A Building Change Detection Method and System Based on Multi-Source Realistic 3D Models

Abstract

The present application provides a method and system for building change detection based on multi-source real-scene three-dimensional models. The method includes: obtaining the triangulation topology data of two-phase real-scene three-dimensional models within the area to be detected, and performing preprocessing and standardization processing; using plane-elevation hierarchical registration technology for automatic registration to achieve coordinate unification without manual intervention; using spatial index difference calculation to identify the three-dimensional change areas of the two-phase real-scene three-dimensional models, and optimizing the three-dimensional change areas by using a multi-scale dilation-erosion combination algorithm; using three-dimensional feature dual-channel filtering for the optimized three-dimensional change areas to verify real building changes; generating four-dimensional spatio-temporal results and outputting a visualization model including elevation change values, material types, and three-dimensional reconstruction parameters. The present application overcomes the machine-readable problem of subtle changes in buildings in large-scale real-scene models, and provides a highly reliable three-dimensional quantitative analysis tool for urban illegal construction monitoring and urban safety management.

Description

Technical Field

[0001] The present invention belongs to the field of urban planning and management, and particularly relates to a method and system for building change detection based on multi-source real-scene three-dimensional models. Background Art

[0002] In the field of urban planning and management, the real-time monitoring and accurate identification of illegal buildings have always been technical difficulties. With the rapid development of high-resolution remote sensing technology, unmanned aerial vehicle oblique photography, and real-scene three-dimensional modeling technology, automated change detection based on stereo images has become possible. Traditional methods mainly rely on two-dimensional pixel comparison of orthophoto images (DOM) or elevation difference analysis of digital surface models (DSM), and manual intervention is used to register two-phase images and mark change areas, mainly including change detection based on the texture features of DOM images, change detection technology based on DSM elevation difference analysis, and illegal building detection methods based on DSM and DOM. However, these methods have problems such as single data dimension, insufficient automation, and serious interference of pseudo-changes. For example, DOM only provides plane texture information and cannot capture the three-dimensional structural changes of buildings (such as height and volume); although DSM contains elevation information, it is easily interfered by non-building changes such as vegetation and temporary objects, and relies on manual selection of homologous points for registration, with low efficiency; existing technologies (such as CN105893972B) need to manually select more than 4 groups of homologous points for rough registration, and then achieve data alignment through pixel-level fine registration. The manual intervention link seriously restricts the feasibility of large-scale applications; traditional methods rely on area thresholds or third-party software to remove vegetation, and cannot effectively distinguish buildings with no actual changes from dynamic noises such as vegetation edges and vehicle movements, with a high false detection rate. In recent years, the popularization of real-scene three-dimensional models (such as OSGB format) has provided new ideas for building change detection. OSGB data completely records the geometric features and texture information of buildings through a triangular mesh structure, and can directly reflect the addition, demolition, renovation, or expansion behaviors of buildings from the three-dimensional space dimension. Summary of the Invention

[0003] Aiming at the problems in the prior art such as strong dependence on manual registration, lack of three-dimensional information analysis dimension, single pseudo-change filtering mechanism, and difficulty in adapting multi-source data, the present invention proposes a method and system for building change detection based on multi-source real-scene three-dimensional models, which not only overcomes the limitations of traditional two-dimensional images and DSM data, but also improves the automation degree and detection accuracy to a new level, providing a reliable technical tool for smart city governance.

[0004] The method includes:

[0005] Obtain the triangular mesh topology data of two-phase real-scene three-dimensional models in the area to be detected, and perform preprocessing and standardization processing on the triangular mesh topology data.

[0006] The plane-elevation hierarchical registration technology is used to automatically register the triangular mesh topological data of two-phase real-scene three-dimensional models after preprocessing and standardization, so as to achieve coordinate unification without manual intervention. The plane-elevation hierarchical registration technology is as follows: First, identify the surface area in the two-phase real-scene three-dimensional models to construct a control point set, and use an improved algorithm ICP (Iterative Closest Point, a precise point cloud registration algorithm) to iteratively calculate the optimal rigid body transformation matrix to realize plane reference correction; then use the moving least squares method to perform three-dimensional spatial interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real-scene three-dimensional models.

