Point cloud correction method and system based on building information model, storage medium and equipment

Abstract

The invention discloses a point cloud correction method based on a building information model. The method comprises the following steps: acquiring original point cloud data to be processed and a corresponding building information model; based on the spatial topological structure of the model, segmenting original point cloud data into a plurality of local point cloud sub-blocks and performing feature extraction to obtain main plane features of the local point cloud sub-blocks; the main plane features of the local point cloud sub-blocks are registered with corresponding space areas and structural members in the building information model, and the target pose of each local point cloud sub-block is obtained; sampling in the original point cloud data to generate a plurality of nodes, and constructing a deformation graph covering the original point cloud data; taking the target pose of each local point cloud sub-block as a position constraint, and constructing an energy function comprising a rigid constraint term, a smooth regularization term and a position constraint term; the energy function is minimized, and optimal transformation parameters of a plurality of nodes in the deformation graph are solved; and based on the parameter, calculating the final position of each point in the original point cloud data, and generating a corrected point cloud.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a point cloud correction method, system, storage medium, and device based on Building Information Modeling (BIM). Background Technology

[0002] With the widespread application of 3D laser scanning technology in fields such as architecture and civil engineering, large-scale point cloud data acquired through Simultaneous Localization and Mapping (SLAM) technology has become a crucial foundation for applications such as digital twins and construction quality inspection. However, when SLAM technology operates in long-distance, weakly textured, or loop-incomplete scenes, it inevitably generates cumulative errors (drift), leading to geometric distortions such as overall bending, twisting, or misalignment in the final generated 3D point cloud model. This drift problem severely affects the absolute accuracy of the point cloud data and its alignment with the design model, limiting its direct application in high-standard scenarios such as precise measurement and reverse modeling.

[0003] In existing technologies, point cloud drift correction methods mainly fall into two categories: one is a global optimization method based on loop closure detection, which relies on the existence of effective loop closures in the scene and has limited effectiveness in scenarios lacking obvious loop closure features, such as long corridors and underground spaces; the other is a correction method based on external positioning references (such as GPS), but it cannot be used indoors or in environments with severe satellite signal obstruction. In recent years, some researchers have attempted to use Building Information Modeling (BIM) as a reference for point cloud registration, but existing methods mostly employ a global rigid registration strategy, which cannot effectively handle the nonlinear and non-rigid characteristics of SLAM drift. When the point cloud is severely distorted, global rigid registration will destroy the local geometry, generating huge registration errors in areas with severe drift, and even leading to registration failure. Summary of the Invention

[0004] Therefore, it is necessary to propose a point cloud correction method based on Building Information Modeling to address the above problems.

[0005] A point cloud correction method based on Building Information Modeling (BIM) includes the following steps: Acquire the raw point cloud data to be processed and the corresponding building information model; Based on the spatial topology of the building information model, the original point cloud data is divided into several local point cloud sub-blocks; Feature extraction is performed on each of the local point cloud sub-blocks to obtain the principal plane features of the local point cloud sub-blocks; The main plane features of the local point cloud sub-blocks are registered with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Several nodes are uniformly sampled from the original point cloud data, and a deformed map covering the original point cloud data is constructed. Using the target pose of each local point cloud sub-block as a position constraint, an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms is constructed. Minimize the energy function to solve for the optimal transformation parameters of several nodes in the deformation graph; Based on the optimal transformation parameters of the aforementioned nodes, the final position of each point in the original point cloud data is calculated to generate the corrected point cloud.

[0006] In the above scheme, the segmentation of the original point cloud data into several local point cloud sub-blocks based on the spatial topology of the building information model specifically includes: The walls and floors in the building information model are used as physical boundaries, and the rooms or grid lines are used as basic units to determine the segmented areas. The original point cloud data is divided into several local sub-blocks; A buffer of predetermined width is set at the cutting boundary of adjacent units so that adjacent local point cloud sub-blocks contain partially identical point cloud data. The width of the buffer is determined based on the estimated cumulative drift and the local feature geometry.

[0007] In the above scheme, the step of registering the main planar features of the local point cloud sub-blocks with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block specifically includes: Obtain the structural component mesh surface within the spatial region corresponding to the local point cloud sub-block in the building information model; An error function based on point-to-plane distance is constructed. By iteratively minimizing the distance between the main plane feature of the local point cloud sub-block and the mesh surface of the structural component, the translation vector of the local point cloud sub-block in the building information model and the rotation matrix representing the three-dimensional spatial orientation are solved, and these are used as the target pose.

