 pixel average downsampling, with a resolution of half that of the top level.

[0104] This process continues until the lowest level meets the lightweight requirements (e.g., Leveln resolution ≤ 10m / pixel).

[0105] Each layer of imagery is cut into tiles of fixed size, and each tile is marked with its corresponding geographic coordinate range to ensure alignment with the spatial range of regular grid blocks at the same level.

[0106] For the sampled point cloud data, the grid blocks are traversed using an octree spatial index to identify continuous void regions without points. The boundary point set of the voids is extracted, and the normal vector and elevation trend of the boundary points are calculated to determine the void type (such as flat areas, undulating terrain areas, and building shadow areas).

[0107] For cavities in flat areas, the inverse distance weighted (IDW) algorithm is used to calculate the elevation of any point within the cavity by weighting the elevation of the cavity boundary points as samples (the weight is inversely proportional to the distance).

[0108] For areas with undulating terrain, the surface is fitted based on the terrain trend of neighboring points (3-5 grids outward from the boundary points) by moving least squares (MLS) or kriging interpolation to fill the void elevation.

[0109] The interpolation direction is constrained by the normal vector of the boundary point (such as the vertical wall) in the building's shadow area to avoid elevation anomalies.

[0110] Hole regions in the image are marked by threshold segmentation (e.g., DN value = 0 or outside the normal range). For each hole region, the window is dynamically adjusted according to the hole size, and corresponding methods are used for holes of different sizes. For example, for small and medium holes (≤5×5 pixels), bilinear interpolation or mean filtering is used to fill them, and the RGB values ​​of neighboring pixels are used to smooth the transition.

[0111] For large holes (>5×5 pixels), texture-based interpolation is used to maintain texture consistency by combining low-level high-resolution images of the same area (such as Level 1 referencing Level 0) or point cloud reflection intensity information.

[0112] Based on terrain relief and line-of-sight parameters, a dynamic LOD strategy is obtained by adjusting various detail levels in real time through a dynamic LOD optimization algorithm. Multiple data blocks are aligned and feature sets are fused to obtain a basic terrain database. Feature set fusion may include: fusing the elevation information of laser point clouds with DEM data to generate a high-precision terrain skeleton; matching remote sensing image textures with ground feature vector data to generate a surface texture layer; and integrating the artificial structure data of the BIM model with the terrain skeleton to form fused data containing both natural terrain and artificial features.

[0113] The steps to obtain a dynamic LOD strategy include:

[0114] The terrain baseline data is divided into spatial grids to generate a multi-level LOD model, and the terrain relief of each spatial grid is calculated. The formula is as follows:

[0115]

[0116] In the formula, It refers to the degree of terrain undulation. It is the maximum elevation value within the spatial grid. It is the minimum elevation value within the spatial grid.

[0117] The line-of-sight distance is calculated based on the camera position and the coordinates of the spatial grid center, expressed by the following formula:

[0118]

[0119] In the formula, It's the viewing distance. It's the camera position. These are the coordinates of the center of the spatial grid.

[0120] Based on the impact of terrain relief on the screen at the current viewing distance, and by introducing a relief correction coefficient, the screen error is obtained, expressed by the formula:

[0121]

[0122] in,

[0123]

[0124] In the formula, It is the baseline fluctuation. It is the fluctuation correction factor. It's a screen error. It is the conversion factor between screen pixels and actual distance.

[0125] The error threshold for each LOD model is set according to the level, and the corresponding level threshold is obtained. The LOD model that satisfies the screen error not being greater than the level threshold is selected as the optimal LOD level.

[0126] When the absolute value of the change in LOD level between adjacent frames is not less than two, linear interpolation is used for transition, expressed by the formula:

[0127]

[0128] In the formula, It is the first Frame transition layers, It is the current frame level. The sequence number of the current transition frame. It is the total number of transition frames. It is the optimal LOD level.

[0129] An incremental computing model is used to perform real-time fusion processing on newly added and changed data from different dynamic scenarios, and then upload the data to the terrain database to obtain basic terrain data.

[0130] The steps for real-time fusion processing of newly added and changed data include:

[0131] Spatial change detection and attribute change detection are performed on the newly added changed data to obtain the bounding boxes of the changed areas and the types of changed data. The newly added changed data includes: disaster early warning scenarios: real-time monitored terrain deformation data (such as laser point cloud updates and radar interferometry data in landslide areas) and disaster images (real-time images transmitted back by UAVs).