[0007] The three-dimensional change regions of the two-phase real-scene three-dimensional models are identified using spatial index difference calculation, and the multi-scale dilation-erosion combination algorithm is used to optimize the three-dimensional change regions. The multi-scale dilation-erosion combination algorithm is as follows: First, use the dilation operation to capture potential change boundaries, and then use an adaptive erosion operator to suppress isolated noise points.

[0008] The optimized three-dimensional change regions are filtered using three-dimensional feature dual channels to verify real building changes. The three-dimensional feature dual channels include a morphology channel and a spectral channel.

[0009] Generate four-dimensional spatio-temporal results and output a visualization model containing elevation change values, material types, and three-dimensional reconstruction parameters.

[0010] Preferably, preprocessing and standardization processing are performed on the triangular mesh topological data. Specifically, the vertex coordinates and texture mapping relationships of the triangular mesh topological data are extracted reversely for preprocessing to obtain the plane offset value, and then three-dimensional spatial difference is performed on the triangular mesh vertices of the first-phase real-scene three-dimensional model in the second-phase real-scene three-dimensional model to obtain the elevation values of the triangular mesh vertices of the first-phase real-scene three-dimensional model in the second-phase real-scene three-dimensional model. The elevation systematic difference of the triangular meshes of the two-phase real-scene three-dimensional models is obtained as the offset value of the coordinate starting points of the two-phase real-scene three-dimensional models, and then a three-dimensional transformation matrix is constructed.

[0011] Preferably, the moving least squares method is used to perform three-dimensional spatial interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real-scene three-dimensional models. Specifically:

[0012] By extracting the elevation difference sample set of the control points in the surface area of the two-phase real-scene three-dimensional models, an adaptive neighborhood window is established with each vertex of the second-phase real-scene three-dimensional model as the center. The moving least squares method is used to perform dynamic Gaussian weighting on the sample points in the neighborhood, construct a local quadratic surface equation and solve the optimal fitting parameters. After predicting the vertex elevation offset amount, the elevation coordinates of the second-phase real-scene three-dimensional model are corrected. The weight function and neighborhood range are iteratively optimized until the residual converges, and finally the elevation offset of the two-phase real-scene three-dimensional models is eliminated.

[0013] Preferably, the improved ICP algorithm is an ICP algorithm constrained by ground features.

[0014] Preferably, the spatial index differential calculation is specifically as follows:

[0015] Construct a tree spatial index database based on the triangulation topology data of two-phase real-scene 3D models. The database includes triangulation vertex coordinates and triangulation patch topology construction, and perform the following 3D differential analysis: Calculate the vertex displacement through the difference in triangulation vertex coordinates at the same geographical location in the two-phase real-scene 3D models; Analyze the percentage change in the area of triangulation patches in the same geographical area in the two-phase real-scene 3D models. When it exceeds 15%, it is regarded as a significant change; Finally, calculate the cumulative value of the volume differences of all changed triangulation patches in the overlapping area of the two-phase real-scene 3D models to obtain the volume change.

[0016] Preferably, first capture potential change boundaries through dilation operations, and then use an adaptive erosion operator to suppress isolated noise points, specifically: Connect and merge adjacent change areas through a dilation operation with a precision of 10 cm, use the adaptive erosion operator to eliminate isolated noise points, and perform region marking, retaining candidate regions with an area > 10 m² and a shape factor > 0.7 to improve the accuracy of change detection.

[0017] Preferably, for the morphological channel, filter the vegetation edge by restricting the maximum width of the minimum bounding rectangle, and use the aspect ratio criterion to exclude temporary enclosures. For the spectral channel, mark the vegetation by calculating the dynamic greening index and determine the material type based on the texture RGB mean clustering.

[0018] This application also proposes a building change detection system based on multi-source real-scene 3D models. The system includes:

[0019] A multi-source real-scene 3D model data input module for obtaining the triangulation topology data of two-phase real-scene 3D models in the area to be detected.

[0020] A preprocessing and standardization module for preprocessing and standardizing the triangulation topology data.

[0021] A full-automatic registration module for automatically registering the triangulation topology data of the two-phase real-scene 3D models after preprocessing and standardization using the plane-elevation hierarchical registration technology to achieve coordinate unification without manual intervention. The plane-elevation hierarchical registration technology is as follows: First, identify the ground surface area in the two-phase real-scene 3D models to construct a control point set, and use the improved ICP algorithm to iteratively calculate the optimal rigid body transformation matrix to correct the plane reference; Then use the moving least squares method to perform 3D spatial interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real-scene 3D models.

