A Seven-Parameter Transformation Geometric Quality Inspection Method and System for Building Components

Abstract

This application provides a seven-parameter transformation geometric quality inspection method and system for building components, belonging to the field of quality inspection technology. The method includes: constructing a point cloud dataset; performing semantic partitioning of the CAD model using the parsing results; establishing partition identification features of the semantic partitions using the CAD model, and performing global search matching of the point cloud dataset using the partition identification features as matching features, configuring the partition mapping using the global search matching results; performing partition feature matching of the CAD model based on the partition mapping using the point cloud feature set, and establishing initial values ​​for the seven parameters; performing recursive iterative optimization to complete fine registration; calculating local residuals based on the fine registration results, and generating quality inspection results using the local residual calculation results. This application can solve the technical problem of insufficient registration accuracy between point cloud data and CAD models in key functional areas in the prior art, achieving the technical effect of improving the registration accuracy of building component quality inspection.

Description

Technical Field

[0001] This application relates to the field of quality inspection technology, and in particular to a seven-parameter conversion geometric quality inspection method and system for building components. Background Technology

[0002] The requirements for geometric accuracy of building components during processing, installation and acceptance are becoming increasingly stringent. Especially against the backdrop of the rapid development of industrialized prefabricated buildings, the precision control of key parts such as the connection relationship between components, positioning holes and structural interfaces has become a core factor in ensuring construction quality and structural safety.

[0003] Currently, the registration accuracy between existing point clouds and CAD models directly determines the accuracy of subsequent error analysis and quality assessment. However, existing technologies generally lack an understanding of the structural semantics of the CAD model during the registration process, typically employing overall rigid alignment or a global error minimization strategy, ignoring the functional and structural regional differences of the components themselves, resulting in a lack of targeted registration. Especially when the component structure is complex, the boundaries are blurred, or there are repeated local features, point cloud data, without semantic partitioning guidance, is prone to misalignment or mismatch in key areas such as connection boundaries, positioning holes, and tongue and groove joints. This local misalignment not only undermines the rigor of the overall alignment but also causes deviation masking or misjudgment amplification during error statistics, thus distorting the final quality inspection results. For example, if a positioning hole area is not correctly aligned, its local residual value may be masked by the overall average error, causing the component to be misjudged as qualified, thereby affecting assembly accuracy and engineering safety.

[0004] In summary, the existing technology suffers from a lack of a regional partitioning and registration mechanism based on structural semantics, which leads to insufficient registration accuracy between point cloud data and CAD models in key functional areas. This further affects the accuracy of identifying local errors in components and the reliability of overall quality inspection results, easily causing assembly deviations, misjudgments in inspections, and potential engineering safety hazards. Summary of the Invention

[0005] The purpose of this application is to provide a seven-parameter conversion geometric quality inspection method and system for building components, in order to solve the technical problems in the prior art where the lack of a regional partitioning registration mechanism based on structural semantics leads to insufficient registration accuracy between point cloud data and CAD models in key functional areas, which further affects the accuracy of local error identification of components and the reliability of overall quality inspection results, and easily causes assembly deviations, inspection misjudgments and engineering safety hazards.

[0006] In view of the above problems, this application provides a method and system for testing the geometric quality of building components using a seven-parameter conversion method.

[0007] Firstly, this application provides a seven-parameter transformation geometric quality inspection method for building components, implemented through a seven-parameter transformation geometric quality inspection system for building components. The method includes: activating a laser scanner to perform multi-view scanning acquisition of the building components; after data authentication of the multi-view scanned point clouds, constructing a point cloud dataset; parsing the CAD model of the building components; performing semantic partitioning of the CAD model using the parsing results; establishing partition identification features of the semantic partitions using the CAD model; performing global search matching of the point cloud dataset using the partition identification features as matching features; configuring partition mapping using the global search matching results; performing feature extraction of the point cloud dataset to establish a point cloud feature set; performing partition feature matching of the CAD model based on the partition mapping using the point cloud feature set; establishing initial values ​​for the seven parameters; performing coarse registration of the CAD model and the point cloud dataset based on the initial values ​​of the seven parameters; performing recursive iterative optimization to complete fine registration; calculating local residuals based on the fine registration results; and generating quality inspection results using the local residual calculation results.

[0008] Preferably, the seven-parameter transformation geometric quality detection method for building components further includes: extracting local geometric key points for each semantic partition of the point cloud feature set and the CAD model according to the partition mapping; performing adaptive neighborhood search of the key points based on the local geometric key points to establish local descriptors; performing similarity matching of the local descriptors of the point cloud feature set and the CAD model under the same semantic partition; and establishing initial values ​​for the seven parameters after global verification based on the similarity matching results.

[0009] Preferably, the seven-parameter transformation geometric quality inspection method for building components further includes: rigidly aligning the point cloud dataset to the CAD model according to the initial values ​​of the seven parameters; performing surface analysis on the CAD model to establish a uniformly segmented mesh; configuring sparse key points of the CAD model using the uniformly segmented mesh; performing a nearest similarity search on the point cloud dataset based on the sparse key points and the local geometric key points; and performing recursive iterative optimization using the nearest similarity search results to complete the fine registration.

