Three-dimensional geographic information engine system and method based on multi-source heterogeneous data fusion

Abstract

This invention relates to the field of 3D modeling and discloses a 3D geographic information engine system and method based on multi-source heterogeneous data fusion. The system includes: a data acquisition module that receives heterogeneous model data; a topology matrix construction module that solves the manifold closure equation and outputs topology correction terms, terminating evolution and completing facets when a stability threshold is exceeded; a topology preservation and simplification module that thins feature edges based on simplification cost scores and locks the boundary topology structure; and a dynamic geometric deformation adaptation module that injects the topology correction terms and simplified mesh structure as topology constraints into the model data to drive boundary mesh deformation. This invention eliminates global rewiring, achieves boundary topology self-healing and improves manifold integrity, reduces computational complexity, and avoids overlapping flickering and facet tearing during large-span switching scheduling at multiple detail levels.

Description

Technical Field

[0001] This invention relates to a three-dimensional geographic information engine system and method based on multi-source heterogeneous data fusion, belonging to the field of three-dimensional modeling technology. Background Technology

[0002] Currently, in large-scale real-world 3D spatial reconstruction, using oblique photogrammetry meshes, aerial laser point clouds, and building information models to reconstruct spatial mesh models is a common approach. This approach integrates multi-source data to achieve feature complementarity, which is used to construct a base mesh model. Its physical basis lies in multi-angle perception and discrete sampling. However, under sensing conditions, there is a physical barrier between optical perspective projection constraints and active laser capture mechanisms. Heterogeneous data exhibits deterministic displacement misalignment and multi-scale structural discontinuity at the edges. Due to differences in sensing principles, topological breaks occur in the boundary mesh. As the scale of the twin scene increases, the misalignment of cross-source mesh splicing exhibits nonlinear growth. The topological matching constraints of traditional solutions fail, and topological degradation phenomena such as gaps, non-manifold topology, and overlapping surfaces are prone to occur at the splicing points of multi-source heterogeneous data. Without cross-source topological linkage control, boundary topological distortion propagates outward along the surface connection path, resulting in geometric surface damage and local voids. This causes polygonal overlap flickering and overall structural cracking when the graphics card rendering pipeline processes spatial view distance switching.

[0003] Conventional approaches, such as improving coordinate registration accuracy or using welding to stitch together geometric reconstructions, cannot absorb the physical distortions caused by sensor capture. Unilaterally improving coordinate accuracy increases the iterative computation load, while directly welding boundaries distorts the original feature constraints and cannot prevent the transmission of topological degradation. Existing reconstruction paths have the following shortcomings in technical practice: different physical hardware acquisition mechanisms lead to a lack of geometric continuity; there are technical bottlenecks at the level of conventional data registration at the bottom layer and backend software modeling and data fusion algorithms. For example, Chinese invention patent application CN118279507A discloses a three-dimensional geological structure modeling method that integrates multi-source heterogeneous data. It uses a deep artificial neural network combined with a multi-point statistical EM iterative algorithm to reconstruct the global features of multi-source data. The global probability field or statistical template matching mechanism implicitly depends on the underlying objective premise of continuous distribution of regional attributes. When facing the high-dynamic nonlinear multi-topological boundary stitching conditions in real-scene 3D geographic information engines, the global statistical mapping method cannot establish a non-rigid residual diffusion buffer at the surface grid boundary, and it is difficult to perceive and absorb the position-sensitive hidden limit of the graphics processor's depth buffer at the millimeter scale.

[0004] Therefore, the technical problem to be solved by this invention is how to retrieve the probability field of the confidence of the sensor properties to construct the boundary topological association constraint, solve the manifold closure equation on the basis of decoupling the adjacency connection of the multi-source boundary grid to generate the topological boundary correction term, and coordinate the control of the cost of grid feature simplification and the topological diffusion of non-rigid deformation displacement residuals to eliminate grid breakage and overlapping scintillation in the multi-source boundary region. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion, comprising: The data acquisition module is used to receive multi-source heterogeneous mesh model data; The topology matrix construction module is connected to the topology preservation and simplification module. It is used to solve the manifold closure equation and output the topology correction term. The matrix evolution is terminated when the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertex exceeds the discrete stability threshold. The mean value of the normal vector of the neighborhood grid is extracted to fill and repair the grid patch in situ and converge the non-manifold topological anomaly in the boundary region. The topology preservation and simplification module is used to calculate the comprehensive simplification cost score based on the linear ratio of the sum of the product of the geometric error metric and the boundary topology association factor. It then discretizes and thins the feature edges based on the comprehensive simplification cost score. When the feature edges are within the boundary of multi-source data, it drives the boundary topology association factor to undergo a weight step switch to lock the grid adjacency topology at the boundary. It also thins and simplifies the grid in the remaining areas to output a simplified grid topology. The dynamic geometric deformation adaptation module is used to inject topology correction terms and simplified mesh topology as boundary topology constraints into multi-source heterogeneous mesh model data, driving the boundary mesh to perform in-situ non-rigid mesh deformation to output a surface-continuous 3D geographic information mesh model.

