Pole tower point cloud and model multi-scale registration method and system based on prior constraint

Abstract

The invention relates to the technical field of three-dimensional image data processing, in particular to a priori constraint-based tower point cloud and model multi-scale registration method and system. The method comprises the following steps: S1, performing standardization processing on multi-source data to enable to-be-registered point cloud data to be matched with a point cloud representation standard of a prior knowledge base of a candidate model; s2, preliminary screening; s3, performing coarse registration, and performing candidate model rapid screening and key point priority sampling strategies; s4, fine registration is carried out, soft constraint optimization is adopted, priori knowledge is quantized into weights, and weighted ICP and nonlinear least square are combined for solving; and S5, performing matching scoring according to the matching parameters, and obtaining an optimal matching result of the to-be-registered point cloud data and the candidate model of coarse registration. A self-learning priori knowledge base is constructed, a high-confidence matching result is fed back to update the priori base, a dynamic weight adjustment strategy and a candidate screening rule are realized, and the adaptability of the system to different pole tower models is improved.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional image data processing technology, and in particular to a method and system for multi-scale registration of pole point clouds and models based on prior constraints. Background Technology

[0002] In the field of 3D modeling and point cloud registration of transmission line towers, existing technologies typically rely on geometric matching or global descriptor methods to align the 3D model of the tower with the measured point cloud. The FBX format tower 3D model and the LAS format tower point cloud exhibit significant differences in data representation, including coordinate systems, units, scale, orientation, and origin position. These heterogeneities make direct registration difficult, usually requiring manual adjustment or reliance on coarse global registration algorithms, such as point-to-point distance or voxel histogram matching. However, these methods often suffer from insufficient accuracy and low efficiency in practical applications, failing to meet the high-precision requirements for tower location and structure in transmission line construction and operation.

[0003] On the other hand, point cloud data itself suffers from problems such as noise, uneven point density, occlusion, and complex geometric structures. Traditional registration methods often rely on geometric feature point matching or global descriptor calculation. However, these methods are prone to falling into the trap of local optima leading to overall misalignment, especially in the correspondence of key structures (such as tower tops, crossarm endpoints, and attachment points), which can easily lead to height shifts, hierarchical misalignments, or crossarm misalignments in the overall model. Furthermore, existing technologies lack effective utilization of tower topology information, crossarm layer count, and other structural features, and cannot embed prior knowledge into the registration process.

[0004] Furthermore, existing methods lack adaptive and intelligent mechanisms, making it impossible to dynamically update the prior library based on historical matching results, and also difficult to adjust candidate model selection and matching strategies according to different tower models. Overall, existing technologies have significant shortcomings in multi-source data compatibility, key structure matching accuracy, automation level, robustness, and adaptability, which limits the application and promotion of high-precision registration of large-scale tower point clouds with models. Summary of the Invention

[0005] Therefore, it is necessary for the present invention to provide a method and system for multi-scale registration of tower point clouds and models based on prior constraints, in order to solve at least one of the above-mentioned technical problems.

[0006] To achieve the above objectives, a multi-scale registration method for pole point clouds and models based on prior constraints is proposed, comprising the following steps: S1, standardization processing of multi-source data, including: S11, Construction of the prior knowledge base for the 3D model of the pole: Extract the geometric element information, preset key points, descriptors and semantic attributes of the preset key points of different 3D models of poles, and store the point cloud data information corresponding to each 3D model of poles as a prior knowledge base in a preset format. S12, Tower point cloud data preprocessing: The tower point cloud data is preprocessed and global geometric features are extracted to construct point cloud data to be registered; wherein, the point cloud data to be registered is matched with the point cloud representation standard of the prior knowledge base; S2, Preliminary screening: Based on the geometric element information and semantic attributes of the prior knowledge base, the point cloud data to be registered is initially matched, and the 3D models of the tower with similarity higher than the preset value are obtained as candidate models. S3, Coarse Registration: After matching the main axis of the tower body direction with the Z-axis of the point cloud data to be registered, feature matching under prior constraints is performed on the preset key points in the prior knowledge base, and coarse registration results are output. Candidate models whose errors meet the preset conditions are selected as candidate models for coarse registration. S4, fine registration, includes: S41, assign differentiated weights to preset key points in the prior knowledge base and construct an initial correspondence; S42, by constructing an optimized objective function, an alignment target is built between the point cloud data to be registered and the candidate model for coarse registration; S43, Solve the optimization objective function to obtain the matching parameters between the point cloud data to be registered and each candidate model for coarse registration; S5. Perform a matching score based on the matching parameters to obtain the best matching result between the point cloud data to be registered and the candidate model for coarse registration.

[0007] Preferably, step S11 specifically includes: S111, Perform geometric analysis on the three-dimensional model of the tower and calculate the geometric element information; the geometric element information includes the geometric center, the base center, the boundary box dimensions, and the principal axis direction; S112, Preset key point extraction, including: extraction of global key points, mesoscale key points and local key points; the global key points include the top and bottom of the tower, the mesoscale key points include the crossarm endpoints, and the local key points include the cover nodes and connection points. S113, Descriptor Calculation: For the extracted global key points, mesoscale key points, and local key points, a point cloud is generated through grid sampling, and then the local descriptor, neighborhood radius, and global descriptor are calculated using FPFH. S114, Extract semantic attributes: including the number of crossarm layers, number of segments, and the pattern and information of the connection points; S115, Perform statistical analysis on different tower 3D models, extract the mean and variance of geometric quantities, construct the corresponding tolerance range, and construct the prior knowledge base.