[0010] Preferably, the seven-parameter transformation geometric quality detection method for building components further includes: establishing a partition weight factor for semantic partitions; performing a nearest similarity search for each semantic partition and calculating the partition iteration residual; performing multi-partition fusion based on the partition weight factor and the partition iteration residual, and iteratively updating the global seven parameters; and completing fine registration based on the final global seven parameters after the convergence condition is met.

[0011] Preferably, the seven-parameter conversion geometric quality inspection method for building components further includes: interactively reading the CAD model of the building component, performing model complexity analysis of the CAD model, and establishing a model complexity index; obtaining the target quality inspection accuracy of the building component, using the target quality inspection accuracy and the model complexity index as matching features, performing viewpoint adaptation matching, and establishing viewpoint adaptation matching results; and completing multi-view scanning initialization based on the viewpoint adaptation matching results.

[0012] Preferably, the seven-parameter transformation geometric quality inspection method for building components further includes: searching for the nearest feature point of the point cloud feature set on the surface of the CAD model based on the fine registration result; calculating the shortest distance between the nearest feature point and the surface of the CAD model, and using the shortest distance as the local residual; performing partitioned residual statistics of the local residual using the semantic partition; and generating quality inspection results using the partitioned residual statistics results.

[0013] Preferably, the seven-parameter transformation geometric quality inspection method for building components further includes: establishing a multi-layer residual threshold index, wherein the multi-layer residual threshold index is configured with color mapping; performing threshold index trigger analysis of local residuals based on the multi-layer residual threshold index, and establishing point colors using the trigger analysis results; generating a residual heatmap based on the point colors, and performing visualization display.

[0014] Preferably, the seven-parameter transformation geometric quality inspection method for building components further includes: performing zonal residual trend analysis based on the local residuals; using the results of the zonal residual trend analysis to perform abnormal mutation identification and establish regional quality early warning.

[0015] Preferably, the seven-parameter transformation geometric quality inspection method for building components further includes: configuring normal residuals, point-to-CAD plane residuals of the point cloud dataset, and point density residuals; performing local residual calculations using the normal residuals, point-to-CAD plane residuals of the point cloud dataset, and point density residuals respectively; and generating quality inspection results using the results of the local residual calculations.

[0016] Secondly, this application also provides a seven-parameter transformation geometric quality inspection system for building components, used to execute the seven-parameter transformation geometric quality inspection method for building components as described in the first aspect, comprising: a data authentication module, used to activate a laser scanner to perform multi-view scanning acquisition of building components, and after data authentication of the multi-view scan acquisition point cloud, construct a point cloud dataset; a semantic partitioning module, used to parse the CAD model of the building components, and use the parsing results to perform semantic partitioning of the CAD model; a search matching module, used to establish partition identification features of the semantic partitions using the CAD model, and use the partition identification features as matching features to perform global search matching of the point cloud dataset, and use the global search matching results to configure partition mapping; a feature matching module, used to perform feature extraction of the point cloud dataset, establish a point cloud feature set, use the point cloud feature set to perform partition feature matching of the CAD model based on the partition mapping, and establish initial values ​​for the seven parameters; a recursive iteration module, used to perform coarse registration of the CAD model and the point cloud dataset according to the initial values ​​of the seven parameters, and then perform recursive iteration optimization to complete fine registration; and a residual calculation module, used to calculate local residuals according to the fine registration results, and use the local residual calculation results to generate quality inspection results.

[0017] The technical solution provided in this application has at least the following technical effects or advantages: by achieving the technical goal of constructing a high-precision geometric quality detection method that integrates semantic partition feature matching and multi-dimensional residual evaluation, the technical effect of improving the registration accuracy, error identification accuracy and detection intelligence level of building component quality detection is achieved.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application 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 merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the seven-parameter conversion geometric quality inspection method for building components used in this application.

[0021] Figure 2 This is a structural schematic diagram of the seven-parameter conversion geometric quality inspection system for building components used in this application.

[0022] Figure labeling: Data authentication module 11, semantic partitioning module 12, search matching module 13, feature matching module 14, recursive iteration module 15, residual calculation module 16. Detailed Implementation

[0023] This application provides a seven-parameter transformation geometric quality inspection method and system for building components. It addresses the technical problem in existing technologies where the lack of a structural semantic-based regional partitioning registration mechanism leads to insufficient registration accuracy between point cloud data and CAD models in key functional areas. This, in turn, affects the accuracy of local error identification and the reliability of overall quality inspection results, easily causing assembly deviations, misjudgments, and potential engineering safety hazards. The application aims to construct a high-precision geometric quality inspection method that integrates semantic partitioning feature matching and multi-dimensional residual evaluation, thereby improving the registration accuracy, error identification accuracy, and intelligent inspection level of building components.

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0025] Example 1, please refer to the appendix. Figure 1 This application provides a seven-parameter transformation geometric quality inspection method for building components, applied to a seven-parameter transformation geometric quality inspection system for building components, specifically including the following steps:

[0026] The laser scanner is activated to perform multi-view scanning and acquisition of building components. After data authentication of the point cloud acquired from the multi-view scanning, a point cloud dataset is constructed.