[0006] Preferably, the dynamic geometric deformation adaptation module includes an error diffusion topology smoothing control module. This module is used to capture high-frequency local displacement residuals generated during the deformation evolution of the in-situ non-rigid mesh in real time. When the absolute value of the high-frequency local displacement residuals exceeds the geometric displacement culling threshold, the error diffusion control process is triggered to calculate the geometric position compensation of each adjacent node. The geometric position compensation is then used to smoothly diffuse the high-frequency local displacement residuals outward along the mesh topology adjacent path, so that the geometric displacement culling threshold is stabilized within the corresponding dynamic change range, and finally, a continuous three-dimensional geographic information mesh model is output.

[0007] Preferably, the topology matrix construction module has a convergence control module inside. The convergence control module is used to perform time-series dynamic response calculation during the process of solving the manifold closure equation. Its limiting logic includes: step S301, measuring the data missing rate in the boundary region of the multi-source heterogeneous grid model under the data sparse condition; step S302, when the data missing rate reaches 45%, extracting the mean normal vector of the known grid in the neighborhood of the boundary region, and in-situ filling and repairing the grid patches in the feedforward dimension to block the computational dead loop.

[0008] Preferably, the topology preservation simplification module is used to retrieve the geometric error metric and the boundary topology association factor when the roaming line-of-sight jumps over a large area, and to cause a step-like weight jump switch of the boundary topology association factor when the feature edge of the grid patch is within the boundary range of multi-source data, so as to forcibly lock the grid adjacency topology structure at the boundary and keep it in a resting state during the adaptive detail level simplification process.

[0009] Preferably, the multi-source heterogeneous mesh model data received by the data acquisition module includes oblique photogrammetry 3D mesh data, airborne laser point cloud data, and structured building information model data.

[0010] Preferably, the topology preservation and simplification module includes a boundary recognition module, which is used to determine whether the feature edges of the mesh patches are within the boundary range of the multi-source data based on the spatial overlap between the oblique photogrammetry 3D mesh data and the airborne laser point cloud data.

[0011] Preferably, when the convergence control module has performed 15 iterations and the rate of change of the discrete curvature of the vertices exceeds the discrete stability threshold, it automatically activates the in-situ completion process of the repair mesh patch under step S302 to output a deterministic closed mesh model to the dynamic geometric deformation adaptation module.

[0012] Preferably, when the topology preservation simplification module thins and simplifies the mesh in non-boundary regions, it protects the manifold structure of the boundary mesh by restricting the disordered reduction of mesh patches at the boundary, thereby eliminating boundary topological cracks.

[0013] Preferably, the error diffusion topology smoothing control module is located in the depth buffer computing architecture of the graphics processor, and the geometric displacement blanking threshold is dynamically adjusted according to the position-sensitive blanking limit of the depth buffer of the graphics processor to eliminate the flickering degradation defect of the boundary mesh when polygons overlap.

[0014] A method for creating a 3D geographic information engine based on multi-source heterogeneous data fusion, used to run a 3D geographic information engine system based on multi-source heterogeneous data fusion, includes the following steps: Step S101: Receive multi-source heterogeneous mesh model data through the data acquisition module; Step S102: Solve the manifold closure equation through the topology matrix construction module and output the topology correction term. When the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertex exceeds the discrete stability threshold, terminate the matrix evolution and extract the mean value of the neighborhood grid normal vector to fill and repair the grid patch in situ, and converge the non-manifold topological anomaly in the boundary region. Step S103: The topology preservation simplification module calculates the comprehensive simplification cost score by the linear ratio of the sum of the product of the geometric error metric and the boundary topology association factor. The feature edges are discretized and thinned according to the comprehensive simplification cost score. When the feature edges are within the boundary of multi-source data, the boundary topology association factor is driven to undergo a weight step switch to lock the grid adjacency topology structure at the boundary. The grid in the remaining area is thinned and simplified to output a simplified grid topology structure. Step S104: The topology correction term and the simplified mesh topology structure are injected into the multi-source heterogeneous mesh model data as boundary topology constraints through the dynamic geometric deformation adaptation module, driving the boundary mesh surface patches to perform in-situ non-rigid mesh deformation to output a continuous three-dimensional geographic information mesh model.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the 3D geographic information engine, the multi-source topology association decoupling module retrieves the confidence probability field of the sensor properties from the accompanying input, uses the spatial 3D bounding box intersection judgment rule to lock the overlapping boundaries of heterogeneous spatial data, converts the laser echo energy and photogrammetric projection residual into topological map constraints, filters the boundary vertex sequence through confidence weights and removes high-frequency noise points, decouples the adjacency connection of multi-source grids at the boundary, and constructs a local boundary topology association matrix. Thus, before the geometric deformation response, it eliminates the displacement distortion caused by the difference in sensor capture mechanism, avoids the defect of non-manifold edges and grid self-intersection caused by the forced merging of conventional spatial coordinates, and realizes the closed constraint output of multi-source boundary data in the pure topological dimension.

