Real estate surveying and mapping three-dimensional construction method based on point cloud model

Abstract

The present application relates to computer graphics image processing and three-dimensional model construction technical field, specifically to the real estate surveying and mapping three-dimensional construction method based on point cloud model. The method constructs boundary topology constraint graph according to boundary point, building contour boundary and interlayer corresponding relationship, carries out multi-scale neighborhood analysis to point cloud model, obtains stable plane, plane intersection line, corner candidate area and point level geometric reliability, and establishes candidate association chain according to boundary closure relationship, adjacent plane angle continuity and interlayer projection relationship; according to position conflict and connection conflict, the conflict propagation path is determined, the point cloud missing conflict and boundary mismatch conflict are distinguished, and local point set redistribution or boundary association adjustment is respectively executed, and then the boundary topology constraint graph is corrected reversely and three-dimensional surface is constructed. The present application can limit the missing area cross-border completion, avoid the boundary deviation diffusion to high reliability point cloud, and make the three-dimensional model keep the boundary not cross, the contour closed and the surface continuous.

Description

Technical Field

[0001] This invention relates to the field of computer graphics and image processing and 3D model construction technology, specifically a method for 3D construction of real estate surveying based on point cloud models. Background Technology

[0002] 3D modeling of real estate surveying typically uses discrete point sets formed by laser scanning, photogrammetry, or multi-source surveying as a foundation. This is followed by denoising, coordinate unification, point cloud segmentation, plane fitting, edge extraction, and surface reconstruction to generate a 3D building model. Conventional processing often involves analyzing the normal vectors, curvature, and density distribution of points in a fixed or multi-scale neighborhood. Points with approximately consistent normal vectors and close spatial distances are grouped into wall, floor, or roof point sets. Building edges are then determined through the intersection of adjacent planes, and closed surfaces are generated using triangular meshes, plane clipping, or regularized contours. For sparse point cloud areas, existing solutions typically fill in missing surfaces using interpolation of neighboring points, extension of adjacent planes, or mesh hole filling, and smoothing processes are used to eliminate isolated points, jagged edges, and local cracks. These solutions rely primarily on the geometric continuity within the point cloud, enabling 3D reconstruction of areas with high point cloud density, minimal occlusion, and clear building boundaries. However, the utilization of boundary lines, building outlines, and inter-layer correspondences is usually limited to post-processing verification or overall alignment.

[0003] To ensure that point cloud models conform to the boundary representation in real estate surveying results, existing technologies often overlay 2D or 3D surveying elements such as boundary points, building outlines, and floor division lines into the point cloud processing flow. Through nearest-neighbor matching, projection snapping, or control point constraints, the extracted planar intersections and corner positions are adjusted to near the surveying boundary. Some solutions establish a correspondence between building outlines and wall point sets. Under rules such as outline closure, orthogonal edges, or consistent projection between upper and lower floors, local point clouds are translated, rotated, resampled, or surface-completed. When point clouds are missing, boundary patches are directly generated according to known outlines and connected to surrounding meshes. This type of processing can reduce the apparent deviation between the model outline and the surveyed lines. However, its matching process typically focuses on a single boundary, a single corner, or a single distance residual, lacking a joint assessment of the internal topological order of the closed outline, the connection relationship between adjacent planes, and the reliability of boundary constraints. When boundary data conflicts with point cloud geometric data, a uniform snapping or uniform relaxation approach is often used.

[0004] Point cloud gaps and boundary ambiguities are prone to occur simultaneously at building exterior corners, eaves-covered areas, adjacent building contact areas, and inter-floor edges. Simply relying on local distances, normal vectors, or plane fitting residuals cannot determine the wall surface or boundary segment to which the missing point belongs, nor can it determine whether the conflict originates from insufficient point cloud sampling or boundary matching errors. Directly supplementing surfaces based on adjacent plane continuity may result in crossing boundary lines or incorrectly connecting adjacent building surfaces. Rigidly snapping the point cloud to the surveyed contour may transmit boundary position deviations to the originally complete, high-confidence wall surface, causing the real plane to shift, tilt, or shrink at corners. Because the existing workflow lacks a collaborative determination mechanism between boundary topological relationships, point cloud geometric confidence, and conflict propagation range, it is difficult to simultaneously maintain boundary non-crossing, contour closure, and surface continuity in occluded and sparse areas, leading to wall adhesion, corner misalignment, local holes, or boundary misalignment in the 3D model. Summary of the Invention

[0005] The purpose of this invention is to provide a method for constructing three-dimensional real estate surveying based on point cloud models, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for constructing 3D models of real estate surveying based on point cloud models includes: constructing a boundary topology constraint diagram based on boundary points, building outline boundaries, and inter-layer correspondences; Multi-scale neighborhood analysis is performed on the point cloud model to be constructed to obtain stable planes, plane intersections, corner candidate regions, and point-level geometric confidence. Based on boundary closure relationships, the continuity of angles between adjacent planes, and inter-layer projection relationships, multiple candidate association chains are established between boundary segments and geometric features; Determine the conflict propagation path based on positional and connection conflicts in the association chain, and distinguish between low-confidence point cloud missing conflicts and high-confidence boundary mismatch conflicts; For different conflict types, perform topologically constrained local point set redistribution or boundary association adjustment, and reverse-correct the boundary topological constraint graph based on the updated plane intersection line to generate a three-dimensional model that satisfies the conditions of no boundary crossing, contour closure and surface continuity.

[0007] Preferably, constructing the boundary topology constraint graph includes: setting the boundary points, building outline corner points and corresponding points between floors as topology nodes, and setting adjacent boundary segments, closed boundary segments and upper and lower layer projection correspondences as topology edges; Write allowed connection objects, prohibited crossing objects, and direction order identifiers for each topological edge; Local subgraphs are established based on the cyclical order of topological nodes within the same closed contour, and cross-contour constraint relationships are formed based on the topological nodes shared by adjacent local subgraphs. For topological edges with repeated projections or adjacency ambiguities, multiple candidate connection states are retained, and the candidate connection states are updated in conjunction with subsequent geometric feature matching results.

[0008] Preferably, the multi-scale neighborhood analysis includes: establishing multiple neighborhoods that expand progressively with the point to be analyzed as the center, and calculating the point distribution covariance, normal vector consistency relationship, and local curvature change relationship in each neighborhood; The set of points that maintain consistent plane affiliation across neighborhood scales is determined as a stable plane; the set of points whose plane affiliation undergoes orderly transformation and whose intersection direction is continuous is determined as candidate points for plane intersection lines; and the intersection region of multiple candidate points for plane intersection lines is determined as a candidate region for a corner. The point-level geometric reliability is generated jointly based on the continuity of point density, the cross-scale stability of normal vectors, and the degree of preservation of geometric categories.

[0009] Preferably, the establishment of the multi-candidate association chain includes: selecting a set of spatially adjacent planar intersection candidate points and the corner candidate region as initial association objects for each boundary segment; According to the directional order of the boundary segments in the closed contour, check the continuity of the included angles of the planes corresponding to adjacent related objects, the consistency of the extension of the intersection lines, and the consistency of the projections of the upper and lower layers. Retain candidate associations that simultaneously satisfy the closed direction relationship and the included angle continuity or the interlayer projection relationship, and connect adjacent candidate associations with shared association objects to form a chain structure; The location where a mutation of associated objects, a reversal of boundary order, or a prohibited crossing relationship is triggered in the chain structure is marked as the conflict starting point.