[0027] Specifically, activating the laser scanner means waking up and starting the laser scanning equipment, putting it into working condition and preparing it for 3D information acquisition of building components. The laser scanner performs multi-view scanning of the building components, scanning the target components from multiple different directions and positions to reduce occlusion and data loss areas, ensuring that the 3D geometric information of each surface of the building component is completely acquired. The number of multi-view scans is determined by both the target quality inspection accuracy and the geometric complexity of the building components. The multi-view scanned point clouds undergo data authentication, including removing noise points, identifying erroneous points, checking for duplicate information, and ensuring temporal and spatial alignment to guarantee the accuracy and reliability of the point cloud data. The authenticated multi-view point clouds are then registered, fused, and uniformly formatted to form a complete 3D point cloud data set, constructing a point cloud dataset for subsequent model comparison and quality inspection analysis.

[0028] The CAD models of building components are analyzed, and the results are used to perform semantic partitioning of the CAD models.

[0029] Specifically, the process involves analyzing the CAD models of building components, performing structural interpretation, and extracting geometric information, topological structures, and component attributes. During analysis, each face, edge, and corner of the CAD model, along with their connections, is identified, thus restoring the CAD model to a computable and analyzable data structure. Based on the obtained geometric structure and component attributes, the CAD model is divided into several sub-regions with specific engineering meanings. For example, it can be divided into edge regions, hole regions, connection regions, or stress regions based on the function of the building component, or into planar regions, curved surface regions, or angular regions based on geometric features. This process elevates the CAD model from a geometric representation to a semantic level, ensuring that each partition not only possesses shape information but also engineering or structural significance. For instance, a metal plate model with bolt holes can be divided into a hole region, an edge region, and a planar main body region. Each region will participate in different matching, registration, or quality evaluation strategies, enabling targeted geometric analysis and processing of complex building components, thereby improving the efficiency and accuracy of subsequent algorithms in terms of precision control and feature extraction.

[0030] The CAD model is used to establish semantic partition identification features, and the partition identification features are used as matching features to perform global search matching of the point cloud dataset. The partition mapping is configured using the global search matching results.

[0031] Specifically, feature information representing the geometric structure or engineering attributes of each semantic partition is extracted, including the partition's area, boundary shape, normal direction, curvature, number of holes, or local geometric texture, to distinguish different semantic regions. The partition identifier features are used as matching features to perform a global search and matching on the point cloud dataset. This search traverses the point cloud dataset using the partition identifier features to identify regions in the point cloud that correspond to a specific semantic partition feature in the CAD model, such as through similarity calculations in high-dimensional feature space and spatial location constraints. The global search and matching results are used to configure partition mappings. Based on the correspondence established in the search and matching, the matching regions found in the point cloud dataset are mapped one-to-one with the semantic partitions in the CAD model, thus establishing a mapping structure. This achieves intelligent transformation from geometric point sets to semantic structures, laying the structural foundation for subsequent registration, detection, and residual analysis.

[0032] Perform feature extraction on the point cloud dataset, establish a point cloud feature set, and use the point cloud feature set to perform partition feature matching of the CAD model based on the partition mapping to establish initial values ​​for seven parameters.

[0033] Specifically, structural analysis is performed on the point cloud data after multi-view scanning and integration to identify representative geometric features, such as edge points, corner points, abrupt changes in surface curvature, or regions with significant local normal changes, reflecting the important geometric structure of the point cloud in space. All extracted key geometric features are then organized into a unified dataset for subsequent matching and analysis, establishing a point cloud feature set.

[0034] Based on the semantic partitioning correspondence, the consistency of structural features is sought between each pair of point cloud and CAD model partitions. By comparing indicators such as point position, orientation, and descriptor similarity, it is determined whether the two match geometrically. The initial alignment relationship between the point cloud and the CAD model in 3D space is calculated, including the rotation angles around the three coordinate axes, the translation distances in the three directions, and an overall scaling factor, establishing initial values ​​for seven parameters.

[0035] After coarse registration of the CAD model and point cloud dataset based on the initial values ​​of the seven parameters, recursive iterative optimization is performed to complete fine registration.

[0036] Specifically, after coarse registration of the CAD model and point cloud dataset based on the initial values ​​of seven parameters, the point cloud data is transformed into the model coordinate system using three rotation parameters, three translation parameters, and one scale parameter. The coarse registration stage primarily addresses the overall orientation and positional offset. However, since the initial matching often contains some error, it cannot meet the requirements for high-precision detection. Therefore, further recursive iterative optimization is necessary. Recursive iterative optimization refers to continuously extracting feature point pairs, calculating the registration error, and repeatedly adjusting the seven parameters based on the error, with each round using the current parameters as the basis for a new round of fitting, thus gradually approaching the optimal matching state. After multiple iterations, when the parameter changes tend to stabilize or the error decreases to a set threshold, it can be considered that the point cloud data and the CAD model have achieved high-precision alignment in three-dimensional space.

[0037] The local residuals are calculated based on the registration results, and the quality inspection results are generated using the local residual calculation results.

[0038] Specifically, local residuals are calculated based on the precise registration results. After high-precision registration between the point cloud dataset and the CAD model, the shortest distance from each point cloud feature point to the nearest point on the CAD model surface in 3D space is calculated based on the registered positional relationship. This shortest distance is the local residual. The local residual reflects the degree of local deviation between the point cloud and the model at a specific location. Quality inspection results are generated using the local residual calculation results. All local residuals are summarized and analyzed to determine whether they are within acceptable error ranges, thereby providing a quantitative evaluation of the geometric accuracy of the entire building component or a local area.

