Instance segmentation and structured topology reconstruction method for complex geometric morphology building

Abstract

The invention discloses an instance segmentation and structured topology reconstruction method for a complex geometric shape building, and relates to the technical field of three-dimensional model generation, and the method comprises the following steps: S1, building point cloud instance segmentation: employing a SoftGroup network to carry out the semantic analysis of an unmanned plane LiDAR point cloud, outputting an independent part instance containing the semantic tag of each building assembly, and carrying out the segmentation of a building point cloud instance; fine segmentation of each building component is realized through a point cloud instance mask; and S2, geometric primitive extraction: grouping instance point clouds according to semantic tags, and executing differential geometric fitting. Through semantic-geometric coupling modeling, unified expression of a special-shaped structure and a regular component is realized, the bottleneck of segmentation and parameterization of a traditional method in a complex building scene is solved, and the method is more efficient and efficient. And the boundary expression of the conical top or the hemispherical top and the supporting structure is simplified by adopting an elliptic curve section, the calculation complexity is reduced, and meanwhile, the integrity of the building framework is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional model generation, and in particular to a method for instance segmentation and structured topology reconstruction of buildings with complex geometric shapes. Background Art

[0002] Instance segmentation and structured topology reconstruction of complex geometric buildings involves extracting semantic building components from unordered point clouds, establishing geometric constraints, and generating regularized 3D models. This technology has significant application value in fields such as the digital preservation of historical buildings, urban planning, and intelligent construction. This technology enables semantic analysis and parametric representation of building components, providing high-precision foundational data for structural health monitoring, heritage restoration, and BIM modeling.

[0003] However, existing methods still have significant drawbacks when dealing with scenarios where multiple types of geometric primitives (such as cones and hemispherical tops) coexist with regular structures (planes and cylinders):

[0004] (1) Traditional point cloud segmentation methods are unable to distinguish semantically between irregular structures (conical tops, hemispherical tops) and regular components, resulting in component segmentation adhesion. In addition, traditional geometric primitive fitting strategies are unable to accurately express the geometric characteristics of complex components.

[0005] (2) The geometric constraint rules in the topology reconstruction process are single, ignoring the multi-level directional combination characteristics of the building structure, resulting in poor model geometric consistency and frequent topological fractures;

[0006] Therefore, a method for instance segmentation and structured topology reconstruction of buildings with complex geometric shapes is proposed to solve the above problems. Summary of the Invention

[0007] The purpose of the present invention is to solve the problems in the prior art and to propose a method for instance segmentation and structured topology reconstruction of buildings with complex geometric shapes.

[0008] A method for instance segmentation and structured topology reconstruction of buildings with complex geometric shapes, comprising the following steps:

[0009] S1. Building point cloud instance segmentation: The SoftGroup network is used to perform semantic analysis on the UAV LiDAR point cloud, outputting independent component instances containing semantic labels of each building component, and achieving refined segmentation of each building component through point cloud instance masks;

[0010] S2, geometric primitive extraction: group instance point clouds according to semantic labels and perform differentiated geometric fitting;

[0011] S3, adjacency matrix construction: Generate the adjacency matrix A=[A ij ], judge Ii and I j Adjacency between two instances;

[0012] S4. Boundary line extraction: For any two adjacent structure instances I i and I j , fracture line modeling based on its structural type combination;

[0013] S5. Feature Vertex Extraction: Extract key building connection points through multi-plane intersection solution and column endpoint interception, combined with bounding box corner point extraction to assist in geometric boundary expression, to form a structured topological connection point set;

[0014] S6. Boundary regularization constraint: Construct a hierarchical constraint model through geometric grouping, main direction extraction and joint optimization.

[0015] Preferably, in step S1, the SoftGroup network is used, the instance segmentation threshold τ = 0.8, and the minimum bounding box size of the instance point cloud is ≥ 1m 3 , eliminating noise components whose size is smaller than a threshold.

[0016] Preferably, in step S2, performing differential geometric fitting includes:

[0017] For plane semantic components, the RANSAC algorithm is used to fit the plane equation:

[0018] a k x+b k y+c k z+d k =0

[0019] Get the normal vector n k =(a k ,b k ,c k ), where a k , b k , c k , d k is the plane parameter;

[0020] Fit the cylindrical axial vector and radius to the column semantic component to construct a cylindrical parameter model;

[0021] For semantic components such as cone tops and hemispherical tops, the coordinates of the highest point are extracted as the structure center, and the radius of the enclosing sphere is recorded.