[0010] Preferably, the linked update of the candidate connection state includes: constructing a set of closed paths composed of candidate topological edges for each local subgraph; Each closed path is mapped to the associated planar intersection line and the corner candidate region to obtain the path start and end spatial closure difference, direction sequence consistency relationship and cross-layer projection offset relationship; A joint comparison is performed on multiple closed paths sharing the same topological edge. Candidate connection states that appear only in a single path and lack continuous geometric support are deleted. Candidate connection states that maintain topological order and have continuous geometric support in adjacent closed paths are locked. The remaining candidate connection states retain a fallback flag.

[0011] Preferably, the generation of the point-level geometric credibility includes: forming a cross-scale feature sequence by combining the plane assignment results of each point at different neighborhood scales, the principal direction of the normal vector, and the curvature category; Structural stability components are generated based on the continuity of category transfer and the consistency of principal direction deflection between adjacent scales, and complete sampling components are generated based on the residual distribution of points and their respective stable planes and the directional concentration of sampling holes in the neighborhood. The structurally stable component and the sampled complete component are propagated in a continuous field along the plane intersection direction, and the propagation is truncated at the prohibited boundary crossing point to obtain the point-level geometric confidence corresponding to the boundary topological constraint graph.

[0012] Preferably, the determination of the conflict propagation path includes: extending bidirectionally from the conflict origin along the adjacent topological edges in the boundary topological constraint graph and the shared plane intersection line in the multiple candidate association chains, and recording the boundary order state, plane attribution state and point-level geometric confidence change corresponding to the extension position; When a conflict extends continuously along a low-confidence region and terminates at a high-confidence stable plane, the propagation range is marked as a point cloud missing conflict region. When a conflict spans multiple high-confidence stable planes and continues to expand along with the same candidate boundary association, the corresponding candidate boundary association is marked as a boundary mismatch conflict object.

[0013] Preferably, the local point set redistribution is performed on the point cloud missing conflict area, including: determining the stable plane and plane intersection line associated with the same closed contour on both sides of the conflict area, and setting the normal vector relationship of the stable plane, the extension direction of the plane intersection line and the prohibited crossing boundary as local reconstruction constraints; Points within the conflict region are redistributed to the corresponding plane point sets according to their residual continuity with each stable plane; In sections without original point support, the completion boundary is defined by the topological correspondence of the intersection lines of the two planes, and the completion point set and the updated plane intersection lines are generated based on the intersection results of the refitted planes.

[0014] Preferably, performing the boundary association adjustment for the boundary mismatch conflict object includes: freezing the high-confidence stable plane and its plane intersection line crossed by the conflict object, and selecting alternative connection states from candidate connection states with a fallback identifier that maintain the closure order and interlayer projection of the frozen geometric features; The affected local subgraph is reconstructed using the alternative connection state, and the consistency of the closed paths before and after reconstruction is checked. When the alternative connection state still triggers the prohibition of crossing relationship, the association between the corresponding boundary segment and the geometric feature is released, the corresponding boundary segment is retained as the boundary to be verified, and no position correction is performed on the frozen geometric feature.

[0015] Preferably, the reverse correction of the boundary topology constraint graph includes: writing back the plane intersection lines obtained after the redistribution of the local point set and the alternative connection states obtained after the boundary association adjustment to the corresponding local subgraph, and regenerating the closed path set and cross-layer projection relationship; For the topological edges that have changed before and after the update, a local iteration range is defined, and within the local iteration range, multiple candidate association, conflict propagation, point set reallocation and boundary association adjustment are performed sequentially. When the locked connection state, stable plane affiliation, and plane intersection connection order remain consistent in two consecutive iterations, the three-dimensional surface is constructed according to the updated closed profile and stable plane boundary.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Boundary points, building outline boundaries, and inter-story correspondences are organized into a boundary topology constraint diagram with closed order, prohibited crossing relationships, and upper / lower floor projection relationships. Stable planes, plane intersections, corner candidate regions, and point-level geometric confidence are obtained through multi-scale neighborhood analysis. Multiple candidate association chains are used to represent the relationship between boundary segments and geometric features. Positional and connection conflicts propagate bidirectionally along topological edges and shared plane intersections. Point cloud missing conflicts and boundary mismatch conflicts are then distinguished based on the confidence distribution within the propagation range. Consequently, low-confidence missing regions undergo local point set redistribution and restricted completion according to adjacent stable planes, intersection extension directions, and prohibited crossing boundaries. High-confidence geometric regions remain locked. Mismatched boundary associations are adjusted through candidate connection state rollback and closed path verification. This process restricts the completed boundaries to topological relationships while preventing biased boundary data from pulling on the complete point cloud. Logically, it blocks cross-boundary surface completion and erroneous wall surface adsorption, maintaining the closure of the building outline, the correctness of corner connections, and the continuity of adjacent surfaces.

[0017] 2. Candidate connection states and multiple candidate association chains in the boundary topology constraint graph can preserve ambiguous relationships between boundaries and geometric features. Under the joint verification of closed paths, direction sequences, adjacent plane angles, and inter-layer projection relationships, erroneous connections lacking continuous geometric support are gradually deleted, and association relationships with stable structural support are locked. Point-level geometric credibility is jointly formed by cross-scale plane assignment, normal vector stability, curvature category, plane residuals, and sampling hole directions, and propagates along plane intersections and is truncated at prohibited boundary crossings, ensuring that the conflict range and the adjustable point set have consistent boundaries. The plane intersections and alternative connection states generated after local point set redistribution are written back to the local subgraph, and the association, conflict propagation, and correction are repeatedly performed within the local scope until the connection states, plane assignments, and intersection order are consistent. This closed-loop processing reduces the participation of irrelevant areas in repeated reconstruction, avoids the spread of local anomalies to the complete wall surface, and ensures that missing point completion, boundary adjustment, and surface construction use the same set of topological and geometric constraints, reducing the probability of wall double layers, edge ghosting, contour breaks, and inter-layer misconnections. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the overall process of constructing a 3D model for real estate surveying based on point cloud models; Figure 2 This is a point-level geometric confidence distribution diagram; Figure 3 A flowchart illustrating the association between boundary topological constraints and point cloud geometric features; Figure 4 A flowchart illustrating the conflict propagation classification, local correction, and closed-loop iteration process. Figure 5 A flowchart illustrating the conflict propagation classification, local correction, and closed-loop iteration process. Figure 6 This is a locally weighted plane fitting residual plot; Figure 7 This is a diagram showing the changes in the local iterative convergence state. Detailed Implementation

[0019] In one embodiment, a method for constructing 3D models of real estate surveying based on point cloud models is provided, including: constructing a boundary topology constraint map based on boundary points, building outline boundaries, and inter-layer correspondences; performing multi-scale neighborhood analysis on the point cloud model to be constructed to obtain stable planes, plane intersections, corner candidate regions, and point-level geometric confidence; establishing multiple candidate association chains between boundary segments and geometric features based on boundary closure relationships, the continuity of angles between adjacent planes, and inter-layer projection relationships; determining conflict propagation paths based on positional conflicts and connection conflicts in the association chains, and distinguishing between low-confidence point cloud missing conflicts and high-confidence boundary mismatch conflicts; performing topology-constrained local point set redistribution or boundary association adjustment for different conflict types, and reversing the boundary topology constraint map based on the updated plane intersections to generate a 3D model that satisfies the conditions of no boundary crossing, outline closure, and surface continuity.