[0008] In the above scheme, the step of constructing an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms, using the target pose of each local point cloud sub-block as a position constraint, specifically includes:

[0009] in, These are rigid constraint terms; This is a regularization smoothing term; For positional constraints; , , These are the weight coefficients for each corresponding item.

[0010] In the above scheme, the rigid constraint term, smoothing regularization term, and position constraint term are determined according to the following formulas: The rigid constraint term is:

[0011] in, Let be the rotation matrix of the j-th node in the deformed graph. Let M be the identity matrix, and M be the total number of nodes in the deformed graph; The smoothing regularization term is:

[0012] in, and Let t be the initial coordinates of adjacent nodes j and k, respectively. j and t k Let N(j) be the translation vector between adjacent nodes j and k, and let N(j) be the set of neighboring nodes of node j. The first weighting coefficient; The position constraint term is:

[0013] in, Let be the 3D coordinates of the i-th point in the point cloud after correction. The correct 3D coordinates of point i in the point cloud in the building information model; Based on the initial coordinates vi of point i, the target pose transformation of its local point cloud sub-block is applied to obtain the correct 3D coordinates of point i in the building information model. .

[0014] In the above scheme, minimizing the energy function and solving for the optimal transformation parameters of several nodes in the deformation graph specifically includes: Assign initial transformation parameters to all nodes j in the deformed graph: rotation matrix Translation vector , wherein the rotation matrix Translation vector The initial values ​​are the identity matrix and the zero vector; The energy function is iterated, and in the k-th iteration, based on the current rotation matrix... Translation vector Calculate the energy function and update the transformation parameters based on the parameter update amounts ΔR and Δt:

[0015] in, , Let represent the rotation matrix and translation vector of node j in the k-th iteration, respectively; , Let represent the rotation matrix and translation vector of node j in the (k+1)th iteration, respectively; ΔR and Δt represent the corresponding parameter update amounts, respectively. When the energy difference between two consecutive iterations of the energy function is less than a preset threshold, the iteration is terminated, and the obtained optimal rotation matrix is... and optimal translation vector As the optimal transformation parameter for node j.

[0016] In the above scheme, the step of calculating the final position of each point in the original point cloud data based on the optimal transformation parameters of the plurality of nodes to generate the corrected point cloud specifically includes: For each point in the original point cloud data Identify the K nearest neighbor nodes that affect the point, forming a node set. ; Calculate this point To neighboring nodes interpolation weights The weight is inversely proportional to the distance from the point to the node, and satisfies: ; Based on nodes Optimal rotation matrix and optimal translation vector The final position of the point is calculated using the following formula. :

[0017] in, For point The set of K nearest neighbors; For point To neighboring nodes Interpolation weights; and For nodes Optimal rotation matrix and optimal translation vector ; Let be the initial position coordinates of the j-th node.

[0018] This application also proposes a point cloud correction system based on building information modeling, the system comprising: a data acquisition unit, a feature extraction unit, a registration unit, a deformation map construction unit, an energy function construction unit, a solution unit, and a correction generation unit; The data acquisition unit is used to acquire the raw point cloud data to be processed and the corresponding building information model; The feature extraction unit is used to divide the original point cloud data into several local point cloud sub-blocks based on the spatial topology of the building information model; and to extract features from each local point cloud sub-block to obtain the main plane features of the local point cloud sub-block. The registration unit is used to register the main planar features of the local point cloud sub-block with the corresponding spatial region and structural component in the building information model to obtain the target pose of each local point cloud sub-block. The deformation map construction unit is used to uniformly sample and generate several nodes in the original point cloud data, and construct a deformation map covering the original point cloud data. An energy function construction unit is used to construct an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms, with the target pose of each local point cloud sub-block as the position constraint. The solving unit is used to minimize the energy function and solve for the optimal transformation parameters of several nodes in the deformation graph; The correction generation unit is used to calculate the final position of each point in the original point cloud data based on the optimal transformation parameters of the plurality of nodes, and generate the corrected point cloud.

