Geometric rule driven engineering structure information model reconstruction method

Abstract

The present application belongs to the technical field of point cloud data processing and building information modeling, and discloses a geometric rule driven engineering structure information model reconstruction method. Obtain building indoor point cloud semantic segmentation result data; extract wall point cloud set in the building indoor point cloud semantic segmentation result data, and obtain cross-room wall information by using a wall information extraction method based on geometric rules; extract non-wall component point cloud set in the preprocessed point cloud data, and output non-wall component entity parameters containing geometric center spatial position, three-dimensional geometric size and rotation angle; convert global structure wall center line, wall thickness information, wall height information and non-wall component entity parameters into structured parameter data and import into a three-dimensional parameterized modeling environment, and finally construct a complete building information model. The method improves the extraction integrity of key components such as walls, and significantly improves the spatial accuracy and topological rationality of the building structure model.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing and building information modeling technology, specifically to a geometric rule-driven method for reconstructing engineering structural information models. Background Technology

[0002] With the increasing demand for structural safety analysis and seismic performance assessment of existing buildings, how to quickly and accurately obtain structural analysis models that reflect the actual state of buildings has become an urgent problem to be solved in the engineering field. Building Information Modeling (BIM), as an important foundation for structural analysis, can provide geometric information and spatial relationships of components, playing a vital supporting role in subsequent mechanical modeling and performance assessment. However, in the context of existing buildings, due to the lack of design drawings, frequent structural modifications, and deviations between the actual construction state and the original design, traditional methods relying on drawings or manual modeling are difficult to obtain accurate and reliable BIM models, limiting their application in engineering.

[0003] With the development of 3D laser scanning technology, Scan-to-BIM technology, which utilizes point cloud data for building information reconstruction, has gradually become a research hotspot. This method acquires 3D point cloud data of a building and extracts component information from it to automatically construct the building model. However, in practical applications, due to the complexity of the building's internal structure, mutual occlusion between components, and the limitations of scanning equipment accuracy, the acquired point cloud data often suffers from problems such as missing, incomplete, and noise interference. Especially in wall junction areas and component boundary positions, the point cloud data is prone to breakage and blurring, resulting in inaccurate extraction of component geometric information and seriously affecting the quality of the building information model.

[0004] Existing methods typically extract building component information through point cloud segmentation and geometric fitting. However, these methods are prone to obtaining incorrect geometric information about components when faced with incomplete point cloud data, making it difficult to capture the true structural form. Furthermore, the repetitive representation of shared walls between different rooms and the geometric deviations between intersecting walls further reduce the spatial consistency and structural rationality of the model. Therefore, how to accurately extract component information and optimize the structure by combining the geometric features and spatial constraints of building components when point cloud data is incomplete has become a key problem urgently needing to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a geometric rule-driven method for reconstructing engineering structural information models, in order to solve the problems in the prior art, such as inaccurate extraction of component geometric information, incorrect wall connection relationships, and discontinuous model structure caused by missing, incomplete, and noise interference point cloud data. This enables high-precision extraction of component information and automated construction of building information models in complex indoor building environments.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a geometric rule-driven method for reconstructing engineering structural information models, comprising the following steps: Step 1: Obtain semantic segmentation results of building interior point cloud; Step 2: Extract the wall point cloud set from the semantic segmentation results of the building interior point cloud, and use the wall information extraction method based on geometric rules to obtain cross-room wall information; Step 3: Extract the non-wall component point cloud set from the preprocessed point cloud data, and perform random uniform downsampling without replacement on the non-wall component point cloud set; use a density-based clustering algorithm to perform spatial connectivity aggregation on each downsampled non-wall component point cloud to achieve automatic instance segmentation of independent non-wall component entities; use principal component analysis to calculate the eigenvalues ​​and eigenvectors of each segmented non-wall component entity, determine the eigenvector corresponding to the largest eigenvalue as the major axis principal direction, the second eigenvector as the width direction, and the eigenvector corresponding to the smallest eigenvalue as the thickness direction; perform one-dimensional projection of the point cloud of the non-wall component entity along the major axis principal direction, width direction, and thickness direction respectively, calculate the three-dimensional geometric dimensions of the non-wall component entity based on the difference between the maximum and minimum values ​​of the projection position, calculate the three-dimensional spatial position of the geometric center of the non-wall component entity based on the average of the maximum and minimum projection values, and determine the rotation angle around the vertical axis by combining the projection of the eigenvector in the horizontal plane, outputting the parameters of the non-wall component entity including the spatial position of the geometric center, the three-dimensional geometric dimensions, and the rotation angle; Step four involves converting the global structural wall centerline, wall thickness information, wall height information, and non-wall component entity parameters into structured parameter data and importing them into a 3D parametric modeling environment. Based on the parameter-driven approach, basic geometry corresponding to the semantic tags of each component is automatically generated. The basic geometry is then automatically assembled in spatial posture and position through scaling, rotation, and translation operations, ultimately automatically constructing a complete building information model.