[0020] In this embodiment, please refer to Figure 1 The input data includes a point cloud model, a set of boundary points, a set of building outline boundaries, and a set of inter-floor correspondences under a unified mapping coordinate system. Each point in the point cloud model has at least three-dimensional coordinates. The set of boundary points records the discrete control positions in the property ownership boundary. The set of building outline boundaries records the directed boundary segments formed by connecting adjacent boundary points, building corner points, or outline control points. The set of inter-floor correspondences records the projection belonging relationships between outline nodes of different floors. Before processing, coordinate reference consistency checks are performed on all types of data. Data that cannot be expressed under the same coordinate reference is marked as data to be verified and not included in automatic reconstruction. When removing outliers from the point cloud model, a joint judgment based on local neighborhood distance distribution and normal vector deviation is used to avoid misjudging wall corners, roof polylines, and narrow component edges as noise. For duplicate points, only data records with complete coordinates, normal vectors, and source identifiers are retained to facilitate subsequent tracking of the original point sets corresponding to boundary conflicts.

[0021] Specifically, after spatial segmentation, the point cloud model forms overlapping local processing areas. The overlapping areas cover the stable planes and candidate intersection lines of adjacent segments, ensuring that walls or contours across segment boundaries are not truncated. Within each local processing area, point sets are read according to the neighborhood scale from small to large. Smaller neighborhoods are used to preserve corners and narrow edges, while larger neighborhoods are used to identify continuous planes such as walls, floors, and roofs. The neighborhood scale is not directly set according to a fixed physical length, but rather a scale sequence is generated based on the median of the local point spacing. When the local point spacing changes, the number of points contained in each scale remains relatively stable, thereby reducing feature bias caused by using the same neighborhood radius for dense and sparse point areas. The multi-scale analysis results are stored as corresponding records between point indexes, scale indexes, fitted plane parameters, normal vector directions, curvature categories, and residual distributions.

[0022] Preferably, the point-level geometric confidence level is calculated according to the following formula:

[0023] ; in, This represents the point-level geometric confidence level of the i-th point. This represents the proportion of the i-th point that maintains the same geometric category across all neighborhood scales. This indicates the degree of consistency of the principal directions of the normal vector across adjacent scales. This represents the normalized residual from a point to its corresponding candidate plane. This indicates the degree to which neighborhood sampling holes are concentrated along a single direction. , , and Let each of the four indicators represent a combination coefficient, and let them be among a set of feasible values. , , , When a point inside a wall , , , At that time, the calculated Approaching the high-confidence interval and locking its planar affiliation, when the occluded edge point... , , , When the calculation results enter the low confidence interval, they are allowed to participate in local redistribution, and the coefficients remain unchanged during the same model processing.

[0024] Please refer to Figure 2 , Figure 2The horizontal axis represents the sampling point number after spatial partitioning and multi-scale neighborhood analysis of the point cloud model, arranged according to the scanning order within the local processing area; the vertical axis represents the point-level geometric confidence, with higher values ​​indicating greater stability in terms of cross-scale plane attribution, principal direction of the normal vector, plane residuals, and sampling integrity. The black main curve corresponds to the point-by-point confidence obtained from the point-level geometric confidence calculation formula. The horizontal reference line represents the discrimination boundary between low, medium, and high confidence intervals. Typical wall points, occlusion corner points, and boundary transition points in the legend are used to indicate the positions of different spatial parts in the formula output. Figure 2 As can be seen, the confidence curves for continuous areas inside the wall and on the floor are generally in a high range with small fluctuations, indicating that these points maintain the same geometric category at different neighborhood scales, have continuous normal vector directions, and low normalized residuals and unidirectional sampling holes. Therefore, their stable plane affiliation should be locked. The curves near the occluded corners and eaves edges show continuous troughs, which correspond to the concentration of sampling holes and the increase in residuals. This indicates that these points are more suitable as candidate ranges for local point set redistribution and restricted completion, rather than directly causing translation of the complete wall surface. The confidence of the boundary transition region is in the middle range, reflecting that it may be in the transition zone between plane, intersection, and corner categories. The candidate state should be retained and further judged in conjunction with subsequent correlation chains and conflict propagation results.

[0025] The extraction of stable planes is based on the consistency of plane affiliation across scales. Within each scale, three feature directions are obtained through the point distribution covariance matrix. The direction corresponding to the smallest eigenvalue is determined as the candidate normal vector. The proportion of the smallest eigenvalue to the sum of all eigenvalues ​​is used to represent local planarity. However, the stable plane is not directly determined by the planarity of a single scale. Instead, the angle between the normal vectors of the same point in adjacent scales, the offset of the fitted plane, and the change of neighborhood members are compared. When the plane affiliation is consistent across multiple consecutive scales and the point-level geometric confidence is continuously distributed, the corresponding point set is determined as the stable plane. Points whose affiliation switches orderly between two planes across scales are retained as intersection candidate points. Regions whose affiliation converges between more than three planes and whose spatial location is concentrated are retained as corner candidate regions. Points whose affiliation changes irregularly and whose residuals are discrete are not assigned a structural category for the time being.

[0026] The set of candidate points for the plane intersection is generated from the common neighborhood of two stable planes. For each pair of spatially adjacent stable planes whose normal vectors are not parallel, the theoretical intersection is obtained according to the analytical expression of the two planes. Then, points located in the neighborhood of the theoretical intersection and whose cross-scale planes are ordered from the first plane to the second plane are projected onto the theoretical intersection. Candidate line segments are formed according to the continuity of the projection parameters. The two ends of the candidate line segments are jointly defined by the point density decrease position, the corner candidate region, or the boundary topological node neighborhood. When two stable planes are only locally close but do not have continuous intersection points, no plane intersection is generated. The corner candidate region is formed by the aggregation of the ends of at least two plane intersection candidate line segments. The direction order, associated plane identifier, and cross-scale stability of the intersection line segments are recorded for multi-candidate association of building outline corner points.

[0027] In a preferred embodiment, please refer to Figure 3 The construction of the boundary topology constraint graph includes: setting boundary points, building outline corner points, and corresponding points between floors as topology nodes; setting adjacent boundary segments, closed boundary segments, and upper and lower floor projection correspondences as topology edges; writing allowed connection objects, prohibited crossing objects, and direction order identifiers for each topology edge; establishing local subgraphs based on the cyclical order of topology nodes within the same closed outline, and forming cross-outline constraint relationships based on the topology nodes shared by adjacent local subgraphs; retaining multiple candidate connection states for topology edges with repeated projections or adjacency ambiguities, and updating the candidate connection states in conjunction with subsequent geometric feature matching results; topology nodes also store source type, floor, closed outline, and movable state, where boundary points are set as non-crossable nodes by default, and whether building outline corner points are allowed to be locally adjusted is determined based on their overlap with boundary points.