[0019] This application also proposes a readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: Acquire the raw point cloud data to be processed and the corresponding building information model; Based on the spatial topology of the building information model, the original point cloud data is divided into several local point cloud sub-blocks; Feature extraction is performed on each of the local point cloud sub-blocks to obtain the principal plane features of the local point cloud sub-blocks; The main plane features of the local point cloud sub-blocks are registered with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Several nodes are uniformly sampled from the original point cloud data, and a deformed map covering the original point cloud data is constructed. Using the target pose of each local point cloud sub-block as a position constraint, an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms is constructed. Minimize the energy function to solve for the optimal transformation parameters of several nodes in the deformation graph; Based on the optimal transformation parameters of the aforementioned nodes, the final position of each point in the original point cloud data is calculated to generate the corrected point cloud.

[0020] This application also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the following steps: Acquire the raw point cloud data to be processed and the corresponding building information model; Based on the spatial topology of the building information model, the original point cloud data is divided into several local point cloud sub-blocks; Feature extraction is performed on each of the local point cloud sub-blocks to obtain the principal plane features of the local point cloud sub-blocks; The main plane features of the local point cloud sub-blocks are registered with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Several nodes are uniformly sampled from the original point cloud data, and a deformed map covering the original point cloud data is constructed. Using the target pose of each local point cloud sub-block as a position constraint, an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms is constructed. Minimize the energy function to solve for the optimal transformation parameters of several nodes in the deformation graph; Based on the optimal transformation parameters of the aforementioned nodes, the final position of each point in the original point cloud data is calculated to generate the corrected point cloud.

[0021] The embodiments of this invention have the following beneficial effects: By semantically segmenting the original point cloud using the spatial topology of the Building Information Model (BIM), the complex global drift problem is decomposed into a registration problem of multiple local regions. Then, by registering the principal plane features of each local point cloud sub-block with precise structural components in the BIM, a high-precision target pose in the BIM coordinate system is obtained for each sub-block, thus introducing the BIM as an absolute truth benchmark. Next, an embedded deformation map covering the global point cloud is constructed, and the aforementioned target pose is used as a position constraint to construct an energy function that integrates rigid constraints, smoothing regularization, and position constraints. Finally, by minimizing this energy function, the optimal parameters driving the global point cloud to undergo non-rigid deformation are solved, ensuring that the corrected point cloud not only accurately aligns with the BIM as a whole, eliminating cumulative drift, but also maintains the integrity of local geometric features and the smooth continuity of global deformation, effectively avoiding the geometric distortion or seam tearing problems that may occur in areas with severe drift in traditional rigid registration methods. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] in: Figure 1 This is a schematic diagram of a point cloud correction method based on building information model in one embodiment. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention; however, it will be apparent to those skilled in the art that the invention may be practiced without one or more of these details; in other instances, certain technical features well-known in the art have not been described in order to avoid confusion with the invention. It should be understood that the invention can be practiced in different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided to make the disclosure thorough and complete and to fully convey the scope of the invention to those skilled in the art.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms, unless the context clearly indicates otherwise. The terms “comprising” and / or “including,” when used in this specification, identify the presence of said features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0027] To fully understand the present invention, a detailed structure will be presented in the following description in order to illustrate the technical solution proposed by the present invention; optional embodiments of the present invention are described in detail below, however, in addition to these detailed descriptions, the present invention may have other embodiments.

[0028] like Figure 1As shown, in one embodiment, a point cloud correction method based on Building Information Modeling (BIM) is provided. This BIM-based point cloud correction method includes steps S101 to S108, which are detailed below: S101. Obtain the raw point cloud data to be processed and the corresponding building information model; Building Information Modeling (BIM) is a data-driven tool applied to engineering design, construction, and management. By integrating digital information such as the geometry, physical properties, and functions of a building project, it enables information sharing and transmission throughout the entire project lifecycle, supporting collaborative design, construction simulation, cost control, and operation and maintenance management, thereby improving the efficiency and quality of the construction industry.

[0029] By simultaneously acquiring raw point cloud data obtained from on-site scanning and BIM models from the design phase, a data link between the actual scene and digital design was established, laying a data foundation for the subsequent use of high-precision BIM models to guide and constrain the correction process of point cloud data.

[0030] S102. Based on the spatial topology of the building information model, the original point cloud data is divided into several local point cloud sub-blocks; Preferably, the semantic and topological information of the BIM model (such as the boundaries and positions of components like walls, columns, and beams) is used to intelligently segment the original point cloud. This not only avoids the problem that traditional uniform grid segmentation may cut off the edges of objects, but also ensures that each sub-block corresponds to a specific building component or spatial region. This overcomes the complexity of the overall point cloud under non-rigid deformation, making subsequent local feature extraction and registration more accurate.