[0008] Preferably, step S2 includes: S21, Height Constraint: Calculate the maximum height difference in the principal direction of the point cloud data to be registered. The height of the three-dimensional model of the tower. height difference , ,in, To allow for height deviation, , The z-axis coordinate of point i in the point cloud data to be registered; S22, Connection Point Constraint: Calculate the number of wires and ground wires in the point cloud data to be registered using clustering. Based on the number of attachment points in the 3D model of the tower Ensure that the number of conductors and ground wires matches the number of connection points; S23, Crossarm Count Constraint: After projecting the point cloud data to be registered, the number of crossarms in the tower body is counted. The number of layers in the 3D model of the tower same; S24. Select the three-dimensional model of the tower that meets the constraints of steps S21 to S23 above as the candidate model.

[0009] Preferably, step S3 includes: S31, obtain the principal direction Z-axis of the point cloud data to be registered through PCA principal component analysis, and rotate the point cloud data to be registered to be consistent with the principal axis direction of the candidate model; S32, Key point matching: The coordinates of the points in the point cloud data to be registered are calculated by clustering. The corresponding key point annotations are called from the prior knowledge base of the candidate model. By comparing the spatial coordinates, the key points of the point cloud data to be registered are initially matched with the key points of the candidate model, and a point-to-point matching relationship is established. S33, based on the descriptors of the preset key points, a coarse matching is performed using the RANSAC random sampling consensus algorithm to obtain 3D tower models with similarity higher than a preset value as candidate models, and the initial rotation matrix of the point cloud data to be registered relative to the candidate models is recorded. Translation vector .

[0010] Preferably, step S33 includes the coarse matching including prior knowledge constraints: If the distance difference between the hook points is greater than a set threshold, the match is considered invalid. If the difference in crossarm length exceeds the tolerance range, the match is rejected. If the preset number of key points or the hierarchical distribution are inconsistent, incorrect candidate models will be directly excluded.

[0011] Preferably, step S41 includes: S411, assign differentiated weights to the preset key points i in the prior knowledge base. The preset key point i includes high-weight key points and ordinary points; S412, using the initial rotation matrix output from coarse registration. Translation vector The point cloud data to be registered is initially aligned with the point cloud of the candidate model; S413, Key Point Correspondence Construction: S4131: For each high-weight key point in the point cloud data to be registered, search for the most similar candidate point in the point cloud of the candidate model. S4132, Local Descriptor Matching: Calculating High-Weight Keypoints Fast point feature histogram descriptors: select the point with the highest similarity between the descriptor of the high-weight key point and the descriptor of the candidate point as the corresponding point; S4133, for each pair of high-weight key points Calculate the matching confidence score , ;in, Indicates proportional to, This represents a local descriptor function. As a similarity metric, This indicates the key points of the point cloud data to be registered. express The corresponding points in the point cloud of the candidate model; S4134 performs downsampling processing on each ordinary point in the point cloud data to be registered, and establishes an auxiliary correspondence.

[0012] Preferably, step S42 includes: S421, each key point in the point cloud data to be registered Calculate point-to-point error : ;in, , Represents the rotation matrix. It is the set of all rotation matrices used to represent all possible rotations of a rigid body in three-dimensional space; , Represents the translation vector. S represents the three-dimensional translation vector; S is the set of point indices participating in the optimization. S422, Constructing the prior constraint loss: Alignment of the center points of the tower top and bottom: The height difference between the center points of the tower top and bottom of the point cloud data to be registered and the center points of the tower top and bottom of the candidate model is minimized, calculated by the following formula: ;in, This represents the Z-axis coordinate of point k in the point cloud data to be registered. The Z-axis coordinate of point k in the candidate model; Key layer alignment: Ensure that the crossarm endpoints and attachment points are aligned with the model in the height direction, calculated using the following formula: ;in, This represents the Z-axis coordinate of point j in the point cloud data to be registered. Represents the Z-axis coordinate of point j in the candidate model; Symmetry and Proportion Constraints: Maintaining the symmetry of the crossarm and the left and right structures of the tower is expressed by distance constraints as follows: ;in, This represents the symmetric distance of point l in the point cloud data to be registered. Represents the symmetric distance of point l in the candidate model; The prior loss is formed by weighted combination: ;in, , as well as This is a coefficient, adjusted according to the tower type; Construct the overall optimization objective function: ,in, is a coefficient.

[0013] Preferably, step S43 includes: using a weighted ICP framework + nonlinear least squares to solve for the optimal rigid body transformation parameters. .