[0007] The wall information extraction based on geometric rules specifically involves: obtaining point cloud contour lines based on the wall point cloud set; extracting the initial wall centerline based on the point cloud contour lines using contour constraint rules; extracting the centerline of a single room wall based on wall orthogonality rules; and obtaining cross-room wall information based on direction consistency rules.

[0008] Project the wall point cloud set along the Z-axis onto a horizontal plane to construct a two-dimensional point set with a strip-shaped distribution feature; calculate the convex hull of the two-dimensional point set to extract a closed outer contour point sequence; connect adjacent points in the closed outer contour point sequence in sequence to construct a point cloud contour line segment set.

[0009] The specific contour constraint rules are as follows: calculate the direction vector and direction angle of each point cloud contour line segment in the point cloud contour line segment set, divide the point cloud contour line segments that are connected end to end and whose direction deviation meets the preset angle threshold into the same line segment set; count all the endpoints in each line segment set, select the two endpoints with the largest spatial distance as the initial wall center line of the corresponding wall, and output the initial wall center line set.

[0010] The wall orthogonality rule specifically involves: identifying all intersecting initial wall centerlines in the initial wall centerline set, and calculating the angle between the intersecting initial wall centerlines; when the angle deviates from a right angle, using the longer initial wall centerline as the reference wall line, and projecting the far endpoint of the shorter initial wall centerline perpendicularly onto the reference wall line to obtain the corresponding projection point; performing correction based on the relative spatial position of the projection point on the reference wall line: if the projection point is located inside the line segment of the reference wall line, then the common intersection of the two initial wall centerlines is directly adjusted to the position of the projection point to achieve orthogonal constraint; if the projection point is located inside the line segment of the reference wall line, then the common intersection of the two initial wall centerlines is directly adjusted to the position of the projection point to achieve orthogonal constraint; if the projection point is located inside the line segment of the reference wall line, then the orthogonal constraint is achieved. On the extended line segment of the reference wall line, a graded fine-tuning is performed based on the deviation value of the included angle and the right angle: when the deviation value of the included angle and the right angle does not exceed a preset threshold, the reference wall line is kept fixed, and only the far end of the shorter wall line is moved to rotate it to be strictly orthogonal to the reference wall line; when the deviation value of the included angle and the right angle is greater than the preset threshold, the end of the reference wall line located on the intersection side is directly adjusted to the projection point position, and the orthogonal relationship is accommodated by changing the end length of the reference wall line; all intersecting initial wall center lines are traversed and adjusted in turn, and the global orthogonality of the overall wall center lines in the room is checked and fine-tuned, and a set of wall center lines for a single room is output.

[0011] The direction consistency rule is as follows: All single-room wall centerline sets are uniformly mapped to the same global horizontal plane to construct a global wall line set; the direction vector of each wall centerline in the global wall line set is normalized, and all direction angles are uniformly mapped to a preset target angle range; based on direction consistency constraints and spatial proximity constraints, multi-level constraint clustering is performed on the direction-normalized wall centerlines, and combined with the result of the overlap relationship of the projection intervals of the endpoints of each wall centerline in the main direction, the global wall line set is divided into several independent global wall sets. Simultaneously, the normal distance between multiple centerlines within each global wall set is calculated as wall thickness information; for each global wall set, the overall direction vector of the global wall set is determined through a main direction estimation method; all original endpoints within the global wall set are projected onto the overall direction vector, and the minimum and maximum values ​​of the projected values ​​are selected as reconstruction endpoints to generate continuous and consistent global structural wall centerlines; combined with the extracted height difference between the floor and ceiling point clouds, wall height information is obtained.