[0031] In some embodiments, based on the spatial topology of the Building Information Model (BIM), the original point cloud data is divided into several local point cloud sub-blocks, specifically including: The walls and floors in the building information model are used as physical boundaries, and the rooms or grid lines are used as basic units to determine the segmented areas. The original point cloud data is divided into several local sub-blocks; A buffer of predetermined width is set at the cutting boundary of adjacent units so that adjacent local point cloud sub-blocks contain some of the same point cloud data. The width of the buffer is determined based on the estimated cumulative drift and the local feature geometry.

[0032] The preferred formula for calculating the width of the buffer is: ,in To estimate the cumulative drift, The local feature geometric scale is denoted by k, and k is the safety factor (values ​​range from 1.5 to 2.0).

[0033] First, by utilizing physical boundaries such as walls and floors in BIM and dividing the data into logical units like rooms or grid lines, each local sub-block strictly corresponds to a real building component or functional space, laying a semantic foundation for subsequent precise component-level registration. Second, buffer zones are set at the cutting boundaries of adjacent units, allowing adjacent sub-blocks to contain partially overlapping point cloud data. This design cleverly addresses potential cumulative errors or boundary mismatches that may arise during subsequent local registration. The buffer width is adaptively determined based on the estimated scan drift and local feature scale, ensuring registration continuity while avoiding unnecessary computational burden due to excessive overlap. Overall, this approach not only achieves efficient and meaningful organization of point clouds but also provides crucial data redundancy and constraints for subsequent global smoothing correction through the overlap buffer mechanism, effectively improving the robustness of the overall correction algorithm and the geometric consistency of the final result.

[0034] S103. Extract features from each local point cloud sub-block to obtain the principal plane features of the local point cloud sub-block; By extracting principal plane features (walls and ground), effective dimensionality reduction and abstraction of the core geometric information of local point clouds are achieved. Building structures are usually composed of a large number of planes. Extracting principal plane features can filter out noise and interference from discrete points, highlighting the geometric features that best represent the posture of the components, and providing a reliable geometric basis for subsequent accurate registration with the corresponding planes in the BIM model.

[0035] S104. Register the main planar features of the local point cloud sub-blocks with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Using the high-precision geometric information of the BIM model as a benchmark, the correct positions and orientations of the point cloud sub-blocks were calculated. This step, by aligning the actual scanned data with the ideal design model, determined the target for local correction and solved the problem of absolute positional deviations in the original point cloud caused by scanning errors or object deformation.

[0036] In some embodiments, the principal planar features of the local point cloud sub-blocks are registered with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block, specifically including: Obtain the structural component mesh surface within the spatial region corresponding to the local point cloud sub-block in the building information model; An error function based on point-to-plane distance is constructed. By iteratively minimizing the distance between the main plane feature of the local point cloud sub-block and the mesh surface of the structural component, the translation vector of the local point cloud sub-block in the building information model and the rotation matrix representing the orientation in three-dimensional space are solved, and these are used as the target pose.

[0037] First, by directly associating the mesh surfaces of the corresponding structural components in the BIM, a high-precision, noise-free geometric reference target is provided for registration, avoiding the uncertainty of point cloud-to-point cloud registration. Second, an error function is constructed using point-to-plane distance and iteratively optimized. This method effectively utilizes the strong geometric constraints of planar features, exhibits good robustness to noise and outliers in the point cloud, and can converge quickly and stably. Finally, the solved translation vector and rotation matrix together constitute a rigid body transformation. This transformation not only aligns the measured point cloud plane with the design model plane in space but also strictly guarantees the rigid invariance of the internal geometry of local sub-blocks. This provides accurate and reliable local anchor point constraints for subsequent global non-rigid deformation, which is a key step in ensuring the overall correction accuracy.

[0038] S105. Generate several nodes by uniformly sampling from the original point cloud data, and construct a deformed map covering the original point cloud data. A gridded control structure for controlling the overall deformation of a point cloud was established. By constructing a deformation map, the complex problem of continuous point cloud deformation is transformed into a problem of solving the transformation parameters of a finite number of nodes, greatly reducing computational complexity. Uniformly sampled nodes can uniformly control the deformation field of the point cloud, ensuring that the distribution of the correction force across the entire point cloud space is continuous and controllable.

