Low-utility land dynamic monitoring management method based on real scene three-dimensional geographic entity

Abstract

The application discloses a low-efficiency land dynamic monitoring management method based on real three-dimensional geographical entities, relates to the technical field of natural resource informatization management, and comprises the following steps: constructing a terrain mutation sensitivity atlas and a vertical risk guide grid, generating a weighted initial value field under the joint driving of an elevation gradient, a reflection intensity and an echo confidence, and forming vertical prior information of geographical entities; based on the vertical prior information of the geographical entities, performing spatiotemporal consistency forced registration, performing local elastic calibration on oblique photography images and laser point cloud data through a multi-scale rigid-flexible mixed constraint mechanism, and generating a vertical error field. Through the construction of the sensitivity atlas, the execution of rigid-flexible registration, the introduction of semantic reconstruction, the implementation of uncertainty propagation and dynamic resampling regulation and control, the vertical error closed-loop correction of the three-dimensional model under complex topography is realized, the accuracy and stability of low-efficiency land identification are improved, and high-credibility technical support is provided for urban space analysis and land decision-making.

Description

Technical Field

[0001] This invention relates to the field of natural resource information management technology, specifically to a method for dynamic monitoring and management of inefficient land use based on real-world three-dimensional geographic entities. Background Technology

[0002] Dynamic monitoring and management of inefficient land use based on real-scene 3D geographic entities refers to using real-scene 3D modeling technology to model the spatial morphology, attribute information, and temporal changes of urban features into semantic geographic entity units in a high-precision, three-dimensional manner. Based on this, it integrates multi-source heterogeneous data such as remote sensing imagery, UAV oblique photography, socio-economic data, and current land use status to construct a continuously updated thematic geographic entity database. This database is used to conduct full-process, automated, and high-frequency dynamic monitoring and identification of land parcels in urban areas that are characterized by low utilization efficiency, scattered spatial forms, and insufficient development intensity. This process achieves a closed-loop management mechanism from data collection, semantic modeling, change identification to visualization by establishing a 3D spatiotemporal data foundation, constructing a thematic knowledge graph of inefficient land use, and integrating an integrated air-space-ground sensing network. Ultimately, it provides real-time, accurate, and intelligent support for land redevelopment, urban renewal, and planning decisions, comprehensively improving the efficiency of intensive use of natural resources and intelligent governance capabilities.

[0003] The existing technology has the following shortcomings:

[0004] In existing technologies, the construction of real-world 3D models typically relies on the joint fusion of heterogeneous data from multiple sources, such as oblique photography, laser point clouds, orthophotos, and digital elevation models, to generate 3D geographic entities with stereoscopic representation capabilities and semantic attributes. However, in areas with abrupt terrain changes, such as elevation fault zones, slope backfill areas, and areas with dense steep slopes, vertical positioning accuracy is easily reduced due to factors such as sensor incident angle, reflection path disturbances, and abnormal echo acquisition. This leads to phenomena such as facade distortion, structural tilting, and boundary twisting of buildings and features in the model. Initially, this type of vertical inaccuracy manifests as subtle geometric shifts, often difficult to detect through conventional checks. However, it will cause significant interference in subsequent inefficient land use identification processes: First, the distortion of the 3D feature outlines can lead to errors in land parcel segmentation algorithms, resulting in false parcels or misconnected boundaries. Second, spatial attributes such as parcel area, volume, and height can be misjudged due to ground surface distortion, leading to systematic shifts in evaluation indicators such as utilization rate, building density, and vacancy rate. Once such errors accumulate during the model building stage, they may lead to cumulative biases in multiple stages such as monitoring, evaluation, and analysis, ultimately weakening the credibility and applicability of the results of identifying inefficient land use based on real-world 3D geographic entities.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for dynamic monitoring and management of inefficient land use based on real-world three-dimensional geographic entities, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities, comprising the following steps:

[0008] Construct a terrain change sensitivity map and a vertical risk guidance grid, and generate a weighted initial field under the joint drive of elevation gradient, reflection intensity and echo confidence, forming the vertical prior information of geographic entities;

[0009] Based on the vertical prior information of geographic entities, spatiotemporal consistency forced registration is performed, and local elastic calibration is carried out on oblique photogrammetric images and laser point cloud data through a multi-scale rigid-flexible hybrid constraint mechanism to generate a vertical error field.

[0010] By introducing semantic contour priors using the vertical error field, contour and volume consistency reconstruction is performed based on the parametric template library of building components to correct building boundary twists and facade deformations.

[0011] Based on the reconstruction results of the contour and volume consistency, a plot unit uncertainty propagation model is established. The area, height and volume parameters are recalculated according to the grid level, spatial attribute confidence layers are generated, and the confidence layer results are written into the thematic geographic entity database.

[0012] In the thematic geographic entity database, areas with low spatial attribute confidence are identified, an adaptive re-sampling strategy is triggered, and UAVs are dispatched to acquire multi-angle perspective data and perform reverse reconstruction. The vertical drift caused by terrain abrupt changes is suppressed by synthetic perspective observation.

[0013] Based on adaptive re-acquisition data and reverse reconstruction results, during the iterative process of constant energy surface in 3D energy balance construction, the local energy change rate and vertical deformation amplitude of the building surface are monitored. When the energy change or vertical displacement of any region exceeds the preset threshold, a threshold event is triggered. Spatial compensation points with time delay are generated in the triggered region to slow down the rate of geometric change. At the same time, the construction weights are redistributed according to the observation quality of image data in different directions to suppress the spread of vertical error and achieve dynamic correction and stable convergence of geographic entity structure.

[0014] Preferably, the steps for generating vertical prior information of geographic entities are as follows:

[0015] The digital elevation model, oblique photogrammetry dense point cloud, ground laser reflection intensity information and laser echo confidence data are acquired and then rasterized after unifying the coordinate system.

[0016] Calculate the elevation gradient factor, reflection intensity fluctuation index, and echo confidence index, and normalize the three indices, then fuse them into a weighted initial value field according to their weights.

[0017] A graded risk map is generated based on the weighted initial value field, and a two-dimensional guiding grid containing comprehensive risk value and directional gradient attribute is constructed.