[0022] A three-dimensional change detection module is used to identify the three-dimensional change regions of two-phase real-scene three-dimensional models by using spatial index difference calculation, and optimize the three-dimensional change regions by using a multi-scale dilation-erosion combination algorithm. The multi-scale dilation-erosion combination algorithm is as follows: first, capture potential change boundaries by dilation operation, and then use an adaptive erosion operator to suppress isolated noise points.

[0023] An intelligent pseudo-change filtering module is used to perform three-dimensional feature dual-channel filtering on the optimized three-dimensional change regions to verify real building changes. The three-dimensional feature dual-channel filtering includes a morphology channel and a spectral channel. In the morphology channel, the vegetation edges are filtered by restricting the maximum width of the minimum bounding rectangle, and temporary enclosures are excluded by using the aspect ratio criterion. In the spectral channel, the vegetation is marked by calculating the dynamic green index, and the material type is determined based on texture RGB mean clustering.

[0024] A multi-level result output module is used to generate four-dimensional spatio-temporal results and output a visualization model including elevation change values, material types, and three-dimensional reconstruction parameters.

[0025] This application also proposes a computer device, including a processor and a memory for storing processor-executable programs. When the processor executes the programs stored in the memory, the above-mentioned building change detection method can be realized.

[0026] This application also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned building change detection method can be realized.

[0027] By integrating the OSGB reverse parsing engine and the hierarchical intelligent registration technology, the present invention breaks through the limitations of insufficient geometric accuracy and error accumulation in traditional three-dimensional change detection. Based on centimeter-level triangular mesh topology reconstruction and multi-modal filtering algorithms, high-precision three-dimensional change recognition is realized, which is significantly better than traditional two-dimensional image analysis methods; an innovative dual-channel three-dimensional feature filtering mechanism is constructed, combined with dual verification of spatial morphology and material properties, effectively suppressing environmental noise interference such as vegetation jitter; the fully automated four-dimensional spatio-temporal modeling process greatly improves the processing efficiency and breaks through the bottleneck of manual verification. Description of the Drawings

[0028] Figure 1 It is a flowchart of a building change detection method based on multi-source real-scene three-dimensional models.

[0029] Figure 2 It is the detection process of illegal buildings in the embodiment.

[0030] Figure 3 It is a block diagram of a building change detection system based on multi-source real-scene three-dimensional models. Detailed Embodiments

[0031] In order to make the objectives, technical solutions, and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0032] This embodiment provides a method for building change detection based on multi-source real-scene 3D models. The flowchart of the method is as Figure 1 shown, and the method includes the following steps:

[0033] First step, obtain the triangular mesh topology data of two phases of real-scene 3D models in the area to be detected, and perform preprocessing and standardization processing on the triangular mesh topology data.

[0034] The preprocessing and standardization processing of the triangular mesh topology data in this step retains the millimeter-level geometric accuracy of the original model, eliminates the format differences of manufacturers, and is compatible with 5 types of manufacturer formats such as ContextCapture and DJI Zhitu.

[0035] Performing preprocessing and standardization processing on the triangular mesh topology data specifically means: reversely extracting the vertex coordinates and texture mapping relationships of the triangular mesh topology data for preprocessing to obtain the plane offset value, and then performing three-dimensional space difference on the vertices of the first-phase triangular mesh in the second-phase model to obtain the elevation value of the vertices of the first-phase triangular mesh in the second phase. The systematic elevation difference between the two phases of triangular meshes is the offset value of the coordinate origin points of the two phases of real-scene 3D models (systematic difference, and only translation can be considered for the plane rectangular coordinate system within a local range), and a three-dimensional transformation matrix is constructed. Subsequent steps, such as registration, differential calculation, and filtering, are all based on this matrix for three-dimensional space analysis and transformation.

[0036] Second step, use the plane-elevation hierarchical registration technology to automatically register the triangular mesh topology data of the two phases of real-scene 3D models after preprocessing and standardization processing to achieve coordinate unification without manual intervention. The plane-elevation hierarchical registration technology is as follows: First, identify the ground surface area in the two phases of real-scene 3D models to construct a control point set, and use an improved algorithm ICP (Iterative Closest Point, a point cloud fine registration algorithm) to iteratively calculate the optimal rigid body transformation matrix to correct the plane reference; then use the moving least squares method to perform three-dimensional space interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate origin base points of the two phases of real-scene 3D models.