[0039] Furthermore, this application also includes: extracting local geometric key points for each semantic partition of the point cloud feature set and the CAD model according to the partition mapping; performing adaptive neighborhood search of the key points based on the local geometric key points to establish local descriptors; performing local descriptor similarity matching of the point cloud feature set and the CAD model under the same semantic partition, and establishing initial values ​​for seven parameters after global verification based on the similarity matching results.

[0040] Specifically, local geometric key points are extracted from each semantic partition of the point cloud feature set and the CAD model based on the partition mapping. Utilizing the semantic region correspondence between the point cloud and the CAD model, representative geometric feature points are identified in each pair of corresponding partitions. Local geometric key points refer to points with significant geometric changes or high information content, such as curvature abrupt change points, corner points, and hole edge points. These points accurately reflect the morphological characteristics of the local structure and serve as important bases for subsequent registration.

[0041] An adaptive neighborhood search is performed on keypoints based on local geometric keypoints. This involves dynamically selecting a set of surrounding points within a certain range around each geometric keypoint as its neighborhood, used to describe the structural features of the geometric keypoint in local space. The adaptive neighborhood search is flexibly adjusted based on factors such as point cloud density, surface curvature, or model scale, selecting smaller neighborhoods in densely detailed regions and expanding the neighborhood range in smooth regions to ensure the accuracy of local feature extraction. Within each neighborhood, quantifiable geometric feature vectors are extracted, such as the projection distribution of points to the principal direction, statistical values ​​of normal angles, or spatial geometric texture patterns. This transforms the original set of points into a numerical feature representation that can be used for matching, establishing a local descriptor.

[0042] Euclidean distance, cosine similarity, or learned distance metrics are used to perform local descriptor similarity matching between point cloud feature sets and CAD models within the same semantic partition, finding the closest feature pairs. Global verification is performed based on the similarity matching results, checking the consistency and spatial rationality of the matching relationships across the entire partition to ensure that the matching results conform to geometric logic and component structure globally, avoiding the accumulation of overall errors due to local mismatches. After local matching and global verification, the initial pose relationship between the point cloud dataset and the CAD model is calculated, including three rotation parameters, three translation parameters, and one scale parameter in 3D space, totaling seven variables. These seven parameters serve as initial values, providing a preliminary alignment basis for subsequent fine registration. Table 1 shows the record of local keypoint matching and the generation of initial values ​​for the seven parameters within the semantic partition.

[0043] Table 1: Record of Local Keypoint Matching and Seven-Parameter Initial Value Generation under Semantic Partitioning

[0044]

[0045] Furthermore, this application also includes: rigidly aligning the point cloud dataset to the CAD model based on the initial values ​​of the seven parameters; performing surface analysis on the CAD model to establish a uniformly segmented mesh; configuring sparse key points of the CAD model using the uniformly segmented mesh; performing a nearest similarity search of the point cloud dataset based on the sparse key points and the local geometric key points; and performing recursive iterative optimization using the nearest similarity search results to complete the fine registration.

[0046] Specifically, the point cloud dataset is rigidly aligned with the CAD model based on seven initial parameters. The point cloud data undergoes coordinate transformation to initially coincide with the coordinate system of the CAD model in 3D space. Rigid alignment maintains the relative structure between points in the point cloud, changing only the overall position and orientation, thus laying the foundation for subsequent fine-grained registration. For example, if there is a rotational error and a displacement difference between the point cloud coordinate system and the model, rigid alignment will initially overlap them.

[0047] Perform surface analysis on the CAD model and divide it into regular, uniform grid regions according to spatial scale or number of patches. Establishing a uniformly divided grid helps to evenly distribute key points across the entire model surface, ensuring spatial balance in subsequent matching.

[0048] By using uniformly segmented grids to configure sparse keypoints in CAD models, one or more feature points are selected at the center of each grid or at representative locations to form a sparse keypoint set for that region. This set is used to guide the matching of point cloud feature points and error calculation, thereby avoiding excessive reliance on dense computing resources.

[0049] Based on sparse keypoints and local geometric keypoints, a nearest-neighbor similarity search is performed on the point cloud dataset. This search searches for local keypoints in the point cloud data that are similar in geometry and spatial location. The nearest-neighbor similarity search considers not only the similarity of geometric descriptors but also the spatial distance and direction between the two points. For example, if the normal direction of a hole edge point on a CAD model is vertically upward, then points with similar spatial orientation and local curvature are searched in the point cloud as candidate matching points.

[0050] Recursive iterative optimization is performed using nearest-neighbor similarity search results. After obtaining a batch of initial matching pairs, the error is calculated, and the seven parameters are updated based on the error results. The updated parameters are then used for matching again and re-optimization. This process is repeated until the matching error is lower than a preset threshold or the parameter changes tend to stabilize, thus completing the recursive iterative optimization. Through multiple rounds of refinement adjustments, the point cloud data and the CAD model achieve a high degree of overlap in three-dimensional space, reaching sub-millimeter registration accuracy.

[0051] Furthermore, this application also includes: establishing a partition weight factor for semantic partitions; performing a nearest similarity search for each semantic partition and calculating the partition iteration residual; performing multi-partition fusion based on the partition weight factor and the partition iteration residual, and iteratively updating the global seven parameters; and completing fine registration based on the final global seven parameters after the convergence condition is met.