[0016] 2. The topology matrix construction module receives the correlation matrix, solves the manifold closure equation according to the manifold rules, determines the monotonically convergent state of the boundary to output the topology correction term, and terminates the matrix evolution when the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertices exceeds the limit. It also extracts the mean normal vector of the known grid in the neighborhood and fills in the transition topology block in the feedforward dimension to block the computational dead loop caused by the data missing rate exceeding the theoretical limit of 45%. In this way, it converges the singular topology of the boundary region of the multi-source grid, so that the system can still output a deterministic closed grid model under the sparse detection condition, avoiding the destruction of geometric surface features caused by conventional post-processing rewiring.

[0017] 3. The topology preservation and simplification module retrieves the geometric error metric and the boundary topology correlation factor. Based on the linear ratio of their product to the sum of the two, it calculates the comprehensive simplification cost score. When the feature edge is within the boundary of multi-source data, it drives the boundary topology correlation factor to undergo a step-like leap, which in turn induces a leap in the comprehensive simplification cost score to forcibly lock the mesh connection relationship at the boundary and keep it at rest during the adaptive detail level simplification process. Only the mesh in the non-boundary area is thinned and simplified, thereby protecting the manifold structure of the boundary mesh when the roaming line-of-sight jumps over a large area and avoiding boundary topology cracks caused by disordered reduction of cross-level scheduling patches. Attached Figure Description

[0018] Figure 1 This is a flowchart of the three-dimensional geographic information engine method for multi-source heterogeneous data fusion according to the present invention. Figure 2 This is a simplified diagram of the mesh thinning of the topology-preserving simplification module of the present invention.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] A 3D geographic information engine system based on multi-source heterogeneous data fusion includes: The data acquisition module is used to receive multi-source heterogeneous mesh model data; The topology matrix construction module is connected to the topology preservation and simplification module. It is used to solve the manifold closure equation and output the topology correction term. The matrix evolution is terminated when the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertex exceeds the discrete stability threshold. The mean value of the normal vector of the neighborhood grid is extracted to fill and repair the grid patch in situ and converge the non-manifold topological anomaly in the boundary region. The topology preservation and simplification module is used to calculate the comprehensive simplification cost score based on the linear ratio of the sum of the product of the geometric error metric and the boundary topology association factor. It then discretizes and thins the feature edges based on the comprehensive simplification cost score. When the feature edges are within the boundary of multi-source data, it drives the boundary topology association factor to undergo a weight step switch to lock the grid adjacency topology at the boundary. It also thins and simplifies the grid in the remaining areas to output a simplified grid topology. The dynamic geometric deformation adaptation module is used to inject topology correction terms and simplified mesh topology as boundary topology constraints into multi-source heterogeneous mesh model data, driving the boundary mesh to perform in-situ non-rigid mesh deformation to output a surface-continuous 3D geographic information mesh model.

[0022] Preferably, the dynamic geometric deformation adaptation module includes an error diffusion topology smoothing control module. This module is used to capture high-frequency local displacement residuals generated during the deformation evolution of the in-situ non-rigid mesh in real time. When the absolute value of the high-frequency local displacement residuals exceeds the geometric displacement culling threshold, the error diffusion control process is triggered to calculate the geometric position compensation of each adjacent node. The geometric position compensation is then used to smoothly diffuse the high-frequency local displacement residuals outward along the mesh topology adjacent path, so that the geometric displacement culling threshold is stabilized within the corresponding dynamic change range, and finally, a continuous three-dimensional geographic information mesh model is output.

[0023] Preferably, the topology matrix construction module has a convergence control module inside. The convergence control module is used to perform time-series dynamic response calculation during the process of solving the manifold closure equation. Its limiting logic includes: step S301, measuring the data missing rate in the boundary region of the multi-source heterogeneous grid model under the data sparse condition; step S302, when the data missing rate reaches 45%, extracting the mean normal vector of the known grid in the neighborhood of the boundary region, and in-situ filling and repairing the grid patches in the feedforward dimension to block the computational dead loop.

[0024] Preferably, the topology preservation simplification module is used to retrieve the geometric error metric and the boundary topology association factor when the roaming line-of-sight jumps over a large area, and to cause a step-like weight jump switch of the boundary topology association factor when the feature edge of the grid patch is within the boundary range of multi-source data, so as to forcibly lock the grid adjacency topology structure at the boundary and keep it in a resting state during the adaptive detail level simplification process.

[0025] Preferably, the multi-source heterogeneous mesh model data received by the data acquisition module includes oblique photogrammetry 3D mesh data, airborne laser point cloud data, and structured building information model data.

[0026] Preferably, the topology preservation and simplification module includes a boundary recognition module, which is used to determine whether the feature edges of the mesh patches are within the boundary range of the multi-source data based on the spatial overlap between the oblique photogrammetry 3D mesh data and the airborne laser point cloud data.

[0027] Preferably, when the convergence control module has performed 15 iterations and the rate of change of the discrete curvature of the vertices exceeds the discrete stability threshold, it automatically activates the in-situ completion process of the repair mesh patch under step S302 to output a deterministic closed mesh model to the dynamic geometric deformation adaptation module.

[0028] Preferably, when the topology preservation simplification module thins and simplifies the mesh in non-boundary regions, it protects the manifold structure of the boundary mesh by restricting the disordered reduction of mesh patches at the boundary, thereby eliminating boundary topological cracks.