[0018] The guiding grid is used as the basis for the structural response of high-risk areas, and a weighted initial value field with spatial risk expression capability is output to form the vertical prior information of geographic entities used to guide subsequent registration and modeling.

[0019] Preferably, the vertical error field generation steps are as follows:

[0020] Based on the grid cells marked as high-risk and extremely high-risk in the weighted initial value field, geographic entities within the coverage area are extracted, and their three-dimensional structure and texture data are extracted.

[0021] After unifying the coordinate system of the oblique photogrammetric images and ground laser point clouds collected in the same area, a point-to-point index relationship is established and rigid-flexible hybrid registration is performed.

[0022] Deformation weight grids are constructed within high-risk areas, deformation rules are set for edge points, internal points and constrained edge points, and elevation gradient factors are introduced as adjustment criteria to complete local elastic registration.

[0023] A vertical error layer is generated on the geographic entity grid, anomaly areas are marked and merged into the weighted initial value field to form a risk-error joint map, and the error information is written into the thematic geographic entity database.

[0024] Preferably, the steps for introducing semantic contour priors using the vertical error field and performing contour-volume consistency reconstruction are as follows:

[0025] Based on the vertical error layer, select geographic entities with boundary misalignment or facade deformation, and extract the boundary point set and roof projection point set;

[0026] Construct a parametric template set for building components and perform template matching, selecting the template with the highest score as the structural prototype of the geographic entity;

[0027] The template control points are fitted to the original point set, and the vertical error information is combined to perform reverse offset and boundary reconstruction on the outlier points to complete the contour reconstruction and volume generation.

[0028] Perform geometric and semantic consistency verification, write the verified reconstruction model into the thematic geographic entity database, and update the modeling quality level and confidence score.

[0029] Preferably, the steps for writing the confidence stratification results into the thematic geographic entity database are as follows:

[0030] Geographic entities are projected onto a two-dimensional plane using an equidistant grid division method, and the area, height, and volume parameters of each grid cell are extracted.

[0031] The vertical error layer is matched with the grid position, the error propagation coefficient is set according to the region type, and a weighted diffusion calculation is performed;

[0032] Based on the propagation error results, the area, height, and volume are corrected, and various spatial attribute error indices are generated;

[0033] Assign confidence levels based on the proportions of various errors and generate spatial attribute confidence layer layers;

[0034] The spatial attribute confidence stratification results and 3D model information are written into the thematic geographic entity database and a queryable structure is formed.

[0035] Preferably, the steps for identifying areas with low spatial attribute confidence in the thematic geographic entity database, triggering an adaptive re-sampling strategy, and performing reverse reconstruction are as follows:

[0036] Read the spatial attribute confidence stratification map from the thematic geographic entity database, filter spatial grid cells with low confidence levels and determine the target areas in the building outline that need to be supplemented;

[0037] By combining building height and surrounding terrain information, a drone flight path is generated for the target area, and high-resolution oblique images covering multiple viewing directions are acquired.

[0038] The acquired oblique images are used to generate dense image point clouds, and reverse reconstruction is performed based on multi-angle observation information. The reconstructed model replaces the original model, the spatial indicators are updated, and the data is written into the thematic geographic entity database.

[0039] Preferably, the steps for dynamically suppressing vertical errors based on adaptive re-acquisition data and reverse reconstruction results are as follows:

[0040] The error manifold is constructed and the model error direction tensor is extracted. The error concentration region is identified, and the phase shift direction and the control point used to constrain geometric changes are determined.

[0041] A constant energy surface iteration mechanism is established at the control site to balance the tension changes between local structural surfaces, and a threshold event is triggered when the surface height change or the energy gradient exceeds the preset range.

[0042] Within the error region where the threshold event occurs, a set of spatial compensation points with time lag characteristics are generated to slow down the local geometric update speed. The modeling influence weights are redistributed according to the clarity and stability of the observation data in each direction, so that the reconstruction process prioritizes the use of high-quality view data, thereby guiding the model to tend to stabilize in the vertical direction.

[0043] Vertical error verification is performed on the dynamically adjusted model, the spatial attribute confidence layer is updated, and the adjustment records, threshold event information and the updated confidence layer are written into the thematic geographic entity database to form a closed-loop management from vertical prior generation to dynamic correction.

[0044] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0045] This invention significantly enhances the early warning capability for potential error areas by constructing a terrain abrupt change sensitivity map and vertical prior information. Combining spatiotemporal consistency-driven registration and rigid-flexible hybrid constraints, it achieves high-fidelity fusion of multi-source data, reducing registration errors between oblique imagery and laser point clouds under complex terrain conditions. Introducing semantic contour priors and parameterized component templates improves the integrity and structural realism of building contour reconstruction, eliminating identification biases caused by boundary kinks and facade distortions. An uncertainty propagation model and spatial attribute confidence layering mechanism ensure the credibility of key indicator calculations (such as area, height, and volume). Adaptive resampling strategies and reverse reconstruction enhance the compensation capability for weak data areas. Furthermore, a phase migration and beam angle weight rearrangement mechanism based on manifold constraints further achieves dynamic real-time error suppression, forming a self-regulating modeling closed loop. Overall, this method significantly improves the accuracy, robustness, and dynamic response capability of real-scene 3D geographic entities in identifying inefficient urban land use, ensuring higher scientific rigor and practicality for spatial analysis, land assessment, and planning decisions based on 3D models. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0047] Figure 1 This is a flowchart of the method for dynamic monitoring and management of inefficient land use based on real-world three-dimensional geographic entities, as described in this invention. Detailed Implementation

[0048] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0049] This invention provides, for example Figure 1 The method for dynamic monitoring and management of inefficient land use based on real-world 3D geographic entities, as shown, includes the following steps:

[0050] Construct a terrain change sensitivity map and a vertical risk guidance grid, and generate a weighted initial field under the joint drive of elevation gradient, reflection intensity and echo confidence, forming the vertical prior information of geographic entities;

[0051] To enable proactive identification and response to vertical error risks in areas of abrupt terrain changes, a terrain abruptness sensitivity map and a vertical risk guiding grid are constructed before modeling. Combined with elevation gradient, laser reflection intensity, and echo confidence, a weighted initial value field is generated to establish the vertical prior information of geographic entities. The specific steps include:

[0052] The acquisition of multi-source geospatial data covering the target urban area mainly includes: (1) 1-meter resolution digital elevation model data, used to describe the undulations of the terrain; (2) dense point clouds in the aerial triangulation results after oblique photogrammetry processing, providing the three-dimensional outlines of buildings and the ground surface; (3) reflection intensity information obtained by ground-based mobile laser scanning, used to identify areas of material change; and (4) multi-echo information provided by laser scanning equipment, used to calculate echo confidence. The above four types of data need to be spatially aligned, unified in projection coordinate system, and spatial pixel matching is performed by bilinear interpolation, so that all data are processed on a 1-meter × 1-meter two-dimensional grid structure. Based on the data alignment, the target area boundary is divided into several equal-width two-dimensional grid units, and each unit records the elevation value and laser echo information. For elevation data, the slope change rate of each pixel in the east-west, north-south and diagonal directions is calculated by the 8-neighbor difference method, and the comprehensive elevation gradient factor of each pixel is constructed on this basis. The elevation gradient factor is used to quantify the intensity of elevation changes within each pixel. The larger the value, the more drastic the terrain change, and it is an important indicator for subsequent assessment of terrain distortion risk.

[0053] Based on the extraction of elevation gradient factors, the reflection intensity of the laser point cloud data within the grid cells is statistically calculated. Specifically, all laser echo points falling within each 1-meter grid cell are extracted, and the average and variance of the reflection intensity values ​​of these points are calculated to form the material response characteristics of that cell. Simultaneously, the number of echoes and the amplitude ratio between the primary echo intensity and the secondary echoes are extracted from each laser echo record to construct an echo confidence index. A higher number of echoes, a higher primary echo intensity, and a clearer waveform compared to the secondary echoes indicate a higher echo confidence index, suggesting more accurate surface morphology identification. The mean, variance, and echo confidence values ​​of the reflection intensity of all grid cells are normalized to establish a two-dimensional response surface: a reflective material fluctuation surface and an echo stability surface. High-value areas on the reflective material fluctuation surface typically correspond to boundaries between dissimilar materials, such as the boundary between buildings and mountains, or bare rock and backfilled land; low-value areas on the echo stability surface typically correspond to areas with weak laser penetration and strong signal attenuation, such as areas with dense vegetation cover or strong reflection obstruction. These two sets of data provide the material and equipment response basis for the next step of risk field calculation.

[0054] After extracting the elevation gradient factor, reflective material fluctuation surface, and echo stability surface, the three indicators were standardized and normalized, mapped to the range of 0-1, and a three-channel multi-factor fusion raster layer was constructed. In each raster cell, weight factors were assigned to the three normalized indicators to form a comprehensive risk index. The weights were set based on empirical test results as follows: elevation gradient factor weight 0.5, reflective intensity fluctuation weight 0.3, and echo confidence weight 0.2. A weighted summation formula was used to fuse the three indicators into a single value, representing the potential vertical misalignment risk of that cell. The weighted initial value field generated after fusion exhibits a continuous surface distribution, with risk values ​​showing a spatial clustering trend. High-risk areas are mainly distributed at cliff edges, slope transition zones, large building foundation boundaries, and the interface between exposed rock and artificial structures. Based on this weighted initial value field, the risk values ​​were divided into five levels according to the natural discontinuity method: very low risk, low risk, medium risk, high risk, and extremely high risk. Each level corresponds to a different color gradation, generating a visualized terrain change sensitivity map to intuitively present high-risk distribution areas.

[0055] Based on the weighted initial value field and the classification results, a vertical risk guidance grid is constructed on a two-dimensional raster layer. This guidance grid uses each raster cell as the smallest unit, and each cell is appended with the following attribute fields: elevation gradient value, mean reflection intensity, echo confidence, comprehensive risk value, risk level, and rate of change of surrounding gradient directions. The guidance grid not only has a static storage function for risk values ​​but also establishes spatial adjacency relationships. In the specific implementation, for each high-risk level cell, the gradient change directions of its four-neighbor and eight-neighbor areas are searched to construct local gradient field information, which guides the selection of directions where constraint strength should be increased during the subsequent model registration stage. For example, high-density point cloud data is preferentially used for fitting and modeling in directions of abrupt changes in vertical gradient values; angle images with less reflection interference are preferentially selected as reconstructed views in areas with significant changes in reflection intensity; and multi-view redundant scanning is performed in areas with low echo confidence to improve modeling accuracy. The entire guidance grid forms a decision structure based on spatial risk response, transforming the weighted initial value field and raster level information into executable modeling control parameters, enabling the geographic entity model to dynamically avoid the problem of error accumulation in high-risk areas from the initial construction stage. By establishing this guiding mesh, each subsequent 3D modeling sub-process can obtain spatial risk references, enabling dynamic registration strategies, adjustable reconstruction density, and flexible accuracy assurance, thereby comprehensively improving vertical accuracy stability.

[0056] Based on the vertical prior information of geographic entities, spatiotemporal consistency forced registration is performed, and local elastic calibration is carried out on oblique photogrammetric images and laser point cloud data through a multi-scale rigid-flexible hybrid constraint mechanism to generate a vertical error field.

[0057] After constructing the weighted initial value field and the vertical risk-guided grid, the spatial distribution of high-risk areas and the risk level information of each geographic entity have been obtained. Based on this prior information, spatiotemporal consistency mandatory registration is performed. A multi-scale rigid-flexible hybrid constraint mechanism is used to perform local elastic calibration on the oblique photogrammetric imagery and laser point cloud data, generating an accurate vertical error field. The process includes the following steps:

[0058] Based on the grid cells marked as "high risk" and "extremely high risk" in the generated weighted initial field, all 3D geographic entities within their coverage areas are extracted, including buildings, roads, surface structures, and elevation boundary areas. For each geographic entity, its corresponding base surface grid, top surface outline, building block edge line, and facade texture block are obtained. Subsequently, oblique photogrammetry images and ground laser point cloud data collected in the same region within the same week are retrieved and uniformly converted to the 2000 geodetic coordinate system. The oblique photogrammetry images are used to generate image point clouds through dense matching, and their spatial 3D point sets are obtained using the reconstructed triangular mesh projection method. Then, the ground laser point clouds are resampled at a resolution of 0.1 meters using a voxel downsampling method to construct a high-density reference point set. An index is established using geographic entity numbers so that each point cloud set can correspond to a specific spatial unit, realizing point-to-point registration preparation.