[0132] Construction monitoring scenarios include: BIM model version updates for the construction area (such as adding new building components), daily construction progress point cloud data, and on-site real-time images.

[0133] The newly added and changed data undergoes format conversion and coordinate correction, and is then spatiotemporally aligned. Format conversion: New point clouds are converted to LAS format, and imagery is converted to TIFF format to ensure compatibility with historical data formats.

[0134] Coordinate correction: Incremental data is unified to the global coordinate system using control points or GNSS (Global Navigation Satellite System) information, eliminating local coordinate system deviations. Time alignment: Based on a sliding time window, the timestamps of incremental data are matched with the time dimension of historical data to ensure data correlation within the same time period.

[0135] Spatial alignment: For changed areas, the newly added point cloud is registered with the corresponding area of ​​the historical terrain model (e.g., the newly added point cloud in the construction area is aligned with the original terrain), or the newly added image is aligned with the historical texture through an image registration algorithm (e.g., ORB feature matching) to ensure that the spatial position deviation is ≤0.5 meters.

[0136] The corresponding fusion algorithm is used to perform fusion based on the data type of the change, and the historical fusion results of the changed area are updated based on the bounding box.

[0137] Geometric incremental fusion (e.g., adding terrain deformation, BIM components): For point cloud / elevation data of changed areas, a weighted average fusion is used: the weight of newly added data (e.g., 0.8) is higher than that of historical data (e.g., 0.2), and the latest observation results are retained first.

[0138] Incremental texture blending (e.g., adding new images): Feathering is used for textures in changed areas: new images and historical textures are blended in overlapping areas by distance weighting to avoid seams.

[0139] Incremental attribute fusion (such as BIM status update): directly replace the attribute fields of the corresponding features in the historical data (such as updating "construction status" from "not started" to "in progress"), and retain the unchanged attributes (such as "building height").

[0140] A 3D terrain model is constructed based on top-level reconstruction technology and 3D terrain data. An optimized terrain model is obtained by repairing geometric holes based on Poisson surface reconstruction and an improved ICP (Iterative Closest Point) algorithm.

[0141] The steps to generate 3D terrain data include:

[0142] Elevation data blocks and image texture data are extracted from the terrain baseline data and spatial reference consistency processing is performed.

[0143] Using the spatial range of elevation data blocks, crop the image texture data to the same range to remove redundant parts.

[0144] Elevation data blocks are converted into corresponding triangular meshes based on different structured data types, and then optimized according to the hole type. Triangular meshes can be divided into regular meshes and irregular triangular meshes. Regular meshes: For structured elevation data (such as DEM), square meshes are generated at fixed intervals (such as 1m×1m), and the Z value of each mesh vertex is the elevation value of the corresponding location.

[0145] Irregular Triangular Mesh (TIN): For unstructured elevation data such as laser point clouds, the Delaunay triangulation algorithm is used to connect discrete points to form triangular surfaces, which better expresses terrain details (such as steep slopes and cliffs).

[0146] For flat areas (elevation variation ≤ threshold, such as 0.5m), the number of triangular faces is simplified (e.g., merging adjacent triangles), while for areas with drastic terrain undulations (e.g., slope ≥ 30°), a high-density grid is retained to balance accuracy and efficiency.

[0147] Traverse the triangular mesh to identify the set of boundary edges without adjacent faces. When the boundary edges form a closed loop and there are no triangular faces inside, it is determined to be a void region, and it is classified into multiple void types according to the cause of void formation.

[0148] Texture coordinates are calculated and assigned to each vertex of the triangular mesh using the affine transformation formula, expressed as follows:

[0149]

[0150]

[0151] In the formula, It refers to the spatial range of the basic terrain data. It is the pixel width of the image texture data. It is the pixel height of the image texture data. yes axis, yes axis, These are texture coordinates.

[0152] Image texture data is attached to the surface of a triangular mesh, and the texture coordinates are spatially aligned with the image pixels through graphics rendering to obtain three-dimensional terrain data.

[0153] For regions where the texture resolution and mesh density do not match (such as low-resolution images mapped to high-detail meshes), bilinear interpolation is used for resampling to avoid texture stretching or blurring.

[0154] For areas where multiple images are stitched together, color balance algorithms (such as histogram matching) are used to unify the color tone and eliminate stitching seams.