[0022] Preferably, in step S3, it is determined that i and I j The adjacency relationship between two instances includes calculating the minimum Euclidean distance min||pq||(p∈I i ,q∈I j ):

[0023]

[0024] Among them, A ij is the adjacency matrix element, representing instance I i and I j Whether they are adjacent, ε is the spatial proximity threshold.

[0025] Preferably, in step S4, for any two adjacent structure instances I i and I j , fracture line modeling based on its structural type combination includes:

[0026] For the intersection line between planes, the two plane equations π are solved simultaneously. i and π j :

[0027] π i :a i x+b i y+c i z+d i =0

[0028] π j :a j x+b j y+c j z+d j =0

[0029] Find the parameterized intersection line L ij (t)=P0+t·(n i ×n j ), where t represents the position of the point on the line, n i 、n j is plane π i and π j Normal vector, P0 is any point on the intersection line of the two surfaces;

[0030] For the intersection line between the conical top or hemispherical top and the boundary of the supporting structure, an elliptical curve segment is used for approximation. The equation in the local coordinate system is: Connecting structure vertex V k An approximate cone skeleton is formed, and the major axis of the elliptical curve segment is the radius R of the enclosing sphere k , the short axis is 0.8R k The ellipse parameters are optimized by the Levenberg-Marquardt algorithm.

[0031] Preferably, in step S5, when the three planes of the roof intersect, the simultaneous plane equations are used to construct a linear equation system, and the Cramer's rule is used to directly solve the three-dimensional intersection coordinate V=π1∩π2∩π3, where π iis a plane equation; for instances with the semantic label of cone top or hemispherical top, directly take the vertex coordinates or the center of the sphere of the fitting model; for instances with the semantic label of cylinder, take the upper and lower endpoints along the cylinder axis; for instance point clouds with the semantic label of vertical wall, project them onto a two-dimensional plane in the direction of the normal vector, calculate the axial bounding box, and extract the four corner points as contour vertices.

[0032] Preferably, in step S6, constructing the hierarchical constraint model includes: first, geometrically grouping the boundary lines into orthogonal constraint groups G ortho and Freeform Group G free ; Then, for the orthogonal constraint group G ortho Extract the main direction and ortho Middle boundary angle θ i Perform K-means clustering (k=2), enforce the orthogonal constraint θ2=θ1+π / 2, and generate the main direction vector set Finally, the extracted building boundary lines are jointly optimized: the objective function E is defined total =E angle +λE conn ,in: Force orthogonal group boundaries to align with the main direction; Maintain endpoint continuity through the adjacency matrix; use the Levenberg-Marquardt algorithm to optimize the boundary endpoint coordinates, free group G free Only participate in E conn Calculations are performed to maintain topological connectivity, ultimately obtaining a building outline structure that is connected end to end and complies with topological connectivity rules.

[0033] Compared with the existing technology, the advantages of the present invention are:

[0034] 1. This invention realizes the unified expression of special-shaped structures and regular components through semantic-geometric coupling modeling, solving the bottleneck of segmentation and parameterization of traditional methods in complex architectural scenes.

[0035] 2. The present invention uses elliptical curve segments to simplify the boundary expression between the conical top or hemispherical top and the supporting structure, reducing the computational complexity while ensuring the integrity of the building skeleton.

[0036] 3. The present invention groups boundaries according to geometric characteristics, imposes principal direction constraints on orthogonal groups, and retains geometric freedom for free groups. By jointly optimizing coupled direction alignment and topological continuity, the geometric distortion of inclined surfaces caused by traditional global orthogonal constraints is overcome. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a principle block diagram of the present invention.

[0038] Figure 2 It is the overall workflow diagram of the present invention.

[0039] Figure 3 This is a flow chart of an example segmentation of the present invention.

[0040] Figure 4 This is a component coding diagram after segmentation of an example of the present invention.

[0041] Figure 5 It is the adjacency matrix graph of the present invention.

[0042] Figure 6 This is a schematic diagram of plane boundary fitting of the present invention.

[0043] Figure 7 This is a schematic diagram of the hemisphere top boundary fitting of the present invention.

[0044] Figure 8 Schematic diagram before and after boundary regularization of the present invention. DETAILED DESCRIPTION

[0045] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0046] Reference Figure 1-2 As shown, a method for instance segmentation and structured topology reconstruction of complex geometric buildings includes the following steps:

[0047] S1. Building point cloud instance segmentation: The SoftGroup network is used to perform semantic analysis on the UAV LiDAR point cloud, outputting independent component instances containing semantic labels of each building component. The point cloud instance mask is used to achieve fine segmentation of each building component. Each building component includes roof plane, roof slope, vertical wall, decorative column, hemispherical dome, conical dome, etc.