[0052] Specifically, different computational weights are assigned to different semantic partitions based on their importance, geometric stability, or feature saliency within the entire building component. These partition weights reflect the priority or credibility of semantic partitions during the matching process. For example, for a component containing large plane areas and multiple connecting holes, higher weights can be assigned to structurally complex and feature-clear connecting hole areas, while lower weights can be assigned to flat but less varied areas, thus guiding the registration process to focus on key regions.

[0053] For each semantic partition, a nearest neighbor similarity search is performed. Within each semantic partition, feature point matching between the point cloud and the CAD model is performed independently, and the geometric error between the matching pairs is calculated, including position difference, normal angle difference, or distance difference. The iterative residual refers to the degree of spatial inconsistency among the matching point pairs under the current seven parameters, reflecting the registration accuracy and helping to capture the error distribution characteristics of different semantic regions during the registration process.

[0054] The iterative residuals calculated from each semantic partition are weighted and integrated according to the partition weight factor to construct a global error function. The seven parameters are then iteratively updated using an optimization algorithm. In each round, the parameters are readjusted based on the current error, with the goal of minimizing the sum of the weighted residuals. Integrating the matching results from multiple partitions avoids the impact of local information bias on the overall registration, thereby enhancing the robustness of the registration results.

[0055] In several iterations, when the update magnitude of the seven parameters is lower than the preset threshold or the overall residual change tends to stabilize, it is determined that the current parameters have reached the optimal state. At this time, the final seven parameters are used to perform rigid transformation to strictly align the point cloud dataset with the CAD model, thereby completing the entire registration process and achieving high-precision spatial consistency.

[0056] Furthermore, this application also includes: interactively reading the CAD model of the building component, performing model complexity analysis of the CAD model, and establishing a model complexity index; obtaining the target quality detection accuracy of the building component, using the target quality detection accuracy and the model complexity index as matching features, performing viewpoint adaptation matching, and establishing viewpoint adaptation matching results; and completing multi-view scanning initialization based on the viewpoint adaptation matching results.

[0057] Specifically, the system interactively retrieves 3D design files of building components from storage media and reads CAD models. These CAD models, or computer-aided design models, are saved in a standard format and contain the geometric shape, dimensional information, and structural features of the components. Interactive reading can involve manually selecting the file path or automatically calling a preset interface to obtain the model for subsequent analysis and processing.

[0058] This process involves performing a model complexity analysis on the loaded CAD model, calculating and evaluating its structural complexity, including geometric dimensions such as the number of boundary faces, surface types, and frequency of detail changes. For example, a window frame model with numerous surface intersections and small-scale details is far more complex than a wall panel model with a single plane. Based on the analysis results, a model complexity index is established to quantify the data processing difficulties and matching challenges that the CAD model may present during the scanning process.

[0059] The system acquires the target quality inspection accuracy of building components, receiving user-defined or standard-required quality inspection accuracy parameters to determine the point density and matching accuracy of subsequent scans. It combines the target quality inspection accuracy with CAD model complexity indicators as input features for generating a multi-view scanning strategy. Viewpoint adaptation matching is then performed, evaluating whether different viewpoint configurations can simultaneously meet the coverage requirements and target accuracy requirements of geometrically complex areas. This generates the optimal combination of scanning viewpoints as the viewpoint adaptation matching result, including parameters such as the position, direction, and scanning distance of each viewpoint. This ensures that all key geometric features of the model are covered as much as possible during the scanning process while meeting accuracy requirements.

[0060] The multi-view scanning initialization is completed based on the viewpoint adaptation and matching results. The working parameters of the laser scanner are controlled, such as automatically adjusting the rotation platform angle, setting the scanning trajectory, or switching the sensor resolution, to ensure that each scan is performed based on the optimal viewpoint, thereby obtaining high-quality point cloud data and achieving efficient and comprehensive 3D information acquisition.

[0061] Furthermore, this application also includes: searching for the nearest feature points of the point cloud feature set on the surface of the CAD model based on the fine registration results; calculating the shortest distance between the nearest feature points and the surface of the CAD model, and using the shortest distance as the local residual; performing partitioned residual statistics of the local residual using the semantic partition; and generating quality inspection results using the partitioned residual statistics results.

[0062] Specifically, based on the fine registration results, the nearest feature point in the point cloud feature set is searched on the surface of the CAD model. That is, with the CAD model surface as a reference, every position on the CAD model is traversed, and the point closest to that position in the point cloud feature set is found. The nearest feature point is the point in three-dimensional space that is closest to the model surface and has characteristic attributes, such as a point with a clear normal direction or curvature.

[0063] The Euclidean distance method is used to calculate the shortest distance between the nearest feature point and the surface of the CAD model. This shortest distance is taken as the local residual, representing the specific numerical value of the alignment accuracy of the point cloud data in that region. The smaller the local residual, the more accurate the registration; conversely, a larger residual indicates the presence of deformation, error, or scanning anomalies in that region.

[0064] Semantic partitioning is used to perform partitioned residual statistics on local residuals. All calculated local residuals are divided according to the previously established semantic partitions, and statistical analysis is performed within each partition. The statistics include indicators such as the mean, maximum, and standard deviation of the residuals, reflecting the distribution of registration errors within each semantic region. Quality inspection results are generated using the partitioned residual statistics. Based on the residual statistical values ​​of each partition, it is determined whether the quality standards are met, thus outputting the geometric accuracy qualification status of the component in each region. Quality inspection results can be presented in the form of charts, color maps, or numerical reports to help users intuitively understand the inspection status of each region and identify potential manufacturing or assembly deviations.