[0155] Weighted blending is applied to overlapping areas (e.g., the closer the distance, the higher the weight) to ensure a natural texture transition.

[0156] It supports the creation of accurate, massive 3D terrain datasets with real-world geographic coordinates from data of different resolutions, sizes, and coordinate systems (including aerial photographs, satellite imagery, and digital elevation models), achieving data compression. The created 3D terrain scene file is a single, independent file, making it easy to manage.

[0157] It supports a mechanism that allows workload sharing among multiple computers on a multi-core or local area network, improving performance through multiple hardware components for rapidly building 3D terrain datasets. This plays a crucial role in the creation of high-resolution 3D terrain data at the provincial level for this project.

[0158] Supports resumeable interruption. This means that if data calculation is stopped during 3D terrain creation, the creation process can continue after reopening the project without needing to recreate the terrain. Interruptions during data processing allow for viewing the generated terrain data through a visualization platform, enabling timely identification and correction of data issues. If the terrain data is normal and can be resumed, there is no need to recalculate using additional time.

[0159] It supports common coordinate systems such as WGS84, CGCS2000, and CGCS2000 projected coordinate system.

[0160] The steps to construct a 3D terrain model include:

[0161] Invalid value removal and data cropping are performed on the elevation data blocks and image texture data that have undergone spatial reference consistency processing in the software tool, and the overlapping effects of multiple images are stitched together according to the color equalization algorithm to obtain preprocessed data.

[0162] Invalid value removal: Remove cloud cover, noise points, and outliers (such as elevation values ​​that are outside the reasonable range) from the image.

[0163] Data cropping: Cropping data based on the spatial extent of the target area (such as administrative boundaries) to remove redundant areas and reduce computational load.

[0164] Using the geographic coordinate system as a reference, a three-dimensional coordinate system is created in the software tool, and the preprocessed data is mapped according to the coordinate relationship to form a basic terrain model.

[0165] The terrain files are created using software tools and the terrain data is stored at the optimal LOD level. The number of rendering requests is reduced by merging adjacent low-precision data blocks using top-level reconstruction technology.

[0166] like Figure 2 As shown, the principle of top-level reconstruction is based on the creation, overlay, and rendering of layers. By reasonably organizing and adjusting layers, developers can achieve detailed and realistic geographic visualization effects. This invention can lightweight process urban 3D model data. After reconstructing the top layer (each rectangular bounding box represents a tile, i.e., one request), it is obvious that the number of requests is greatly reduced, accelerating the rendering process.

[0167] Image data is applied to the surface of the base terrain model using texture mapping technology, and combined with a dynamic LOD strategy, the matching texture resolution is automatically switched under different viewpoints.

[0168] Edge detection is used to identify the boundaries and attributes of typical land features from image data to obtain land feature characteristics, which are then fused with the basic terrain model to obtain a three-dimensional terrain model.

[0169] The steps to obtain ground feature characteristics include:

[0170] The impact data is subjected to contrast stretching and image noise removal processing, and appropriate band combinations are selected for different types of impact.

[0171] For different features of land cover, select the corresponding edge detection algorithm to obtain the detection results, and perform threshold segmentation to binarize edge pixels and non-edge pixels.

[0172] The Hough transform is used to detect straight lines and curves at discrete edge points, connecting broken edge segments, and then eliminating and filling them through morphological operations to obtain the boundary contours of ground features.

[0173] Extract image features within the boundary contour area of ​​land features to distinguish land feature types.

[0174] Vegetation: Areas where the Normalized Difference Vegetation Index (NDVI) is calculated are identified as vegetation. Texture features (such as the rough texture of tree canopies) are then used to further distinguish between trees and grasslands.

[0175] Architecture: Analyze spectral characteristics (mostly low reflectance gray / dark colors) and shape characteristics (regular geometric shapes, aspect ratio close to 1), and further subdivide building types by roof texture (such as flat roofs and pitched roofs).

[0176] Roads: Main roads and secondary roads are identified by utilizing high reflectivity (asphalt / cement pavement has high brightness in the visible light band) and linear features (long strip shape, width-to-height ratio > 5) combined with lane line texture.

[0177] The feature of a land feature is obtained by associating its boundary outline with its corresponding attribute information.

