A multi-source point cloud data fusion method suitable for building digitization

Abstract

The application discloses a multi-source point cloud data fusion method suitable for building digitization, and relates to the technical field of building digitization modeling, and comprises the following steps: extracting component boundary sampling points in a projection overlapping area, dividing a repeated facade component according to a normal vector flip position, and solving two groups of coordinate transformation parameters by using component boundary sampling points on both sides of the current repeated facade component along a facade main direction; mapping the current repeated facade component to a target source point cloud coordinate system according to the two groups of coordinate transformation parameters, positioning a cross-component misregistration according to a boundary projection interval change direction, and determining component boundary sampling points corresponding to the cross-component misregistration according to a boundary normal offset direction; removing the component boundary sampling points corresponding to the cross-component misregistration, solving coordinate transformation parameters for fusion by using remaining component boundary sampling points, and generating a building digitization point cloud according to the coordinate transformation parameters for fusion. The application is favorable for reducing the influence of repeated facade component misregistration on building point cloud fusion.

Description

Technical Field

[0001] This invention relates to the field of building digital modeling technology, and specifically to a multi-source point cloud data fusion method suitable for building digitalization. Background Technology

[0002] In digital architectural modeling, multi-source point cloud fusion requires converting discrete spatial samples formed under different observation conditions into a unified architectural geometric representation. Due to differences in observation angles, sampling densities, ranging errors, and occlusion ranges among different acquisition devices, the same building facade exhibits varying boundary integrity and local geometric stability in point clouds from different sources. Existing processing workflows often rely on spatial overlap relationships to determine the registration area and extract corresponding features within that area for coordinate transformation, making it difficult to adequately distinguish between spatial overlap and the shared origin of components.

[0003] In scenarios involving the preservation of ancient buildings, surveying of historical buildings, and digital archiving of existing buildings, facade components often exhibit repetitive arrangement characteristics. Adjacent components share high similarities in local morphology, edge contours, and projected structures, and features extracted from overlapping spatial areas are prone to cross-component correspondences. When such erroneous correspondences are entered into the registration solution, they are absorbed by the overall residual adjustment process, leading to local component boundary offsets in the fused point cloud even when the overall error is under control. This affects the geometric reliability of the digital point cloud in component measurement, defect comparison, and repair archiving. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-source point cloud data fusion method suitable for building digitization, so as to solve the problems in the background art mentioned above.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides a multi-source point cloud data fusion method suitable for building digitization, comprising:

[0007] Acquire multi-source building point clouds containing the acquisition pose, select the target source point cloud, project the multi-source building point cloud onto the building facade fitting plane, and determine the projection overlap area by the coincident projection.

[0008] Extract component boundary sampling points in the overlapping projection area, divide the repeated facade components according to the normal vector flip position, and solve two sets of coordinate transformation parameters by using the component boundary sampling points on both sides of the main facade direction of the current repeated facade component.

[0009] The current repeating facade components are mapped to the target source point cloud coordinate system according to two sets of coordinate transformation parameters. First, the cross-component misalignment is located according to the direction of change of boundary projection spacing, and then the component boundary sampling point corresponding to the cross-component misalignment is determined according to the direction of boundary normal offset.

[0010] Remove the component boundary sampling points corresponding to cross-component misalignment, use the remaining component boundary sampling points to solve for the coordinate transformation parameters used for fusion, and generate a digital point cloud of the building based on the coordinate transformation parameters used for fusion.

[0011] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0012] This invention projects multi-source building point clouds onto a building facade fitting plane and determines the overlapping projection area based on coincident projection, ensuring that the data range before fusion corresponds to the actual coverage relationship of the building facade. Furthermore, it divides duplicate facade components based on the normal vector flip position, solves the coordinate transformation parameters using the boundary sampling points of the components on both sides of the current duplicate facade component, and identifies the component boundary sampling points corresponding to cross-component misalignment by combining the direction of boundary projection spacing change and the direction of boundary normal offset. This reduces the situation where duplicate facade components enter incorrect correspondences due to local morphological similarities. After removing the component boundary sampling points corresponding to cross-component misalignment, the coordinate transformation parameters used for fusion are solved again, which reduces the impact of pseudo-identical components on the overall registration solution, making the generated building digital point cloud more suitable for component boundary measurement, defect comparison, and repair documentation. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0014] Figure 1 This is a flowchart illustrating a multi-source point cloud data fusion method suitable for building digitization, provided as an embodiment of the present invention. Detailed Implementation

[0015] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make the description of this application more complete and comprehensive, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The drawings are merely illustrative illustrations of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0016] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of the exemplary embodiments disclosed in this application. However, those skilled in the art will recognize that the technical solutions disclosed in this application can be practiced with one or more specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the disclosure of this application.

[0017] Example 1

[0018] like Figure 1 As shown in the figure, this embodiment discloses a multi-source point cloud data fusion method suitable for building digitization, the method including:

[0019] S101: Acquire multi-source building point clouds containing the acquisition pose, select the target source point cloud, project the multi-source building point cloud onto the building facade fitting plane, and determine the projection overlap area by the coincident projection.

[0020] The building facade fitting plane is a plane formed by plane fitting of sampling points belonging to the same building facade outer surface in the target source point cloud.

[0021] The multi-source building point clouds originate from different measurement results of the same ancient building facade. Each source point cloud contains the 3D coordinates and acquisition pose of the sampling points. The acquisition pose includes a rotation matrix and a translation vector. The rotation matrix represents the orientation relationship between the acquisition device's coordinate system and the measurement coordinate system. The translation vector represents the positional relationship between the center of the acquisition device and the measurement coordinate system. The system reads the 3D coordinates, source identifier, and acquisition pose of the sampling points from each source point cloud, and maintains the correspondence between the sampling points and the source identifier.