[0037] In this step, the surface area is the vertices of the 20% triangular patches with the lowest local Z values. The specific implementation of using the improved algorithm ICP to iteratively calculate the optimal rigid body transformation matrix for plane datum correction is to extract the triangular mesh vertices in the surface area of the two-phase real-scene 3D models to construct a control point set, and use the improved ICP algorithm to iteratively calculate the optimal rigid body transformation matrix. The transformation matrix includes a rotation matrix R and a translation vector T, aligning the plane coordinates of the second-phase model to the first-phase datum to achieve plane datum correction without manual intervention.

[0038] The three-dimensional spatial interpolation of the second-phase model is performed using the moving least squares method to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real-scene 3D models. Specifically:

[0039] By extracting the elevation difference sample set of the control points in the surface area of the two-phase real-scene 3D models, an adaptive neighborhood window is established with each vertex of the second-phase real-scene 3D model as the center. The moving least squares method is used to perform dynamic Gaussian weighting on the sample points within the neighborhood, construct a local quadratic surface equation and solve for the optimal fitting parameters, predict the vertex elevation offset, and then correct the elevation coordinates of the second-phase real-scene 3D model. The weight function and neighborhood range are iteratively optimized until the residual converges, finally eliminating the elevation offset between the two-phase real-scene 3D models. That is, the moving least squares method is used to perform elevation interpolation of the first-phase triangular mesh vertices in the triangular mesh of the second-phase model. The plane and elevation coordinates of the triangular mesh vertices in the two phases are inconsistent. The planes are unified by interpolating the elevation while keeping the elevations different. The interpolated elevation is subtracted from the elevation of the first-phase triangular mesh vertices as the elevation difference. The elevation differences are statistically analyzed, and the gross errors are removed and used as the elevation system difference between the two-phase models, which is corrected as a systematic error in the later processing.

[0040] The improved ICP algorithm is the ICP algorithm constrained by ground features. The ICP (Iterative Closest Point) algorithm constrained by ground features is an improved point cloud registration method specifically used for plane datum correction of 3D models or point cloud data. Its core idea is to screen the ground feature points as the control point set and perform registration only using the stable geometric features of the ground area, thus avoiding the interference of dynamic or non-rigid structures such as buildings and vegetation. This algorithm is especially suitable for the automated registration scenario of urban real-scene 3D models, such as the coordinate alignment without manual intervention in building change detection.

[0041] This step eliminates the elevation offset caused by the building projection error and achieves centimeter-level spatial consistency. The plane-elevation hierarchical registration technology can directly utilize the triangular mesh compared with the traditional method, thus retaining the model details, significantly improving the accuracy, and requiring no manual intervention in the whole process.

[0042] In the third step, use spatial index difference calculation to identify the 3D change regions of the two-phase real-scene 3D models, and optimize the 3D change regions using a multi-scale dilation-erosion combination algorithm. The multi-scale dilation-erosion combination algorithm is as follows: first, use the dilation operation to capture potential change boundaries, and then use an adaptive erosion operator to suppress isolated noise points.

[0043] In this step, when using spatial index difference calculation to identify the 3D change regions of the two-phase real-scene 3D models, a tree spatial index database is constructed based on the triangulation network topology data of the two-phase real-scene 3D models. The database includes the vertex coordinates of the triangulation network and the topology construction of the triangular patches. Perform the following 3D difference analysis: calculate the vertex displacement by the difference in the vertex coordinates of the triangulation network at the same geographical location in the two-phase real-scene 3D models; analyze the percentage change in the area of the triangular patches in the same geographical region in the two-phase real-scene 3D models. When it exceeds 15%, it is regarded as a significant change; finally, calculate the cumulative value of the volume differences of all the changed triangular patches in the overlapping region of the two-phase real-scene 3D models to obtain the volume change.

[0044] The tree spatial index database is constructed based on the vertex coordinates of the triangulation network and the patch topology extracted from the OSGB format, and is used to efficiently manage the 3D transformation matrix and support spatial difference calculation.

[0045] Vertex displacement: the difference in the vertex coordinates (ΔX, ΔY, ΔZ) of the triangulation network at the same geographical location in the two-phase OSGB models. Area change rate: the percentage change in the area of the triangular patches in the same geographical region in the two-phase OSGB models, used to mark the deformation of the building. Volume change: the cumulative value of the volume differences of all the changed triangular patches in the overlapping region of the two-phase OSGB models, reflecting the overall increase or decrease of the building.