[0178] Building: Boundary polygon + attributes (Type = Residential / Commercial, Estimated number of floors, Roof material).

[0179] Road: Centerline / Boundary Line + Attributes (Type = Main Road / Secondary Road, Width, Road Surface Material).

[0180] Vegetation: Boundary polygon + attributes (type = woodland / grassland, coverage).

[0181] The steps for Poisson surface reconstruction include:

[0182] High curvature vertices are extracted from the cavity boundary as feature points, and a local point cloud dataset is constructed by combining it with the point cloud data of the surrounding complete area. Then, noise reduction and downsampling are performed.

[0183] Using the normal vectors of the local point cloud dataset as constraints, the implicit surface function is obtained by solving the Poisson equation, and the formula is expressed as:

[0184]

[0185] In the formula, It is a point cloud index. It is the total number of point clouds involved in surface reconstruction. It is the first The weight of each point cloud, It is the first The normal vector of a point cloud. It is the Dirac function. It is the first The three-dimensional coordinates of a point cloud. It is an implicit function Changes in curvature in space.

[0186] Isosurfaces matching the curvature of the surrounding mesh are extracted from the implicit surface function to generate a mesh for filling holes. The reconstructed surface is then smoothed using Laplacian smoothing, as expressed by the formula:

[0187]

[0188] In the formula, It is the vertex The set of neighboring vertices, It is the number of vertices in the domain. It is a smoothing factor. It is the first vertex in the neighborhood vertex set. vertex coordinates It is the smoothed first The new coordinates of each vertex.

[0189] The steps for repairing geometric holes using the improved ICP algorithm include:

[0190] The point cloud surrounding the hole repair area is selected as the source point cloud, and the vertices of the repaired triangular mesh surface are used as target points. Initial alignment is performed through coarse registration. The coarse registration can utilize the feature-point-based SAC-IA algorithm.

[0191] To address outlier interference, key feature points in the source point cloud are weighted using a formula:

[0192]

[0193] in, It is the source point In the nearest neighbor of the target point set, For variables Translation vector Perform minimization optimization. It is the first The weights of key feature points It is the source point The coordinates after rotation and translation.

[0194] Key feature points can include vertices at abrupt changes in terrain and edge points of ground features.

[0195] The convergence conditions are dynamically adjusted based on the current registration error, and normal vector constraints are introduced to ensure that the direction of the vertex normal vector is consistent with the surrounding terrain trend during the registration process.

[0196] The process involves iteratively calculating the optimal transformation matrix between the source point cloud and the target point until the deviation between the vertex coordinates of the repaired area and the original point cloud is within a preset threshold.

[0197] Optimize the terrain model, publish it to the GIS platform and adapt it to mobile devices, and use edge computing to perform lightweight processing and cross-platform deployment.

[0198] It can publish 3D geospatial services, including terrain services, vector services, and 3D mesh model services, supporting streaming and dynamic loading. The GIS platform supports WebGIS and mobile device adaptation, achieves cross-platform deployment through lightweight edge computing processing, and also supports application needs in multiple fields such as disaster early warning, engineering construction, and smart cities, possessing multi-industry generalization capabilities.

[0199] Example 2: Principle of 3D Base Map Production

[0200] Use SkylineGlobeGlobeTerraBuilder to create a realistic, georeferenced, and accurate 3D data scene. The generated 3D terrain dataset can be used locally as a base terrain or published to remote clients. It can be seamlessly integrated with existing terrain data, providing the foundational terrain for 3D spatial visualization.

[0201] like Figure 3 As shown, the generated 3D terrain data can be used as image data or as elevation data.

[0202] 3D base map data creation:

[0203] Image data: Data year: 2025.

[0204] Image type: Satellite imagery.

[0205] Image resolution: 2 meters, better than 1 meter, 0.5 meters (multiple terrain data sets were created separately).

[0206] Data source: Collected from the construction unit.

[0207] Elevation data: Elevation resolution: 5 meters for the entire province.

[0208] Data requirements: If the DEM is updated, it must be updated before creating the terrain-level 3D base map.

[0209] Data source: Collected from the construction unit.

[0210] like Figure 4 As shown, the above data is used to create satellite data-based terrain-level 3D reality construction and aerial photography-based terrain-level 3D reality engineering files (MPT). The project .tbp file must be retained.