[0014] Preferably, step S5 includes: S51, ICP overlap score : N is the number of points involved in the registration. S52, Descriptor Consistency Score : Among them, Represents cosine similarity, where, A descriptor representing point k in the point cloud data to be registered. The descriptor of point k in the candidate model; K is the number of point pairs participating in the descriptor matching; S53, Consistency Score : ;in, The height difference of the attachment points representing the candidate models. The height difference of the attachment points in the point cloud data to be registered; M is the number of attachment points participating in the structural feature comparison. S54, Construct the comprehensive scoring function: α, β, γ are normalized weighting coefficients that control the importance of the three indicators (α+β+γ=1). S55, select the candidate model with the highest coarse registration score as the best matching result.

[0015] The present invention also provides a multi-scale registration system for pole point clouds and models based on prior constraints, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the multi-scale registration method for pole point clouds and models based on prior constraints as described above.

[0016] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for multi-scale registration of pole point clouds and models based on prior constraints as described above.

[0017] This invention presents a method and system for multi-scale registration of tower point clouds and models based on prior constraints. Addressing the problems of heterogeneous multi-source data, point cloud noise interference, unstable matching of key structures, underutilization of topological information, and lack of adaptive mechanisms in existing technologies, this invention proposes a method for multi-scale registration of tower point clouds and models based on prior knowledge constraints. The core objective of this method is to achieve high-precision, automated, and robust registration of tower models with measured point clouds. Compared with existing technologies, this invention has at least the following advantages: (1) High accuracy in matching key structures: By extracting key points at multiple scales, such as the geometric center of the tower, the center point of the tower top / bottom, the endpoints of the crossarms, and nodes, and combining local and global descriptors, geometric, topological, and semantic prior information is deeply embedded into the registration process, thereby achieving high-precision correspondence of key structures. Compared with existing registration methods that rely solely on geometric features or global descriptors, this invention significantly reduces the risks of tower top offset, crossarm misalignment, and overall structural asymmetry, and improves the stability and reliability of the registration results.

[0018] (2) The registration process is robust and highly automated: In the coarse registration stage, the reliability of the initial transformation is improved by rapidly screening candidate models and prioritizing key point sampling. In the fine registration stage, soft constraint optimization is adopted, prior knowledge is quantified into weights, and combined with weighted ICP and nonlinear least squares solution, so that the registration process can still converge stably even when there is noise, uneven density, or partial occlusion in the point cloud. The entire registration process is highly automated, requiring no large amount of manual intervention, which greatly improves the registration efficiency and ease of operation.

[0019] (3) Enhanced self-learning capability improves system adaptability: It has a self-learning mechanism that feeds back high-confidence matching results to update the prior library, and can dynamically adjust the weight strategy and candidate model selection rules, thereby enhancing the system's adaptability to different tower types and complex environments. Compared with the lack of adaptive mechanism in existing technologies, this invention can continuously optimize matching accuracy and robustness, and achieve long-term stable operation.

[0020] (4) Comprehensive scoring strategy improves matching reliability: By integrating multi-dimensional information such as geometric overlap, local and global descriptor consistency, and structural semantic consistency, a comprehensive and accurate scoring judgment is made on candidate matches, effectively improving the reliability of matching results. Existing technologies mostly rely on a single indicator for evaluation, which easily leads to insufficient accuracy and stability. However, this invention significantly enhances the comprehensiveness and reliability of matching judgment.

[0021] (5) This invention takes into account registration accuracy, computational efficiency, automation level and system adaptability, and overcomes the shortcomings of existing technologies in multi-source data processing, key structure matching, utilization of prior constraints and self-learning ability. It is especially suitable for high-precision registration of tower point clouds and three-dimensional models and digital twin applications in complex environments, and can significantly improve the efficiency and reliability of engineering applications. Attached Figure Description

[0022] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart illustrating a method for multi-scale registration of tower point clouds and models based on prior constraints provided by the present invention.

[0024] Figure 2 This is a schematic diagram of the hardware structure of a system running a multi-scale registration method for tower point clouds and models based on prior constraints, according to an embodiment of the present invention.

[0025] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0026] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] Furthermore, the accompanying drawings are merely illustrative of the invention 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. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0028] To achieve the above objectives, please refer to Figure 1 and Figure 2 This invention provides a method for multi-scale registration of tower point clouds and models based on prior constraints, comprising the following steps: S1, standardization processing of multi-source data, including: S11, Construction of the prior knowledge base for the 3D model of the pole: Extract the geometric element information, preset key points, descriptors and semantic attributes of the preset key points of different 3D models of poles, and store the point cloud data information corresponding to each 3D model of poles as a prior knowledge base in a preset format. S12, Tower point cloud data preprocessing: The tower point cloud data is preprocessed and global geometric features are extracted to construct point cloud data to be registered; wherein, the point cloud data to be registered is matched with the point cloud representation standard of the prior knowledge base; S2, Preliminary screening: Based on the geometric element information and semantic attributes of the prior knowledge base, the point cloud data to be registered is initially matched, and the 3D models of the tower with similarity higher than the preset value are obtained as candidate models. S3, Coarse Registration: After matching the main axis of the tower body direction with the Z-axis of the point cloud data to be registered, feature matching under prior constraints is performed on the preset key points in the prior knowledge base, and coarse registration results are output. Candidate models whose errors meet the preset conditions are selected as candidate models for coarse registration. S4, fine registration, includes: S41, assign differentiated weights to preset key points in the prior knowledge base and construct an initial correspondence; S42, by constructing an optimized objective function, an alignment target is built between the point cloud data to be registered and the candidate model for coarse registration; S43, Solve the optimization objective function to obtain the matching parameters between the point cloud data to be registered and each candidate model for coarse registration; S5. Perform a matching score based on the matching parameters to obtain the best matching result between the point cloud data to be registered and the candidate model for coarse registration.