[0012] The beneficial effects of this invention are as follows: Compared with the prior art, this invention, by introducing contour constraint rules, transforms incomplete component information in point clouds into recoverable geometric structure problems, effectively compensating for the impact of missing point cloud data on modeling accuracy and improving the completeness of extracting key components such as walls; through wall orthogonal rules and direction consistency rules, it achieves geometric consistency correction for intersecting walls and walls spanning rooms, significantly improving the spatial accuracy and topological rationality of building structure models; by combining density clustering algorithms and principal component analysis methods, it achieves automatic instance segmentation and parameterized expression of discrete components such as columns, doors, and windows, improving the completeness of information extraction for multiple types of components; at the same time, this invention adopts a parameter-driven modeling approach, which can realize the automatic generation and efficient assembly of building components, improve the construction efficiency and accuracy of building information models, and provide reliable data support for existing building structure analysis and seismic assessment. Attached Figure Description

[0013] Figure 1 This is an overall technical flowchart of an embodiment of the present invention; Figure 2 This is a schematic diagram of the wall contour extraction process according to an embodiment of the present invention; (a) is a two-dimensional projection of the wall point cloud, (b) is the outer contour point sequence, (c) is the point cloud contour line segment, and (d) is the initial wall centerline. Figure 3 This is a schematic diagram of wall geometry correction according to an embodiment of the present invention; (a) is the wall before correction, and (b) is the wall after correction; Figure 4 This is a schematic diagram of a cross-room wall according to an embodiment of the present invention. Detailed Implementation

[0014] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. This embodiment is only used to illustrate the present invention and is not intended to limit the scope of protection of the present invention.

[0015] 1. Preprocessing and extraction of information from individual walls; First, obtain the point cloud semantic segmentation result data, which includes three-dimensional spatial coordinate information and component category labels (such as walls, columns, floors, doors, windows, etc.).

[0016] Next, the wall point cloud set is extracted from the point cloud semantic segmentation result data, denoted as P ={( x i , y i , z i Since occlusion and equipment errors can cause discontinuities in the point cloud surface, this embodiment aggregates the wall point cloud. PProjecting along the vertical direction (Z-axis) onto the horizontal plane (XOY plane) generates a two-dimensional point set with a certain thickness and a strip-shaped spatial distribution. P ’ ={( x i , y i )},like Figure 2 As shown in (a).

[0017] Subsequently, the convex hull constraint algorithm was used to process the two-dimensional point set. P ’ Perform boundary search to extract the closed outer contour point sequence to fill in the missing areas of the point cloud. C ,like Figure 2 As shown in (b). The sequence... C Connect the convex hull points in the graph in adjacent order to construct a set of point cloud contour segments, denoted as . L ={ l 1, l 2,…, l m},like Figure 2 As shown in (c).

[0018] For sets L Each point cloud outline segment l k Its endpoint coordinates are c k ( x k , y k )and c k+1 ( x k+1 , y k+1 The direction vector is calculated using equation (1). : (1) And calculate the corresponding direction angle using equation (2). θ k : (2) Traverse all point cloud contour segments. When two adjacent segments are connected end to end and their orientation angles deviate from the criterion (3): (3) In the formula, θ a , θ b They are line segments a , bDirection angle, Δ θ max If the angle threshold is set to the line segment angle, then point cloud contour line segments that meet the condition will be grouped into the same line segment set. S j For each set of line segments S j The system counts all line segment endpoints and selects the two endpoints with the largest spatial distance as the initial wall centerline endpoints for the corresponding wall entity. This completes the single wall boundary completion and centerline extraction. Vertical stretching then forms a complete wall surface. Figure 2 As shown in (d), the entire process outputs the initial set of wall centerlines.

[0019] 2. Geometric orthogonal correction of intersecting walls within the room; Based on the initial set of wall centerlines output from the above steps, the automatic identification of intersection relationships within the room is first performed, extracting the endpoint information of the intersecting initial wall centerlines. Let the two identified intersecting initial wall centerlines be... L 1=( P 1, P 2) and L 2 = ( P 2, P 3), of which P 2 represents the common intersection of the center lines of the two initial walls.