[0022] The target source point cloud is selected from multi-source building point clouds. The target source point cloud is used to provide a unified coordinate reference. During implementation, the ground 3D laser scan point cloud constrained by measurement control points is used as the target source point cloud. The UAV oblique photogrammetry reconstructed point cloud and the close-range photogrammetry point cloud are transformed to the target source point cloud coordinate system according to the acquisition pose and then participate in subsequent projection, component boundary extraction and point cloud fusion.

[0023] The building facade fitting plane is formed by fitting sampling points from the target source point cloud that belong to the same building facade's outer surface. Sampling points belonging to the same building facade's outer surface refer to sampling points located on the geometric envelope of the facade's outer side facing the same direction. Taking the dougong (bracket set) facade as an example, sampling points are distributed along the outer contour of the dougong, the outer edges of the beams and rafters, the outer sides of the walls between columns, and the outer contours under the eaves. These sampling points are used to characterize the overall spatial orientation of the ancient building facade.

[0024] The fitted plan representation of the building facade is as follows:

[0025]

[0026] in, This represents the fitted plane of the building facade. This represents the unit normal vector of the fitted plane of the building facade. This represents the spatial coordinates of a point in the target source point cloud coordinate system. Represents the plane constant term. This indicates the matrix transpose.

[0027] Let the sampling points involved in the plane fitting be... The number of sampling points is The mean coordinates of the sampling points are:

[0028]

[0029] The coordinate covariance matrix is:

[0030]

[0031] right Perform eigenvalue decomposition, sort the eigenvalues ​​in ascending order of their numerical values, and select the eigenvector corresponding to the first-ranked eigenvalue as the eigenvector. The plane constant term is:

[0032] Therefore, the building facade fitting plane is determined by the actual facade sampling points in the target source point cloud. This plane reflects the overall orientation of the ancient building facade and maintains the measurement benchmark in the target source point cloud coordinate system.

[0033] When projecting multi-source building point clouds onto the building facade fitting plane, each source point cloud is first transformed to the target source point cloud coordinate system based on its acquisition pose. Let the sampling points in any source point cloud be... The rotation matrix corresponding to the acquired pose is The translation vector corresponding to the acquired pose is The coordinates of the sampled points after transformation are:

[0034] in, This represents the coordinates of the sampling point in the target source point cloud coordinate system. For the sampling points of the target source point cloud itself, the rotation matrix is ​​the identity matrix, and the translation vector is the zero vector. Then... Projecting along the normal direction of the building facade fitting plane onto the building facade fitting plane yields the projection point. :

[0035] Each source point cloud, after the above processing, forms a corresponding set of projection points. The set of projection points retains the coverage area of ​​the source point cloud on the facade of the ancient building. This processing transforms spatial overlap judgment into projection overlap judgment within the same building facade, which can reduce the mis-inclusion of non-homogeneous components caused by the intersection of 3D bounding boxes.

[0036] When determining the overlapping projection area, a two-dimensional coordinate system is established within the building facade fitting plane. The first coordinate axis of the two-dimensional coordinate system lies within the building facade fitting plane and is set along the lateral extension direction of the ancient building facade. The second coordinate axis of the two-dimensional coordinate system lies within the building facade fitting plane and is perpendicular to the first coordinate axis. Each projection point... Convert to 2D projected coordinates .

[0037] To ensure the reproducibility of the projected coverage area, a unified regular grid is established for the building facade fitting plane. First, the median nearest neighbor distance of each source point cloud within the building facade fitting plane is calculated. Then, the median of the median nearest neighbor distances of all source point clouds is taken as the unified grid side length. The unified regular grid uses the origin of the two-dimensional coordinate system of the building facade fitting plane as its origin. If a projection point of a corresponding source point cloud exists within a certain grid, that grid is recorded as an occupied grid of the corresponding source point cloud. Adjacent occupied grids are connected to form the projected coverage area of ​​the corresponding source point cloud. A common occupied area is calculated for the projected coverage areas of each source point cloud, and the building facade range corresponding to the common occupied area is taken as the projected overlapping area. The projected overlapping area determined in this way has a unified grid reference and can accommodate facade forms with grooves and gaps in components such as brackets and window lattices.

[0038] S102: Extract component boundary sampling points in the overlapping projection area, divide the repeated facade components according to the normal vector flip position, and solve two sets of coordinate transformation parameters by using the component boundary sampling points on both sides of the current repeated facade component along the main direction of the facade.

[0039] Wherein, the main direction of the facade is the direction with the highest variance in the projection coordinates of the sampling points of the component boundary in the fitted plane of the building facade, and the positive direction of the main direction of the facade is the direction in which the projection coordinates increase.

[0040] The component boundary sampling points are sampling points located at the edges of the ancient building facade components. The normal vector flip position is the position on the point cloud intercept where the sign of the inner product of the normal vectors of adjacent sampling points changes. Repeating facade components are ancient building facade components that are repeated along the main direction of the facade. For ancient building protection scenarios, repeating facade components include continuous brackets, continuous window lattices, continuous balustrades, or continuous beams.