[0028] The allowed connection objects are used to define the geometric feature categories that a boundary segment can be associated with. For example, the outer contour boundary of a wall can be associated with the plane intersection line formed by the plane of the outer wall and the ground, floor slab, or adjacent outer wall. The inter-floor boundary can be associated with the lateral intersection line formed by the outer wall and floor slab of the upper and lower floors. The prohibited crossing objects are composed of boundary lines, independent contours of adjacent buildings, and common boundaries that do not belong to the same closed contour. The direction sequence identifier is written according to the unified bypass direction of the closed contour, so that the front-to-back relationship of adjacent boundary segments can be compared with the extension direction of the candidate plane intersection line. The cross contour constraint relationship is used to express the situation where adjacent surveying units share the same corner or boundary but have different closed paths. The candidate connection state saves the undetermined association possibility between the boundary segment and multiple candidate intersection lines, without forcibly selecting the nearest candidate object in the initial matching stage.

[0029] In a preferred embodiment, the association cost between the boundary segment and the geometric feature is determined according to the following formula: ; in, This represents the association cost between the b-th boundary segment and the q-th geometric candidate object. This represents the deviation of the boundary segment from the normalized space of the candidate object. This indicates the amount of inconsistency between the boundary direction and the candidate intersection direction. This represents the inconsistency in the relationship between the angle between adjacent planes of candidate objects and the boundary transition. This indicates the deviation of the candidate object in the projection of the upper and lower layers. , , and Let the non-negative combination coefficients sum to 1, and let the boundary segment be a single-layer outer contour with no inter-layer corresponding information. Then, the remaining coefficients are normalized to improve performance when there is a clear inter-layer correspondence. The value of is determined when the spatial distance is close but the directional order is opposite. Smaller and When the spatial distance is relatively large, the association cost remains high and the object does not enter a locked state. When the spatial distance is slightly large but the closing direction, included angle, and inter-layer projection are all continuous, the corresponding candidate object can be retained in the association chain.

[0030] The process of establishing multiple candidate association chains selects the set of candidate points of spatially adjacent plane intersections and candidate regions of corners for each boundary segment as the initial association objects. The search range of the initial association objects is generated based on the relative position of the nodes at both ends of the boundary segment and the bounding range of the point cloud. A uniform distance range covering the entire point cloud model is not used. Then, according to the directional order of the boundary segment in the closed contour, the angular continuity, intersection extension consistency and upper and lower layer projection consistency of the planes corresponding to the adjacent association objects are checked. Candidate associations that simultaneously satisfy the closed direction relationship and the angular continuity or inter-layer projection relationship are retained. Adjacent candidate associations are connected by shared association objects to form a chain structure. Each association item in the chain structure stores the predecessor association, successor association, shared plane, shared corner and association cost. The position where the association object mutation, boundary order reversal or prohibited crossing relationship is triggered is marked as the conflict starting point.

[0031] Specifically, when two consecutive building outline boundaries form a corner in a two-dimensional projection, the two plane intersection lines associated with them should intersect within the same corner candidate area or extend to intersect within a local tolerance range. The order of the external wall planes associated with the two intersection lines should be consistent with the outline's direction of travel. If the two boundary segments are associated with the wall intersection lines of adjacent buildings, even if their spatial distance is small, a connection conflict will occur due to the prohibition of crossing or the discontinuity of the shared plane. When a single boundary segment is roughly parallel to the candidate intersection line but is offset as a whole, a positional conflict occurs. When the candidate intersection line is interrupted in the point cloud occlusion area but both sides of the interruption are associated with the same pair of stable planes, it is not directly identified as a connection conflict. Instead, the interruption area is recorded as the area to be propagated for analysis, and the subsequent confidence distribution determines whether it belongs to point cloud missing or boundary mismatch.

[0032] Preferably, for each local subgraph, a set of closed paths composed of candidate topological edges is constructed. Each closed path is mapped to the associated planar intersection and corner candidate regions to obtain the spatial closure difference between the beginning and end of the path, the consistency relationship of the direction sequence, and the cross-layer projection offset relationship. Multiple closed paths sharing the same topological edge are jointly compared. Candidate connection states that appear only in a single path and lack continuous geometric support are deleted. Candidate connection states that maintain topological order and have continuous geometric support in adjacent closed paths are locked. The remaining candidate connection states retain a fallback indicator. The closed path is not limited to the complete outer contour of the building, but also includes local closed loops composed of shared boundaries, local corners, and inter-layer projection edges, so as to use local topological relationships to check candidate connections when the complete contour is occluded.

[0033] The inconsistency of a closed path is calculated using the following formula: ; in, This represents the inconsistency degree of the p-th closed path. Let represent the displacement vector formed by the q-th candidate geometric line segment in the path along the path direction. This represents the difference in closure between the beginning and end of the space. This indicates the number of candidate line segments in the path whose directional order is opposite to the boundary topological order. This indicates the total number of candidate line segments contained in the path. This represents the normalized projection offset of the path between corresponding nodes on adjacent floors. , and They are respectively represented as , and The corresponding combination coefficients sum to 1, when a single-layer path does not involve inter-layer projection. For example, if four candidate line segments are closed at both ends but one of them is reversed in direction, the first term is close to 0 while the second term remains non-zero. Thus, the candidate connection with the reversed direction will not be incorrectly locked due to the closure of the space.

[0034] The linked update of candidate connection status is performed after each closed path verification. If a candidate connection participates in multiple closed paths at the same time, its stability is judged based on the change in inconsistency of each path. When the inconsistency of multiple related paths decreases synchronously after a candidate connection is removed, and there are other candidate connections that maintain the topological connectivity of all paths, the removed candidate connection is marked as having no continuous geometric support. When a candidate connection maintains consistent directional order, consistent shared plane, and continuous inter-layer projection in multiple adjacent closed paths, it is marked as locked. In the subsequent local point set reassignment, the locked state only allows updating the specific position of the associated geometric object and does not allow changing to other building outlines. Candidate connections with rollback indicators are used for boundary mismatch conflict handling, so that erroneous associations can be undone without regenerating the entire boundary topology constraint diagram.

[0035] Furthermore, the generation of point-level geometric credibility includes: forming a cross-scale feature sequence by combining the plane assignment results, principal direction of normal vector, and curvature category of each point at different neighborhood scales; generating structural stability components based on the continuity of category transfer and consistency of principal direction deflection between adjacent scales; generating sampled complete components based on the residual distribution of the point and its belonging stable plane and the directional concentration of sampling holes in the neighborhood; propagating the structural stability components and sampled complete components along the plane intersection direction and truncating the propagation at the prohibited boundary crossing point to obtain point-level geometric credibility corresponding to the boundary topology constraint map. The continuous field propagation is only carried out between points sharing the same stable plane or the same candidate intersection line to avoid high credibility being transmitted from adjacent point sets that do not belong to the same building surface to the occluded area.

[0036] In the process of generating structural stability components, the geometric category changes in the cross-scale feature sequence are encoded into four states: plane preservation, orderly transformation from plane to intersection line, orderly transformation from intersection line to corner, and disordered jump. The plane preservation state corresponds to the interior of the wall or floor slab, the orderly transformation from plane to intersection line corresponds to the position near the edge of the plane, the orderly transformation from intersection line to corner corresponds to the neighborhood of the building corner, and the disordered jump state corresponds to noise, mixed sampling, or incorrect plane assignment. The consistency of the principal direction deflection of the normal vector is determined by comparing the sequence of principal direction angles of adjacent scales. If the angle remains stable as the scale increases, the structural stability component is increased. If the angle changes alternately between multiple directions, the structural stability component is decreased. The sampling complete component considers the number of neighborhood points, the direction of voids, and the residual skewness at the same time, so that the points that are uniformly sparse but structurally continuous can be distinguished from the missing points formed by unilateral occlusion.