[0059] Forced registration of point cloud data for spatiotemporal consistency is performed. For areas with a "low risk" level, a rigid registration strategy is adopted. Specifically, the point cloud of geographic entity boundaries is extracted, the centroid coordinates and principal direction vectors are calculated, and a transformation matrix is ​​constructed accordingly. Rotation and translation operations are then used to align the oblique photogrammetry point cloud to the laser point cloud reference frame. In "medium risk," "high risk," and "extremely high risk" areas, a flexible registration strategy is used. This involves establishing a free-deformation weighted grid within each geographic entity component, dividing the original point cloud into several local regional units. Within each unit, three types of control points are defined: edge points, interior points, and constrained edge points. Edge points are subject to strong constraints, allowing only vertical fine-tuning; interior points allow multi-directional elastic changes; and constrained edge points restrict horizontal displacement, allowing only vertical dynamic adjustments. An elevation gradient weight field is introduced as a deformation adjustment factor. Higher deformation resistance is applied to high-gradient areas, while lower gradient areas are given greater elasticity, achieving a registration strategy based on terrain perception.

[0060] During the registration process, the offset values ​​of each pair of spatially corresponding points on the Z-axis (vertical direction) are recorded in real time. For each geographic entity, a vertical difference layer is overlaid on its base grid. This layer uses 1-meter × 1-meter resolution raster encoding, and each raster cell records the average vertical difference and its standard deviation. After recording on all geographic entities, the layers are integrated into a regional-level vertical error layer and smoothed. Specifically, a bidirectional third-order Gaussian filter kernel is applied to the error layer to preserve the macroscopic error trend while eliminating local noise disturbances. In addition, outliers with a difference greater than 1.0 meter are removed and marked as "extreme error zones," which will be forcibly included in the reconstruction process in subsequent steps. This error layer fully presents the vertical offsets generated by different types of geographic entities during multi-source data alignment and can be used to determine which areas require structural reconstruction and which can maintain the current modeling results.

[0061] Cross-analysis is performed based on the error layer results and the risk-guided grid to label each geographic entity and record its error characteristics. The error range is divided into three categories: low error (less than 0.2 meters), medium error (0.2 to 0.5 meters), and high error (greater than 0.5 meters). Each error category is assigned a corresponding processing strategy: low-error areas can be directly used for indicator calculation; medium-error areas require boundary correction before contour reconstruction; and high-error areas require complete contour and volume consistency reconstruction. Based on this, the error layer is merged with the original weighted initial value field to form a risk-error joint map, allowing subsequent processing to dynamically select correction paths based on actual error conditions. Furthermore, the error information, risk level, displacement, and positioning accuracy of each geographic entity are written into the thematic geographic entity database, achieving spatial representation and tracking management of error data.

[0062] By introducing semantic contour priors using the vertical error field, contour and volume consistency reconstruction is performed based on the parametric template library of building components to correct building boundary twists and facade deformations.

[0063] After registering the oblique photogrammetry point cloud and laser point cloud data and generating a vertical error layer, semantically driven contour and volume consistency reconstruction is required to correct building outline offsets and facade geometric distortions caused by error accumulation. This ensures the geometric accuracy of geographic entities and the integrity of their semantic representation. The process includes the following steps:

[0064] Based on the error distribution information recorded in the vertical error layer, target objects with structural offsets, boundary misalignments, or facade deformations are selected from all geographic entities. The selection criteria include: incomplete boundary closures in the base surface outline, angles between boundary vertices deviating from the regular right-angled grid by more than 15 degrees, and areas in the facade grid where the surface normal direction changes abruptly by more than 30 degrees. For geographic entities meeting any of these conditions, their base boundary point set, facade grid edge point set, and roof projection point set are extracted, and a reconstruction input queue is constructed according to the geographic entity number. Based on this, for each object to be reconstructed, its base boundary point set is projected onto a two-dimensional plane using directional projection to form a building plan outline sketch. A point set sequential rearrangement algorithm is used to reorganize the outline points clockwise to ensure boundary closure, while identifying abnormal side length segments and corner misalignment points, marking them as geometrically abnormal point sets as key areas for subsequent geometric correction.

[0065] To achieve consistency between geometric accuracy and architectural semantic structure during contour reconstruction, a parametric template set for building components was constructed and structural matching was performed. This template set includes rectangular, T-shaped, L-shaped, cross-shaped blocks, symmetrical tower components, multi-tiered building blocks, and pitched roof components. Each component template consists of five parameters: number of vertices, boundary segment length ratio, inter-face angle, block height range, and roof structure type. For each geographic entity to be reconstructed, an initial template set was first determined based on the number of sides of its base surface contour. Then, the template parameters were compared one-to-one based on the distance ratio between boundary points and the vertex interior angle values. Three indicators—Euclidean distance mean square error, mean angle deviation, and side length ratio matching degree—were used for scoring and ranking. Finally, the template with the highest score was selected as the structural prototype for the geographic entity. During template matching, in cases where multiple component types are possible, the type containing the largest building volume was prioritized, and the second-best template was recorded for subsequent iterations.

[0066] Perform a consistent reconstruction of the outline and volume. First, spatially fit the selected parametric template to the location of the original geographic entity. The fitting process uses a control point-driven approach, mapping the template vertices to the corresponding positions of the original boundary point set. For the previously identified geometric anomaly point set, in conjunction with the error values ​​in the vertical error layer, perform reverse offset correction along the Z-axis direction for each point. For boundary points with error values ​​between 0.3 meters and 0.6 meters, perform fine-tuning; for points with errors exceeding 0.6 meters, perform complete boundary reconstruction. Boundary reconstruction transforms irregular boundaries into regular polygons through line segment rearrangement and angle correction. After the planar outline reconstruction is completed, construct the volume structure in the vertical direction based on the principal direction vector of the original roof projection point set, forming a complete three-dimensional closed block. For building entities with roof structures, construct pitched roofs or tiered roof structures based on the roof parameters in the template, and smoothly connect the top edge lines through interpolation to ensure a continuous transition between the top and the facade, avoiding sharp angles or abrupt height changes.