[0041] The main orientation of the facade is determined by the projected coordinates of the component boundary sampling points, and the initial extraction of these sampling points requires a truncation direction. To ensure the feasibility of the implementation process, the first direction, sorted by the variance of the coordinates of all projected points within the overlapping projection area, is used as a temporary truncation direction to form an initial point cloud cutoff and extract the initial component boundary sampling points. Subsequently, the main orientation of the facade is determined by sorting the first direction based on the variance of the projected coordinates of the initial component boundary sampling points, and then the formal extraction of the component boundary sampling points is completed according to the main orientation of the facade. This transitional calculation is only used to determine the truncation direction and does not change the final definition of the main orientation of the facade.

[0042] Projecting the component boundary sampling points onto the building facade fitting plane yields the projected coordinates of the component boundary sampling points. Let the projected coordinates of the component boundary sampling points be... The number of component boundary sampling points is The mean of the projected coordinates is:

[0043]

[0044] The projected coordinate covariance matrix is:

[0045]

[0046] right Eigenvalue decomposition is performed, and the eigenvalues ​​are sorted in descending order of numerical value. The eigenvector corresponding to the first-ranked eigenvalue is selected as the principal orientation of the facade. The positive direction of the principal orientation is the direction in which the projected coordinates increase. For horizontally continuous bracket sets, the principal orientation of the facade corresponds to the direction in which the bracket sets are arranged. For vertically continuous window frames, the principal orientation of the facade corresponds to the direction in which the window frames are arranged.

[0047] Adjacent repeating facade components are similar in shape, and relying solely on local shape similarity can easily lead to pseudo-identical component boundary sampling points. By extracting component boundaries through three-dimensional normal transformation and solving the coordinate transformation parameters from samples on both sides of the current repeating facade component, subsequent misalignment identification can have a dual-sided geometric basis.

[0048] Specifically, the step of extracting component boundary sampling points in the projection overlap area and dividing repeating facade components according to the normal vector flip position includes:

[0049] The overlapping area of ​​the projection is intercepted along a direction perpendicular to the main direction of the facade, and a point cloud intercept line is formed within the intercepted area;

[0050] When truncating the overlapping area of ​​projections along a direction perpendicular to the main direction of the facade, the fitting plane of the building facade is used as the reference plane for truncating. Each truncation forms a strip-shaped truncating range. Sampling points within the same truncating range are arranged in order of their projected positions on the fitting plane of the building facade, forming a point cloud truncating line. The point cloud truncating line is used to express the geometric transitions at the edges of local components. Taking the dougong (bracket set) facade as an example, the point cloud truncating line passes through the edge of the dougong, the outline of the arch eye, and the intersection of the beam and rafter. These locations have changes in the surface direction of the components, so the normal vector of the sampling points on the point cloud truncating line will show continuous changes.

[0051] After the point cloud cutoff line is formed, the three-dimensional coordinates of each sampling point on the cutoff line in the target source point cloud coordinate system are retained, as well as the projected coordinates of each sampling point in the building facade fitting plane. The three-dimensional coordinates are used for normal vector calculation, and the projected coordinates are used to determine the arrangement order of the sampling points on the point cloud cutoff line. This method allows the point cloud cutoff line to simultaneously incorporate three-dimensional geometric information and the facade projection arrangement relationship.

[0052] Calculate the sign of the dot product of the normal vectors of adjacent sampling points on the point cloud cutoff line, and take the position where the sign of the dot product of the normal vectors changes as the position where the normal vectors are flipped.

[0053] The calculation of the sign of the dot product of the normal vectors of adjacent sampling points on the point cloud cutoff line, and the determination of the position where the sign of the dot product changes as the position where the normal vector is flipped, includes:

[0054] Select neighborhood sampling points centered on each sampling point on the point cloud cutoff line, and calculate the coordinate covariance matrix of the neighborhood sampling points;

[0055] For any sampling point on a point cloud cutoff line, neighboring sampling points are selected according to the order of the point cloud cutoff lines. Neighboring sampling points include the adjacent sampling points located before and after the given sampling point. Let the neighboring sampling points be... The number of neighboring sampling points is The mean coordinates of the neighboring sampling points are:

[0056]

[0057] The coordinate covariance matrix of the neighborhood sampling points is:

[0058]

[0059] Coordinate covariance matrix This characterizes the local spatial distribution of neighborhood sampling points. Neighborhood sampling points in the middle of a wall typically exhibit planar distribution characteristics. The distribution direction of neighborhood sampling points at the edges of components may change due to the folding structures of the outer contours of brackets, beam edges, or window lattice edges. Extracting the normal vector from the coordinate covariance matrix of the neighborhood sampling points allows component boundary identification to be completed based on local three-dimensional geometric changes.

[0060] Eigenvalue decomposition is performed on the coordinate covariance matrix. The eigenvector corresponding to the first eigenvalue in ascending order is taken as the normal vector of the sampling point, and the position where the sign of the inner product of the normal vector changes along the intercept of the point cloud is taken as the position of normal vector flip.

[0061] For coordinate covariance matrix Eigenvalue decomposition is performed to obtain eigenvalues ​​and their corresponding eigenvectors. The eigenvalues ​​are sorted in ascending order, and the eigenvector corresponding to the first eigenvalue is used as the normal vector of the sampling point. To unify the normal vector direction, the normal vector of the sampling point is adjusted so that its dot product with the unit normal vector of the fitted plane of the building facade is positive. If the dot product is zero, the normal vector direction of the sampling point is determined along the unified normal vector direction of adjacent points on the point cloud intercept.