[0048] S2, geometric primitive extraction: group instance point clouds according to semantic labels and perform differentiated geometric fitting;

[0049] S3, adjacency matrix construction: Generate the adjacency matrix A=[A ij ], judge I i and I j Adjacency between two instances;

[0050] S4. Boundary line extraction: For any two adjacent structure instances I i and I j , fracture line modeling based on its structural type combination;

[0051] S5. Feature Vertex Extraction: Extract key building connection points through multi-plane intersection solution and column endpoint interception, combined with bounding box corner point extraction to assist in geometric boundary expression, to form a structured topological connection point set;

[0052] S6. Boundary regularization constraint: Construct a hierarchical constraint model through geometric grouping, main direction extraction and joint optimization.

[0053] In step S1, the SoftGroup network is used, the instance segmentation threshold τ = 0.8, and the minimum bounding box size of the instance point cloud is ≥ 1m 3 , eliminating noise components whose size is smaller than the threshold. The SoftGroup network can simultaneously process irregular structures (conical tops, hemispherical tops) and regular structures (planes, cylinders), and achieve fine segmentation of each building component through point cloud instance masks to eliminate adhesion between components.

[0054] In step S2, performing differential geometric fitting includes:

[0055] For plane semantic components (roof plane, roof slope, vertical wall), the RANSAC algorithm is used to fit the plane equation:

[0056] a k x+b k y+c k z+d k =0

[0057] Get the normal vector n k =(a k ,b k ,c k ), where a k , b k , c k , d k is the plane parameter;

[0058] Fit the cylindrical axial vector and radius to the column semantic component (decorative column) to construct the column parameter model;

[0059] For semantic components such as cone tops and hemispherical tops, the coordinates of the highest point are extracted as the structure center, and the radius of the enclosing sphere is recorded.

[0060] In the step S3, it is determined that i and I j The adjacency relationship between two instances includes calculating the minimum Euclidean distance min||pq||(p∈I i ,q∈I j ):

[0061]

[0062] Among them, Aij is the adjacency matrix element, representing instance I i and I j Whether they are adjacent, ε is the spatial proximity threshold.

[0063] In step S4, for any two adjacent structure instances I i and I j , fracture line modeling based on its structural type combination includes:

[0064] For the intersection line between planes, the two plane equations π are solved simultaneously. i and π j :

[0065] π i :a i x+b i y+c i z+d i =0

[0066] π j :a j x+b j y+c j z+d j =0

[0067] Find the parameterized intersection line L ij (t)=P0+t·(n i ×n j ), where t represents the position of the point on the line, n i 、n j is plane π i and π j Normal vector, P0 is any point on the intersection line of the two surfaces;

[0068] For the intersection line between the conical top or hemispherical top and the boundary of the supporting structure, an elliptical curve segment is used for approximation. The equation in the local coordinate system is: Connecting structure vertex V k An approximate cone skeleton is formed, and the major axis of the elliptical curve segment is the radius R of the enclosing sphere k , the short axis is 0.8R k The ellipse parameters are optimized by the Levenberg-Marquardt algorithm.

[0069] In step S5, when the three planes of the roof intersect, the linear equations are constructed by simultaneous plane equations, and the three-dimensional intersection coordinates V=π1∩π2∩π3 are directly solved by Cramer's rule, where π iis a plane equation; for instances with the semantic label of cone top or hemispherical top, directly take the vertex coordinates or sphere center of the fitting model, and the coordinate error shall not exceed 5% of the radius of the bounding sphere; for instances with the semantic label of cylinder, take the upper and lower endpoints along the cylinder axis; for instance point clouds with the semantic label of vertical wall, project them onto a two-dimensional plane in the direction of the normal vector, calculate the axial bounding box, and extract the four corner points as contour vertices, with a corner point spacing error of <0.02m.