[0067] To ensure the reconstruction results meet accuracy standards in both geometric structure and semantic features, model consistency verification and data update operations are conducted. Geometric consistency verification includes two parts: First, an overlap analysis is performed between the reconstructed model and the laser point cloud, the density of point cloud falling on the model surface is statistically analyzed, and the average distance between the point cloud and the model surface is calculated; second, the difference in the contour boundaries between the reconstructed model and the original point cloud is calculated, and the closure error is checked to see if it is less than 0.2 meters. For semantic consistency verification, the structural template category of the current model is compared using the already classified building type labels. If they are inconsistent, it is judged as a semantic conflict, and it is necessary to revert to the template selection step for rematching. Models that pass consistency verification will be marked as "structure reconstructed" and the original geographic entity model data will be overwritten. Geometric information such as boundary coordinates, block height, facade normal direction, and roof type will be written into the thematic geographic entity database, and the modeling quality level and confidence score of the geographic entity will be updated for confidence screening in subsequent indicator calculation stages.

[0068] Based on the reconstruction results of the contour and volume consistency, a plot unit uncertainty propagation model is established. The area, height and volume parameters are recalculated according to the grid level, spatial attribute confidence layers are generated, and the confidence layer results are written into the thematic geographic entity database.

[0069] After completing the consistency reconstruction of building outlines and volumes, it is necessary to further quantify the reliability of attribute data within the plot. To this end, this invention establishes a plot unit uncertainty propagation model, recalculates area, height, and volume at the grid level, generates spatial attribute confidence layers, and writes the layering results into a thematic geographic entity database. Specifically, the following steps are included:

[0070] Based on the reconstructed 3D model of the geographic entities, each building entity is projected onto a 2D plane in space. A 1m x 1m equally spaced grid is used to divide the basic outline of each building into several spatial grid units. For each grid unit, its corresponding basic surface outline boundary, top surface projection boundary, building height value, volume boundary closure status, and error propagation range are extracted. This information is written into the grid attribute set, providing a structural foundation for the subsequent error propagation model. The building height is the difference between the Z-value of the top point and the Z-value of the foundation center, and the volume is calculated only when the 3D model closure condition is met. This operation ensures that each grid unit has independent geometric representation and index calculation capabilities.

[0071] Based on the grid structure, the vertical error layer generated in the previous stage is imported, and each grid cell is matched according to its position coordinates to extract the vertical error value at its location. Then, combined with building boundary information, the error values ​​are propagated according to the spatial structure. The grid is divided into three types of regions: first, the geometric core region, i.e., the internal region within 2 meters of the building's centroid; second, the edge-sensitive region, i.e., the outer edge region near the outline boundary; and third, the height abrupt change region, i.e., the region where the Z-value fluctuates more than 0.5 meters in the facade direction. Different error propagation weights are set for each type of region: the propagation coefficient is set to 0.5 for the core region, 1.0 for the edge-sensitive region, and 1.5 for the height abrupt change region. The propagation method adopts the weighted neighborhood diffusion method, spreading from the high error point to the surrounding area. During the diffusion process, error attenuation is calculated based on the propagation coefficient to achieve a structural distribution estimation of the vertical error within the grid.

[0072] Based on the error propagation results, the area, height, and volume values ​​of each grid cell are recalculated, and their error magnitudes are labeled. When recalculating the area, boundary error is used as an influencing factor, and the original area value is adjusted according to the distance of the grid cell from the building edge; the closer the area deviates from the boundary line, the larger the area correction value. When recalculating the height, the height value of each grid cell is adjusted to the difference between the original Z-value and its propagation error value to reflect the true vertical offset; the height difference before and after correction is recorded as a height error identifier. Volume is calculated using a three-dimensional column volume calculation method after correcting both area and height, and the volume offset value caused by the propagation error for each column is recorded. Finally, correction values, error ranges, and error proportions for the three indicators (area, height, and volume) are generated for each grid cell.

[0073] Based on the error proportion of each indicator, the spatial attribute confidence level of the grid cells is evaluated. The area error proportion, height error proportion, and volume error proportion are each divided into five levels: within 2% is extremely high confidence, 2%-5% is high confidence, 5%-10% is medium confidence, 10%-20% is low confidence, and above 20% is extremely low confidence. The confidence levels of the three attributes are weighted and synthesized, with area weighted at 0.3, height at 0.4, and volume at 0.3, generating a comprehensive confidence index for each grid cell. The confidence index is mapped to discrete level labels, generating a spatial attribute confidence layer in a five-color "red-orange-yellow-green-blue" grading format, and this layer is overlaid on the boundary of the two-dimensional projected building. This layer not only visually displays the confidence distribution of building spatial indicators but can also be used to filter out abnormal areas and determine locations requiring review or supplementary sampling.

[0074] The generated spatial attribute confidence stratification results are written into the thematic geographic entity database. An independent confidence data structure is established for each geographic entity, including the following fields: grid number, grid center coordinates, area correction value, area error value, height correction value, height error value, volume correction value, volume error value, comprehensive confidence index, confidence level, and color identifier. These fields are bound to the original building model data, enabling the database to simultaneously represent 3D geometry, attribute indicators, and confidence level data, and supporting the retrieval of areas with specific confidence levels through queries. The database also supports area filtering based on confidence levels, providing data filtering criteria for subsequent land parcel identification, redevelopment assessment, and planning feasibility analysis.

[0075] In the thematic geographic entity database, areas with low spatial attribute confidence are identified, an adaptive re-sampling strategy is triggered, and UAVs are dispatched to acquire multi-angle perspective data and perform reverse reconstruction. The vertical drift caused by terrain abrupt changes is suppressed by synthetic perspective observation.