[0062] Set the point cloud intercept line first The normal vector of each sampling point is , No. The normal vector of each sampling point is The dot product of the normal vectors of adjacent sampling points is represented as:

[0063]

[0064] in, Indicates the sign of the dot product of normal vectors. Represents a symbolic function. When... At that time, the corresponding sampling position is used as the position to flip the normal vector. The adjacent positions are compared according to the arrangement order along the point cloud cutoff line. .when When a sign change occurs, the corresponding sampling position is taken as the normal vector flip position. The normal vector flip position indicates the location where the surface direction of the component changes. For the dougong (bracket set) facade, the normal vector flip position can correspond to the turning point of the outer contour of the dougong, the edge of the arch eye, or the intersection of the beam and rafter.

[0065] The sampling point segment located between the flipped positions of adjacent normal vectors is taken as the component boundary sampling point, and the repeated facade components are divided according to the repeated arrangement of the sampling point segment in the main direction of the facade.

[0066] Following the arrangement order of the point cloud cutoff lines, the sampling point segments between the flipped positions of adjacent normal vectors are extracted. These sampling point segments, located within the local extension range of the same component boundary, are used as component boundary sampling points. After performing the same processing on multiple point cloud cutoff lines, the component boundary sampling points belonging to the same component boundary form a continuous distribution within the building facade fitting plane.

[0067] Subsequently, the projected centroid coordinates of each sampling point segment on the main direction of the facade are calculated, and the points are sorted according to these coordinates. Let the sorted projected centroid coordinates be... Calculate the difference in barycenter coordinates between adjacent projections. The median of the coordinate differences is taken as the reference for the spacing between adjacent repeating components. . No. The sequence number of each sampling point segment is:

[0068]

[0069] in, Indicates the first The sequence number of each sampling point segment. Indicates the first The centroid coordinates of the projection of each sampling point segment on the main direction of the facade. This represents the coordinates of the centroid of the first-ranked projection. This represents the rounding function. Sampling point segments with adjacent sequence numbers and the same normal vector flip type are classified as repeating facade components arranged repeatedly along the main direction of the facade. The normal vector flip type is determined by the direction of the sign change of the inner product at the flipped positions of the normal vectors at both ends of the sampling point segment.

[0070] For continuous bracket sets, the sampling point segments corresponding to the edges of the bracket sets repeat along the direction under the eaves. By sorting the projection centroid coordinates and comparing the normal vector flip type, adjacent bracket sets can be divided into different repeating facade components, while maintaining the arrangement order of each repeating facade component. This method provides an object basis for subsequent judgment of cross-component misalignment between adjacent repeating facade components.

[0071] The two sets of coordinate transformation parameters are rigid body coordinate transformation parameters from the source point cloud coordinate system to the target source point cloud coordinate system.

[0072] Rigid body coordinate transformation parameters include rotation matrices and translation vectors. Let the sampling points in the source point cloud coordinate system be... The rotation matrix is The translation vector is The coordinates of the sampling point after rigid body coordinate transformation are: in, This represents the transformed coordinates in the target source point cloud coordinate system. Rigid body coordinate transformation parameters are used to express the spatial relationship between the source point cloud coordinate system and the target source point cloud coordinate system. Ancient building facade components correspond to the same solid structure in different source point clouds; using rigid body coordinate transformation parameters can preserve the component shape and the relative position of component boundaries.

[0073] Specifically, the step of solving for two sets of coordinate transformation parameters by sampling points of the component boundaries on both sides of the main direction of the facade of the current repeating facade component includes:

[0074] Using the projected coordinates of the center of gravity of the current repeating facade component in the main direction of the facade as the boundary, determine the component boundary sampling points on the positive side of the main direction of the facade;

[0075] The current repeating facade components are selected from repeating facade components that have both forward-adjacent and reverse-adjacent repeating facade components along the main facade direction. Repeating facade components located at the end of the repeating sequence participate in the selection of samples on both sides as adjacent reference components, but are not used as current repeating facade components for solving the two sets of coordinate transformation parameters.

[0076] After selecting the current repeating facade component, calculate the mean coordinates of the component boundary sampling points within the current repeating facade component, and project the mean coordinates onto the building facade fitting plane to obtain the centroid of the current repeating facade component. Then, project the centroid of the current repeating facade component onto the main direction of the facade to obtain the projected coordinates of the centroid of the current repeating facade component in the main direction of the facade. Using the projected coordinates as the boundary, the component boundary sampling points located on the positive side of the main direction of the facade are determined as the component boundary sampling points on the positive side.

[0077] To maintain a local correspondence between the two sets of coordinate transformation parameters and the current repeating facade components, the component boundary sampling points on the positive side are taken from the repeating facade components adjacent to the current repeating facade components along the main direction of the facade. For transverse continuous brackets, the positive side corresponds to the adjacent bracket area of ​​the current bracket along the bracket arrangement direction.

[0078] Using the same projected coordinates as the boundary, determine the component boundary sampling points on the opposite side of the main direction of the facade, and use the component boundary sampling points on the positive side to solve a set of coordinate transformation parameters;

[0079] Continuing with the projection coordinates of the centroid of the current repeating facade member in the main direction of the facade as the boundary, the component boundary sampling points located on the opposite side of the main direction of the facade are determined as the component boundary sampling points on the opposite side. The component boundary sampling points on the opposite side are taken from the repeating facade members adjacent to the current repeating facade member along the main direction of the facade.