[0039] S106. Using the target pose of each local point cloud sub-block as the position constraint, construct an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms. Among them, the rigid constraint term ensures the rigidity of the local sub-block, prevents the sub-block from undergoing unnatural twisting or stretching during the correction process, and maintains the geometric integrity of the building components. Among them, the rigid constraint term forces that the transformation of each control node in the deformation diagram be as close as possible to a rigid rotation (the rotation matrix is ​​orthogonal and the determinant is 1), which effectively prevents non-physical deformations (such as scaling and shearing) in local areas and ensures that the basic rigidity characteristics of the internal geometry of the point cloud sub-block are not destroyed during the correction process.

[0040] The smoothing regularization term acts between adjacent control nodes, forcing the nodes to transform spatially neighboring points as consistently as possible, thus propagating smoothness throughout the deformation field. This ensures natural transitions from different local sub-block boundaries, eliminates splicing cracks or distortions that may result from independent registration, and achieves seamless global deformation.

[0041] The position constraint term serves as a bridge connecting the optimization process with the previous local registration results. It projects each point onto the BIM coordinate system through the target pose (i.e., the local optimal rigid transformation) of its sub-block, resulting in a target position. The optimization process strives to make the final deformation of the point cloud approximate these target positions, thereby ensuring accurate alignment between the overall point cloud and the BIM model in key local areas.

[0042] In some embodiments, the target pose of each local point cloud sub-block is used as the position constraint to construct an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms, specifically including:

[0043] in, These are rigid constraint terms; This is a regularization smoothing term; For positional constraints; , , These are the weight coefficients for each corresponding item.

[0044] In some embodiments, the rigid constraint term, the smoothing regularization term, and the position constraint term are determined according to the following formula: The rigid constraint terms are:

[0045] in, Let be the rotation matrix of the j-th node in the deformed graph. Let M be the identity matrix, and M be the total number of nodes in the deformed graph; The smoothing regularization term is:

[0046] in, and Let t be the initial coordinates of adjacent nodes j and k, respectively. j and t k Let N(j) be the translation vector between adjacent nodes j and k, and let N(j) be the set of neighboring nodes of node j. The first weighting coefficient; The position constraints are:

[0047] in, Let be the 3D coordinates of the i-th point in the point cloud after correction. The correct 3D coordinates of point i in the point cloud in the building information model; Based on the initial coordinates vi of point i, the target pose transformation of its local point cloud sub-block is applied to obtain the correct 3D coordinates of point i in the building information model. .

[0048] By assigning adaptive weights to these three terms and minimizing them, this energy function intelligently balances three sometimes conflicting objectives: accurate local alignment, local shape preservation, and smooth global deformation. It guides the entire point cloud in a physically reasonable and geometrically continuous manner, smoothly and accurately deforming it from its original, error-laden scan state to a state highly consistent with the BIM design model, thus outputting a high-quality, corrected point cloud that can be directly used for comparative analysis.

[0049] S107. Minimize the energy function and solve for the optimal transformation parameters of several nodes in the deformation diagram; By solving the global energy minimization problem, the optimal node transformation parameters that balance local precise alignment and global smoothness were obtained. This step utilizes a numerical optimization algorithm to find the node transformation solution that makes the deformed graph smoothest (i.e., minimizes the regularization term) while satisfying rigidity and position constraints, thus realizing the transfer from local constraints to the global optimal solution.

[0050] In some embodiments, minimizing the energy function and solving for the optimal transformation parameters of several nodes in the deformation graph specifically includes: Assign initial transformation parameters to all nodes j in the deformed graph: rotation matrix Translation vector , where the rotation matrix Translation vector The initial values ​​are the identity matrix and the zero vector; The energy function is iterated, and in the k-th iteration, based on the current rotation matrix... Translation vector Calculate the energy function and update the transformation parameters based on the parameter update amounts ΔR and Δt:

[0051] in, , Let represent the rotation matrix and translation vector of node j in the k-th iteration, respectively; , Let represent the rotation matrix and translation vector of node j in the (k+1)th iteration, respectively; ΔR and Δt represent the corresponding parameter update amounts, respectively. The iteration terminates when the energy difference between two consecutive iterations of the energy function is less than a preset threshold, and the resulting optimal rotation matrix is ​​obtained. and optimal translation vector As the optimal transformation parameter for node j.