[0020] Using the coordinates of the endpoints of each line segment, the direction vectors of the two center lines are calculated using equation (4): (4) And calculate the included angle at the intersection according to equation (5). α : (5) When the included angle α When the angle deviates from a right angle (i.e., not equal to 90°), it indicates that the intersection point is affected by noise or missing data, resulting in geometric deviation. Geometric correction is then performed: using the longer initial wall centerline (assuming it is...) L 1) As the reference wall line, the shorter initial wall centerline ( L 2) The distal endpoint P 3-way reference wall line L 1. Project the perpendicular line to obtain the projection point. P 3 ’ .

[0021] According to the projection point P 3 ’ Falling on the reference wall line L The spatial topological relationships on 1 are adjusted hierarchically: if the projection point P 3 ’Located on line segment L Inside 1, the common intersection is directly connected. P Adjust the coordinates of 2 to the projection point P 3 ’ Position, forced intersecting positions satisfy orthogonal constraints; projection points P 3 ’ Falling on the reference wall line L On the extension line of 1, fine-tuning is performed in stages according to the magnitude of the included angle deviation: when the deviation of the included angle from the right angle is less than or equal to the preset threshold of 15°, the baseline wall line is maintained. L The spatial position of line 1 remains fixed; only the endpoints of the short line are adjusted. P The spatial coordinates of 3 make the initial wall centerline L 2 and L 1. Maintain strict orthogonality; when the deviation of the included angle from the right angle is greater than 15°, directly use the reference wall line. L 1's common endpoint P 2. Adjust to projection point P 3 ’ Place.

[0022] After projecting and traversing the endpoints of all intersecting initial wall centerlines within the room, a global closure and orthogonality check is performed on all initial wall centerlines for the entire room. Geometric normalization is then fine-tuned while maintaining the overall accuracy of the room's outline, outputting a set of wall centerlines for a single room. A comparison of the effects before and after correction is provided. Figure 3 (a) and Figure 3 As shown in (b).

[0023] 3. Continuous correction for wall consistency across rooms; After obtaining the set of wall centerlines for each individual room, all line segments are mapped to the horizontal XOY plane of the global 3D Cartesian coordinate system to construct a global wall line set. Because shared walls between adjacent rooms may result in redundant representations or axis misalignment / breakage when extracted independently for each room, such as... Figure 4 The walls in P 2 P 5 and P 3 P 6 and P 1 P 2 and P 3 P As shown in Figure 4, this embodiment performs cross-directional consistency fusion: First, iterate through the global set of wall lines and normalize the direction vector of the center line of each single room wall by mapping the direction angle to [-90°]. ° 90 °The range allows centerlines extracted from different rooms that have opposite topological orientations but belong to the same physical wall to have completely consistent directional representations.

[0024] Next, directional consistency constraints and spatial proximity constraints are introduced to cluster the global wall line set: for any two wall centerlines L i and L j Calculate the angle Δ between the directions of the two. θ ij and normal distance d normal When Δ is satisfied θ ij Less than the preset angle threshold of 15 ° and d normal If the distance is less than the preset distance threshold of 0.3m, it is determined to be a candidate line segment belonging to the same wall entity.

[0025] Subsequently, to prevent the centerlines of parallel but non-adjacent isolation walls in space from being misclassified, the endpoints of candidate line segments are projected onto their main directions, and the projection intervals on the one-dimensional axis are calculated. By determining whether two projection intervals overlap or whether their distance is less than a connectivity threshold, the final aggregation is completed, thereby dividing the global wall line set into several independent global wall sets. S k Each set corresponds to an independent wall in the actual physical space. At the same time, the average normal distance between multiple centerlines within each global wall set is calculated, and this average normal distance is directly defined as the physical thickness information of that wall.

[0026] For each global wall set S k The global orientation vector is determined using a principal orientation estimation method. Then, all original endpoints of the scattered initial wall centerlines within this set are projected onto this global orientation vector, and the global minimum value of the projected value is obtained by searching on the corresponding one-dimensional coordinate axis. P gmin With global maximum value P gmax Using these two extreme points as the endpoints after reconstruction, a continuous and consistent global structural wall centerline is generated, enabling the merging and seamless extension of collinear fracture fragments.

[0027] Finally, the point cloud of the overall ceiling and the point cloud of the floor of the interior building are extracted, and the height difference between the two in the vertical Z-axis direction is calculated. This height difference is used as the height information of the wall corresponding to the center line of the global structure wall.