[0070] In step S6, constructing the hierarchical constraint model includes: first, geometrically grouping the boundary lines into orthogonal constraint groups G ortho (including roof plane, vertical wall) and free form group G free (including roof slope, etc.); then, for the orthogonal constraint group G ortho Extract the main direction and ortho Middle boundary angle θ i Perform K-means clustering (k=2), enforce the orthogonal constraint θ2=θ1+π / 2, and generate the main direction vector set The objective function is defined as the weighted sum of the direction alignment term (calculating the least square deviation between the direction vector of the orthogonal group boundary and the main direction) and the adjacent endpoint continuity term (constraining the adjacent boundary endpoint coincidence error to be less than 0.05m) (the weight of the direction term accounts for ≥ 60%); finally, the extracted building boundary lines are jointly optimized: define the objective function E total =E angle +λE conn ,in: Force orthogonal group boundaries to align with the main direction; The continuity of endpoints is maintained by the adjacency matrix; the Levenberg-Marquardt algorithm is used to optimize the boundary endpoint coordinates. The algorithm iteration termination condition is that the total energy function decreases by ΔE<1e-4 or the number of iterations exceeds 100 times. The free group G free Only participate in E conn Calculations are performed to maintain topological connectivity, ultimately obtaining a building outline structure that is connected end to end and complies with topological connectivity rules.

[0071] Example

[0072] S1. Building point cloud instance segmentation

[0073] Given an input point cloud The network first predicts the semantic labels of each point in parallel and an offset vector towards the instance center The semantic branch is implemented through cross entropy loss Optimize classification ability, offset branch through L1 loss Constrain geometric consistency:

[0074]

[0075] Where N is the total number of points in the point cloud, p i is the coordinate of the point in three-dimensional space, p i =(x i ,y i , z i );s i is the semantic label of the i-th point, is the category set of semantic labels, δ i For point p i The offset vector to the center of the instance, represents the three-dimensional real space, For point p i The geometric center of the instance, y i,c is the semantic truth label.

[0076] SoftGroup introduces a probability-aware soft grouping strategy, for each seed point p j , dynamically aggregate neighboring points p that meet the following conditions i :

[0077]

[0078] When the above three points are met, p i ,p j belong to the same instance, where d(p i ,p j ) is point p i to p j Euclidean distance; r dynamic =α·std({||p i -p j ||2}) is the adaptive radius, where α is the scaling factor, std is the standard deviation of the Euclidean distance between all point pairs in the point cloud; sim(·) is the semantic similarity function, τ s and τ δ is the learnable threshold.

[0079] For the generated candidate regions, a lightweight U-Net network is used to predict the instance score S m ∈[0,1], and merge overlapping candidate instances through non-maximum suppression (NMS), the network structure is as follows Figure 3 As shown, the instance mask M m Determined by the following formula:

[0080]

[0081] in, is the seed point pj is the soft neighborhood of , and γ is the confidence threshold.

[0082] S2. Geometric primitive extraction

[0083] For multiple components of the same semantic type, independent fitting is required based on instance labels. k Corresponding to a structural unit, the fitting method is as follows:

[0084] For plane semantic components (including vertical walls, roof planes, and roof slopes), RANSAC is used to fit the plane equation:

[0085] a k x+b k y+c k z+d k =0

[0086] Get the normal vector n k =(a k ,b k ,c k ), where a k , b k , c k , d k is the plane parameter.

[0087] For column semantic components, fit the column axis v k With radius r k .

[0088] For components such as cone tops and hemispherical tops, the highest point is extracted as the structural center, and the radius of the enclosing sphere and the highest point information are recorded.

[0089] S3. Adjacency matrix construction

[0090] In the topological relationship modeling of building components, the adjacency matrix is ​​generated based on the spatial proximity constraint, defining two structural instances I i and I j The adjacency matrix of Figure 4 and Figure 5 As shown:

[0091]

[0092] Among them A ij is the adjacency matrix element, representing instance I i and I j Whether they are adjacent. ε is the spatial proximity threshold.

[0093] S4. Boundary line extraction

[0094] After completing the geometric fitting of the structural facets and determining the adjacency relationship, it is necessary to further transform the spatial relationship between adjacent structures into explicit boundary segments and feature vertices to construct the building topology skeleton.

[0095] For any two adjacent structure instances I i and I j , and perform fracture line modeling based on the combination of structural types. For planar structures (including vertical walls, roof planes, and roof slopes), directly combine the fitting equations of the two patches:

[0096] π i :a i x+b i y+c i z+d i =0

[0097] π j :a j x+b j y+c j z+d j =0

[0098] Find the parametric form of the intersection line:

[0099] L ij (t)=P0+t·(n i ×n j )

[0100] Among them, t is the position of the point on the straight line, n i 、n j is plane π i and π j Normal vector, P0 is any point on the intersection line of the two surfaces.