[0052] This invention starts with an initial identity matrix and zero vector, and iteratively refines the transformation parameters of each node. In each iteration, the algorithm calculates the total cost of the energy function based on the current parameters and finds a parameter update amount that reduces this cost, thereby driving the node transformation to evolve in a direction that satisfies all constraints (rigidity, smoothness, and positional alignment). This incremental update method decomposes the complex nonlinear optimization problem into a series of more manageable steps, ensuring the stability and controllability of the solution process.

[0053] S108. Based on the optimal transformation parameters of several nodes, calculate the final position of each point in the original point cloud data and generate the corrected point cloud.

[0054] The optimal transformation parameters of the solved nodes are interpolated or propagated to every point in the original point cloud, achieving precise relocation of the entire point cloud data. This method not only significantly reduces the systematic errors and random noise of the original point cloud data, enabling high-precision geometric registration with the BIM model, but also preserves the detailed features of the point cloud, generating a high-quality corrected point cloud that conforms to design specifications and reflects actual site details. This provides reliable data support for subsequent applications such as construction quality inspection and progress management.

[0055] In some embodiments, based on the optimal transformation parameters of several nodes, the final position of each point in the original point cloud data is calculated to generate a corrected point cloud, specifically including: For each point in the original point cloud data Identify the K nearest neighbor nodes that affect the point, forming a node set. ; Calculate this point To neighboring nodes interpolation weights The weight is inversely proportional to the distance from the point to the node, and satisfies: ; Based on nodes Optimal rotation matrix and optimal translation vector The final position of the point is calculated using the following formula. :

[0056] in, For point The set of K nearest neighbors; For point To neighboring nodes Interpolation weights; and For nodes Optimal rotation matrix and optimal translation vector ; Let be the initial position coordinates of the j-th node.

[0057] This invention first adaptively determines the K nearest neighbor control nodes for each point in the point cloud. This ensures that the transformation of each point is only directly affected by nodes in its local region, making the deformation local and avoiding the invalid or erroneous influence of distant nodes. Secondly, it calculates interpolation weights based on inverse distance ratios; points closer to a node are more significantly affected by that node's transformation, which aligns with physical intuition and ensures a natural spatial transition of the deformation field. Most importantly, the final position of each point is not determined by the transformation of a single node, but rather by the weighted average of the optimal transformations of all its K nearest neighbors.

[0058] This calculation formula ensures that the transformations experienced by adjacent points within a local area are highly consistent, thus strictly maintaining the local rigidity and global smoothness sought in the optimization process and effectively preventing cracks or discontinuous distortions between nodes. Finally, the set of node parameters obtained from the aforementioned global optimization is efficiently and robustly converted into a dense and continuous displacement field covering the entire point cloud. This allows the original point cloud to be precisely moved to its correct position in a way that preserves geometric details, eliminates accumulated errors, and aligns with the BIM model as a whole, generating high-quality correction results that can be directly used for subsequent analysis.

[0059] In some embodiments, missing regions in the point cloud after correction are detected, and intelligent point cloud completion is performed using a pre-trained deep neural network, while geometric constraints from the BIM model are introduced during the completion process.

[0060] The above-mentioned intelligent point cloud completion using pre-trained deep neural networks specifically includes: Detecting holes or low-density regions in point clouds based on voxelization methods; Pre-trained deep neural networks employ generative adversarial networks or Transformer architectures; The training data for the network is constructed as follows: ideal point clouds are generated using BIM components as ground truth, and random masks are applied to them to simulate scan occlusion as network input; If a missing region is detected to be located on a wall plane defined in the building information model, the completed point cloud generated by the neural network is projected onto the plane equation Ax+By+Cz+D=0 where the wall is located.

[0061] Preferably, after local rigid registration, a construction error judgment step is also included: Calculate the average deviation distance between the local point cloud sub-blocks and the structural components of the building information model after registration; If the average deviation distance is greater than the preset construction error threshold, it is determined that there is a construction error in the area, and one of the following operations will be performed automatically: reduce the weight of the corresponding position constraint term in the energy function, or mark the area as an abnormal area for subsequent analysis.