[0037] In this embodiment, please refer to Figure 4The conflict propagation path extends bidirectionally from the conflict origin along adjacent topological edges in the boundary topological constraint graph and the intersection line of the shared plane in the multi-candidate association chain. During the expansion, the boundary order state, plane belonging state, and point-level geometric confidence change corresponding to the current position are recorded. The propagation does not use a fixed number of layers for truncation, but stops when it reaches the locked topological node, inside the high confidence stable plane, when crossing the boundary is prohibited, or when there is no shared geometric object. When the conflict extends continuously along the low confidence region and terminates at the high confidence stable plane, the propagation range is marked as a point cloud missing conflict region. When the conflict crosses multiple high confidence stable planes and continues to extend with the same candidate boundary association, the corresponding candidate boundary association is marked as a boundary mismatch conflict object. When both types of features exist at the same time, the propagation path is segmented at the confidence change position.

[0038] The cumulative amount of conflict propagation paths is calculated using the following formula: ; in, Indicates the propagation path The cumulative amount, where u represents the point cloud nodes or associated nodes traversed by the path. This represents the point-level geometric confidence level corresponding to the node. This is a discrete flag indicating when a node triggers an abnormal boundary order or prohibits crossing relationships; it is 0 when no boundary is triggered and 1 when a boundary is triggered. A discrete flag indicating a sudden change in the association state of a node relative to its predecessor; it is 0 when there is no change and 1 when there is a change. , and As a non-negative coefficient, if the path continuously passes through low-confidence points and does not trigger boundary anomalies, the cumulative amount mainly comes from... It is then classified as point cloud missing propagation. If the path crosses high-confidence points but continuously triggers the same boundary association anomaly, then... and The dominant cumulative amount is included in the boundary mismatch propagation.

[0039] Please refer to Figure 5 , Figure 5 The horizontal axis represents the propagation steps taken as the path extends from the conflict origin along the intersection of the boundary topological edge and the shared plane; the vertical axis represents the cumulative amount of the conflict propagation path, which measures the degree to which the current propagation path is affected by low-confidence points, boundary order anomalies, and abrupt changes in associated states. Figure 5 The multiple curves in the graph correspond to point cloud missing propagation, boundary mismatch propagation, and mixed propagation paths, respectively. The line shape reflects the change in cumulative amount as the propagation steps increase. The legend indicates different propagation types, and reference markers are used to help determine the intervals where the path reaches segmentation processing or termination of analysis. From Figure 5As can be seen, the point cloud missing propagation curve accumulates gradually with the propagation steps in the low confidence region, and the growth slows down after entering the high confidence stable planes on both sides. This indicates that this type of conflict is mainly caused by insufficient point-level geometric confidence, and it is suitable to perform local point set redistribution according to adjacent stable planes and prohibit crossing boundaries. The boundary mismatch propagation curve continues to rise when crossing high confidence points. The main reason is not point cloud instability, but the abnormal boundary order and the continuous appearance of associated mutation markers. This indicates that the stable plane should be frozen and alternative states should be selected from the fallback candidate connections. Hybrid paths usually exhibit low confidence propagation in the early stage, followed by a slope change due to associated mutations. It is necessary to process them in segments at the confidence mutation locations to avoid confusing occlusion missing with boundary mismatch as the same problem.

[0040] The boundary of the point cloud missing conflict region is jointly determined by the propagation termination position, the prohibited crossing boundary, and the support range of the stable plane. During processing, the stable planes and plane intersection lines associated with the same closed contour on both sides of the conflict region are determined. The normal vector relationship of the stable plane, the extension direction of the plane intersection line, and the prohibited crossing boundary are set as local reconstruction constraints. The points in the conflict region are redistributed to the corresponding plane point sets according to the residual continuity relationship with each stable plane. In the section without original point support, the completion boundary is limited by the topological correspondence of the plane intersection lines on both sides. The completion point set and the updated plane intersection line are generated based on the intersection result of the refitted plane. The completion point set has a generation source identifier and is not confused with the original measurement points. It can be deleted and regenerated when the boundary association changes in the future.

[0041] Local point set redistribution does not simply assign each point within a conflict region to the nearest plane. Instead, it establishes a set of candidate planes and compares the orthogonal residuals from a point to a candidate plane, the angle between the point's original normal vector and the plane's normal vector, the plane membership continuity of adjacent points, and the boundary prohibition relationship. When a point is located in the intersection neighborhood of two candidate planes, its intersection candidate attribute is retained without being forced into a single plane. When assigning a point to a candidate plane would cause it to cross a boundary with the same locally connected point set, the corresponding assignment result is excluded. Weighted fitting is performed on the redistributed plane point set, with high-confidence original points receiving higher weights and low-confidence points receiving lower weights. The completion points do not participate in determining the plane parameters in the first fitting and are only generated based on the plane intersection relationship after the plane parameters have stabilized.

[0042] No. The local weighted fitting objective of each candidate plane is determined according to the following formula: ; in, This represents the fitting target value for the g-th candidate plane. This represents the set of original points temporarily assigned to the g-th candidate plane. This represents the coordinate vector of the i-th point. Describes the unit normal vector of the candidate plane. Indicates transpose. This represents the offset of the candidate plane. Represents the set of locally adjacent pairs of points. Let $\frac{i}{g}$ represent the attribute variable of the $i$-th point to the $g$-th candidate plane, taking the value 1 when it is attributed and 0 when it is not. This represents the neighborhood continuity constraint coefficient. When an isolated low-confidence point is close to a certain plane but its surrounding points all belong to another plane, the second term suppresses isolated assignment. When a continuous band of points is close to the same plane, the first and second terms together maintain the plane assignment of that band of points. Please refer to [reference needed]. Figure 6 , Figure 6 The horizontal axis represents the sampling point number participating in local point set redistribution and candidate plane fitting, arranged in order of locally connected point zones; the vertical axis represents the weighted residual contribution from a point to the candidate plane, with lower values ​​indicating a more stable geometric relationship between the point and the current candidate plane. The black main curve represents the point-by-point residual contribution calculated according to the local weighted fitting objective formula. Other interval markers are used to distinguish between points inside the wall, low-confidence edge points, and intersection neighborhood points. The horizontal reference line is used to help determine whether the residual is in a stable, transitional, or anomalous range. Figure 6 As can be seen, the residual contribution of points inside the wall is generally low and continuous, indicating that high-confidence original points have a strong supporting role in the fitting process and can stably determine the normal vector and offset of the candidate plane. In the point sequence near the shading area, under the eaves, or the boundary of adjacent buildings, the residual contribution shows a local increase, corresponding to the low confidence interval. This indicates that these points should not be directly used as high-weight supporting points to guide the plane parameters. Instead, they should be judged in conjunction with the neighborhood continuity constraint to determine whether to reassign them to adjacent planes, retain them as intersection candidate points, or enter the completion process. When individual isolated high residual points do not form a continuous point band, their impact on the overall fitting target is suppressed by the neighborhood continuity term, which can reduce the incorrect assignment caused by noise points. A continuously increasing residual point band suggests that the local plane assignment may have a real transformation, and the boundary and intersection structure need to be preserved.