[0029] Preferably, the error diffusion topology smoothing control module is located in the depth buffer computing architecture of the graphics processor, and the geometric displacement blanking threshold is dynamically adjusted according to the position-sensitive blanking limit of the depth buffer of the graphics processor to eliminate the flickering degradation defect of the boundary mesh when polygons overlap.

[0030] A method for creating a 3D geographic information engine based on multi-source heterogeneous data fusion, used to run a 3D geographic information engine system based on multi-source heterogeneous data fusion, includes the following steps: Step S101: Receive multi-source heterogeneous mesh model data through the data acquisition module; Step S102: Solve the manifold closure equation through the topology matrix construction module and output the topology correction term. When the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertex exceeds the discrete stability threshold, terminate the matrix evolution and extract the mean value of the neighborhood grid normal vector to fill and repair the grid patch in situ, and converge the non-manifold topological anomaly in the boundary region. Step S103: The topology preservation simplification module calculates the comprehensive simplification cost score by the linear ratio of the sum of the product of the geometric error metric and the boundary topology association factor. The feature edges are discretized and thinned according to the comprehensive simplification cost score. When the feature edges are within the boundary of multi-source data, the boundary topology association factor is driven to undergo a weight step switch to lock the grid adjacency topology structure at the boundary. The grid in the remaining area is thinned and simplified to output a simplified grid topology structure. Step S104: The topology correction term and the simplified mesh topology structure are injected into the multi-source heterogeneous mesh model data as boundary topology constraints through the dynamic geometric deformation adaptation module, driving the boundary mesh surface patches to perform in-situ non-rigid mesh deformation to output a continuous three-dimensional geographic information mesh model.

[0031] Example 1: In the scenario of a 3D geographic information engine used to construct a large-scale digital twin city, the system processes triangular mesh data, surface laser point cloud data, and urban building information models from oblique photogrammetry sensors in real time. Due to differences in sensing principles and spatiotemporal misalignment in data acquisition, nonlinear geometric misalignment exists in the spatial boundary regions of the various geographic data sources, with displacements reaching up to 1.5m. If Euclidean distance retrieval is used to forcibly merge the heterogeneous boundary vertices, it will lead to self-entanglement of triangular faces and non-manifold topological degradation defects in the merged boundary regions. This system uses a spatial bounding box intersection algorithm to lock the overlapping boundary regions and extract the initial discrete topological boundary vertices of the polygonal meshes within the regions. The system reads raw sensor measurement auxiliary metadata associated with each geospatial data point sequence and generates a sensor property confidence probability field. In practice, the specific generation and mapping procedure for the sensor property confidence probability field is as follows: The system extracts the raw infrared echo energy intensity physical quantity from the airborne laser point cloud data and the projection residual spatial geometric quantity from the multi-view geometric collinearity equation in oblique photogrammetry. It then uses a minimum-maximum normalization function to convert the echo energy intensity and projection residual into unified dimensionless relative scalar values. Next, through a pre-defined sensor property joint probability distribution model, the normalized positive weight of the echo energy and the negative penalty term of the projection residual are linearly superimposed and weighted, thereby generating a probability field in the grid. At each three-dimensional spatial coordinate position of the boundary vertex, a dimensionless scalar confidence value ranging from 0 to 1 is mapped, representing the confidence level of the physical property of the current sampled data point. This constitutes the sensing physical property confidence probability field. The multi-source topology association decoupling unit obtains the probability amplitude of the sensing physical property confidence probability field at the corresponding vertex spatial coordinate position. This amplitude is used as a feedforward constraint weight to filter the initial discrete topology boundary vertex sequence, eliminating abnormal perturbation points below the confidence threshold, and constructing a local boundary topology association matrix. This matrix quantifies the adjacency geometric topology relationship between boundary vertices of different data sources. During mesh thinning simplification, in order to eliminate geometric error metrics with the dimension of length squared and those belonging to dimensionless weights... To address the dimensional conflict between boundary topological correlation factors, which prevent direct summation, the system performs dimensionless normalization bridging before computation: the retrieved geometric error metric is divided in real-time by the maximum geometric error baseline value within the current local neighborhood grid, transforming it into a dimensionless relative geometric error coefficient between 0 and 1. After this transformation, the dimensionless relative geometric error coefficient can be directly multiplied and summed with other dimensionless boundary topological correlation factors, resulting in a comprehensive simplified cost score that is completely consistent in both physical dimensions and mathematical logic. The topology-preserving simplification module uses dimensionless normalization to calculate the comprehensive simplified cost score. The formula for calculating the comprehensive simplified cost score is as follows: ,in, Set as a comprehensive simplified cost score. Let the relative geometric error coefficient be generated by dividing the retrieved geometric error metric by the maximum geometric error reference value within the local neighborhood grid. Let it be a dimensionless boundary topological correlation factor. Set as the first adjustment coefficient. Set as the second adjustment coefficient, in the non-boundary region, and The values ​​are all fixed at 1. When the boundary recognition module determines that the feature edge of the mesh patch is within the boundary range of multi-source data, the boundary topology association factor is triggered. A weight step switch occurs, at which point the factor value jumps from the initial baseline value of 1 to 10,000,000, while the second adjustment coefficient... The value is then synchronously rewritten to 0.000001 via a conditional trigger, causing the second adjustment coefficient in the denominator to... Related factors with boundary topology The product term approaches zero, leading to a reduced overall simplification cost score. Numerical transitions are generated to forcibly lock the mesh adjacency topology at the boundary. The co-occurrence topology matrix building element receives the local boundary topological incidence matrix and performs iterative solutions to the manifold closure constraint equations to obtain convergent solutions where the boundary gap area approaches zero. Topological boundary correction terms are generated. Specifically, the manifold closure constraint equations are constructed based on the discrete Laplace-Beltrami operator, and their matrix expression is a linear system. ,in, The Laplace topological weight matrix of the boundary mesh is determined by the local boundary topological incidence matrix. This is the correction vector for the three-dimensional coordinates of the boundary vertices to be solved, i.e., the topological boundary correction term. The geometric offset constraint vector characterizes the initial opening of the multi-source boundary gap. During the solution process, the system uses the sparse matrix conjugate gradient iteration method to perform a constrained least-squares solution to the equation. In each iteration, the convergence gradient of the physical integral area of ​​the repair patch in the mesh boundary region is calculated, thereby solving for the precise spatial displacement vector of each boundary vertex and outputting it as the topological boundary correction term. The convergence control module performs time-sequential dynamic response calculation when solving the linear system equation constructed by the discrete Laplace Beltrami operator. The matrix expression of the linear system equation is as follows: ,in, Let the boundary mesh Laplace topological weight matrix be defined by the local boundary topological incidence matrix. Let be the correction vector for the spatial coordinates of the boundary vertices to be solved. Let the geometric bias constraint vector characterize the initial opening of the gap at the multi-source boundary. The system performs a constrained least-squares solution to this equation using the sparse matrix conjugate gradient iteration method. During each iteration, the data missing rate in the boundary region of the multi-source heterogeneous mesh model is measured. When the data missing rate reaches 45% and the solution iteration reaches 15, if the rate of change of the discrete curvature of the vertex continuously exceeds the preset divergence critical threshold of 0.15, the convergence control module terminates the matrix evolution and extracts the mean value of the normal vector of the known manifold mesh in the neighborhood of the boundary region. It then fills in the transitional mesh topology block in situ within the feedforward dimension to output a deterministic closed mesh model to the dynamic geometric deformation adaptation module and prevent the solution from falling into an infinite loop.