[0029] Preferably, step S11 specifically includes: S111, Perform geometric analysis on the three-dimensional model of the tower (e.g., FBX three-dimensional model) and calculate the geometric element information; the geometric element information includes the geometric center, base center, boundary box dimensions, and principal axis direction; S112, Preset key point extraction, including: extraction of global key points, mesoscale key points and local key points; the global key points include the top and bottom of the tower, the mesoscale key points include the crossarm endpoints, and the local key points include the cover nodes and connection points. S113, Descriptor Calculation: For the extracted global keypoints, mesoscale keypoints, and local keypoints, a point cloud is generated through grid sampling, and then the local descriptor, neighborhood radius, and global descriptor are calculated using Fast Point Feature Histogram (FPFH). S114, Extract semantic attributes: including the number of crossarm layers, number of segments, and the pattern and information of the connection points; S115, Perform statistical analysis on different tower 3D models, extract the mean and variance of geometric quantities, construct the corresponding tolerance range, and construct the prior knowledge base.

[0030] In a specific example, step S11, extracting the prior knowledge base of the tower's 3D model, is a crucial step in constructing the prior knowledge base, guiding the registration of point clouds and the model, and improving matching accuracy and robustness. It specifically includes the following: 1) Geometric element information extraction: Perform geometric analysis on the 3D model of the tower, calculate the geometric center, base center (based on axis-aligned bounding box AABB), bounding box dimensions and PCA principal axis direction, to provide basic information for the spatial position, scale and attitude constraints of the model.

[0031] 2) Multi-scale keypoint extraction: Global key points: including the top and bottom of the tower, used to constrain the overall height range; Mesoscale key points: mainly the endpoints of the crossarms, which can be obtained by horizontal slicing or projection detection of the model; Local key points: Covered nodes and connection points, which can be obtained through automatic detection by curvature threshold or manual annotation.

[0032] The extraction of multi-scale key points ensures that the features of the model at different structural levels can be used for correspondence and constraints.

[0033] 3) Geometric and Topological Descriptor Calculation: For the extracted key points, a point cloud is first generated through mesh sampling. Then, local descriptors are calculated using FPFH. FPFH is calculated for each key point (tower top, tower bottom, crossarm endpoint, node) using an appropriate neighborhood radius (adjusted according to the tower size). Simultaneously, a global descriptor is calculated using voxel histograms to reflect the overall topological structure. By combining local and global descriptors, a multi-level, multi-scale feature representation is formed for feature matching in subsequent registration.

[0034] 4) Topology and Semantic Attribute Extraction: Extract the topology and semantic attributes of the tower, including the number of crossarm layers, segments, and attachment point patterns. This information can be obtained through manual annotation or automatically identified using a deep learning-based segmentation network to enhance the semantic constraints during the registration process. For example, it can be obtained through manual annotation.

[0035] 5) Statistical Template and Index Construction: Perform statistical analysis on similar model towers to extract the mean and variance of geometric quantities such as crossarm spacing and floor height, and construct corresponding tolerance ranges. Store the above geometric element information, key points, descriptors, and semantic attributes uniformly in a priori knowledge base, which can be stored in JSON format for easy retrieval and retrieval.

[0036] Through the above steps, a complete prior knowledge base for the FBX model is formed, providing reliable geometric, topological, and semantic constraints for subsequent multi-scale registration of tower point clouds and models.

[0037] Preferably, step S2 includes: S21, Height Constraint: Calculate the maximum height difference in the principal direction of the point cloud data to be registered. The height of the three-dimensional model of the tower. height difference , ,in, Allowable height deviation (e.g., 5%). , The z-axis coordinate of point i in the point cloud data to be registered; S22, Connection Point Constraint: Calculate the number of wires and ground wires in the point cloud data to be registered using clustering. Based on the number of attachment points in the 3D model of the tower Ensure that the number of conductors and ground wires matches the number of connection points; S23, Crossarm Count Constraint: After projecting the point cloud data to be registered, the number of crossarms in the tower body is counted. The number of layers in the 3D model of the tower same; S24. Select the three-dimensional model of the tower that meets the constraints of steps S21 to S23 above as the candidate model.

[0038] Preferably, step S3 includes: S31, obtain the principal direction Z-axis of the point cloud data to be registered through PCA principal component analysis, and rotate the point cloud data to be registered to be consistent with the principal axis direction of the candidate model; S32, Key point matching: The coordinates of the points in the point cloud data to be registered are calculated by clustering. The corresponding key point annotations are called from the prior knowledge base of the candidate model. By comparing the spatial coordinates, the key points of the point cloud data to be registered are initially matched with the key points of the candidate model, and a point-to-point matching relationship is established. S33, based on the descriptors of the preset key points, a coarse matching is performed using the RANSAC random sampling consensus algorithm to obtain 3D tower models with similarity higher than a preset value as candidate models, and the initial rotation matrix of the point cloud data to be registered relative to the candidate models is recorded. Translation vector .