[0046] In addition, in this step, the operation of first using the dilation operation to capture potential change boundaries and then using an adaptive erosion operator to suppress isolated noise points is specifically as follows: connect and merge adjacent change regions through a 10-cm precision dilation operation, use an adaptive erosion operator (noise sensitivity < 0.3) to eliminate isolated noise points, and perform region marking, retaining candidate regions with an area > 10 m² and a shape factor > 0.7 to improve the accuracy of change detection.

[0047] In the fourth step, use 3D feature dual-channel filtering for the optimized 3D change regions to verify real building changes. The 3D feature dual-channel filtering includes a morphological channel and a spectral channel.

[0048] In this step, for the morphological channel, the edges of vegetation are filtered by restricting the maximum width of the minimum bounding rectangle, and temporary enclosures are excluded using the aspect ratio criterion. For the spectral channel, vegetation is marked by calculating the dynamic greening index, and the material type is determined based on the texture RGB mean clustering. The three-dimensional feature dual-channel filtering eliminates the jitter of vegetation edges based on the maximum width threshold of the minimum bounding rectangle, and verifies the changes of real buildings by combining the dynamic greening index with the composite criteria of aspect ratio and volume change rate.

[0049] Step 5: Generate four-dimensional spatio-temporal results and output a visualization model containing elevation change values, material types, and three-dimensional reconstruction parameters.

[0050] This step generates results in the CityGML format compatible with BIM, including three-dimensional boundaries with elevation change values, volume change rates, material types, attribute fields of timestamps (supporting spatio-temporal backtracking), and visualization interfaces supporting direct calls from Revit / ArcGIS, and can generate a heat map for illegal construction warnings.

[0051] This embodiment realizes coordinate unification without manual intervention through local lowest triangular mesh plane registration and elevation datum correction. Combining three-dimensional morphological feature analysis (determination of the maximum width threshold of polygons) and stereoscopic color models (dynamic calculation of the R / G / B ratio), while retaining areas with significant elevation changes, it accurately eliminates non-building interference factors such as vegetation jitter and vehicle displacement. Finally, through spatial interpolation calculation of the native triangular mesh structure of OSGB, the extraction of building change features with centimeter-level accuracy is achieved.

[0052] Another embodiment is to detect the illegal construction and additional floors in a certain community. The detection process is in the single-file processing mode as follows Figure 2 The specific implementation steps are:

[0053] Step 1: Data preparation and full-automatic parsing

[0054] The input data is OSGB data produced by different manufacturers in two phases of a certain community. The first-phase OSGB data is generated by ContextCapture software, and the path is D:\Data\Old\CC_2022, which includes metadata.xml and the Data folder; the second-phase OSGB data is generated by DJI Smart3D software, and the path is D:\Data\New\DJI_2023, which includes metadata.xml and the terra_osgbs folder.

[0055] Step 2: Pretreatment and standardization processing

[0056] Call the open-source OSG library to parse two-phase OSGB data, extract the triangular mesh vertex coordinates, textures, and metadata, and perform automatic standardization processing to unify the coordinate system of the two-phase data to WGS84 and the elevation datum to EGM96, eliminating the manufacturer format differences. Subsequently, generate a gridded three-dimensional point cloud with a resolution of 0.1 meters directly based on the OSGB triangular mesh data.

[0057] Step 3: Fully automatic planar and elevation registration

[0058] Automatically extract the local lowest triangular mesh of the ground / water surface area from the two-phase OSGB data (excluding non-ground triangular patches such as buildings and vegetation). Match the two-phase ground triangular meshes through the ICP (Iterative Closest Point) algorithm, calculate the planar offset (average X offset of 0.3 meters and Y offset of 0.2 meters), and complete the planar registration. Based on the registered planar coordinates, uniformly select 1000 ground reference points from the aforementioned ground triangular mesh vertices of the first-phase OSGB, interpolate the elevation values of these points in the second-phase OSGB data, calculate the average elevation system difference (Δh = 0.15 meters), and automatically correct the elevation datum of the second-phase data.