[0043] For occluded sections without original point support, the completion range is formed by extending the endpoints of the intersection lines of the two stable planes along the direction of the intersection lines. If the two extended intersection lines intersect within the same corner candidate region, the intersection point is taken as the completion boundary corner. If the two intersection lines do not have a direct intersection point due to missing point clouds, the intersection region is limited by the corresponding corner nodes in the boundary topology constraint diagram. The theoretical intersection line of the two planes is obtained without changing the parameters of the high-confidence stable plane. The completion points are generated along the theoretical intersection line and inside the adjacent plane according to the local original point spacing. The generation density changes smoothly with the density of the adjacent original points, does not cross the prohibited crossing boundary, and does not extend to the surface of adjacent buildings where no topological correspondence has been established. After completion, the intersection line endpoints and corner candidate regions are recalculated.

[0044] In a preferred embodiment, when the conflict propagation result is a boundary mismatch conflict object, the high-confidence stable plane and its plane intersection line crossed by the conflict object are frozen. An alternative connection state corresponding to the frozen geometric feature and the interlayer projection is selected from the candidate connection states with the fallback indicator. The affected local subgraph is reconstructed with the alternative connection state, and the consistency of the closed path before and after reconstruction is checked. When the alternative connection state still triggers the prohibited crossing relationship, the association between the corresponding boundary segment and the geometric feature is released, and the corresponding boundary segment is retained as the boundary to be verified. No position correction is performed on the frozen geometric feature. The freezing operation includes keeping the normal vector, offset, support point set and intersection line connection order of the stable plane unchanged, and only allowing the candidate boundary association to switch between fallback states.

[0045] The selection of alternative connection states begins with candidate objects spatially adjacent to the conflict boundary segment. Candidate intersections that belong to the other side of the prohibited object are excluded, as are candidate connections that cause the closed contour direction to be reversed. Then, the spatial closure difference between the beginning and end of the local closed path, the continuity of the shared plane, and the projection offset between the upper and lower layers are compared after the addition of the alternative connection. If multiple alternative connections meet the topological conditions, the connection state that can form a continuous association chain with the predecessor and successor boundary segments is selected first. If there is no alternative connection that meets the conditions, the topological position and mapping source information of the boundary segment are retained, but its adsorption effect on the point cloud surface is canceled so that the high confidence wall will not be translated or tilted due to local deviation of the boundary data. The boundary to be verified is saved in the model output with a different state label than the confirmed boundary.

[0046] Furthermore, the reverse correction of the boundary topology constraint graph includes: writing back the plane intersection lines obtained after the local point set redistribution and the alternative connection states obtained after the boundary association adjustment to the corresponding local subgraph, and regenerating the closed path set and cross-layer projection relationship; limiting the local iteration range for the topological edges that have changed before and after the update, and sequentially performing multi-candidate association, conflict propagation, point set redistribution and boundary association adjustment within the local iteration range; when the locked connection state, stable plane affiliation and plane intersection line connection order are consistent in two consecutive iterations, constructing a three-dimensional surface according to the updated closed contour and stable plane boundary, retaining the original topological edge identifier during the write-back, and only updating its associated object, state identifier and local geometric position, so as to facilitate comparison of the source of change before and after the update.

[0047] The local iteration range extends from the changed topological edge to the adjacent topological node to the nearest locked connection state, and along the shared stable plane to the position where the point-level geometric confidence is continuous and the plane affiliation has not changed. The iteration range does not cross the boundary and is prohibited from crossing the boundary. It does not extend to the region that has no shared plane, shared corner or shared inter-layer correspondence with the conflict. In each iteration, the stable plane, plane intersection line and corner candidate region in the affected region are re-extracted, but the structural features outside the iteration range are retained unchanged. If the local point set redistribution changes the position of the plane intersection line, the corresponding multi-candidate association chain is updated and the relevant closed path is recalculated. If the boundary association adjustment does not change any geometric features, only the local subgraph is updated and the stable plane is not refitted.

[0048] The convergence state of the iteration is determined by the following formula: ; in, This represents the state change during the t-th local iteration. This represents the percentage of locked connection states that have changed relative to the (t-1)th iteration. This indicates the percentage of stable plane attribution points that have changed. The numbers a, b, and c represent the percentage of plane intersections where the connection order changes. , and The corresponding combination coefficients sum to 1, when two consecutive iterations , and When all are 0, the connection state, plane affiliation, and intersection order remain consistent and the local iteration is terminated. When only low-confidence edge points change back and forth between adjacent planes, the corresponding points are restored to undetermined affiliation and their participation in subsequent plane parameter updates is stopped to avoid local loops.

[0049] Please refer to Figure 7 , Figure 7 The horizontal axis represents the iteration number after the local subgraph is written back, and the vertical axis represents the change in local iteration state. Lower values ​​indicate a more stable locked connection state, stable plane affiliation, and plane intersection connection order. The bars represent the combined change obtained from the convergence state criterion in each iteration. The bar segments correspond to the contributions of connection state changes, plane affiliation changes, and intersection order changes, respectively. The legend illustrates the proportion of each contribution in different iteration rounds. Figure 7As can be seen, the state change was relatively high after the first iteration, mainly due to the adjustment of candidate associations in conflict boundary segments and the redistribution of low-confidence point sets. This indicates that the local subgraph still needs to regenerate the set of closed paths and verify the cross-layer projection relationship during the initial write-back. In subsequent iterations, the contributions of connection state changes and intersection order changes gradually decreased, indicating that erroneous candidate connections were deleted or reverted, and the correspondence between the stable plane and the completed boundary began to converge. When the plane assignment changes in subsequent iterations are reduced to only a small number of low-confidence edge points changing repeatedly, Figure 7 The corresponding state change is close to the termination interval, and the relevant points can be restored to undetermined ownership and their participation in subsequent plane parameter updates can be stopped, as described in the original text.

[0050] In this embodiment, the construction of the three-dimensional surface is based on the stable plane boundary and the updated closed contour. For each stable plane, the associated locked plane intersection line, boundary topological edge, and completion boundary are collected. After projecting them onto the local coordinate system of the plane, a directed boundary loop is formed. For planes with inner contours, hole loops are retained. Independent boundary loops are generated for the prohibited crossing boundaries between adjacent buildings. Boundaries that are spatially close but topologically disconnected are not merged. Then, triangular surfaces are generated in each directed boundary loop. The vertices of the triangles preferentially use high confidence edge points, stable intersection line endpoints, and corresponding topological node points in the original point cloud. Completion points are only used for occluded sections. Planes on both sides of a common intersection line share the same set of intersection line vertices to maintain consistent surface connectivity.

[0051] For cases where the outlines of upper and lower floors correspond, the corresponding points between floors and the projected topological edges jointly define the relationship between the intersection lines of the exterior walls of adjacent floors. When the outlines of upper and lower floors remain continuous, the corresponding intersection lines are projected onto a unified vertical reference plane and their planar affiliation and closure direction are compared. The outline of the upper floor is not directly copied to the lower floor. When the building has setbacks, cantilevered sections, or local outline changes, the continuous and changing sections are determined through the correspondence between floors. Continuous sections share vertical planar constraints, while changing sections establish independent local sub-graphs. If the point cloud of a certain floor is missing but the adjacent floor has a high-confidence intersection line, the intersection line of the adjacent floor is only used as a candidate projection relationship and not as a direct completion position. The completion position is still jointly determined by the stable plane of this floor, the boundary topology, and the conflict propagation results.