[0080] When solving for a set of coordinate transformation parameters using component boundary sampling points on the positive side, the component boundary sampling points on the positive side of the source point cloud are taken as the points to be transformed, and the corresponding component boundary neighborhood in the target source point cloud is taken as the reference object. The component boundary neighborhood at the corresponding position is determined as follows: A repeating facade component with the same sequence number as the positive side component is selected in the target source point cloud. The points to be transformed are projected onto the building facade fitting plane, and the sampling point with the highest projection distance among the component boundary sampling points of this repeating facade component is selected as the reference point. A local plane is fitted using sampling points in the neighborhood of the reference point, and the unit normal vector of the local plane is used as... .

[0081] Establish point-surface constraint equations based on the point-surface distance from the point to be transformed to the reference object:

[0082]

[0083] in, This indicates the component boundary sampling points on the positive side of the source point cloud. This represents the reference point in the neighborhood of the corresponding component boundary in the target source point cloud. This represents the normal vector of the neighborhood of the corresponding component boundary in the point cloud from which the target originates. Let represent the rotation matrix to be determined. Let represent the translation vector to be determined. The objective function is:

[0084]

[0085] in, , The rotation matrix and translation vector are solved by iterative linearization. When the difference between the objective function of two adjacent iterations is less than the numerical resolution corresponding to the accuracy of the sampling point coordinate record, the current rotation matrix and translation vector are output, forming a set of coordinate transformation parameters.

[0086] Another set of coordinate transformation parameters is solved by sampling points of the component boundary on the opposite side, and the parameters solved in the two cases are used as two sets of coordinate transformation parameters.

[0087] Point-to-surface constraint equations are established using component boundary sampling points on the reverse side. The construction method for these equations is consistent with that of the point-to-surface constraint equations on the forward side. Component boundary sampling points on the reverse side of the source point cloud are used as points to be transformed. The corresponding component boundary neighborhood in the target source point cloud is used as a reference object. The rotation matrix and translation vector are solved using the same objective function to form another set of coordinate transformation parameters. The set of coordinate transformation parameters corresponding to the forward side and the other set corresponding to the reverse side together serve as two sets of coordinate transformation parameters.

[0088] The two sets of coordinate transformation parameters are obtained from the boundary sampling points of adjacent components on both sides of the main direction of the current repeating facade component. The correspondence of the current repeating facade component itself is verified by subsequent mapping differences. This process can reduce the absorption of cross-component misalignment after the repeating component itself participates in the parameter solution, so that the subsequent misalignment judgment has a stable geometric basis.

[0089] S103: Map the current repeating facade components to the target source point cloud coordinate system according to two sets of coordinate transformation parameters. First, locate the cross-component misalignment according to the direction of change of boundary projection spacing, and then determine the component boundary sampling point corresponding to the cross-component misalignment according to the direction of boundary normal offset.

[0090] The component boundary sampling points of the current repeating facade component are obtained from the source point cloud. Two sets of coordinate transformation parameters are derived from the component boundary sampling points on the positive and negative sides of the main facade direction. By mapping the current repeating facade component using these two sets of coordinate transformation parameters, two mapped positions of the current repeating facade component in the target source point cloud coordinate system can be obtained. The difference between the two mapped positions is used to reflect whether the current repeating facade component is incorrectly mapped to adjacent repeating facade components.

[0091] The direction of change in boundary projection spacing refers to the directional spacing change of the boundary projection lines of the current repeating facade components relative to the corresponding component boundary projection lines in the target source point cloud along the main direction of the facade after mapping using two sets of coordinate transformation parameters. The direction of boundary normal offset refers to the offset direction of the mapped component boundary relative to the corresponding component boundary in the target source point cloud along the normal vector direction. The direction of change in boundary projection spacing is used to locate the cross-component positional offset along the main direction of the facade. The direction of boundary normal offset is used to confirm the boundary geometric offset corresponding to this positional offset.

[0092] The method for determining the corresponding component boundary projection lines in the target source point cloud is as follows: First, based on the component boundary sampling points in the target source point cloud, the repeating facade components in the target source point cloud are divided using the same normal vector flipping position and facade main direction sorting method as described above. Then, based on the current repeating facade component's sequence number in the source point cloud's repeating facade component sequence, repeating facade components with the same sequence number are selected from the target source point cloud's repeating facade component sequence. The component boundary sampling points of the selected repeating facade components are projected onto the building facade fitting plane to form the corresponding component boundary projection lines in the target source point cloud.

[0093] In the scenario of protecting ancient building bracket sets, the current repeated facade component can be a single bracket set. If the current bracket set in the source point cloud corresponds correctly to the same bracket set in the target source point cloud, then after mapping according to the two sets of coordinate transformation parameters, the directional spacing change direction of the boundary projection line of the current bracket set relative to the boundary projection line of the corresponding bracket set in the target source point cloud remains within the boundary range of the current bracket set. If the current bracket set in the source point cloud is incorrectly mapped to an adjacent bracket set, then the boundary projection line after the two mappings will show a state of offset towards the adjacent bracket set, and the boundary normal offset direction will correspond to the cross-component offset direction.

[0094] Specifically, determining the component boundary sampling points corresponding to cross-component misalignment includes:

[0095] The current repeating facade components are mapped to the target source point cloud coordinate system according to two sets of coordinate transformation parameters, forming two boundary projection lines in the building facade fitting plane;

[0096] The component boundary sampling points in the current repeating facade components are used as the sampling points to be mapped. Let the sampling points to be mapped be... The two sets of coordinate transformation parameters are denoted as follows: and The coordinates of the sampling points to be mapped after mapping according to the first set of coordinate transformation parameters are:

[0097]

[0098] The coordinates of the sampling points to be mapped after mapping according to the second set of coordinate transformation parameters are:

[0099]

[0100] in, and All are located in the target source point cloud coordinate system. Then, and The points are projected onto the building facade fitting plane. The projection calculation method is the same as that used for the aforementioned projection of sampling points onto the building facade fitting plane. The set of projection points corresponding to the first set of coordinate transformation parameters forms one boundary projection line, and the set of projection points corresponding to the second set of coordinate transformation parameters forms another boundary projection line.