[0032] During the iterative solution process, the system calculates the discrete average rate of curvature change of the vertices in real time. If the rate of curvature change continuously exceeds a preset threshold and the number of iterations reaches 15, the topological hard truncation mechanism is forcibly activated, terminating the matrix evolution. The mean normal vector of the known manifold mesh in the neighborhood is extracted, and a transitional mesh topology block is generated in situ. The dynamic geometric deformation adaptation unit receives the topological boundary correction term. Within the constraint framework of the topological boundary correction term, the mesh normal realignment calculation is performed on the boundary vertices at the intersection, and local non-rigid deformation compensation is applied. During the deformation process, the error diffusion self-healing control module monitors the local displacement residuals generated by the non-rigid deformation in real time. When the displacement residuals... When the absolute value exceeds 0.05m, the error propagation self-healing control process is activated, based on the calculation formula. The geometric position compensation amount of each adjacent node is calculated. ,in, The preset damping attenuation coefficient is 0.12. For the first The system utilizes the shortest path steps in the grid topology between each adjacent node and the current central residual node, and the system uses geometric position compensation. The concentrated geometric alignment stress is attenuated exponentially outward along the mesh topology connection path, ultimately outputting a 3D mesh model with a closed topological manifold and continuous surfaces in the graphics processor. An adaptive hardware mapping mechanism is established when the error propagation topology smoothing control module determines the geometric displacement hidden surface removal threshold. The formula for calculating the geometric displacement hidden surface removal threshold is as follows: ,in, Set as the geometric displacement blanking threshold. The safety and stability coefficient is set and its value is fixed within a range of 1.2 to 1.5. Let the minimum fragment depth resolution be calculated by the graphics processor (GPU) during the initialization phase by retrieving the near clipping plane distance, far clipping plane distance, and total number of bits in the depth buffer. Before the early depth testing phase in the rasterization pipeline, the GPU's internal compute shader periodically checks the vertex indices and topological adjacency network skeleton of the current mesh model. When it detects that the absolute value of the high-frequency local displacement residuals generated during the in-situ non-rigid mesh deformation evolution exceeds the geometric displacement hidden surface threshold, the GPU will proceed with the calculation. At that time, the error diffusion control process is automatically triggered to calculate the geometric position compensation of each adjacent node, and the geometric position compensation is written back into the vertex cache to update the node geometric position in situ. In the rasterization stage, the depth conflict flicker caused by polygon overlap is eliminated. According to actual measurement, this process eliminates the geometric cracks in the splicing area, and the boundary topology maintains a continuous manifold under the condition of frequent jumps in the overall roaming view distance.