[0052] In another preferred embodiment, for the adjacent building contact area, local sub-graphs are established according to their respective boundary points and building outlines. Shared boundary nodes are only used to express spatial contact relationships and do not automatically establish surface connectivity relationships. If the point clouds on both sides form a single continuous plane in space, the continuous point set is divided into two planar support regions along the boundary boundary according to the prohibited crossing objects. Points in the neighborhood of the boundary boundary that cannot be determined remain in an undetermined state and do not participate in the fitting of the plane parameters on either side. When the point cloud on one side is missing, it is prohibited to use the planar point set of the building on the other side to cross the boundary to complete it. Only the confirmed stable plane and the closed outline of the current side are allowed to generate completion points. This allows adjacent buildings to maintain their own independent three-dimensional boundaries even when the point cloud sampling is continuous.

[0053] For areas under the eaves or at roof transitions, a stable plane may be formed by the exterior wall plane, the roof plane, and the eaves plane. When extracting plane intersections, candidate intersections between different plane pairs are retained, and internal intersections unrelated to the exterior contour are filtered out using the allowed connection objects of the building outline boundary. When the sparse point cloud at the eaves causes a gap between the wall and roof intersections, the conflict propagation extends along the wall edge and the roof edge respectively. If both sides terminate at the high-confidence stable plane, the gap area is identified as a point cloud missing conflict area, and it is segmented and completed according to the plane intersection relationship between the wall and the eaves, and between the eaves and the roof. The wall is not directly extended to the roof, and the topological level corresponding to the eaves transition is retained.

[0054] In a preferred embodiment, when processing the building outline boundary formed by historical real estate surveying results, the boundary data may have local offsets but the closure order is still correct. At this time, the initial multi-candidate association chain retains the association status of the outline boundary and multiple adjacent plane intersection lines. If the position conflict extends along the complete boundary segment and continuously crosses the high-confidence stable plane, and the adjacent boundary segment forms a closed path with the same set of stable planes, the boundary segment is identified as a boundary mismatch conflict object. The alternative intersection line consistent with the closed path is selected by candidate connection backtracking. When the historical boundary and all candidate intersection lines cannot form a closed local sub-graph that does not cross, the boundary segment is set as the boundary to be verified, and the wall is not cut according to the historical boundary position.

[0055] After local redistribution, completion, and boundary association adjustments, the point cloud model saves three types of data layers: the original measurement point layer retains the input coordinates and source identifiers unchanged; the redistribution relationship layer records the plane affiliation, confidence level, and conflict type of each original point involved in the processing in each iteration; and the construction result layer records the final relationships of stable planes, plane intersections, completion points, topological nodes, and 3D surfaces. When a local area needs to be reprocessed, the corresponding original measurement points are read according to the topological edge identifiers and stable plane identifiers, instead of using the already generated triangular surfaces as new measurement inputs, thus avoiding the cumulative changes in point coordinates caused by multiple reconstructions. When the reconstruction range changes, the completion points are deleted and regenerated based on the new plane intersections and topological boundaries.

[0056] Specifically, before outputting the 3D model, the following checks are performed: boundary non-crossing check, contour closure check, and surface continuity check. The boundary non-crossing check projects each triangular facet onto the local coordinate system of the associated stable plane and determines whether its edge crosses the prohibited topological edge. The contour closure check traverses the 3D surface boundary according to the direction order in the local subgraph and confirms that each non-to-be-verified node has a definite predecessor edge and successor edge. The surface continuity check compares the coordinate index and connection order of the triangular vertices on both sides of the intersection line of the shared plane. If a local inconsistency caused by the completion point is found, the corresponding local iteration range is returned to regenerate the completion boundary. If an open contour caused by the boundary to be verified is found, the open state and the verification mark are retained, and facets without point cloud basis are not used to force closure.

[0057] In one engineering application embodiment, the input point cloud covers the building's exterior walls, roof, and partially occluded corners. The building's outline boundary consists of multiple closed contours, where adjacent contours share one boundary node. During processing, two local subgraphs are constructed and the shared node relationship is written. After performing multi-scale neighborhood parsing on the point cloud, stable planes of the exterior walls and roof, several plane intersections, and candidate corner regions are obtained. The point-level geometric confidence near the occluded corner is lower than that inside the adjacent wall. After establishing multiple candidate association chains between the boundary segment and the intersection, a connection conflict is detected near the shared boundary node. The conflict propagates along the low-confidence occlusion area and terminates on both sides of the high-confidence walls. Therefore, local redistribution and completion are performed according to the intersection of the two walls and the prohibition of crossing the boundary. The two closed contours form independent surface boundaries.

[0058] In another engineering application embodiment, the building wall in the point cloud model is complete, but a boundary segment of the input building outline is biased towards the inside of the wall. The boundary segment forms a positional conflict with the nearest plane intersection line. The conflict propagation path continuously passes through multiple high-confidence wall points and always expands with the same candidate boundary. Thus, the association is identified as a boundary mismatch conflict object. During the processing, the stable plane of the wall and the corresponding plane intersection line are frozen. An alternative connection that maintains the closed order with the adjacent boundary segment is selected from the candidate connection states with the rollback indicator. After regenerating the local closed path, the alternative connection is locked. The original wall point coordinates and plane parameters do not change. The source information and association adjustment records of the offset input boundary segment are retained.

[0059] Preferably, when the point cloud occlusion area simultaneously crosses the floor boundary, the plane of the current floor's exterior wall, the plane of the adjacent floor's exterior wall, the floor slab surface, and the inter-floor projection topology edge are all included in the local subgraph. Conflict propagation extends along the intersection line of the current floor's plane and the inter-floor projection relationship, respectively. If the intersection line of the current floor is missing but the projection relationship between the upper and lower floors is continuous, the intersection line of the adjacent floor is not directly used to replace the intersection line of the current floor. Instead, the theoretical intersection line of the stable plane of the current floor's exterior wall and the floor slab surface is used as the main candidate, and the projection position of the adjacent floor is used as the topology check condition. Only when the theoretical intersection line and the projection relationship maintain the same direction and connection order in the same local closed path is the complete boundary generated. This avoids the building setback or inter-floor contour change being mistakenly treated as the upper and lower floors completely overlapping.

[0060] In this embodiment, all processing records are stored in units of local subgraphs. Each local subgraph is associated with the input data range, stable plane set, plane intersection set, candidate connection state set, conflict propagation path set, and local iteration record. Local subgraphs establish reference relationships through shared topological nodes, shared stable planes, or inter-layer projection edges. When a local subgraph is updated, only the changed topological edges, plane intersections, and candidate connection states are notified to the adjacent local subgraphs with associated references. The adjacent local subgraphs determine whether the closed path needs to be re-verified based on the locked state, thereby maintaining data consistency between local regions and avoiding unrelated point cloud regions from repeatedly participating in plane fitting, conflict propagation, and surface reconstruction.