[0101] Both boundary projection lines originate from the current repeating facade component. The difference lies in that the two boundary projection lines are constrained by coordinate transformation parameters obtained from the sampling points of the boundary components on both sides of the current repeating facade component. Therefore, the mapping differences of the current repeating facade component under the geometric constraints on both sides are preserved. This method can reduce the masking of cross-component misalignment after the current repeating facade component itself participates in the same parameter solution.

[0102] Calculate the directed spacing between the two boundary projection lines and the corresponding component boundary projection lines in the target source point cloud. Determine the direction of change of the boundary projection spacing based on the sign change of the directed spacing along the main direction of the facade, and locate the cross-component misalignment.

[0103] The boundary projection line of the corresponding component in the target source point cloud is formed by projecting the component boundary sampling points at the corresponding positions of the component in the target source point cloud and the current repeating facade component. For the first boundary projection line, the directed spacing between the first boundary projection line and the corresponding component boundary projection line in the target source point cloud in the main direction of the facade is calculated. For the second boundary projection line, the directed spacing is calculated in the same way.

[0104] Let the unit vector of the principal direction of the surface be... The centroid of the first boundary projection line is The centroid of the second boundary projection line is The centroid of the corresponding component boundary projection line in the target source point cloud is The directed spacings of the two boundary projection lines relative to the corresponding component boundary projection lines in the target source point cloud are as follows:

[0105]

[0106]

[0107] in, This represents the directed spacing of the first boundary projection line. This indicates the directed spacing of the second boundary projection line. The change in the sign of the directed spacing along the main direction of the elevation is used to indicate the change in the offset direction of the boundary projection line relative to the corresponding component boundary projection line in the target source point cloud after two mappings.

[0108] The positioning of cross-component misalignment includes:

[0109] Calculate the difference in projection coordinates between the centroids of the two boundary projection lines and the centroids of the corresponding component boundary projection lines in the target source point cloud in the main direction of the elevation.

[0110] The centroids of the two boundary projection lines and the centroids of the corresponding component boundary projection lines in the target source point cloud are projected onto the main direction of the elevation. The projection coordinates of the centroid of the first boundary projection line on the main direction of the elevation are as follows: The projection coordinates of the centroid of the second boundary projection line on the principal direction of the elevation are: The projection coordinates of the centroid of the corresponding component boundary projection line in the target source point cloud along the principal direction of the elevation are: The difference in projected coordinates is expressed as:

[0111]

[0112]

[0113] in, and These represent the differences in projection coordinates between the centroids of the two boundary projection lines and the centroids of the corresponding component boundary projection lines in the target source point cloud, along the principal direction of the elevation. The projection coordinate differences have the same directional meaning as the directional spacing, used to express the changes in the projected position of the component boundary after two mappings.

[0114] Arrange the projected coordinate differences according to the solution order of the two sets of coordinate transformation parameters, and locate the cross-component misalignment when the sign of the projected coordinate difference changes from the current repeated facade component to the adjacent repeated facade component.

[0115] Following the solution order of one set of coordinate transformation parameters corresponding to the sampling points of the component boundary on the positive side and another set of coordinate transformation parameters corresponding to the sampling points of the component boundary on the negative side, the projected coordinate difference is calculated. and The arrangement is performed. The projection range of the current repeating facade component is determined by the outer envelope of the component boundary sampling points of the current repeating facade component within the building facade fitting plane. The projection range of adjacent repeating facade components is determined by the outer envelope of adjacent repeating facade components along the main facade direction within the building facade fitting plane. The outer envelope is determined by the grid boundary occupied by the corresponding component boundary sampling points within the building facade fitting plane, and the grid side length is consistent with the grid side length of the projection overlap area.

[0116] When the difference in the projected coordinates of the centroids of two boundary projection lines, after being arranged, points towards the centroid of the adjacent repeating facade member of the current repeating facade member, and the centroid of the corresponding boundary projection line falls within the projection range of the adjacent repeating facade member, cross-member misalignment is identified. For example, continuous bracket sets are arranged along the main direction of the facade. The boundary projection range of the current bracket set is located between the projection ranges of two adjacent bracket sets. If the difference in the projected coordinates of the centroids of the two boundary projection lines indicates that the boundary projection line has shifted from the projection range of the current bracket set to the projection range of the adjacent bracket set, then the correspondence of the current bracket set exhibits cross-member offset under the constraints of the coordinate transformation parameters on both sides.

[0117] Calculate the normal projection difference between the mapped component boundary and the corresponding component boundary in the target source point cloud. Determine the boundary normal offset direction based on the sign continuation direction of the normal projection difference, and determine the component boundary sampling point corresponding to the cross-component misalignment.