[0033] Example 2: The experiment aims to verify the geometric closure stability and deformation repair accuracy of this technical solution in heterogeneous geographic data boundary areas. The experimental environment is built on a high-performance 3D geographic information processing platform, configured with a processor with a main frequency of 3.5GHz and a graphics processor with 16GB of video memory. The data source is selected from three sets of typical real-world 3D meshes containing boundary misalignment. The original maximum geometric crack width is distributed in the range of 0.2m to 2.5m, and there is local non-manifold topological degradation. To ensure the engineering rationality of the experimental design, the system presets the following parameter decision logic: the boundary association search radius is set to 2.0m, which is determined based on the nominal positioning drift limit of the sensor in complex urban canyon scenes, to ensure coverage of all potential boundary geometric misalignment areas; the confidence probability field sampling interval is set to 0.05m, which can effectively capture the surface features of the mesh while ensuring computational efficiency; the iterative convergence threshold is set to 1e-6, which is based on the general engineering tolerance setting of the graphics rendering engine for visual continuity, to ensure the processed boundary The visual error is below 0.001m. Furthermore, the other control thresholds and iteration limits used in this invention have clear engineering physical boundaries and test derivation basis: the number of solution iterations reaching 15 is determined based on the curvature inflection point analysis of the boundary convergence curve. After exceeding 15 iterations, the residual convergence slope is lower than the extreme value, and continuing to solve will generate unnecessary processor overhead; the data missing rate of 45% is the topological geometric limit to maintain the discrete mesh manifold from structural collapse. If it exceeds 45%, the mean of the neighborhood normal vector will lose statistical significance; the threshold of 0.15 for the rate of change of vertex discrete curvature is the critical point of geometric nonlinear distortion divergence obtained by second-order difference testing on typical misaligned mesh surfaces; the absolute value boundary of displacement residuals of 0.05m is calculated based on the minimum blanking limit of the single-precision floating-point bit depth of the graphics card under standard digital twin view distance switching; and the damping attenuation coefficient of 0.12 is the optimal window value obtained by fitting the laboratory stress relaxation curve, which can ensure monotonically exponential attenuation of stress without triggering secondary topological distortion.

[0034] The experiment set up three control groups: the experimental group adopted the proposed technical solution, constructing a local boundary topological correlation matrix and performing non-rigid deformation compensation; the first control group only used the forced mesh stitching algorithm, i.e., directly merging boundary vertices; the second control group lacked the error diffusion self-healing control process, only performing geometric alignment. Under the typical working condition of a boundary crack width of 1.5m, the first control group resulted in 42.5% of the boundary mesh faces becoming self-intersecting, causing rendering depth conflicts; the second control group, after aligning the boundary vertices, had residual displacement errors. The mean value is 0.18m, which still cannot close the manifold boundary. The experimental group's data records show that when the input residual is... The damping attenuation coefficient is 0.18m. The geometric position compensation amount of each adjacent node when the value is 0.12 and the number of adjacent node steps is 3. The calculated result is 0.0072m. Based on this compensation amount, iterations were performed, and the boundary gap area increased from the initial 35.2cm². 2 It dropped to 0.04cm 2 This effectively achieves topological closure, with a damping attenuation coefficient Under the out-of-range condition set to 0.35, although the convergence speed is improved, the excessive stress diffusion gradient leads to a 12.4% decrease in geometric accuracy of the neighborhood mesh, proving that the limited coefficient range is the optimal working window. The experimental results confirm that by constructing a multi-source topological correlation matrix and coordinating an error diffusion self-healing mechanism, displacement distortion caused by multi-source heterogeneous spatial data can be eliminated before the geometric deformation response. This technical solution achieves a dynamic balance between manifold closure and surface deformation when processing heterogeneous geospatial mesh stitching, eliminating the boundary overlap flickering phenomenon.

[0035] Example 3: In the application scenario of constructing a city-level large-scale real-scene 3D geospatial digital twin system, the system needs to fuse triangular mesh data, surface laser point cloud data, and urban building information models in real time. Due to geometric misalignment and inconsistent spatial sampling intervals caused by different sensor detection principles, the 3D geographic information engine system exhibits mesh patch tearing and depth buffer overlap flickering defects in the boundary area when performing multi-source mesh boundary stitching. This system performs boundary repair operations, locks the boundary overlap area through a spatial bounding box intersection algorithm, extracts the initial discrete topological boundary vertex sequence of all polygonal meshes within the area, and reads the sensor auxiliary metadata associated with each geospatial data source to generate a sensor property confidence probability field. The multi-source topology association decoupling unit obtains the value of the sensor property confidence probability field at the corresponding vertex spatial coordinate position. This value is used as a feedforward constraint weight to filter the initial discrete topological edges. The system extracts the boundary vertex sequence, removes anomalous perturbation points with a confidence level below 0.4, and constructs a local boundary topological correlation matrix. Each element of the matrix quantifies the adjacency geometric topological correlation strength between boundary vertices of different data sources. The topological matrix construction unit receives the local boundary topological correlation matrix and performs iterative solutions to the manifold closure constraint equations to obtain a convergent solution where the boundary gap area approaches zero. During the solution process, the system calculates the discrete average rate of curvature change of the grid vertices in real time. If the rate of curvature change exceeds 0.15 continuously and the number of iterations reaches 15, the topological structure is determined to be in a distorted divergent state, and the system forcibly terminates the matrix evolution. At the same time, it extracts the mean normal vector of the known manifold grid in the boundary neighborhood, fills in the transitional grid topological block based on the mean, and closes the fault in the geometric dimension. The dynamic geometric deformation adaptation unit receives the topological boundary correction term and performs grid normal realignment calculation on the boundary vertices at the boundary within the constraint framework.