[0065] Furthermore, this application also includes: establishing a multi-layer residual threshold index, wherein the multi-layer residual threshold index is configured with color mapping; performing threshold index trigger analysis of local residuals based on the multi-layer residual threshold index, and establishing point colors using the trigger analysis results; generating a residual heatmap based on the point colors, and performing visualization display.

[0066] Specifically, multiple level intervals are set according to the magnitude of the error to establish a multi-level residual threshold index, which serves as the residual threshold hierarchy. The multi-level residual threshold index is configured with color mapping, assigning a color to each threshold level to visualize the error magnitude at different levels.

[0067] The local residuals of each point, calculated in practice, are compared with multi-level residual thresholds to determine their respective error level range. Threshold-triggered analysis is a classification mechanism; when the residual value of a point reaches a certain threshold level, the corresponding classification label is triggered, ensuring that each point is accurately classified according to its error magnitude. Point colors are established using the trigger analysis results. Based on the triggered error level label, each point in the point cloud is automatically assigned a corresponding color identifier, allowing the geometric accuracy information of each point to be directly presented through color.

[0068] After summarizing the color information of all points, a color error distribution map is generated in 3D space. The residual heatmap visually reflects the magnitude of error on the component surface in different areas through color changes, forming a thermal distribution that gradually transitions from cool to warm colors. Visualizing this heatmap involves displaying it in a user interface or 3D software, allowing for the intuitive identification of abnormal areas.

[0069] Furthermore, this application also includes: performing partitioned residual trend analysis based on the local residuals; using the results of the partitioned residual trend analysis to perform abnormal mutation identification and establish a regional quality early warning system.

[0070] Specifically, based on the local residuals between the point cloud and the CAD model, temporal or spatial trend analysis is performed on the error data of each region according to semantic partitioning. Local residuals refer to the shortest distance between a single point and its corresponding position on the model surface, reflecting the magnitude of the registration error. Partition residual trend analysis examines whether the local residuals exhibit characteristics such as continuous increase, drastic fluctuations, or local clustering within a specific region.

[0071] Anomaly identification is performed using the results of partitioned residual trend analysis. Based on the abnormal error changes found in the trend analysis, it is further determined whether there are error abrupt change points, i.e., areas where the residual value changes drastically over a short distance or within a short period of time. Anomalies indicate that the area may have quality problems such as structural defects, scan occlusion, or component misalignment. After identifying the abnormal area, an early warning message is issued, marking the area as a quality risk area requiring key attention. This can be presented through prompts, color markings, or automatic reports, establishing a regional quality early warning system.

[0072] Furthermore, this application also includes: configuring normal residuals, point-to-CAD plane residuals of the point cloud dataset, and point density residuals; performing local residual calculations using the normal residuals, point-to-CAD plane residuals of the point cloud dataset, and point density residuals respectively, and generating quality inspection results using the results of the local residual calculations.

[0073] Specifically, this involves configuring normal residuals, point-to-CAD plane residuals from the point cloud dataset, and point density residuals, introducing multiple evaluation metrics to more comprehensively assess the matching quality between point cloud data and CAD models. Normal residuals refer to the angle difference between the normal direction of a point in the point cloud and the normal direction of its corresponding CAD model surface, used to measure the degree of fit between the point cloud surface orientation and the model. Point-to-CAD plane residuals refer to the vertical distance between a point in the point cloud and its corresponding theoretical plane in the CAD model, reflecting whether the point deviates from its ideal planar position. Point density residuals refer to the difference between the number of points per unit area or volume in the point cloud and the expected point density of the theoretical model, reflecting whether the point cloud coverage is uniform or whether there are missing areas.

[0074] Local residual calculations are performed using the normal residual, the point-to-CAD plane residual from the point cloud dataset, and the point density residual, respectively. These calculations determine the error values ​​for the point in terms of normal angle, geometric position, and planar distribution. This helps reveal problems that traditional geometric error detection methods cannot detect. For example, although the geometric position deviation of a point may be small, a large deviation in the normal direction could affect assembly accuracy; or, although the average residual of a planar area may be acceptable, a low point density indicates incomplete scanning or data defects. The results of the local residual calculations are used to generate quality inspection results. Based on preset quality standards, the system determines whether the area meets the accuracy requirements, thus outputting a pass / fail inspection conclusion.

[0075] In summary, the seven-parameter transformation geometric quality inspection method for building components provided in this application has the following technical effects: by achieving the technical goal of constructing a high-precision geometric quality inspection method that integrates semantic partition feature matching and multi-dimensional residual evaluation, it achieves the technical effect of improving the registration accuracy, error identification accuracy and inspection intelligence level of building component quality inspection.