[0039] Preferably, step S33 includes the coarse matching including prior knowledge constraints: If the distance difference between the hook points is greater than a set threshold, the match is considered invalid. If the difference in crossarm length exceeds the tolerance range, the match is rejected. If the preset number of key points or the hierarchical distribution are inconsistent, incorrect candidate models will be directly excluded.

[0040] Before performing fine registration, coarse registration needs to be completed first to ensure that the point cloud and the tower model point cloud are roughly aligned in overall position, scale, and orientation. The goal of coarse registration is to use prior knowledge to impose preliminary constraints on the point cloud, narrowing the search space and providing a reliable initial solution for subsequent fine registration. In a specific instance, step S3 may include: 1) Input: LAS point cloud: actual tower point cloud data to be registered; Candidate FBX model point cloud and its prior knowledge.

[0041] 2) Initial rotation based on PCA: Principal Component Analysis (PCA) has already determined the principal orientation of the point cloud (generally the tower direction), and the rotation angle of the point cloud around the Z-axis can be calculated. Rotating the point cloud to align with the principal axis of the FBX model eliminates large-scale rotational uncertainties. This step can significantly reduce the computational load during subsequent matching and improve the stability of matching convergence.

[0042] 3) Keypoint matching: Extracting keypoints from the point cloud: Hook point extraction: By performing DBSCAN clustering on the tower point cloud and combining it with distance constraints, the nearest point corresponding to each conductor and ground wire is selected as the hook point.

[0043] Tower vertex extraction: Take the point with the maximum height in the point cloud, and at the same time combine the x and y distances from the tower center to exclude outlier points that deviate from the target.

[0044] Crossarm endpoints: obtained through clustering and edge detection.

[0045] The corresponding key point annotations are called from the FBX model prior library. By comparing spatial coordinates, the key points in the point cloud are initially matched with the key points in the model, and a point-to-point matching relationship is established to provide candidate pairs for subsequent RANSAC coarse registration.

[0046] 4) Feature matching under prior constraints: Local Feature Descriptors (FPFHs) are extracted from the point cloud, and coarse matching is performed using the RANSAC random sampling consensus algorithm. Prior knowledge constraints are introduced during the matching process. If the distance difference between the attachment points is greater than the set threshold, the match is considered invalid. If the difference in crossarm length is too large (exceeding the tolerance range), the match will be rejected. If the number of key points or the hierarchical distribution are inconsistent, incorrect candidate models are directly excluded.

[0047] 5) Coarse registration output results: The point cloud and the FBX model point cloud are initially aligned in terms of overall rotation, scale, and position. Based on prior knowledge, five candidate matching results with the smallest errors are selected as the initial input for fine registration, thereby ensuring that the subsequent optimization process converges faster and the results are more stable.

[0048] Preferably, step S41 includes: S411, assign differentiated weights to the preset key points i in the prior knowledge base. The preset key point i includes high-weight key points and ordinary points; S412, using the initial rotation matrix output from coarse registration. Translation vector The point cloud data to be registered is initially aligned with the point cloud of the candidate model; S413, Key Point Correspondence Construction: S4131: For each high-weight key point in the point cloud data to be registered, search for the most similar candidate point in the point cloud of the candidate model. S4132, Local Descriptor Matching: Calculating High-Weight Keypoints Fast point feature histogram descriptors: select the point with the highest similarity between the descriptor of the high-weight key point and the descriptor of the candidate point as the corresponding point; S4133, for each pair of high-weight key points Calculate the matching confidence score , ;in, Indicates proportional to, This represents a local descriptor function. As a similarity metric, This indicates the key points of the point cloud data to be registered. express The corresponding points in the point cloud of the candidate model; : for point The corresponding index in the FBX candidate model point cloud; S4134 performs downsampling processing on each ordinary point in the point cloud data to be registered, and establishes an auxiliary correspondence.

[0049] Preferably, step S42 includes: S421, each key point in the point cloud data to be registered Calculate point-to-point error : ;in, , Represents the rotation matrix. Let be the set of all rotation matrices; , Represents the translation vector. S represents the three-dimensional translation vector; S is the set of point indices participating in the optimization. S422, Constructing the prior constraint loss: Alignment of the center points of the tower top and bottom: The height difference between the center points of the tower top and bottom of the point cloud data to be registered and the center points of the tower top and bottom of the candidate model is minimized, calculated by the following formula: ;in, This represents the Z-axis coordinate of point k in the point cloud data to be registered. The Z-axis coordinate of point k in the candidate model; Key layer alignment: Ensure that the crossarm endpoints and attachment points are aligned with the model in the height direction, calculated using the following formula: ;in, This represents the Z-axis coordinate of point j in the point cloud data to be registered. Represents the Z-axis coordinate of point j in the candidate model; Symmetry and Proportion Constraints: Maintaining the symmetry of the crossarm and the left and right structures of the tower is expressed by distance constraints as follows: ;in,

[0050] This represents the symmetric distance of point l in the point cloud data to be registered. Represents the symmetric distance of point l in the candidate model; The prior loss is formed by weighted combination: ;in, , as well as This is a coefficient, adjusted according to the tower type; Construct the overall optimization objective function: ,in, is a coefficient.