[0076] To enhance the reliability of geographic entity data in high-risk areas, targeted data augmentation and structural reconstruction are required in low-confidence areas. This implementation method uses confidence-layered data already written into the thematic geographic entity database to identify areas with low spatial attribute confidence and triggers an adaptive re-acquisition strategy using unmanned aerial vehicles (UAVs). Combined with multi-angle image reconstruction, vertical error correction is achieved. This process includes the following steps:

[0077] Based on the spatial attribute confidence stratification layer generated and written to the database, all spatial grid cells marked with "low confidence" and "very low confidence" are extracted. The identification process treats each geographic entity as the basic processing object, reading its confidence stratification map one by one and calculating the proportion of low-confidence grid cells. If this proportion exceeds 20% of the total grid cells of the building outline, the entity is marked as an area with insufficient data confidence. Subsequently, spatial analysis is performed on these low-confidence areas to extract their specific locations within the building structure, including foundation surfaces, roof edges, facade junctions, and intersections of adjacent terrain abrupt boundary lines. For marked areas, based on their error directionality records, combined with image acquisition direction and lighting conditions recorded during the original modeling stage, the presence of image occlusion, data voids, low-texture areas, or strong reflection interference sources is analyzed. After confirming that the area meets the requirements for visualization restoration and data acquisition feasibility, a closed polygon is constructed as the target area boundary for UAV flight path design, for subsequent data acquisition path scheduling. This processing flow differs from traditional fixed-area re-sampling methods; it is a target guidance mechanism that actively identifies and locates based on confidence stratification results.

[0078] For the aforementioned enclosed boundary area, an adaptive flight and multi-view image acquisition scheme for UAVs was developed. During flight design, multiple flight paths were generated based on the actual building height, surrounding elevation changes, and occlusion structure information. Each path covered different facades and top edges of the building, ensuring complementary perspectives. The UAV, equipped with a high-precision oblique imaging device, acquired high-resolution oblique image data using low-altitude, close-range, continuous heading shooting. During flight shooting, the camera focal length and angle were adjusted in real time. Side-angle shooting and multi-image overlapping strategies were employed for building corners, shadowed areas, and reflective glass areas to capture structural details to the greatest extent possible. After acquisition, all oblique image data were categorized by timestamp and processed for image correction, color equalization, and geometric alignment. Subsequently, a dense 3D image point cloud was reconstructed using a dense image matching method. In the reconstructed point cloud data, the shooting angle and view vector of each point were marked, and the number of observations from multiple angles at that point was counted to establish an observation overlap map. This observation overlap map was used to select the optimal direction for subsequent reverse modeling.

[0079] Based on the collected multi-angle oblique image point cloud data, a reverse reconstruction process was performed to restore the structure and correct the morphology of low-confidence areas. First, the observation points in each direction were spatially clustered according to their angle vectors, into a main view group and an auxiliary view group. The clearest image with the smallest error in the main view group was selected as a reference, and the building outline and boundary shape from that view were extracted and projected in 3D space to form an initial structural framework. Next, the facade structure change trend was extracted from the auxiliary view group, and boundary point information was superimposed layer by layer to refine the structural framework, especially at roof corners, balcony protrusions, and window sill recesses, where accurate surface meshes were formed through point cloud morphology refinement. After establishing the surface meshes, the surface normal direction was calculated, and all reconstructed surfaces were stitched together into a complete closed block. After completing the block reconstruction, the spatial difference between it and the original low-confidence area model block was calculated, mainly analyzing the change in the vertical direction. If the maximum vertical error of the reconstructed model does not exceed 0.2 meters, the average error is controlled within 0.1 meters, and the structural integrity is free from issues such as broken edges or overlapping patches, then the reconstruction is considered valid. The new reconstructed volume replaces the original low-confidence area model, and the spatial indicators of this volume are recalculated, updating attributes such as area, volume, and height, and generating a new confidence stratification map. This reconstructed confidence stratification map is then written back into the thematic geographic entity database for a comparative analysis with the original confidence map. If the confidence level is significantly improved and there are no longer any extremely low-confidence marker units, it is recorded as "reconstruction successful"; if significantly low-confidence units still exist, it is marked as "partially repaired" and queued for subsequent higher-resolution data collection.

[0080] Based on adaptive re-acquisition data and reverse reconstruction results, during the iterative process of constant energy surface in 3D energy balance construction, the local energy change rate and vertical deformation amplitude of the building surface are monitored. When the energy change or vertical displacement of any region exceeds the preset threshold, a threshold event is triggered. Spatial compensation points with time delay are generated in the triggered region to slow down the rate of geometric change. At the same time, the construction weights are redistributed according to the observation quality of image data in different directions to suppress the spread of vertical error and achieve dynamic correction and stable convergence of geographic entity structure.

[0081] After completing multi-angle data acquisition and reverse reconstruction using UAVs, the remaining structural errors need to be dynamically adjusted to achieve closed-loop correction of the vertical error. This step introduces a manifold constraint mechanism and an energy control model, adjusting the modeling path through phase shifting, and combining micro-delay injection and observation direction rearrangement to ultimately achieve continuous and stable convergence of the model in the vertical dimension. The steps of this method are as follows:

[0082] An error manifold is constructed and a phase migration target is set. Based on the spatial comparison between the reverse-reconstructed 3D model and point cloud data in each direction, the distance values ​​from all points to the model surface are extracted to form a complete residual dataset. In 3D space, with each observation point as the center, the normal directions of its corresponding model patch are connected, and its corresponding normal offset value is calculated. A tensor structure based on distance, direction, and density is established. This tensor structure is the error manifold, and each layer represents the energy accumulation distribution of an error direction. Through the analysis of the tensor structure, error concentration regions are identified, and the error growth direction and diffusion rate are calculated. High gradient distribution points of the residuals are extracted at the manifold boundary, and the error phase migration direction is determined by combining their position and direction change trends. Each migration direction is assigned a "control point" on the structural surface as the trigger basis for dynamic correction. Unlike existing modeling methods that use single model error analysis, this step constructs a high-dimensional error structure manifold to identify potential error paths from the spatial topology, possessing the ability of structure perception and direction prediction.