[0059] Step 4: Extraction of three-dimensional change regions and noise filtering

[0060] Compare the elevation values point by point from the two-phase OSGB data at a grid resolution of 0.1 meters. Set the elevation change threshold to 2.2 meters (ΔH = H_New - H_Old), mark the grid points with elevation differences exceeding the threshold, and generate an initial change region mask map. Then perform intelligent noise filtering. Geometric filtering applies the erosion algorithm (3×3 structuring element) to merge adjacent grid points (connected merging of adjacent change regions to avoid missed detections caused by being excluded due to not reaching the threshold), and exclude regions with an area < 3 square meters (not building changes, such as cars, etc.); polygon width analysis calculates the maximum width of the change region polygon and excludes strip-shaped regions with a width ≤ 1.5 meters (such as vegetation edges and unchanged building edges ---- caused by photographic projection differences and three-dimensional reconstruction errors); greening coefficient filtering extracts the RGB values from the OSGB texture, marks regions with a greening coefficient ≥ 90% (G > R and G > B), and automatically excludes vegetation interference. Finally, apply the Douglas-Peucker algorithm to the remaining regions (such as a rooftop addition area with an area of 12.5 square meters) for vector fitting, with a boundary error < 0.05 meters.

[0061] Step 5: Result output

[0062] Output the textured OSGB change area model (format: Area01_Change.osgb), mark the height change value (ΔH = 3.1 m), and generate a Shapefile file (Area01_Change_Final.shp) at the same time. Its attribute table includes area (12.5 square meters), average height difference (3.1 m), polygon width (5.2 m), greening coefficient (12%), and classification label (illegal building addition - confirmed).

[0063] The present invention also provides an embodiment for monitoring illegal construction in urban areas. The detection process is as follows Figure 2 in the batch processing mode, and the specific steps are as follows

[0064] Step 1: Batch parsing and standardization of multi-source OSGB data

[0065] The input data are 10 OSGB datasets in ContextCapture format in the old data (path: E:\Batch\Old) and 15 OSGB datasets in DJI Terra format in the new data (path: E:\Batch\New). The preprocessing operations include multi-threaded parsing, that is, traversing all OSGB data, calling the open-source OSG library to parse the triangulation network, texture, and metadata; data standardization, unifying the coordinate system to CGCS2000 and the elevation datum to the local elevation system, and automatically repairing the data structure differences of different manufacturers; dynamic scheduling optimization, organizing OSGB data blocks based on the quadtree structure, and real-time scheduling processing nodes to reduce memory occupancy.

[0066] Step 2: Automatic batch registration and change detection

[0067] Match the old and new data blocks according to the geographic range metadata (such as old data BlockA and new data Block C). The overlapping area ratio ≥ 75%, and the data pair with the most recent time and the highest resolution is preferred. Extract the local lowest triangulation network (ground reference) for each pair of data, calculate the planar offset (average X / Y offset ≤ 0.5 m) through the ICP algorithm, interpolate the elevation value based on the vertices of the ground triangulation network, and automatically correct the reference deviation (Δh ≤ 0.2 m). Compare the elevation of each grid point (threshold 2.2 m), fuse geometric and color filtering, eliminate areas with an area < 3 square meters, filter strip-shaped noise with a polygon width ≤ 1.5 m, and mark vegetation areas with a greening coefficient ≥ 90%.

[0068] Step 3: Multi-level result output and visualization

[0069] The phased results include CityZone_1_AllChange.osgb, which contains all three-dimensional change regions (in OSGB format), with attributes including elevation difference and area; CityZone_2_Filtered.shp is a vector file that removes isolated small regions and strip-shaped noises; CityZone_3_NonVegetation.osgb is a three-dimensional model file that further removes vegetation regions (greening coefficient < 90%). The final result is CityZone_Final_Fitted.shp, an optimized vector boundary, and the attribute table includes area, average height difference, maximum width of the polygon, greening coefficient, and classification labels (such as "illegal construction - to be verified", "legal renovation - already filed"); CityZone_3D_ChangeViewer is an interactive three-dimensional visualization tool that supports overlaying and comparing the old and new OSGB models and highlights the changed regions.

[0070] Another embodiment also provides a building change detection system based on multi-source real-scene three-dimensional models. The system block diagram is as Figure 3 shown, and the system includes:

[0071] A multi-source real-scene three-dimensional model data input module for obtaining the triangulation topology data of two-phase real-scene three-dimensional models within the area to be detected.