[0051] Preferably, step S43 includes: using a weighted ICP framework + nonlinear least squares to solve for the optimal rigid body transformation parameters. .

[0052] Initial alignment is performed between the point cloud and the FBX model point cloud in terms of overall rotation, scale, and position. Based on prior knowledge, candidate matches with smaller errors (e.g., the five smallest errors, or results with errors below a certain threshold) are selected as initial input for fine-tuning, ensuring faster convergence and more stable results in subsequent optimization processes. In a specific example, the fine-tuning in step S4 may include: 1) Input: LAS point cloud: actual tower point cloud data to be registered; Coarsely registered candidate FBX model point clouds and their prior information; Coarse registration output: Initial rotation matrix Translation vector .

[0053] 2) Keypoint weighting and matching construction: In the fine registration stage, in order to ensure high-precision alignment between the tower point cloud and the model at key parts of the structure, it is necessary to assign differentiated weights to the key points and establish an initial correspondence.

[0054] (a) Key point weight allocation: Based on the importance of the tower structure, the points in the point cloud are divided into two categories: key points and ordinary points. a) Key points (high weight): including: tower top, crossarm endpoints, and connection points; assigned higher weight. This ensures that these points dominate the estimation of rigid body transformations in the optimization objective; the weights can be dynamically adjusted based on the importance of the points or the confidence level of the descriptor matching, for example: ; b) Ordinary points (low weight): Non-critical points on the tower, assigned lower weights. It is used to provide overall geometric information, but does not dominate the optimization: ; This differentiated weight allocation allows key structural points to dominate the correction of rotation and translation during the optimization process, thereby improving the matching accuracy of the tower top, crossarm, and nodes.

[0055] (b) Based on the correspondence between coarse registration and refined points: a) Initial alignment input: Initial rigid body transformation using coarse registration output. , The LAS point cloud is roughly aligned with the FBX model point cloud. This initial alignment ensures that the point cloud and the model are close in overall rotation, scale, and position, which facilitates subsequent refinement of the correspondence.

[0056] b) Key point mapping: Precise matching is performed on structural key points (tower top, crossarm endpoints, connection points, etc.). For each LAS point cloud key point Search for the most similar candidate points in the model point cloud. ; Local descriptor matching: Calculate FPFH descriptors and select the point with the highest descriptor similarity as the counterpart; Calculate the matching confidence for each pair of key points. Point pairs with higher confidence levels have greater weight in subsequent optimizations. ;in Indicates proportional to, This represents a local descriptor function. This is a similarity metric.

[0057] c) Construction of ordinary point correspondences: Non-critical points are first downsampled to establish auxiliary correspondences to provide overall geometric constraints: nearest neighbor sampling matching is used. Lower weights are assigned to ensure that critical structures are not excessively disturbed during optimization, while also helping to reduce overall shape errors.

[0058] 3) Constructing the optimization objective: The core of fine registration is to achieve high-precision alignment between the LAS point cloud and the candidate FBX model in terms of geometry, topology, and semantics by optimizing the objective function. The optimization objective typically consists of two parts: point-to-point weighted error and prior constraint loss.

[0059] a) Point-to-point weighted error: For each key point in the point cloud... (From LAS point cloud), assuming its corresponding FBX model points are Introducing weighting coefficients This indicates the credibility or importance of each point (critical points such as the top of the tower, the center of the bottom of the tower, the end of the crossarm, and the connection point can be assigned higher weights). The point-to-point error is then... : ;in, , Represents the rotation matrix. It is the set of all rotation matrices used to represent all possible rotations of a rigid body in three-dimensional space; , Represents the translation vector. S represents the three-dimensional translation vector; S is the set of point indices participating in the optimization. : for point The corresponding index in the FBX model point cloud; This ensures that the geometry of the point cloud matches the model as closely as possible, which is the foundation of ICP.

[0060] b) Prior Constraint Loss: In addition to point-to-point errors, precise registration requires incorporating prior knowledge of the tower structure and adding geometric and topological constraints to make the results more consistent with the actual tower shape. Constraints include: (a) Alignment of tower top / bottom center points: Ensure that the height difference between the center points of the LAS point cloud tower top and bottom and the center points of the model tower top and bottom is minimized, calculated using the following formula:

[0061] (b) Key layer alignment: Ensure that the crossarm endpoints and attachment points are aligned with the model in the height direction, calculated using the following formula: ; (c) Symmetry and Proportion Constraints: To maintain the symmetry of the crossarm and the left and right structures of the tower, distance constraints are used as follows: ; These constraints are combined in a weighted manner to form the prior loss: ;in, , as well as This is a coefficient, adjusted according to the tower type; c) Comprehensive Optimization Objective: Combining point-to-point errors with prior constraints, the overall optimization objective function is obtained: The objective function takes into account both the geometric overlap of the point cloud and the prior structural constraints of the tower, and is the core of the fine registration solution.