[0083] An iterative process using constant energy surfaces is employed to mitigate error fluctuations. After determining the phase migration direction, this direction is used as the energy regulation entry point to construct a constant energy surface model. The constant energy surface is a regional equilibrium structure constructed within the 3D mesh model, ensuring that the tension between each facet does not exceed the boundary set value during its changes. An initial energy threshold is set, and iterative adjustments are performed based on the curvature distribution of the triangular mesh facets and the boundary connection state. In each iteration, the curvature change, area deformation rate, and node stretching distance of all structural faces in the vertical direction are measured. If the height difference between two consecutive facets in the vertical direction exceeds 0.2 meters, and the corresponding normal angle is greater than 10 degrees, an energy abrupt change is identified in that region, triggering a "threshold crossing event." This event records the time, location, node number, and current iteration number of the abrupt change. Through continuous iterative optimization, energy is smoothly diffused within the model structure, reducing the occurrence of sudden high-error regions. This step not only constructs a structural surface energy regulation mechanism but also provides temporal and spatial conditions for subsequent dynamic injection and observation direction rearrangement.

[0084] A threshold overflow event is an instantaneous response mechanism triggered when a physical quantity, geometric parameter, or energy index exceeds a preset threshold range during the dynamic modeling or error control of geographic entities. In this invention, threshold overflow events mainly occur during constant energy surface iteration and manifold energy constraint processes. In each iteration, the system monitors the vertical height change, normal angle change rate, stress transmission rate, and energy gradient magnitude of each grid patch in the model. When any of these parameters exceeds the allowable stable range, such as a vertical displacement exceeding 0.2 meters, a normal angle change greater than 10 degrees, or a local energy surge exceeding 15% of the average energy, a threshold overflow event is determined to have occurred. At this time, the threshold overflow event is recorded as a trigger unit containing location, time, direction of change, and magnitude of change, used to identify that the current local area of ​​the model is in an abnormal dynamic state. In other words, a threshold overflow event is essentially a sign of energy or geometric imbalance within the model, reflecting potential unstable factors in the model evolution process.

[0085] The primary function of threshold events is dynamic intervention and real-time correction. When a threshold event is triggered, the modeling process ceases its original evolution and immediately initiates micro-delay injection and beam angle weight rearrangement mechanisms. Specifically, micro-delay injection generates buffer points with temporal lags in abrupt change regions, temporarily slowing the geometric update rate in those regions and thus suppressing further error propagation. Beam angle weight rearrangement redistributes modeling influence based on observation quality in different directions, allowing high-confidence perspectives to dominate local reconstruction and correcting geometric shifts caused by low-quality observations. Through the triggering and response to threshold events, the model can automatically adjust its modeling path at each energy anomaly node, achieving dynamic control effects of local self-correction and global stable convergence. The innovation of this mechanism lies in its transformation of error control from static compensation to a time-driven intelligent feedback process, ensuring that geographic entities maintain vertical stability and geometric continuity during multi-source data fusion and spatiotemporal updates.

[0086] Based on the threshold event, micro-delay point injection and observation direction weight reordering are performed. First, several new spatial points are generated in the neighborhood of the error mutation point. The spatial positions of these points are set on the prediction path of the current model structure change trend, with the time label slightly lagging behind the current iteration frame, and the delay time is controlled between 0.1 and 0.3 seconds. Each delay point is given an error absorption weight, which determines its response priority in the structure fitting process. The introduction of delay points can effectively suppress further divergence of the model structure at high error points, acting as an "energy buffer layer". At the same time, based on the previous point cloud acquisition angle, the observation direction vector corresponding to each point is statistically analyzed, and the observation angle stability, texture richness, and structural integrity score of each direction are calculated. The observation directions are reordered based on the score, and the directions with high stability and high integrity are given higher modeling weights to form the dominant directions in the point cloud projection process. Through the dual control of delay point injection and weight reordering, the structure regulation is more precise, and the fitting stability of the model in the vertical direction is significantly improved. This mechanism is different from the traditional method of simply increasing the point cloud density to solve the error problem. Instead, it actively guides the modeling path away from high error directions.

[0087] The model after adjustment is validated for quality and written to the database. After adjustment, the latest model structure is spatially compared with the model before adjustment. The comparison parameters include changes in the mean vertical error, changes in the number of elevation change points, changes in the proportion of model volume error, and the structural surface kink index. If the adjusted model is better than the pre-adjustment state in all the above indicators, and the mean vertical error is reduced by more than 30%, the adjustment is considered successful. Subsequently, the area, volume, and height of each grid cell of the model are recalculated, and an updated spatial attribute confidence layer map is generated. This layer is bound to the corresponding geographic entity and written to the database, recording the following information: the time and location of all triggered threshold events, the number of injection points, the directional weight rearrangement matrix, the number of constant energy surface iterations, and the energy fluctuation trajectory. All adjustment parameters form a complete adjustment chain log for future model change analysis and strategy optimization.

[0088] This invention significantly enhances the early warning capability for potential error areas by constructing a terrain abrupt change sensitivity map and vertical prior information. Combining spatiotemporal consistency-driven registration and rigid-flexible hybrid constraints, it achieves high-fidelity fusion of multi-source data, reducing registration errors between oblique imagery and laser point clouds under complex terrain conditions. Introducing semantic contour priors and parameterized component templates improves the integrity and structural realism of building contour reconstruction, eliminating identification biases caused by boundary kinks and facade distortions. An uncertainty propagation model and spatial attribute confidence layering mechanism ensure the credibility of key indicator calculations (such as area, height, and volume). Adaptive resampling strategies and reverse reconstruction enhance the compensation capability for weak data areas. Furthermore, a phase migration and beam angle weight rearrangement mechanism based on manifold constraints further achieves dynamic real-time error suppression, forming a self-regulating modeling closed loop. Overall, this method significantly improves the accuracy, robustness, and dynamic response capability of real-scene 3D geographic entities in identifying inefficient urban land use, ensuring higher scientific rigor and practicality for spatial analysis, land assessment, and planning decisions based on 3D models.