[0101] In order to simplify the geometric modeling cost of complex curved surface structures while ensuring the integrity and clarity of the building skeleton structure, the following approximate method is used to express the structure of the cylinder and cone top or hemispherical top structure combination in the specific building point cloud: For the instance of the cone top or hemispherical top, its highest point is extracted as the structure vertex V k , use the adjacency matrix to find the columns and plane roof structures adjacent to the component, and regard them as the support boundary below the vertex; for the intersection lines between the column and the plane, and between the hemispherical top and the plane, use elliptical curve segments to fit their boundary lines, which are expressed as the approximate boundary of the cone base, expressed in the local coordinate system as:

[0102]

[0103] Where (x0, y0) is the coordinate of the center of the ellipse after projection of the cone base, and a and b are the lengths of the semi-axes of the ellipse.

[0104] Connect the boundary elliptical curve segment with the highest point to construct an approximate cone structure, avoiding the reconstruction of the complete free-form surface. At the same time, it can participate in the topology structure optimization, and the results of plane intersection line extraction and hemispherical top boundary line extraction are as follows: Figure 6 and Figure 7 shown.

[0105] S5. Feature vertex extraction

[0106] To form a structured topological model, it is necessary to extract corner points from the intersections as line segment endpoints or facet connection points. The extraction method is as follows:

[0107] For the case where the three planes of the roof intersect, solve the three plane equations and get the intersection coordinates V = π1∩π2∩π3, where π i is a plane equation; for instances with semantic labels of cone top or hemispherical top, directly take the vertex coordinates or sphere center of the fitting model; for cylinders, take the upper and lower endpoints along the cylinder axis; extract the vertical plane boundary points, that is, calculate the axial bounding box of the two-dimensional projection of the patch point set, and extract the four corner points as the contour vertices.

[0108] S6. Boundary Regularization Constraints

[0109] The boundaries are grouped by geometric primitive type, and an optimization model is established to jointly constrain the boundary direction and connectivity. The method flow is as follows:

[0110] Let the input boundary set be Each boundary L i Associated geometric primitive type t i ; Divide the boundary into orthogonal constraint groups G according to type characteristics ortho (including planes, elevations, etc. that require orthogonal / parallel constraints) and free-form group G free (including types such as inclined planes that do not require directional constraints), meeting the following requirements:

[0111]

[0112] (1) To G ortho Perform principal direction analysis on the boundaries in:

[0113] 1) Calculate the direction vectors of each boundary in are the endpoint coordinates;

[0114] 2) Direction angle θ i =arctan(v i ) Perform K-means clustering (k = 2) to obtain the main direction angles θ1 and θ2, which satisfy the orthogonality constraint θ2 = θ1 + π / 2;

[0115] 3) Construct a set of main direction unit vectors where d k =(cosθ k , sinθ k ).

[0116] (2) Define the objective function, including the direction regularization term and the connectivity constraint term:

[0117] 1) Directional regularization term: Force G ortho The middle boundary is aligned with the main direction;

[0118]

[0119] 2) Connectivity constraint: Maintain the endpoint consistency of adjacent boundaries through the adjacency matrix A;

[0120]

[0121] 3) The overall optimization objective is a weighted summation:

[0122] E total =E angle +λE conn

[0123] Among them, λ is the balance weight coefficient, and the optimization variables are the endpoint coordinates of all boundaries.

[0124] (3) E total It is transformed into a nonlinear least squares problem and iteratively optimized using the Levenberg-Marquardt algorithm.

[0125] Free Group G free The boundary only participates in E conn The calculation is done to maintain topological connectivity, and its direction is not explicitly constrained. Finally, the building outline structure that is connected end to end and conforms to the topological connectivity rules is obtained. The effect is as follows Figure 8 shown.

[0126] It is understood from common technical knowledge that the present invention may be implemented by other embodiments that do not depart from its spirit or essential features. Therefore, the embodiments disclosed above are, in all respects, merely illustrative and not exclusive. All modifications within the scope of the present invention or equivalent to the scope of the present invention are intended to be encompassed by the present invention.

Claims

1. A method for instance segmentation and structured topology reconstruction of buildings with complex geometric shapes, characterized by: The following steps are involved: S1. Building point cloud instance segmentation: The SoftGroup network is used to perform semantic analysis on the UAV LiDAR point cloud, outputting independent component instances containing semantic labels of each building component, and achieving refined segmentation of each building component through point cloud instance masks; S2, geometric primitive extraction: group instance point clouds according to semantic labels and perform differentiated geometric fitting; S3, adjacency matrix construction: Generate the adjacency matrix A=[A ij ], judge I i and I j Adjacency between two instances; S4. Boundary line extraction: For any two adjacent structure instances I i and I j , fracture line modeling based on its structural type combination; S5. Feature Vertex Extraction: Extract key building connection points through multi-plane intersection solution and column endpoint interception, combined with bounding box corner point extraction to assist in geometric boundary expression, to form a structured topological connection point set; S6. Boundary regularization constraint: Construct a hierarchical constraint model through geometric grouping, main direction extraction and joint optimization.