[0118] After the misalignment across components is located, the normal projection difference between the mapped component boundary and the corresponding component boundary in the target source point cloud is calculated. Let the sampling points on the mapped component boundary be... The reference point on the boundary of the corresponding component in the target source point cloud is... The normal vector of the corresponding component boundary in the target source point cloud is The normal projection difference is expressed as:

[0119]

[0120] in, Indicates the first The normal projection difference corresponding to each boundary sampling point. Statistical analysis of the arrangement order of sampling points along the component boundary. The sign of the normal projection difference within the same sampling point segment is counted, and the non-zero sign with the dominant number is taken as the sign continuation direction of that sampling point segment. If the number of positive signs and negative signs within the same sampling point segment is the same, the sign continuation direction of that sampling point segment is determined according to the sign direction that is continuous with the adjacent sampling point segments. If the sign direction of adjacent sampling point segments still cannot be determined, the sum of the absolute values ​​of the normal projection differences corresponding to the positive signs and the sum of the absolute values ​​of the normal projection differences corresponding to the negative signs are calculated respectively, and the sign corresponding to the sum of the absolute values ​​is taken as the sign continuation direction of that sampling point segment. The boundary normal offset direction is determined by this sign continuation direction.

[0121] If the boundary normal offset direction and the boundary projection spacing change direction correspond to the same adjacent repeated facade member on the same spatial side, then the corresponding sampling point is determined as the member boundary sampling point corresponding to the cross-member misalignment. The boundary normal offset direction is used to confirm the boundary geometric offset corresponding to the change in projection position. This process combines the position offset in the main direction of the facade with the member boundary normal offset, which can reduce the number of pseudo-identical boundaries formed by similar repeated members entering the subsequent fusion.

[0122] S104: Remove the component boundary sampling points corresponding to cross-component misalignment, use the remaining component boundary sampling points to solve for the coordinate transformation parameters used for fusion, and generate a digital point cloud of the building based on the coordinate transformation parameters used for fusion.

[0123] This step follows the previously determined component boundary sampling points corresponding to cross-component misalignment. After removing these component boundary sampling points, the remaining component boundary sampling points are used to solve for the final coordinate transformation parameters. The coordinate transformation parameters used for fusion are still rigid body coordinate transformation parameters from the source point cloud coordinate system to the target source point cloud coordinate system. Unlike the two sets of coordinate transformation parameters mentioned above, the coordinate transformation parameters used for fusion are obtained from the component boundary sampling points after excluding cross-component misalignment. These parameters are used to uniformly transform the multi-source building point clouds to the target source point cloud coordinate system and generate a digital building point cloud.

[0124] Specifically, the generation of the building digital point cloud includes:

[0125] Remove the component boundary sampling points corresponding to cross-component misalignment from the component boundary sampling points, substitute the remaining component boundary sampling points into the point-surface constraint equation, and solve for the coordinate transformation parameters used for fusion.

[0126] Remove the previously identified cross-component misalignment component boundary sampling points from the component boundary sampling point set. The deleted sampling points form the remaining component boundary sampling point set. Using the remaining component boundary sampling points as the points to be transformed, and the corresponding component boundary neighborhood in the target source point cloud as the reference object, establish the point-surface constraint equations:

[0127]

[0128] in, Indicates the boundary sampling points of the remaining components. This represents the reference point in the neighborhood of the corresponding component boundary in the target source point cloud. This represents the normal vector of the neighborhood of the corresponding component boundary in the point cloud from which the target originates. This represents the rotation matrix used for fusion. This represents the translation vector used for fusion. It is solved through multiple sets of point-surface constraint equations. and The coordinate transformation parameters used for fusion are obtained. After the component boundary sampling points corresponding to cross-component misalignment are excluded from the final solution sample, the degree to which the final coordinate transformation parameters are affected by pseudo-identical duplicate components is reduced, and the risk of component-level boundary offset decreases accordingly.

[0129] The multi-source building point cloud is transformed to the target source point cloud coordinate system according to the coordinate transformation parameters used for fusion. The coordinates of the sampled points of the multi-source building point cloud in the target source point cloud coordinate system are then fused to generate a digital building point cloud.

[0130] Each non-target source point cloud is transformed to the target source point cloud coordinate system according to the coordinate transformation parameters used for fusion. Let the sampling points in the non-target source point cloud be... The rotation matrix used for fusion is The translation vector used for fusion is The transformed coordinates of the sampling points are:

[0131]

[0132] in, This represents the transformed sampled point in the target source point cloud coordinate system. The sampled points in the target source point cloud are maintained within the target source point cloud coordinate system.

[0133] After coordinate transformation, the coordinates of sampling points in the multi-source building point clouds within the target source point cloud coordinate system are fused. The sampling point coordinate fusion uses a unified rule grid to determine the same spatial neighborhood. Specifically, sampling points in the target source point cloud coordinate system are projected onto the building facade fitting plane, using the same rule grid used to determine the overlapping projection area. Sampling points located within the same grid and whose distance to the building facade fitting plane belongs to the same distance layer are considered as sampling points within the same spatial neighborhood. The layer width of the distance layer is taken as the side length of the unified grid.

[0134] The sampling point coordinate fusion is accomplished using measurement accuracy weighting. The measurement accuracy weighting is determined based on the adjustment error of the control points in the corresponding source point cloud. If the source point cloud has no adjustment error of control points, the ranging error given by the calibration of the acquisition equipment is used. Let the first... The mean square error of the source point cloud to which each sampling point belongs is The corresponding measurement accuracy weight is:

[0135] Let the first one be in the same spatial neighborhood. The transformed sampling points are The number of sampling points participating in the fusion is The fused coordinates are:

[0136]

[0137] in, This represents the coordinates of the fused sampling points. The above fusion process is performed neighborhood-by-neighbor on the multi-source building point clouds in the target source point cloud coordinate system to form a digital building point cloud. This digital building point cloud is located in the target source point cloud coordinate system and preserves the spatial relationships of the ancient building facade components. For ancient building preservation scenarios, the digital building point cloud can be used for component damage measurement, component deformation comparison, and restoration archive modeling.