[0028] 4. Extraction of entity parameters for non-wall components and automatic BIM construction; Based on preprocessed point cloud data, point cloud sets of non-wall components (including columns, floor slabs, doors, and windows) excluding walls are isolated and extracted. When the point cloud size of a single category exceeds a set high-density threshold (200,000 points in this example), random uniform downsampling without replacement is performed on it to reduce computational complexity while preserving spatial geometric contour features.

[0029] The downsampled point clouds of each non-wall component are input into the density-based DBSCAN clustering algorithm. By setting the neighborhood search radius and the minimum number of core points, and relying on the connectivity of the point clouds in three-dimensional space, the point clouds belonging to the same component entity are automatically aggregated into an independent point cluster, thereby achieving automatic instance segmentation of discrete non-wall component entities.

[0030] For each segmented non-wall component entity point cloud, principal component analysis is used to calculate the eigenvalues ​​and eigenvectors of its covariance matrix. The eigenvector corresponding to the largest eigenvalue is determined as the principal direction of the major axis (length direction) of the non-wall component entity, the second eigenvector is determined as the width direction, and the eigenvector corresponding to the smallest eigenvalue is determined as the thickness direction.

[0031] After determining the three principal axes, the point cloud of the non-wall component entity is projected onto these three directional vectors respectively, forming a one-dimensional coordinate projection sequence. The three-dimensional geometric dimensions (length, width, height) of the non-wall component entity in each direction are calculated according to equation (6): (6) In the formula, P max This represents the maximum value of the projected coordinates along this direction. P min This represents the minimum value of the projected coordinates along this direction.

[0032] At the same time, the three-dimensional spatial position of the geometric center of the non-wall component entity is calculated according to equation (7): (7) Assuming the rotation angles of the non-wall component entity around the X and Y axes are both 0 when it is placed vertically, the deflection angle around the Z axis is calculated using the projection direction vector of the major axis principal direction in the horizontal plane (XOY plane), and this angle is taken as the spatial rotation angle of the non-wall component entity. The output then includes the parameters of the non-wall component entity, including its spatial geometric center position, three-dimensional geometric dimensions, and rotation angle.

[0033] Finally, the global structural wall centerline, wall thickness, wall height, and non-wall component entity parameters output from the above steps are converted into standard structured parameter data and imported into a 3D parametric modeling environment. Using a parameter-driven approach, the basic geometry is scaled using 3D geometric dimensions, its spatial orientation is adjusted using rotation angles, and it is translated and placed to the target design position using geometric center coordinates. This achieves automatic generation and precise spatial assembly of all building components, ultimately automatically constructing a building information model that reflects the actual state of the building.

[0034] The above description is only a preferred embodiment of the present invention. For those skilled in the art, various modifications or substitutions can be made without departing from the technical solution of the present invention, and all such modifications or substitutions should fall within the protection scope of the present invention.

Claims

1. A geometric rule-driven method for reconstructing engineering structural information models, characterized in that, The steps include the following: Step 1: Obtain semantic segmentation results of building interior point cloud; Step 2: Extract the wall point cloud set from the semantic segmentation results of the building interior point cloud, and use the wall information extraction method based on geometric rules to obtain cross-room wall information; Step 3: Extract the non-wall component point cloud set from the preprocessed point cloud data, and perform random uniform downsampling without replacement on the non-wall component point cloud set; use a density-based clustering algorithm to perform spatial connectivity aggregation on each downsampled non-wall component point cloud to achieve automatic instance segmentation of independent non-wall component entities; use principal component analysis to calculate the eigenvalues ​​and eigenvectors of each segmented non-wall component entity, determine the eigenvector corresponding to the largest eigenvalue as the major axis principal direction, the second eigenvector as the width direction, and the eigenvector corresponding to the smallest eigenvalue as the thickness direction; perform one-dimensional projection of the point cloud of the non-wall component entity along the major axis principal direction, width direction, and thickness direction respectively, calculate the three-dimensional geometric dimensions of the non-wall component entity based on the difference between the maximum and minimum values ​​of the projection position, calculate the three-dimensional spatial position of the geometric center of the non-wall component entity based on the average of the maximum and minimum projection values, and determine the rotation angle around the vertical axis by combining the projection of the eigenvector in the horizontal plane, outputting the parameters of the non-wall component entity including the spatial position of the geometric center, the three-dimensional geometric dimensions, and the rotation angle; Step four involves converting the global structural wall centerline, wall thickness information, wall height information, and non-wall component entity parameters into structured parameter data and importing them into a 3D parametric modeling environment. Based on the parameter-driven approach, basic geometry corresponding to the semantic tags of each component is automatically generated. The basic geometry is then automatically assembled in spatial posture and position through scaling, rotation, and translation operations, ultimately automatically constructing a complete building information model.