[0211] The data processing flow is as follows:

[0212] In the project settings, set the maximum and minimum elevation values, and remove invalid elevation values ​​to prevent abnormal terrain data display due to invalid values:

[0213] Load elevation and image data into TerraBuilder software. For better operation, the view is a planar view, which can load multiple images / elevation data at different resolutions.

[0214] Create new pyramid files for the source data to enable efficient terrain browsing later.

[0215] Attribute settings: such as resolution settings, visibility distance settings, etc.

[0216] Editing includes cropping, feathering settings, and calculation range settings.

[0217] Invalid value removal: Removes black / white borders and invalid elevation values ​​from images.

[0218] Create MPT: Click CreateMPT to create a 3D terrain. During the creation process, you can pause the calculation program and load the created 3D terrain into the visualization platform to check for display problems.

[0219] Continue creating MPT: Continue creating based on an existing MPT.

[0220] The terrain creation dialog box displays the total number of terrain layers, the current calculation layer, the total calculation time, the time to be used, the estimated data size, and the size of the generated data. It also supports operations such as canceling the calculation, adding integrated runs, and reducing the number of calculation threads.

[0221] Terrain data tiles:

[0222] Powerful terrain and map service technologies enable the "streaming" transmission of geospatial datasets, effectively solving the problem of rapid access to massive terrain data, improving the browsing speed and display effect of 3D scenes, and facilitating the rapid publishing and access of 3D data.

[0223] During the creation of 3D terrain (MPT), the original data is integrated into a single file without compressing the display accuracy, and then lightweighted, typically with a compression ratio greater than 3:1. Simultaneously, the visualization platform can dynamically read terrain tiles based on the user's viewing angle and scale, ensuring smooth browsing and handling of TB-level or even PB-level terrain data.

[0224] Example 3: Constructing a 3D terrain model also includes constructing a 3D city model:

[0225] 3DML production:

[0226] Urban 3D model data generally uses the common OSGB format. This format stores multiple .osgb files in multiple folders, making the data scattered and difficult to manage. If some data is lost during data migration or copying, it will be difficult to recover and trace. Therefore, quantization and integration processing are required.

[0227] SkylineGlobeTerraExplorerPro provides a highly efficient OGSB to 3dml converter. It is easy to use and has a fast processing time. Generally, 10GB of OGSB can be successfully converted into 3dml data within 10 minutes.

[0228] OSGB address: The OSGB folder containing the data can be generated into a TXT file, which can convert multiple scattered .osgb data into a single 3dml file.

[0229] Enter the path to the TXT file.

[0230] Input the 3DML output location.

[0231] Export the coordinate system (OSGB original coordinate system).

[0232] Click the "Create" button to generate a 3dml file. You can check the data processing progress by observing the progress bar.

[0233] 3DTiles production:

[0234] We used CesiumLab to lightweight OSGB data, reducing the amount of data by about one-third without compromising the accuracy of the data display, while also improving data refresh efficiency.

[0235] It can handle multiple data processing tasks without requiring a large investment of human resources.

[0236] In the "Input File" section, select the "data" folder where OSGB is stored. The metadata for the city's 3D model data should be located in the "data" folder or in the same directory to facilitate automatic identification of the data's spatial reference by the software.

[0237] Enable "Rebuild Top Layer" and "Backside Cropping" to perform lightweight data processing and ensure efficient data browsing.

[0238] Select "Hash" as the storage type.

[0239] Set the 3DTiles data output path.

[0240] In summary, this invention eliminates the heterogeneity of multi-source data through format conversion, coordinate system unification, and resolution adaptation; it achieves precise spatiotemporal data alignment and rapid matching of spatially overlapping areas using a sliding time window and R-tree indexing; it designs an incremental calculation mode for dynamic scenes, processing only newly added / changed data, and updates historical results through local fusion algorithms to avoid global recalculation. For geometrically hollow areas, a continuous implicit surface is constructed using the normal vectors of surrounding point clouds through Poisson surface reconstruction, generating a smooth filling mesh; Laplacian smoothing is used to optimize surface details, and smoothness and terrain features are balanced by iteratively adjusting vertex positions; combined with an improved ICP algorithm (introducing point weights and normal vector constraints), high-precision registration between the repaired area and the original model is achieved, ensuring that the deviation is within a preset threshold. This improves the geometric integrity and accuracy of the 3D terrain model. Based on terrain undulation and view distance parameters, the optimal LOD level is dynamically calculated, retaining high detail in complex terrain (high undulation) and near-field areas, while reducing the level in flat areas and distant areas. Low-precision data blocks are merged through top-level reconstruction technology, reducing rendering requests, improving efficiency, and enhancing visualization smoothness and interactive experience. It improves data fusion efficiency, and the constructed 3D terrain model has high integrity, high realism and efficient deployment capability.