[0036] To eliminate local displacement residuals caused by non-rigid deformation, the error diffusion self-healing control module monitors the displacement state of the current node in real time. When the displacement residual... When the absolute value exceeds 0.05m, the system activates the self-healing control process, based on the calculation formula. The geometric position compensation amount of each adjacent node is calculated. ,in, For the first The geometric position compensation amount of each adjacent node. This represents the real-time displacement residual of the current monitoring node. The preset damping attenuation coefficient is set to 0.12. This coefficient is obtained by fitting the stress relaxation curve of the real-world mesh splicing in the laboratory. For the first The system calculates the shortest path steps between each adjacent node and the current central residual node. During this process, the error propagation topology smoothing control module achieves this by parallel data binding between the GPU's computation shader and the early depth testing computation architecture in the rasterization pipeline. The system allocates a structured buffer in the GPU's memory to permanently store the vertex indices and topological adjacency network skeleton of the current mesh model. When non-rigid geometric deformations evolve, the computation shader quickly retrieves and calculates the number of adjacent path steps by directly addressing this structured buffer. Simultaneously, based on the position-sensitive blanking limit, the memory bit depth precision of the current depth buffer is dynamically matched. If the absolute value of the high-frequency local displacement residual exceeds the blanking limit, the compute shader will calculate the geometric position compensation amount before the pipeline executes fragment shading rendering. The reverse write operation updates the node's geometric position in situ by writing to the vertex buffer, thereby achieving topology smoothing at the hardware pipeline level. The system utilizes geometric position compensation. The concentrated geometric alignment stress is attenuated and absorbed outward along the mesh topology connection path according to an exponential gradient, ensuring that the three-dimensional mesh surface achieves topological manifold closure and geometrical continuity. According to actual measurements, this processing procedure eliminates geometric cracks at the interface and effectively avoids mesh flicker when switching between multiple detail levels of data from an overall perspective. The average geometric continuity error of the repaired boundary region is controlled within 0.01m, realizing manifold closure and surface deformation repair of multi-source heterogeneous mesh data.

[0037] Example 4: During the manifold closure and surface repair of multi-source heterogeneous mesh data in a 3D geographic information engine, the initial mesh model data exhibits geometric discontinuities in the spatial boundary regions due to differences in sensor detection principles and spatiotemporal misalignment during data acquisition. The system deploys a pre-calibration procedure to ensure the determinism of the fusion process. A spatial bounding box intersection algorithm is used to locate overlapping mesh boundary regions, and the corresponding original confidence probability field and confidence threshold are extracted based on the metadata of each sensor device. The setting is based on the data fitting results in the laboratory for different geometric misalignment conditions. When the confidence level of the vertex geometric distribution in the boundary region is lower than 10%, the setting is determined by the following criteria: When this occurs, the system removes the vertex and uses the mean of the normal vectors of the adjacent high-confidence grids. To fill and generate a transitional topology, and to eliminate residual local displacement during the non-rigid deformation adaptation process, the system introduces an error diffusion self-healing control procedure in the deformation evolution monitoring module. This means that when residual displacement at the boundary vertices is detected... When the absolute value is greater than 0.05m, the vertex coordinates are corrected according to the damping control logic, and the geometric position compensation is adjusted. The calculation formula is: ,in, For the first Geometric position compensation of each adjacent vertex This represents the real-time displacement residual between the current vertex and its theoretical alignment position. The preset damping attenuation coefficient is 0.12. For the first The number of shortest path steps for mesh topology connections between adjacent nodes and residual nodes at the geometric deformation center.

[0038] Based on the aforementioned self-healing compensation actions, the system attenuates and absorbs stress exponentially along the mesh topological connectivity path, smoothing out stress abrupt changes caused by local geometric corrections throughout the entire boundary neighborhood. This ensures the geometric continuity and topological integrity of the 3D mesh model surface. After completing the above calibration procedures, the system performs consistency verification on the repaired boundary region. When the boundary crack area is reduced to 0.04... When the normal vector consistency index of each grid patch reaches the preset threshold, the physical steady state of the model repair is confirmed. The closed-loop procedure not only solves the topological anomaly problem at the intersection of multi-source data, but also controls the degree of geometric distortion in the deformation compensation process by introducing a damping attenuation mechanism. This ensures that the rendering continuity index of the model surface remains within the engineering tolerance range when the 3D geographic information engine processes high-density, large-scale heterogeneous geographic data switching.