2. The geometric rule-driven engineering structure information model reconstruction method according to claim 1, characterized in that, The wall information extraction based on geometric rules specifically involves: obtaining point cloud contour lines based on the wall point cloud set; extracting the initial wall centerline based on the point cloud contour lines using contour constraint rules; extracting the centerline of a single room wall based on wall orthogonality rules; and obtaining cross-room wall information based on direction consistency rules.

3. The geometric rule-driven engineering structure information model reconstruction method according to claim 2, characterized in that, Project the wall point cloud set along the Z-axis onto a horizontal plane to construct a two-dimensional point set with a strip-shaped distribution feature; calculate the convex hull of the two-dimensional point set to extract a closed outer contour point sequence; connect adjacent points in the closed outer contour point sequence in sequence to construct a point cloud contour line segment set.

4. The geometric rule-driven engineering structure information model reconstruction method according to claim 2, characterized in that, The specific contour constraint rules are as follows: calculate the direction vector and direction angle of each point cloud contour line segment in the point cloud contour line segment set, divide the point cloud contour line segments that are connected end to end and whose direction deviation meets the preset angle threshold into the same line segment set; count all the endpoints in each line segment set, select the two endpoints with the largest spatial distance as the initial wall center line of the corresponding wall, and output the initial wall center line set.

5. The geometric rule-driven engineering structure information model reconstruction method according to claim 4, characterized in that, The wall orthogonality rule specifically involves: identifying all intersecting initial wall centerlines in the initial wall centerline set, and calculating the angle between the intersecting initial wall centerlines; when the angle deviates from a right angle, using the longer initial wall centerline as the reference wall line, and projecting the far endpoint of the shorter initial wall centerline perpendicularly onto the reference wall line to obtain the corresponding projection point; performing correction based on the relative spatial position of the projection point on the reference wall line: if the projection point is located inside the line segment of the reference wall line, then the common intersection of the two initial wall centerlines is directly adjusted to the position of the projection point to achieve orthogonal constraint; if the projection point is located inside the line segment of the reference wall line, then the common intersection of the two initial wall centerlines is directly adjusted to the position of the projection point to achieve orthogonal constraint; if the projection point is located inside the line segment of the reference wall line, then the orthogonal constraint is achieved. On the extended line segment of the reference wall line, a graded fine-tuning is performed based on the deviation value of the included angle and the right angle: when the deviation value of the included angle and the right angle does not exceed a preset threshold, the reference wall line is kept fixed, and only the far end of the shorter wall line is moved to rotate it to be strictly orthogonal to the reference wall line; when the deviation value of the included angle and the right angle is greater than the preset threshold, the end of the reference wall line located on the intersection side is directly adjusted to the projection point position, and the orthogonal relationship is accommodated by changing the end length of the reference wall line; all intersecting initial wall center lines are traversed and adjusted in turn, and the global orthogonality of the overall wall center lines in the room is checked and fine-tuned, and a set of wall center lines for a single room is output.

6. The geometric rule-driven engineering structure information model reconstruction method according to claim 5, characterized in that, The direction consistency rule is as follows: All single-room wall centerline sets are uniformly mapped to the same global horizontal plane to construct a global wall line set; the direction vector of each wall centerline in the global wall line set is normalized, and all direction angles are uniformly mapped to a preset target angle range; based on direction consistency constraints and spatial proximity constraints, multi-level constraint clustering is performed on the direction-normalized wall centerlines, and combined with the result of the overlap relationship of the projection intervals of the endpoints of each wall centerline in the main direction, the global wall line set is divided into several independent global wall sets. Simultaneously, the normal distance between multiple centerlines within each global wall set is calculated as wall thickness information; for each global wall set, the overall direction vector of the global wall set is determined through a main direction estimation method; all original endpoints within the global wall set are projected onto the overall direction vector, and the minimum and maximum values ​​of the projected values ​​are selected as reconstruction endpoints to generate continuous and consistent global structural wall centerlines; combined with the extracted height difference between the floor and ceiling point clouds, wall height information is obtained.

Images