[0061] The final generated 3D model includes the original point cloud index, stable plane parameters, plane intersection connection relationships, corner node relationships, boundary topological constraint relationships, completion point source identifiers, boundary identifiers to be verified, and triangular surface connection relationships. Each triangular surface can be traced back to the corresponding stable plane and closed contour. Each completion segment can be traced back to the corresponding missing point cloud conflict area, the stable planes on both sides, and the prohibited crossing boundary. Each boundary association adjustment can be traced back to the original candidate connection state, the alternative connection state, and the related closed path. The data relationships together constitute a repeatable 3D construction process for the point cloud model and provide a consistent topological expression for boundary boundaries, building contours, and inter-layer correspondences.

Claims

1. A method for constructing 3D models of real estate surveying based on point cloud models, characterized in that, include: Construct a boundary topology constraint diagram based on boundary points, building outline boundaries, and inter-story correspondences; Multi-scale neighborhood analysis is performed on the point cloud model to be constructed to obtain stable planes, plane intersections, corner candidate regions, and point-level geometric confidence. Based on boundary closure relationships, the continuity of angles between adjacent planes, and inter-layer projection relationships, multiple candidate association chains are established between boundary segments and geometric features; Determine the conflict propagation path based on positional and connection conflicts in the association chain, and distinguish between low-confidence point cloud missing conflicts and high-confidence boundary mismatch conflicts; For different conflict types, perform topologically constrained local point set redistribution or boundary association adjustment, and reverse-correct the boundary topological constraint graph based on the updated plane intersection line to generate a three-dimensional model that satisfies the conditions of no boundary crossing, contour closure and surface continuity.

2. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 1, characterized in that, Constructing the boundary topology constraint graph includes: setting the boundary points, building outline corner points and corresponding points between floors as topology nodes, and setting adjacent boundary segments, closed boundary segments and upper and lower layer projection correspondences as topology edges; Write allowed connection objects, prohibited crossing objects, and direction order identifiers for each topological edge; Local subgraphs are established based on the cyclical order of topological nodes within the same closed contour, and cross-contour constraint relationships are formed based on the topological nodes shared by adjacent local subgraphs. For topological edges with repeated projections or adjacency ambiguities, multiple candidate connection states are retained, and the candidate connection states are updated in conjunction with subsequent geometric feature matching results.

3. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 1, characterized in that, The multi-scale neighborhood analysis includes: establishing multiple neighborhoods that expand progressively with the point to be analyzed as the center, and calculating the point distribution covariance, normal vector consistency relationship, and local curvature change relationship in each neighborhood; The set of points that maintain consistent plane affiliation across neighborhood scales is determined as a stable plane; the set of points whose plane affiliation undergoes orderly transformation and whose intersection direction is continuous is determined as candidate points for plane intersection lines; and the intersection region of multiple candidate points for plane intersection lines is determined as a candidate region for a corner. The point-level geometric reliability is generated jointly based on the continuity of point density, the cross-scale stability of normal vectors, and the degree of preservation of geometric categories.

4. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 1, characterized in that, The establishment of the multi-candidate association chain includes: selecting a set of spatially adjacent planar intersection candidate points and the corner candidate region as initial association objects for each boundary segment; According to the directional order of the boundary segments in the closed contour, check the continuity of the included angles of the planes corresponding to adjacent related objects, the consistency of the extension of the intersection lines, and the consistency of the projections of the upper and lower layers. Retain candidate associations that simultaneously satisfy the closed direction relationship and the included angle continuity or the interlayer projection relationship, and connect adjacent candidate associations with shared association objects to form a chain structure; The location where a mutation of associated objects, a reversal of boundary order, or a prohibited crossing relationship is triggered in the chain structure is marked as the conflict starting point.

5. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 2, characterized in that, The linked update of the candidate connection state includes: constructing a set of closed paths composed of candidate topological edges for each local subgraph; Each closed path is mapped to the associated planar intersection line and the corner candidate region to obtain the path start and end spatial closure difference, direction sequence consistency relationship and cross-layer projection offset relationship; A joint comparison is performed on multiple closed paths sharing the same topological edge. Candidate connection states that appear only in a single path and lack continuous geometric support are deleted. Candidate connection states that maintain topological order and have continuous geometric support in adjacent closed paths are locked. The remaining candidate connection states retain a fallback flag.

6. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 3, characterized in that, The generation of the point-level geometric credibility includes: forming a cross-scale feature sequence by combining the plane assignment results of each point at different neighborhood scales, the principal direction of the normal vector, and the curvature category; Structural stability components are generated based on the continuity of category transfer and the consistency of principal direction deflection between adjacent scales, and complete sampling components are generated based on the residual distribution of points and their respective stable planes and the directional concentration of sampling holes in the neighborhood. The structurally stable component and the sampled complete component are propagated in a continuous field along the plane intersection direction, and the propagation is truncated at the prohibited boundary crossing point to obtain the point-level geometric confidence corresponding to the boundary topological constraint graph.

7. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 4, characterized in that, The determination of the conflict propagation path includes: extending bidirectionally from the conflict origin along the adjacent topological edges in the boundary topological constraint graph and the shared plane intersection line in the multi-candidate association chain, and recording the boundary order state, plane belonging state and point-level geometric confidence change corresponding to the extension position; When a conflict extends continuously along a low-confidence region and terminates at a high-confidence stable plane, the propagation range is marked as a point cloud missing conflict region. When a conflict spans multiple high-confidence stable planes and continues to expand along with the same candidate boundary association, the corresponding candidate boundary association is marked as a boundary mismatch conflict object.

8. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 7, characterized in that, The local point set redistribution is performed for the point cloud missing conflict region, including: determining the stable planes and plane intersection lines associated with the same closed contour on both sides of the conflict region, and setting the normal vector relationship of the stable planes, the extension direction of the plane intersection line and the prohibited crossing boundary as local reconstruction constraints; Points within the conflict region are redistributed to the corresponding plane point sets according to their residual continuity with each stable plane; In sections without original point support, the completion boundary is defined by the topological correspondence of the intersection lines of the two planes, and the completion point set and the updated plane intersection lines are generated based on the intersection results of the refitted planes.

9. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 7, characterized in that, Perform the boundary association adjustment for the boundary mismatch conflict object, including: freezing the high confidence stable plane and its plane intersection line crossed by the conflict object, and selecting alternative connection states from the candidate connection states with fallback indicators that maintain the closure order and inter-layer projection of the frozen geometric features; The affected local subgraph is reconstructed using the alternative connection state, and the consistency of the closed paths before and after reconstruction is checked. When the alternative connection state still triggers the prohibition of crossing relationship, the association between the corresponding boundary segment and the geometric feature is released, the corresponding boundary segment is retained as the boundary to be verified, and no position correction is performed on the frozen geometric feature.

10. The method for constructing three-dimensional real estate surveying based on point cloud models according to claim 5, characterized in that, The reverse correction of the boundary topology constraint graph includes: writing back the plane intersection lines obtained after the redistribution of the local point set and the alternative connection states obtained after the boundary association adjustment to the corresponding local subgraph, and regenerating the closed path set and cross-layer projection relationship; For the topological edges that have changed before and after the update, a local iteration range is defined, and within the local iteration range, multiple candidate association, conflict propagation, point set reallocation and boundary association adjustment are performed sequentially. When the locked connection state, stable plane affiliation, and plane intersection connection order remain consistent in two consecutive iterations, the three-dimensional surface is constructed according to the updated closed profile and stable plane boundary.

Images