[0138] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for fusing multi-source point cloud data suitable for building digitization, characterized in that, include: Acquire multi-source building point clouds containing the acquisition pose, select the target source point cloud, project the multi-source building point cloud onto the building facade fitting plane, and determine the projection overlap area by the coincident projection. Extract component boundary sampling points in the overlapping projection area, divide the repeated facade components according to the normal vector flip position, and solve two sets of coordinate transformation parameters by using the component boundary sampling points on both sides of the main facade direction of the current repeated facade component. The current repeating facade components are mapped to the target source point cloud coordinate system according to two sets of coordinate transformation parameters. First, the cross-component misalignment is located according to the direction of change of boundary projection spacing, and then the component boundary sampling point corresponding to the cross-component misalignment is determined according to the direction of boundary normal offset. Remove the component boundary sampling points corresponding to cross-component misalignment, use the remaining component boundary sampling points to solve for the coordinate transformation parameters used for fusion, and generate a digital point cloud of the building based on the coordinate transformation parameters used for fusion.

2. The method according to claim 1, characterized in that, The building facade fitting plane is a plane formed by plane fitting of sampling points belonging to the same building facade outer surface in the target source point cloud.

3. The method according to claim 1, characterized in that, The step of extracting component boundary sampling points in the overlapping projection area and dividing repeating facade components according to the normal vector flip position includes: The overlapping area of ​​the projection is intercepted along a direction perpendicular to the main direction of the facade, and a point cloud intercept line is formed within the intercepted area; Calculate the sign of the dot product of the normal vectors of adjacent sampling points on the point cloud cutoff line, and take the position where the sign of the dot product of the normal vectors changes as the position where the normal vectors are flipped. The sampling point segment located between the flipped positions of adjacent normal vectors is taken as the component boundary sampling point, and the repeated facade components are divided according to the repeated arrangement of the sampling point segment in the main direction of the facade.

4. The method according to claim 3, characterized in that, The calculation of the sign of the dot product of the normal vectors of adjacent sampling points on the point cloud cutoff line, and the position where the sign of the dot product of the normal vectors changes as the position where the normal vectors are flipped, includes: Select neighborhood sampling points centered on each sampling point on the point cloud cutoff line, and calculate the coordinate covariance matrix of the neighborhood sampling points; Eigenvalue decomposition is performed on the coordinate covariance matrix. The eigenvector corresponding to the first eigenvalue in ascending order is taken as the normal vector of the sampling point, and the position where the sign of the inner product of the normal vector changes along the intercept of the point cloud is taken as the position of normal vector flip.

5. The method according to claim 1, characterized in that, The main direction of the facade is the direction with the highest variance in the projected coordinates of the sampling points of the component boundary within the fitted plane of the building facade, and the positive direction of the main direction of the facade is the direction in which the projected coordinates increase.

6. The method according to claim 1, characterized in that, Both sets of coordinate transformation parameters are rigid body coordinate transformation parameters from the source point cloud coordinate system to the target source point cloud coordinate system.

7. The method according to claim 1, characterized in that, The method involves using the component boundary sampling points on both sides of the main direction of the facade of the current repeating facade component to solve for two sets of coordinate transformation parameters, including: Using the projected coordinates of the center of gravity of the current repeating facade component in the main direction of the facade as the boundary, determine the component boundary sampling points on the positive side of the main direction of the facade; Using the same projected coordinates as the boundary, determine the component boundary sampling points on the opposite side of the main direction of the facade, and use the component boundary sampling points on the positive side to solve a set of coordinate transformation parameters; Another set of coordinate transformation parameters is solved by sampling points of the component boundary on the opposite side, and the parameters solved in the two cases are used as two sets of coordinate transformation parameters.

8. The method according to claim 1, characterized in that, The determination of the component boundary sampling points corresponding to cross-component misalignment includes: The current repeating facade components are mapped to the target source point cloud coordinate system according to two sets of coordinate transformation parameters, forming two boundary projection lines in the building facade fitting plane; Calculate the directed spacing between the two boundary projection lines and the corresponding component boundary projection lines in the target source point cloud. Determine the direction of change of the boundary projection spacing based on the sign change of the directed spacing along the main direction of the facade, and locate the cross-component misalignment. Calculate the normal projection difference between the mapped component boundary and the corresponding component boundary in the target source point cloud. Determine the boundary normal offset direction based on the sign continuation direction of the normal projection difference, and determine the component boundary sampling point corresponding to the cross-component misalignment.

9. The method according to claim 8, characterized in that, The positioning of cross-component misalignment includes: Calculate the difference in projection coordinates between the centroids of the two boundary projection lines and the centroids of the corresponding component boundary projection lines in the target source point cloud in the main direction of the elevation. Arrange the projected coordinate differences according to the solution order of the two sets of coordinate transformation parameters, and locate the cross-component misalignment when the sign of the projected coordinate difference changes from the current repeated facade component to the adjacent repeated facade component.

10. The method according to claim 1, characterized in that, The generation of the building's digital point cloud includes: Remove the component boundary sampling points corresponding to cross-component misalignment from the component boundary sampling points, substitute the remaining component boundary sampling points into the point-surface constraint equation, and solve for the coordinate transformation parameters used for fusion. The multi-source building point cloud is transformed to the target source point cloud coordinate system according to the coordinate transformation parameters used for fusion. The coordinates of the sampled points of the multi-source building point cloud in the target source point cloud coordinate system are then fused to generate a digital building point cloud.

Images