[0241] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0242] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for multi-source GIS data fusion and 3D terrain construction based on a streaming approach, characterized in that, include: Collect multi-source GIS data, including topographic data, tilt data, image data, and elevation data; The multi-source GIS data is processed according to the streaming data processing architecture to obtain different data blocks. The detail level of the different data blocks is adjusted by the dynamic LOD optimization algorithm to obtain the dynamic LOD strategy. The different data blocks and the newly added and changed data are fused to obtain the terrain basic data. Based on the terrain basic data, three-dimensional terrain data with image texture and elevation relief information is generated. The steps to obtain the dynamic LOD strategy include: The terrain data is divided into spatial grids to generate a multi-level LOD model, and the terrain relief of each spatial grid is calculated. Calculate the line-of-sight distance based on the camera position and the coordinates of the spatial grid center; The screen error is obtained by introducing an undulation correction coefficient based on the impact of the terrain relief on the screen at the current viewing distance. The error threshold for each LOD model is set according to the level to obtain the corresponding level threshold, and the LOD model that satisfies the screen error not being greater than the level threshold is selected as the optimal LOD level. When the absolute value of the change in LOD level between adjacent frames is not less than two, linear interpolation is used for transition. A three-dimensional terrain model is constructed based on the top-level reconstruction technology and the three-dimensional terrain data. An optimized terrain model is obtained by repairing geometric holes based on Poisson surface reconstruction and an improved ICP algorithm. The steps for performing the Poisson surface reconstruction include: High curvature vertices are extracted from the geometric void boundary as feature points, and a local point cloud dataset is constructed by combining it with the point cloud data of the surrounding complete area, followed by denoising and downsampling processing. The implicit surface function is obtained by solving the Poisson equation using the normal vector of the local point cloud dataset as a constraint. Extract isosurfaces that match the curvature of the surrounding mesh from the implicit surface function, generate a mesh to fill the voids, and perform Laplacian smoothing on the reconstructed surface to obtain the void repair area. The steps for repairing geometric holes using the improved ICP algorithm include: The point cloud surrounding the cavity repair area is selected as the source point cloud, and the vertices of the cavity filling mesh surface are selected as target points. Initial alignment is performed through coarse registration. To resolve outlier interference, point weights are introduced for key feature points in the source point cloud. The convergence conditions are dynamically adjusted based on the current registration error, and normal vector constraints are introduced to ensure that the direction of the vertex normal vector is consistent with the surrounding terrain trend during the registration process. The iteration continues until the optimal transformation matrix between the source point cloud and the target point is calculated iteratively until the deviation between the vertex coordinates of the hole repair area and the original point cloud is within a preset threshold. The optimized terrain model is published to a GIS platform and adapted for mobile devices, then lightweighted through edge computing and deployed across platforms.

2. The method for multi-source GIS data fusion and 3D terrain construction based on stream processing as described in claim 1, characterized in that, The steps for generating 3D terrain data with image texture and elevation relief information include: Elevation data blocks and image texture data are extracted from the terrain baseline data, and spatial reference consistency processing is performed. Using the spatial range of the elevation data block, the image texture data is cropped to the same range to remove redundant parts; The elevation data blocks are converted into corresponding triangular meshes according to different structured data, and then optimized accordingly based on the hole type. Texture coordinates are calculated using the affine transformation formula and assigned to each vertex of the triangular mesh; The image texture data is attached to the surface of the triangular mesh, and the texture coordinates are spatially aligned with the image pixels through graphics rendering to obtain the three-dimensional terrain data.

3. The method for multi-source GIS data fusion and 3D terrain construction based on streaming as described in claim 1, characterized in that, The formula for Laplace smoothing is expressed as follows: ； In the formula, It is the vertex The set of neighboring vertices, It is the number of vertices in the domain. It is a smoothing factor. It is the first vertex in the neighborhood vertex set. vertex coordinates It is the smoothed first The new coordinates of each vertex.

Images