[0089] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for dynamic monitoring and management of inefficient land use based on real-scene 3D geographic entities, characterized in that, Includes the following steps: Construct a terrain change sensitivity map and a vertical risk guidance grid, and generate a weighted initial field under the joint drive of elevation gradient, reflection intensity and echo confidence, forming the vertical prior information of geographic entities; Based on the vertical prior information of geographic entities, spatiotemporal consistency forced registration is performed, and local elastic calibration of oblique photogrammetric images and laser point cloud data is carried out through a multi-scale rigid-flexible hybrid constraint mechanism to generate a vertical error field. By introducing semantic contour priors using the vertical error field, contour and volume consistency reconstruction is performed based on the parametric template library of building components to correct building boundary twists and facade deformations. Based on the reconstruction results of the contour and volume consistency, a plot unit uncertainty propagation model is established. The area, height and volume parameters are recalculated according to the grid level, spatial attribute confidence layers are generated, and the confidence layer results are written into the thematic geographic entity database. In the thematic geographic entity database, areas with low spatial attribute confidence are identified, an adaptive re-sampling strategy is triggered, and drones are dispatched to acquire multi-angle view data and perform reverse reconstruction. The vertical drift caused by terrain abrupt changes is suppressed by synthetic view observation. Based on adaptive resampling data and reverse reconstruction results, during the iterative process of constant energy surface in 3D energy balance construction, the local energy change rate and vertical deformation amplitude of the building surface are monitored. When the energy change or vertical displacement of any region exceeds the preset threshold, a threshold event is triggered. Spatial compensation points with time delay are generated in the triggered region to slow down the rate of geometric change. At the same time, the construction weights are redistributed according to the observation quality of image data in different directions to suppress the spread of vertical error.

2. The method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities according to claim 1, characterized in that, The steps for generating vertical prior information for geographic entities are as follows: The digital elevation model, oblique photogrammetry dense point cloud, ground laser reflection intensity information and laser echo confidence data are acquired and then rasterized after unifying the coordinate system. Calculate the elevation gradient factor, reflection intensity fluctuation index, and echo confidence index, and normalize the three indices, then fuse them into a weighted initial value field according to their weights. A graded risk map is generated based on the weighted initial value field, and a two-dimensional guiding grid is constructed. The guiding grid is used as the basis for the structural response of high-risk areas, and a weighted initial value field with spatial risk expression capability is output to form the vertical prior information of geographic entities used to guide subsequent registration and modeling.

3. The method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities according to claim 2, characterized in that, The steps for generating the vertical error field are as follows: Based on the grid cells marked as high-risk and extremely high-risk in the weighted initial value field, geographic entities within the coverage area are extracted, and their three-dimensional structure and texture data are extracted. After unifying the coordinate system of the oblique photogrammetric images and ground laser point clouds collected in the same area, a point-to-point index relationship is established and rigid-flexible hybrid registration is performed. Deformation weight grids are constructed within high-risk areas, deformation rules are set for edge points, internal points and constrained edge points, and elevation gradient factors are introduced as adjustment criteria to complete local elastic registration. A vertical error layer is generated on the geographic entity grid, anomaly areas are marked and merged into the weighted initial value field to form a risk-error joint map, and the error information is written into the thematic geographic entity database.

4. The method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities according to claim 3, characterized in that, The steps for introducing semantic contour priors using the vertical error field and performing contour-volume consistency reconstruction are as follows: Based on the vertical error layer, select geographic entities with boundary misalignment or facade deformation, and extract the boundary point set and roof projection point set; Construct a parametric template set for building components and perform template matching, selecting the template with the highest score as the structural prototype of the geographic entity; The template control points are fitted to the original point set, and the vertical error information is combined to perform reverse offset and boundary reconstruction on the outlier points to complete the contour reconstruction and volume generation. Perform geometric and semantic consistency verification, write the verified reconstruction model into the thematic geographic entity database, and update the modeling quality level and confidence score.

5. The method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities according to claim 4, characterized in that, The steps to write the confidence stratification results into the thematic geographic entity database are as follows: Geographic entities are projected onto a two-dimensional plane using an equidistant grid division method, and the area, height, and volume parameters of each grid cell are extracted. The vertical error layer is matched with the grid position, the error propagation coefficient is set according to the region type, and a weighted diffusion calculation is performed; Based on the propagation error results, the area, height, and volume are corrected, and various spatial attribute error indices are generated; Assign confidence levels based on the proportions of various errors and generate a spatial attribute confidence layer; The spatial attribute confidence stratification results and 3D model information are written into the thematic geographic entity database and a queryable structure is formed.

6. The method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities according to claim 5, characterized in that, The steps for identifying areas with low spatial attribute confidence in the thematic geographic entity database, triggering an adaptive re-sampling strategy, and performing reverse reconstruction are as follows: Read the spatial attribute confidence stratification map from the thematic geographic entity database, filter spatial grid cells with low and very low confidence levels, and determine the target areas in the building outlines that need to be supplemented with data. By combining building height and surrounding terrain information, a drone flight path is generated for the target area, and high-resolution oblique images covering multiple viewing directions are acquired. The acquired oblique images are used to generate dense image point clouds, and reverse reconstruction is performed based on multi-angle observation information. The reconstructed model replaces the original model, the spatial indicators are updated, and the data is written into the thematic geographic entity database.

7. The method for dynamic monitoring and management of inefficient land use based on real-scene three-dimensional geographic entities according to claim 6, characterized in that, The steps for dynamic suppression of vertical error based on adaptive re-acquisition data and reverse reconstruction results are as follows: Construct an error manifold and extract the model error direction tensor to identify error concentration regions, determine the phase shift direction, and identify control points for constraining geometric changes; A constant energy surface iteration mechanism is established at the control site to balance the tension changes between local structural surfaces, and a threshold event is triggered when the surface height change or the energy gradient exceeds the preset range. Within the error region where the threshold event occurs, a set of spatial compensation points with time lag characteristics are generated to slow down the local geometric update speed. The modeling influence weights are redistributed according to the clarity and stability of the observation data in each direction, so that the reconstruction process prioritizes the use of high-quality view data, thereby guiding the model to tend to stabilize in the vertical direction. Vertical error verification is performed on the dynamically adjusted model, the spatial attribute confidence layer is updated, and the adjustment records, threshold event information and the updated confidence layer are written into the thematic geographic entity database to form a closed-loop management from vertical prior generation to dynamic correction.

Images