2. The method for instance segmentation and structured topology reconstruction of complex geometric buildings according to claim 1, characterized in that: In step S1, the SoftGroup network is used, the instance segmentation threshold τ = 0.8, and the minimum bounding box size of the instance point cloud is ≥ 1m 3 , eliminating noise components whose size is smaller than a threshold.

3. The method for instance segmentation and structured topology reconstruction of complex geometric buildings according to claim 1, characterized in that: In step S2, performing differential geometric fitting includes: For plane semantic components, the RANSAC algorithm is used to fit the plane equation: a k x+b k y+c k z+d k =0 Get the normal vector n k =(a k ,b k ,c k ), where a k , b k , c k , d k is the plane parameter; Fit the cylindrical axial vector and radius to the column semantic component to construct a cylindrical parameter model; For semantic components such as cone tops and hemispherical tops, the coordinates of the highest point are extracted as the structure center, and the radius of the enclosing sphere is recorded.

4. The method for instance segmentation and structured topology reconstruction of a building with complex geometric shapes according to claim 1, characterized in that: In the step S3, it is determined that i and I j The adjacency relationship between two instances includes calculating the minimum Euclidean distance min||pq||(p∈I i ,q∈I j ): Among them, A ij is the adjacency matrix element, representing instance I i and I j Whether they are adjacent, ε is the spatial proximity threshold.

5. The method for instance segmentation and structured topology reconstruction of complex geometric buildings according to claim 1, characterized in that: In step S4, for any two adjacent structure instances I i and I j , fracture line modeling based on its structural type combination includes: For the intersection line between planes, the two plane equations π are solved simultaneously. i and π j : π i :a i x+b i y+c i z+d i =0 π j :a j x+b j y+c j z+d j =0 Find the parameterized intersection line L ij (t)=P0+t·(n i ×n j ), where t represents the position of the point on the line, n i 、n j is plane π i and π j Normal vector, P0 is any point on the intersection line of the two surfaces; For the intersection line between the conical top or hemispherical top and the boundary of the supporting structure, an elliptical curve segment is used for approximation. The equation in the local coordinate system is: Connecting structure vertex V k An approximate cone skeleton is formed, and the major axis of the elliptical curve segment is the radius R of the enclosing sphere k , the short axis is 0.8R k The ellipse parameters are optimized by the Levenberg-Marquardt algorithm.

6. The method for instance segmentation and structured topology reconstruction of complex geometric buildings according to claim 1, characterized in that: In step S5, when the three planes of the roof intersect, the linear equations are constructed by simultaneous plane equations, and the three-dimensional intersection coordinates V=π1∩π2∩π3 are directly solved by Cramer's rule, where π i is a plane equation; for instances with the semantic label of cone top or hemispherical top, directly take the vertex coordinates or the center of the sphere of the fitting model; for instances with the semantic label of cylinder, take the upper and lower endpoints along the cylinder axis; for instance point clouds with the semantic label of vertical wall, project them onto a two-dimensional plane in the direction of the normal vector, calculate the axial bounding box, and extract the four corner points as contour vertices.

7. The method for instance segmentation and structured topology reconstruction of a building with complex geometric shapes according to claim 1, characterized in that: In step S6, constructing the hierarchical constraint model includes: first, geometrically grouping the boundary lines into orthogonal constraint groups G ortho and Freeform Group G free ; Then, for the orthogonal constraint group G ortho Extract the main direction and ortho Middle boundary angle θ i Perform K-means clustering (k=2), enforce the orthogonal constraint θ2=θ1+π / 2, and generate the main direction vector set Finally, the extracted building boundary lines are jointly optimized: the objective function E is defined total =E angle +λE conn ,in: Force orthogonal group boundaries to align with the main direction; Maintain endpoint continuity through the adjacency matrix; use the Levenberg-Marquardt algorithm to optimize the boundary endpoint coordinates, free group G free Only participate in E conn Calculations are performed to maintain topological connectivity, ultimately obtaining a building outline structure that is connected end to end and complies with topological connectivity rules.

Images