[0062] This application also proposes a point cloud correction system based on building information modeling, which includes: a data acquisition unit, a feature extraction unit, a registration unit, a deformation map construction unit, an energy function construction unit, a solution unit, and a correction generation unit; The data acquisition unit is used to acquire the raw point cloud data to be processed and the corresponding building information model; The feature extraction unit is used to segment the original point cloud data into several local point cloud sub-blocks based on the spatial topology of the building information model; and to extract features from each local point cloud sub-block to obtain the main plane features of the local point cloud sub-block. The registration unit is used to register the main planar features of the local point cloud sub-blocks with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. The deformation map construction unit is used to uniformly sample and generate several nodes in the original point cloud data, and construct a deformation map covering the original point cloud data. The energy function construction unit is used to construct an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms, with the target pose of each local point cloud sub-block as the position constraint. The solver element is used to minimize the energy function and solve for the optimal transformation parameters of several nodes in the deformation diagram. The correction generation unit is used to calculate the final position of each point in the original point cloud data based on the optimal transformation parameters of several nodes, and generate the corrected point cloud.

[0063] This application also proposes a readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps: Acquire the raw point cloud data to be processed and the corresponding building information model; Based on the spatial topology of Building Information Modeling, the original point cloud data is divided into several local point cloud sub-blocks. Feature extraction is performed on each local point cloud sub-block to obtain the principal plane features of the local point cloud sub-block; Register the main planar features of the local point cloud sub-blocks with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Several nodes are generated by uniformly sampling from the original point cloud data, and a deformed map covering the original point cloud data is constructed. Using the target pose of each local point cloud sub-block as the position constraint, an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms is constructed. Minimize the energy function to solve for the optimal transformation parameters of several nodes in the deformed graph; Based on the optimal transformation parameters of several nodes, the final position of each point in the original point cloud data is calculated to generate the corrected point cloud.

[0064] This application also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps: Acquire the raw point cloud data to be processed and the corresponding building information model; Based on the spatial topology of Building Information Modeling, the original point cloud data is divided into several local point cloud sub-blocks. Feature extraction is performed on each local point cloud sub-block to obtain the principal plane features of the local point cloud sub-block; Register the main planar features of the local point cloud sub-blocks with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Several nodes are generated by uniformly sampling from the original point cloud data, and a deformed map covering the original point cloud data is constructed. Using the target pose of each local point cloud sub-block as the position constraint, an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms is constructed. Minimize the energy function to solve for the optimal transformation parameters of several nodes in the deformed graph; Based on the optimal transformation parameters of several nodes, the final position of each point in the original point cloud data is calculated to generate the corrected point cloud.

[0065] Those skilled in the art will understand that implementing all or part of the processes in the above embodiments can be accomplished by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0066] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The above embodiments merely illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application's patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be used to limit the scope of the present invention. Therefore, equivalent variations made according to the claims of this invention are still within the scope of this invention.

Claims

1. A point cloud correction method based on Building Information Modeling, characterized in that, The method includes: Acquire the raw point cloud data to be processed and the corresponding building information model; Based on the spatial topology of the building information model, the original point cloud data is divided into several local point cloud sub-blocks; Feature extraction is performed on each of the local point cloud sub-blocks to obtain the principal plane features of the local point cloud sub-blocks; The main plane features of the local point cloud sub-blocks are registered with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block. Several nodes are uniformly sampled from the original point cloud data, and a deformed map covering the original point cloud data is constructed. Using the target pose of each local point cloud sub-block as a position constraint, an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms is constructed. Minimize the energy function to solve for the optimal transformation parameters of several nodes in the deformation graph; Based on the optimal transformation parameters of the aforementioned nodes, the final position of each point in the original point cloud data is calculated to generate the corrected point cloud.

2. The point cloud correction method based on Building Information Modeling according to claim 1, characterized in that, The spatial topology based on the building information model divides the original point cloud data into several local point cloud sub-blocks, specifically including: The walls and floors in the building information model are used as physical boundaries, and the rooms or grid lines are used as basic units to determine the segmented areas. The original point cloud data is divided into several local sub-blocks; A buffer of predetermined width is set at the cutting boundary of adjacent units so that adjacent local point cloud sub-blocks contain partially identical point cloud data. The width of the buffer is determined based on the estimated cumulative drift and the local feature geometry.

3. The point cloud correction method based on Building Information Modeling according to claim 1, characterized in that, The step of registering the principal planar features of the local point cloud sub-blocks with the corresponding spatial regions and structural components in the building information model to obtain the target pose of each local point cloud sub-block specifically includes: Obtain the structural component mesh surface within the spatial region corresponding to the local point cloud sub-block in the building information model; An error function based on point-to-plane distance is constructed. By iteratively minimizing the distance between the main plane feature of the local point cloud sub-block and the mesh surface of the structural component, the translation vector of the local point cloud sub-block in the building information model and the rotation matrix representing the three-dimensional spatial orientation are solved, and these are used as the target pose.