[0072] A preprocessing and standardization module for preprocessing and standardizing the triangulation topology data;

[0073] A full-automatic registration module for automatically registering the triangulation topology data of the two-phase real-scene three-dimensional models after preprocessing and standardization using the plane-elevation hierarchical registration technology to achieve coordinate unification without manual intervention. The plane-elevation hierarchical registration technology is as follows: First, identify the ground surface regions in the two-phase real-scene three-dimensional models to construct a control point set, and use the improved ICP algorithm to iteratively calculate the optimal rigid body transformation matrix to correct the plane reference; then use the moving least squares method to perform three-dimensional space interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real-scene three-dimensional models.

[0074] A three-dimensional change detection module for using spatial index difference calculation to identify the three-dimensional change regions of the two-phase real-scene three-dimensional models and optimizing the three-dimensional change regions using a multi-scale dilation-erosion combination algorithm. The multi-scale dilation-erosion combination algorithm is as follows: First, use the dilation operation to capture potential change boundaries, and then use the adaptive erosion operator to suppress isolated noise points.

[0075] An intelligent pseudo-change filtering module is used to perform three-dimensional feature dual-channel filtering on the optimized three-dimensional change area to verify real building changes. The three-dimensional feature dual-channel filtering includes a morphology channel and a spectral channel. In the morphology channel, the vegetation edge is filtered by restricting the maximum width of the minimum bounding rectangle, and temporary enclosures are excluded using the aspect ratio criterion. In the spectral channel, vegetation is marked by calculating the dynamic green index, and the material type is determined based on texture RGB mean clustering.

[0076] A multi-level result output module is used to generate four-dimensional spatio-temporal results and output a visualization model including elevation change values, material types, and three-dimensional reconstruction parameters.

[0077] Another embodiment also provides a computer device, including a processor and a memory for storing programs executable by the processor. When the processor executes the programs stored in the memory, the building change detection method in the above embodiments can be implemented.

[0078] Another embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the building change detection method in the above embodiments can be implemented.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described by referring to the preferred embodiments of the present invention, those of ordinary skill in the art should understand that various changes can be made in form and details without departing from the spirit and scope of the present invention defined by the appended claims.

Claims

1. A building change detection method based on multi-source real-scene three-dimensional models, characterized in that, The method includes: Obtaining the triangular mesh topology data of two-phase real scene three-dimensional models within the area to be detected, and performing preprocessing and standardization processing on the triangular mesh topology data; Adopting the plane-elevation hierarchical registration technology to automatically register the triangular mesh topology data of the two-phase real scene three-dimensional models after preprocessing and standardization processing, so as to achieve coordinate unification without manual intervention. The plane-elevation hierarchical registration technology is as follows: First, identify the ground surface area in the two-phase real scene three-dimensional models to construct a control point set, and use the improved ICP algorithm to iteratively calculate the optimal rigid body transformation matrix to realize plane reference correction; Then, use the moving least squares method to perform three-dimensional space interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real scene three-dimensional models; Using spatial index difference calculation to identify the three-dimensional change area of the two-phase real scene three-dimensional models, and optimizing the three-dimensional change area by using a multi-scale dilation-erosion combination algorithm. The multi-scale dilation-erosion combination algorithm is as follows: First, use the dilation operation to capture potential change boundaries, and then use an adaptive erosion operator to suppress isolated noise points; Using three-dimensional feature dual-channel filtering for the optimized three-dimensional change area to verify real building changes. The three-dimensional feature dual-channel filtering includes a morphology channel and a spectral channel; Generating four-dimensional spatio-temporal results and outputting a visualization model including elevation change values, material types, and three-dimensional reconstruction parameters.

2. The building change detection method according to claim 1, characterized in that, Performing preprocessing and standardization processing on the triangular mesh topology data. Specifically, reversely extracting the vertex coordinates and texture mapping relationships of the triangular mesh topology data for preprocessing to obtain the plane offset value, and then performing three-dimensional space difference on the triangular mesh vertices of the first-phase real scene three-dimensional model in the second-phase real scene three-dimensional model to obtain the elevation values of the triangular mesh vertices of the first-phase real scene three-dimensional model in the second-phase real scene three-dimensional model, obtaining the elevation systematic difference of the triangular meshes of the two-phase real scene three-dimensional models as the offset value of the coordinate starting points of the two-phase real scene three-dimensional models, and finally constructing a three-dimensional transformation matrix.