[0076] Example 2: Based on the same inventive concept as the seven-parameter transformation geometric quality inspection method for building components in the foregoing embodiments, this application also provides a seven-parameter transformation geometric quality inspection system for building components. Please refer to the appendix. Figure 2 The system includes: a data authentication module 11, used to activate the laser scanner to perform multi-view scanning acquisition of building components, and after authenticating the point cloud acquired from the multi-view scanning, construct a point cloud dataset; a semantic partitioning module 12, used to parse the CAD model of the building components and perform semantic partitioning of the CAD model using the parsing results; a search matching module 13, used to establish partition identification features of the semantic partitions using the CAD model, and perform global search matching of the point cloud dataset using the partition identification features as matching features, and configure partition mapping using the global search matching results; a feature matching module 14, used to perform feature extraction of the point cloud dataset, establish a point cloud feature set, and perform partition feature matching of the CAD model based on the partition mapping using the point cloud feature set, and establish initial values ​​for seven parameters; a recursive iteration module 15, used to perform coarse registration of the CAD model and the point cloud dataset according to the initial values ​​of the seven parameters, and then perform recursive iteration optimization to complete fine registration; and a residual calculation module 16, used to calculate local residuals according to the fine registration results, and generate quality inspection results using the local residual calculation results.

[0077] Furthermore, the seven-parameter transformation geometric quality inspection system for building components is also used for: extracting local geometric key points for each semantic partition of the point cloud feature set and the CAD model according to the partition mapping; performing adaptive neighborhood search of key points based on the local geometric key points to establish local descriptors; performing similarity matching of local descriptors of the point cloud feature set and the CAD model under the same semantic partition, and establishing initial values ​​for the seven parameters after global verification based on the similarity matching results.

[0078] Furthermore, the seven-parameter transformation geometric quality inspection system for building components is also used for: rigidly aligning the point cloud dataset to the CAD model based on the initial values ​​of the seven parameters; performing surface analysis on the CAD model to establish a uniformly segmented mesh; configuring sparse key points of the CAD model using the uniformly segmented mesh; performing a nearest similarity search on the point cloud dataset based on the sparse key points and the local geometric key points; and performing recursive iterative optimization using the nearest similarity search results to complete the fine registration.

[0079] Furthermore, the seven-parameter transformation geometric quality inspection system for building components is also used for: establishing a partition weight factor for semantic partitions; performing a nearest similarity search for each semantic partition and calculating the partition iteration residual; performing multi-partition fusion based on the partition weight factor and the partition iteration residual, and iteratively updating the global seven parameters; and completing fine registration based on the final global seven parameters after the convergence condition is met.

[0080] Furthermore, the seven-parameter conversion geometric quality inspection system for building components is also used for: interactively reading the CAD model of the building component, performing model complexity analysis of the CAD model, and establishing a model complexity index; obtaining the target quality inspection accuracy of the building component, using the target quality inspection accuracy and the model complexity index as matching features, performing viewpoint adaptation matching, and establishing viewpoint adaptation matching results; and completing multi-view scanning initialization based on the viewpoint adaptation matching results.

[0081] Furthermore, the seven-parameter transformation geometric quality inspection system for building components is also used for: searching for the nearest feature point of the point cloud feature set on the surface of the CAD model based on the fine registration result; calculating the shortest distance between the nearest feature point and the surface of the CAD model, and using the shortest distance as the local residual; performing partitioned residual statistics of the local residual using the semantic partition; and generating quality inspection results using the partitioned residual statistics results.

[0082] Furthermore, the seven-parameter conversion geometric quality inspection system for building components is also used for: establishing multi-layer residual threshold indicators, wherein the multi-layer residual threshold indicators are configured with color mapping; performing threshold indicator trigger analysis of local residuals based on the multi-layer residual threshold indicators, and establishing point colors using the trigger analysis results; generating residual heatmaps based on the point colors, and performing visualization display.

[0083] Furthermore, the seven-parameter transformation geometric quality inspection system for building components is also used to: perform zonal residual trend analysis based on the local residuals; and use the results of the zonal residual trend analysis to perform abnormal mutation identification and establish regional quality early warning.

[0084] Furthermore, the seven-parameter transformation geometric quality inspection system for building components is also used to: configure the normal residual, the point-to-CAD plane residual of the point cloud dataset, and the point density residual; perform local residual calculations using the normal residual, the point-to-CAD plane residual of the point cloud dataset, and the point density residual respectively; and generate quality inspection results using the results of the local residual calculations.

[0085] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The seven-parameter conversion geometric quality inspection method and specific examples for building components in the foregoing Embodiment 1 are also applicable to the seven-parameter conversion geometric quality inspection system for building components in this embodiment. Through the foregoing detailed description of the seven-parameter conversion geometric quality inspection method for building components, those skilled in the art can clearly understand the seven-parameter conversion geometric quality inspection system for building components in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. 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 this application. Therefore, this application 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 disclosed herein.

[0087] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A seven-parameter conversion geometry quality detection method for building components, characterized in that, The method comprises: activating a laser scanner to perform multi-view scanning acquisition of the building component, after data authentication of the multi-view scanning acquisition point cloud, constructing a point cloud dataset; parsing a CAD model of the building component, and performing semantic partitioning of the CAD model using the parsing result; establishing partition identification features of the semantic partitioning using the CAD model, and performing global search matching of the point cloud dataset using the partition identification features as matching features, and configuring partition mapping using the global search matching result; performing feature extraction of the point cloud dataset, establishing a point cloud feature set, performing partition feature matching of the CAD model based on the partition mapping using the point cloud feature set, and establishing seven-parameter initial values; after coarse registration of the CAD model and the point cloud dataset according to the seven-parameter initial values, performing recursive iteration optimization to complete fine registration; calculating local residuals according to the fine registration result, and generating quality detection results using the local residual calculation result; performing partition feature matching of the CAD model based on the partition mapping using the point cloud feature set, and establishing seven-parameter initial values, comprising: extracting local geometric key points from the point cloud feature set and each semantic partition of the CAD model according to the partition mapping; performing adaptive neighborhood search of the key points based on the local geometric key points, and establishing local descriptors; performing local descriptor similarity matching of the point cloud feature set and the CAD model under the same semantic partition, and establishing seven-parameter initial values after global verification according to the similarity matching result.