[0062] 4) Optimization Solution: A weighted ICP framework combined with nonlinear least squares is used. This involves incorporating point-to-point residuals and prior constraint residuals into a single overall optimization objective, balancing them with weights, and then solving for the optimal rigid body transformation parameters using nonlinear least squares. .

[0063] Those skilled in the art can design optimized solution methods as needed, which will not be elaborated upon in this application. For example, the iterative steps may include: updating the correspondence between points. Calculate point-to-point residuals and prior constraint residuals; minimize the objective function and update the rotation. With translation Convergence condition: The error decreases less than the threshold or the maximum number of iterations is reached.

[0064] 5) Output: After fine registration, output the relevant parameters of each model to provide data support for subsequent matching and scoring.

[0065] Preferably, step S5 includes: S51, ICP overlap score : N is the number of points involved in the registration. S52, Descriptor Consistency Score : Among them, Represents cosine similarity, where, A descriptor representing point k in the point cloud data to be registered. The descriptor of point k in the candidate model; K is the number of point pairs participating in the descriptor matching; S53, Consistency Score : ;in, The height difference of the attachment points representing the candidate models. The height difference of the attachment points in the point cloud data to be registered represents the height difference of the attachment points; M is the number of attachment points participating in the structural feature comparison. S54, Construct the comprehensive scoring function: α, β, γ are normalized weighting coefficients that control the importance of the three indicators (α+β+γ=1). S55, select the candidate model with the highest coarse registration score as the best matching result.

[0066] The present invention also provides a multi-scale registration system for pole point clouds and models based on prior constraints. The system is built on a computer system and specifically includes a memory 61, a processor 62, and a computer program 63 stored in the memory 61 and executable on the processor 62. When the processor 62 executes the computer program 63, it implements the steps of the multi-scale registration method for pole point clouds and models based on prior constraints as described above.

[0067] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer.

[0068] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0069] The memory can be an internal storage unit, such as a hard drive or RAM; it can also be an external storage device, such as an external hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, or a Flash Card. Furthermore, the memory may include both internal and external storage units. The memory is used to store the computer program and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or will be output.

[0070] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for multi-scale registration of pole point clouds and models based on prior constraints as described above.

[0071] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0072] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0073] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0074] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / system and method can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0075] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0076] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0077] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0078] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0079] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for multi-scale registration of tower point clouds and models based on prior constraints, characterized in that, Includes the following steps: S1, standardization processing of multi-source data, including: S11, Construction of the prior knowledge base for the 3D model of the pole: Extract the geometric element information, preset key points, descriptors and semantic attributes of the preset key points of different 3D models of poles, and store the point cloud data information corresponding to each 3D model of poles as a prior knowledge base in a preset format. S12, Tower point cloud data preprocessing: The tower point cloud data is preprocessed and global geometric features are extracted to construct point cloud data to be registered; wherein, the point cloud data to be registered is matched with the point cloud representation standard of the prior knowledge base; S2, Preliminary screening: Based on the geometric element information and semantic attributes of the prior knowledge base, the point cloud data to be registered is initially matched, and the 3D models of the tower with similarity higher than the preset value are obtained as candidate models. S3, Coarse Registration: After matching the main axis of the tower body direction with the Z-axis of the point cloud data to be registered, feature matching under prior constraints is performed on the preset key points in the prior knowledge base, and coarse registration results are output. Candidate models whose errors meet the preset conditions are selected as candidate models for coarse registration. S4, fine registration, includes: S41, assign differentiated weights to preset key points in the prior knowledge base and construct an initial correspondence; S42, by constructing an optimized objective function, an alignment target is built between the point cloud data to be registered and the candidate model for coarse registration; S43, Solve the optimization objective function to obtain the matching parameters between the point cloud data to be registered and each candidate model for coarse registration; S5. Perform a matching score based on the matching parameters to obtain the best matching result between the point cloud data to be registered and the candidate model for coarse registration.

2. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 1, characterized in that, Step S11 specifically includes: S111, Perform geometric analysis on the three-dimensional model of the tower and calculate the geometric element information; the geometric element information includes the geometric center, the base center, the boundary box dimensions, and the principal axis direction; S112, Preset key point extraction, including: extraction of global key points, mesoscale key points and local key points; the global key points include the top and bottom of the tower, the mesoscale key points include the crossarm endpoints, and the local key points include the cover nodes and connection points. S113, Descriptor Calculation: For the extracted global key points, mesoscale key points, and local key points, a point cloud is generated through grid sampling, and then the local descriptor, neighborhood radius, and global descriptor are calculated using FPFH. S114, Extract semantic attributes: including the number of crossarm layers, number of segments, and the pattern and information of the connection points; S115, Perform statistical analysis on different tower 3D models, extract the mean and variance of geometric quantities, construct the corresponding tolerance range, and construct the prior knowledge base.

3. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 2, characterized in that, Step S2 includes: S21, Height Constraint: Calculate the maximum height difference in the principal direction of the point cloud data to be registered. The height of the three-dimensional model of the tower. height difference , ,in, To allow for height deviation, , The z-axis coordinate of point i in the point cloud data to be registered; S22, Connection Point Constraint: Calculate the number of wires and ground wires in the point cloud data to be registered using clustering. Based on the number of attachment points in the 3D model of the tower Ensure that the number of conductors and ground wires matches the number of connection points; S23, Crossarm Count Constraint: After projecting the point cloud data to be registered, the number of crossarms in the tower body is counted. The number of layers in the 3D model of the tower same; S24. Select the three-dimensional model of the tower that meets the constraints of steps S21 to S23 above as the candidate model.

4. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 3, characterized in that, Step S3 includes: S31, obtain the principal direction Z-axis of the point cloud data to be registered through PCA principal component analysis, and rotate the point cloud data to be registered to be consistent with the principal axis direction of the candidate model; S32, Key point matching: The coordinates of the points in the point cloud data to be registered are calculated by clustering. The corresponding key point annotations are called from the prior knowledge base of the candidate model. By comparing the spatial coordinates, the key points of the point cloud data to be registered are initially matched with the key points of the candidate model, and a point-to-point matching relationship is established. S33, based on the descriptors of the preset key points, a coarse matching is performed using the RANSAC random sampling consensus algorithm to obtain 3D tower models with similarity higher than a preset value as candidate models, and the initial rotation matrix of the point cloud data to be registered relative to the candidate models is recorded. Translation vector .

5. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 4, characterized in that, Step S33 includes the coarse matching including prior knowledge constraints: If the distance difference between the hook points is greater than a set threshold, the match is considered invalid. If the difference in crossarm length exceeds the tolerance range, the match is rejected. If the preset number of key points or the hierarchical distribution are inconsistent, incorrect candidate models will be directly excluded.

6. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 4, characterized in that, Step S41 includes: S411, assign differentiated weights to the preset key points i in the prior knowledge base. The preset key point i includes high-weight key points and ordinary points; S412, using the initial rotation matrix output from coarse registration. Translation vector The point cloud data to be registered is initially aligned with the point cloud of the candidate model. S413, Key Point Correspondence Construction: S4131: For each high-weight key point in the point cloud data to be registered, search for the most similar candidate point in the point cloud of the candidate model. S4132, Local Descriptor Matching: Calculating High-Weight Keypoints Fast point feature histogram descriptors: select the point with the highest similarity between the descriptor of the high-weight key point and the descriptor of the candidate point as the corresponding point; S4133, for each pair of high-weight key points Calculate the matching confidence score , ;in, Indicates proportional to, This represents a local descriptor function. As a similarity index, This indicates the key points of the point cloud data to be registered. express The corresponding points in the point cloud of the candidate model; S4134 performs downsampling on each ordinary point in the point cloud data to be registered, and establishes an auxiliary correspondence.

7. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 6, characterized in that, Step S42 includes: S421, each key point in the point cloud data to be registered Calculate point-to-point error : ;in, , Represents the rotation matrix. Let be the set of all rotation matrices; , Represents the translation vector. S represents the three-dimensional translation vector; S is the set of point indices participating in the optimization. S422, Constructing the prior constraint loss: Alignment of the center points of the tower top and bottom: The height difference between the center points of the tower top and bottom of the point cloud data to be registered and the center points of the tower top and bottom of the candidate model is minimized, calculated by the following formula: ;in, This represents the Z-axis coordinate of point k in the point cloud data to be registered. The Z-axis coordinate of point k in the candidate model; Key layer alignment: Ensure that the crossarm endpoints and attachment points are aligned with the model in the height direction, calculated using the following formula: ;in, This represents the Z-axis coordinate of point j in the point cloud data to be registered. Represents the Z-axis coordinate of point j in the candidate model; Symmetry and Proportion Constraints: Maintaining the symmetry of the crossarm and the left and right structures of the tower can be represented by distance constraints as follows: ;in, This represents the symmetric distance of point l in the point cloud data to be registered. Represents the symmetric distance of point l in the candidate model; The prior loss is formed by weighted combination: ;in, , as well as This is a coefficient, adjusted according to the tower type; Construct the overall optimization objective function: ,in, is a coefficient.

8. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 7, characterized in that, Step S43 includes: using a weighted ICP framework + nonlinear least squares to solve for the optimal rigid body transformation parameters. .

9. The method for multi-scale registration of tower point clouds and models based on prior constraints according to claim 8, characterized in that, Step S5 includes: S51, ICP overlap score : N is the number of points involved in the registration. S52, Descriptor Consistency Score : Among them, among them Represents cosine similarity, where, A descriptor representing point k in the point cloud data to be registered. The descriptor of point k in the candidate model; K is the number of point pairs participating in the descriptor matching; S53, Consistency Score : ;in, The height difference of the attachment points representing the candidate models. The height difference of the attachment points in the point cloud data to be registered represents the height difference of the attachment points; M is the number of attachment points participating in the structural feature comparison. S54, Construct the comprehensive scoring function: α, β, γ are normalized weighting coefficients, and the importance of the three indicators is controlled by α + β + γ = 1. S55, select the candidate model with the highest coarse registration score as the best matching result.

10. A multi-scale registration system for pole point clouds and models based on prior constraints, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for multi-scale registration of tower point cloud and model based on prior constraints as described in any one of claims 1 to 9.

Images