3. The building change detection method according to claim 1, characterized in that, The specific process of using the moving least squares method to perform three-dimensional space interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real scene three-dimensional models is as follows: By extracting the elevation difference sample set of the control points in the ground surface area of the two-phase real scene three-dimensional models, establishing an adaptive neighborhood window centered on each vertex of the second-phase real scene three-dimensional model, using the moving least squares method to perform dynamic Gaussian weighting on the sample points within the neighborhood, constructing a local quadratic surface equation and solving the optimal fitting parameters, predicting the vertex elevation offset and then correcting the elevation coordinates of the second-phase real scene three-dimensional model, iteratively optimizing the weight function and neighborhood range until the residual converges, and finally eliminating the elevation offset of the two-phase real scene three-dimensional models.

4. The building change detection method according to claim 1, characterized in that The improved ICP algorithm is an ICP algorithm constrained by ground features.

5. The building change detection method according to claim 1, characterized in that, The specific process of the spatial index difference calculation is as follows: Construct a tree space index database based on the triangulation network topology data of two-phase real-scene 3D models. The database includes the vertex coordinates of the triangulation network and the topology construction of triangular patches, and perform the following 3D differential analysis: Calculate the vertex displacement by the difference in the vertex coordinates of the triangulation network at the same geographical location in the two-phase real-scene 3D models; Analyze the percentage change in the area of triangular patches in the same geographical area in the two-phase real-scene 3D models. When it exceeds 15%, it is regarded as a significant change; Finally, calculate the cumulative value of the volume differences of all changed triangular patches in the overlapping area of the two-phase real-scene 3D models to obtain the volume change.

6. The building change detection method according to claim 1, characterized in that First, use the dilation operation to capture potential change boundaries, and then use the adaptive erosion operator to suppress isolated noise points. Specifically: Connect and merge adjacent change regions through a dilation operation with a precision of 10 cm, use the adaptive erosion operator to eliminate isolated noise points, and perform region marking. Retain candidate regions with an area > 10 m² and a shape factor > 0.7 to improve the accuracy of change detection.

7. The building change detection method according to claim 1, characterized in that The morphological channel filters the vegetation edges by restricting the maximum width of the minimum bounding rectangle and excludes temporary enclosures using the aspect ratio criterion. The spectral channel marks the vegetation by calculating the dynamic greening index and determines the material type based on texture RGB mean clustering.

8. A building change detection system based on a multi-source real-scene three-dimensional model, characterized in that, The system includes: A multi-source real-scene 3D model data input module for obtaining the triangulation network topology data of two-phase real-scene 3D models within the area to be detected; A preprocessing and standardization module for preprocessing and standardizing the triangulation network topology data; A fully automatic registration module for automatically registering the triangulation network topology data of the two-phase real-scene 3D models after preprocessing and standardization using the plane-elevation hierarchical registration technology to achieve coordinate unification without manual intervention. The plane-elevation hierarchical registration technology is as follows: First, identify the ground surface area in the two-phase real-scene 3D models to construct a control point set, and use the improved ICP algorithm to iteratively calculate the optimal rigid body transformation matrix to correct the plane reference; Then, use the moving least squares method to perform 3D spatial interpolation on the second-phase model to eliminate the elevation offset caused by the coordinate starting base points of the two-phase real-scene 3D models. A 3D change detection module for using spatial index difference calculation to identify the 3D change regions of the two-phase real-scene 3D models, and optimizing the 3D change regions using the multi-scale dilation-erosion combination algorithm. The multi-scale dilation-erosion combination algorithm is as follows: First, use the dilation operation to capture potential change boundaries, and then use the adaptive erosion operator to suppress isolated noise points; An intelligent pseudo-change filtering module for performing 3D feature dual-channel filtering on the optimized 3D change regions to verify real building changes. The 3D feature dual-channel filtering includes a morphological channel and a spectral channel. The morphological channel filters the vegetation edges by restricting the maximum width of the minimum bounding rectangle and excludes temporary enclosures using the aspect ratio criterion. The spectral channel marks the vegetation by calculating the dynamic greening index and determines the material type based on texture RGB mean clustering; A multi-level result output module for generating four-dimensional spatio-temporal results and outputting a visualization model containing elevation change values, material types, and 3D reconstruction parameters.

9. A computer device, comprising a processor and a memory for storing processor-executable programs, characterized in that, When the processor executes the program stored in the memory, it implements the building change detection method described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the building change detection method described in any one of claims 1-7.

Images