[0039] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion, characterized in that, include: The data acquisition module is used to receive multi-source heterogeneous mesh model data; The topology matrix construction module is connected to the topology preservation and simplification module. It is used to solve the manifold closure equation and output the topology correction term. The matrix evolution is terminated when the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertex exceeds the discrete stability threshold. The mean value of the normal vector of the neighborhood grid is extracted to fill and repair the grid patch in situ and converge the non-manifold topological anomaly in the boundary region. The topology preservation and simplification module is used to calculate the comprehensive simplification cost score based on the linear ratio of the sum of the product of the geometric error metric and the boundary topology association factor. It then discretizes and thins the feature edges based on the comprehensive simplification cost score. When the feature edges are within the boundary of multi-source data, it drives the boundary topology association factor to undergo a weight step switch to lock the grid adjacency topology at the boundary. It also thins and simplifies the grid in the remaining areas to output a simplified grid topology. The dynamic geometric deformation adaptation module is used to inject topology correction terms and simplified mesh topology as boundary topology constraints into multi-source heterogeneous mesh model data, driving the boundary mesh to perform in-situ non-rigid mesh deformation to output a surface-continuous 3D geographic information mesh model.

2. The three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The dynamic geometric deformation adaptation module has an internal error propagation topology smoothing control module; The error diffusion topology smoothing control module is used to capture the high-frequency local displacement residuals generated during the deformation evolution of the in-situ non-rigid mesh in real time. When the absolute value of the high-frequency local displacement residuals exceeds the geometric displacement blanking threshold, the error diffusion control process is triggered to calculate the geometric position compensation of each adjacent node. The geometric position compensation is used to smoothly diffuse the high-frequency local displacement residuals outward along the mesh topology adjacent path, so that the geometric displacement blanking threshold is stabilized within the corresponding dynamic change range, and finally the surface continuous three-dimensional geographic information mesh model is output.

3. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The topology matrix construction module contains a convergence control module, which is used to perform time-series dynamic response calculation during the solution of the manifold closure equation. Its constraint logic includes: step S301, measuring the data missing rate in the boundary region of the multi-source heterogeneous grid model under the data sparse condition; step S302, when the data missing rate reaches 45%, extracting the mean normal vector of the known grid in the neighborhood of the boundary region, and in-situ filling and repairing the grid patches in the feedforward dimension to prevent the computational dead loop.

4. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The topology preservation simplification module is used to retrieve the geometric error metric and the boundary topology association factor when the roaming line-of-sight jumps over a large area. When the feature edges of the grid patch are within the boundary of multi-source data, the boundary topology association factor undergoes a step-like weight jump switch to forcibly lock the grid adjacency topology structure at the boundary, so that it remains in a resting state during the adaptive detail level simplification process.

5. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The multi-source heterogeneous mesh model data received by the data acquisition module includes oblique photogrammetry 3D mesh data, airborne laser point cloud data, and structured building information model data.

6. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The topology preservation and simplification module includes a boundary recognition module, which determines whether the feature edges of the mesh patches are within the boundary range of the multi-source data based on the spatial overlap between the oblique photogrammetry 3D mesh data and the airborne laser point cloud data.

7. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 3, characterized in that, When the convergence control module has performed 15 iterations and the rate of change of the discrete curvature of the vertices exceeds the discrete stability threshold, it automatically activates the in-situ completion process of the repair mesh patch under step S302 to output a deterministic closed mesh model to the dynamic geometric deformation adaptation module.

8. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, When the topology-preserving simplification module thins and simplifies the mesh in non-boundary regions, it protects the manifold structure of the boundary mesh by restricting the disordered reduction of mesh patches at the boundary, thereby eliminating boundary topological cracks.

9. A three-dimensional geographic information engine system based on multi-source heterogeneous data fusion according to claim 2, characterized in that, The error diffusion topology smoothing control module is located in the depth buffer computing architecture of the graphics processor, and the geometric displacement blanking threshold is dynamically adjusted according to the position-sensitive blanking limit of the depth buffer of the graphics processor to eliminate the flickering degradation defect of the boundary mesh when polygons overlap.

10. A method for a three-dimensional geographic information engine based on multi-source heterogeneous data fusion, used to run the three-dimensional geographic information engine system based on multi-source heterogeneous data fusion as described in claim 1, characterized in that, Includes the following steps: Step S101: Receive multi-source heterogeneous mesh model data through the data acquisition module; Step S102: Solve the manifold closure equation through the topology matrix construction module and output the topology correction term. When the solution iteration reaches 15 times and the rate of change of the discrete curvature of the vertex exceeds the discrete stability threshold, terminate the matrix evolution and extract the mean value of the neighborhood grid normal vector to fill and repair the grid patch in situ, and converge the non-manifold topological anomaly in the boundary region. Step S103: The topology preservation simplification module calculates the comprehensive simplification cost score by the linear ratio of the sum of the product of the geometric error metric and the boundary topology association factor. The feature edges are discretized and thinned according to the comprehensive simplification cost score. When the feature edges are within the boundary of multi-source data, the boundary topology association factor is driven to undergo a weight step switch to lock the grid adjacency topology structure at the boundary. The grid in the remaining area is thinned and simplified to output a simplified grid topology structure. Step S104: The topology correction term and the simplified mesh topology structure are injected into the multi-source heterogeneous mesh model data as boundary topology constraints through the dynamic geometric deformation adaptation module, driving the boundary mesh surface patches to perform in-situ non-rigid mesh deformation to output a continuous three-dimensional geographic information mesh model.

Images