2. The seven-parameter transformed geometry quality detection method for building components according to claim 1, characterized in that, after coarse registration of the CAD model and the point cloud dataset according to the seven-parameter initial values, performing recursive iteration optimization to complete fine registration, comprising: rigidly aligning the CAD model according to the seven-parameter initial values; performing surface analysis on the CAD model, and establishing a uniform segmentation grid; configuring sparse key points of the CAD model using the uniform segmentation grid; performing near similarity search of the point cloud dataset according to the sparse key points and the local geometric key points; performing recursive iteration optimization using the near similarity search result to complete fine registration.

3. The seven-parameter transformed geometry quality detection method for building components according to claim 2, characterized in that, after coarse registration of the CAD model and the point cloud dataset according to the seven-parameter initial values, performing recursive iteration optimization to complete fine registration, comprising: establishing partition weight factors of the semantic partitioning; performing near similarity search on each semantic partition respectively, and calculating partition iteration residuals; performing multi-partition fusion based on the partition weight factors and the partition iteration residuals, and iteratively updating global seven parameters; when the convergence condition is met, completing fine registration according to the final global seven parameters.

4. The seven-parameter transformed geometry quality detection method for building components according to claim 1, characterized in that, before activating the laser scanner to perform multi-view scanning acquisition of the building component, comprising: interactively reading a CAD model of the building component, performing model complexity analysis on the CAD model, and establishing a model complexity index; obtaining a target quality detection accuracy of the building component, performing view angle adaptation matching using the target quality detection accuracy and the model complexity index as matching features, and establishing a view angle adaptation matching result; completing multi-view scanning initialization according to the view angle adaptation matching result.

5. The seven-parameter transformed geometry quality detection method for building components according to claim 1, characterized in that, after coarse registration of the CAD model and the point cloud dataset according to the seven-parameter initial values, performing recursive iteration optimization to complete fine registration, comprising: According to the fine registration result, searching for the nearest feature point of the point cloud feature set on the surface of the CAD model; Calculating the shortest distance between the nearest feature point and the surface of the CAD model, and taking the shortest distance as the local residual; Using the semantic partition to perform partition residual statistics of the local residual; Using the partition residual statistics result to generate the quality detection result.

6. The seven-parameter transformed geometry quality detection method for building components according to claim 5, characterized in that, The use of partition residual statistics result to generate quality detection result, including: Establishing a multi-layer residual threshold index, and the multi-layer residual threshold index is configured with a color mapping; According to the multi-layer residual threshold index, performing threshold index triggering analysis of the local residual, and establishing a point color using the triggering analysis result; Based on the point color, generating a residual heat map and performing visual display.

7. The seven-parameter transformed geometry quality detection method for building components according to claim 5, characterized in that, The use of the semantic partition to perform partition residual statistics of the local residual, including: According to the local residual, performing partition residual trend analysis; Using the partition residual trend analysis result to perform abnormal mutation identification and establishing regional quality early warning.

8. The seven-parameter transformed geometry quality detection method for building components according to claim 1, characterized in that, The use of the semantic partition to perform partition residual statistics of the local residual, including: According to the local residual, performing partition residual trend analysis; Using the partition residual trend analysis result to perform abnormal mutation identification and establishing regional quality early warning.

9. A seven-parameter conversion geometry quality detection system for building components, characterized by, The use of the semantic partition to perform partition residual statistics of the local residual, including: Configuring the normal residual, the point-to-CAD plane residual of the point cloud data set, and the point density residual; Using the normal residual, the point-to-CAD plane residual of the point cloud data set, and the point density residual to perform local residual calculation respectively, and using the local residual calculation result to generate the quality detection result. The steps for implementing the seven-parameter conversion geometry quality detection method for building components according to any one of claims 1-8, comprising: A data authentication module for activating a laser scanner to perform multi-view scanning and collection of building components, and constructing a point cloud data set after data authentication of the multi-view scanning and collection point cloud; A semantic partition module for analyzing a CAD model of the building component and performing semantic partition of the CAD model using the analysis result; A search matching module for establishing partition identification features of the semantic partition using the CAD model, and performing global search matching of the point cloud data set using the partition identification features as matching features, and configuring partition mapping using the global search matching result; A feature matching module for performing feature extraction of the point cloud data set, establishing a point cloud feature set, performing partition feature matching of the CAD model based on the partition mapping using the point cloud feature set, and establishing seven-parameter initial values; A recursive iteration module for performing recursive iteration optimization after coarse registration of the CAD model and the point cloud data set according to the seven-parameter initial values, and completing fine registration; A residual calculation module for calculating local residual according to the fine registration result, and generating a quality detection result using the local residual calculation result.

Images