4. The point cloud correction method based on Building Information Modeling according to claim 3, characterized in that, The construction of an energy function, which uses the target pose of each local point cloud sub-block as a position constraint and includes rigid constraint terms, smoothing regularization terms, and position constraint terms, specifically includes: in, These are rigid constraint terms; This is a regularization smoothing term; For positional constraints; , , These are the weight coefficients for each corresponding item.

5. The point cloud correction method based on Building Information Modeling according to claim 4, characterized in that, The rigid constraint term, smoothing regularization term, and position constraint term are determined using the following formulas: The rigid constraint term is: in, Let be the rotation matrix of the j-th node in the deformed graph. Let M be the identity matrix, and M be the total number of nodes in the deformed graph; The smoothing regularization term is: in, and Let t be the initial coordinates of adjacent nodes j and k, respectively. j and t k Let N(j) be the translation vector between adjacent nodes j and k, and let N(j) be the set of neighboring nodes of node j. This is the first weighting coefficient; The position constraint term is: in, Let be the 3D coordinates of the i-th point in the point cloud after correction. The correct 3D coordinates of point i in the point cloud in the building information model; Based on the initial coordinates vi of point i, the target pose transformation of its local point cloud sub-block is applied to obtain the correct 3D coordinates of point i in the building information model. .

6. The point cloud correction method based on Building Information Modeling according to claim 5, characterized in that, Minimizing the energy function and solving for the optimal transformation parameters of several nodes in the deformation graph specifically includes: Assign initial transformation parameters to all nodes j in the deformed graph: rotation matrix Translation vector , wherein the rotation matrix Translation vector The initial values ​​are the identity matrix and the zero vector; The energy function is iterated, and in the k-th iteration, based on the current rotation matrix... Translation vector Calculate the energy function and update the transformation parameters based on the parameter update amounts ΔR and Δt: in, , Let represent the rotation matrix and translation vector of node j in the k-th iteration, respectively; , Let represent the rotation matrix and translation vector of node j in the (k+1)th iteration, respectively; ΔR and Δt represent the corresponding parameter update amounts, respectively. When the energy difference between two consecutive iterations of the energy function is less than a preset threshold, the iteration is terminated, and the obtained optimal rotation matrix is... and optimal translation vector As the optimal transformation parameter for node j.

7. The point cloud correction method based on Building Information Modeling according to claim 6, characterized in that, The step of calculating the final position of each point in the original point cloud data based on the optimal transformation parameters of the plurality of nodes, and generating the corrected point cloud, specifically includes: For each point in the original point cloud data Identify the K nearest neighbor nodes that affect the point, forming a node set. ; Calculate this point To neighboring nodes interpolation weights The weight is inversely proportional to the distance from the point to the node, and satisfies: ; Based on nodes Optimal rotation matrix and optimal translation vector The final position of the point is calculated using the following formula. : in, For point The set of K nearest neighbors; For point To neighboring nodes Interpolation weights; and For nodes Optimal rotation matrix and optimal translation vector ; Let be the initial position coordinates of the j-th node.

8. A point cloud correction system based on Building Information Modeling, characterized in that, The system includes: a data acquisition unit, a feature extraction unit, a registration unit, a deformation map construction unit, an energy function construction unit, a solution unit, and a correction generation unit; The data acquisition unit is used to acquire the raw point cloud data to be processed and the corresponding building information model; The feature extraction unit is used to divide the original point cloud data into several local point cloud sub-blocks based on the spatial topology of the building information model; and to extract features from each local point cloud sub-block to obtain the main plane features of the local point cloud sub-block. The registration unit is used to register the main planar features of the local point cloud sub-block with the corresponding spatial region and structural component in the building information model to obtain the target pose of each local point cloud sub-block. The deformation map construction unit is used to uniformly sample and generate several nodes in the original point cloud data, and construct a deformation map covering the original point cloud data. An energy function construction unit is used to construct an energy function containing rigid constraint terms, smoothing regularization terms, and position constraint terms, with the target pose of each local point cloud sub-block as the position constraint. The solving unit is used to minimize the energy function and solve for the optimal transformation parameters of several nodes in the deformation graph; The correction generation unit is used to calculate the final position of each point in the original point cloud data based on the optimal transformation parameters of the plurality of nodes, and generate the corrected point cloud.

9. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the processor performs the steps of the method as described in any one of claims 1 to 7.

10. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images