Vision-based constructional engineering quality detection and evaluation system

Abstract

The invention relates to the technical field of constructional engineering quality detection, and particularly discloses a visual sense-based constructional engineering quality detection and evaluation system, which comprises the following steps of: constructing a time-space fusion three-dimensional point cloud containing color and time information by fusing multiple time-space point cloud data; analyzing a local geometric topology and a global context relationship of the point cloud, and intelligently segmenting the point cloud into component point cloud clusters of different engineering semantics; analyzing the connection state and the assembly integrity of the permanent member point cloud cluster, and deducing the current specific construction stage state according to the process logic; then, dynamically calling matched staged three-dimensional design specification point cloud data from a design library organized according to progress according to the state of the stage; carrying out weight self-adaptive registration comparison on the to-be-accepted component point cloud and the staged standard point cloud, and generating a Boolean type conformity matrix based on a staged tolerance threshold value; according to the method, the problem of automatic detection misjudgment caused by construction dynamic change, design change and standard mismatching is solved.

Description

Technical Field

[0001] This invention relates to the field of building engineering quality inspection technology, and specifically to a vision-based building engineering quality inspection and evaluation system. Background Technology

[0002] In modern construction project management, achieving refined and automated testing and evaluation of construction quality is a core requirement for improving project quality, ensuring construction safety, and controlling project costs and schedules. Traditional quality inspection methods mainly rely on quality inspectors using tools such as straightedges and feeler gauges for on-site manual sampling. This approach suffers from problems such as low efficiency, limited coverage, strong subjectivity, and poor record traceability, making it difficult to meet the needs of large and complex projects for full-sample, high-frequency testing.

[0003] The technical problem solved by this invention is: how to overcome the inherent "semantic gap" between the dynamic physical state of the construction site and the static design model, that is, to avoid the problem that the automated detection system misjudges the state that conforms to the latest construction reality as a quality defect due to the failure to update design changes in a timely manner, the mixing of temporary structures and permanent components, and the different acceptance standards at different construction stages, thereby generating a large number of invalid alarms and seriously interfering with the true quality assessment. Summary of the Invention

[0004] The purpose of this invention is to provide a vision-based building engineering quality inspection and evaluation system to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: A vision-based building engineering quality inspection and evaluation system includes: The multi-source spatiotemporal point cloud fusion module is used to collect three-dimensional point cloud data streams from different spatiotemporal locations in the construction scene and construct a spatiotemporally fused three-dimensional point cloud that integrates color intensity information and collection timestamps. The engineering semantic perception and segmentation module, based on the local geometric topology and global context relationship of points in the spatiotemporal fusion 3D point cloud, segments the point cloud into component point cloud clusters with different engineering semantics through semantic understanding. The engineering semantics include permanent components to be accepted, temporary construction components, and unconstructed areas. The construction phase status inference module analyzes the connection status and assembly integrity between components represented by the component point cloud clusters corresponding to the permanent components to be accepted, and deduces the specific construction phase status of the construction scenario at the current time point based on the predefined construction progress logic rules. The phased design specification dynamic retrieval module dynamically retrieves phased 3D design specification point cloud data that matches the construction objectives of the specific construction stage from the 3D design specification library organized according to the construction progress sequence, based on the specific construction stage status. The phase adaptability quality comparison and evaluation module performs phase adaptability registration between the point cloud cluster of permanent components to be accepted and the point cloud data of the phased 3D design specifications, and quantifies the differences based on the acceptance tolerance threshold applicable to the current construction phase, outputting the phase-limited quality compliance judgment result.

[0006] As a further aspect of the present invention: the construction process of the spatiotemporal fusion three-dimensional point cloud is as follows: Extract a set of feature points with stable geometric features from the 3D point cloud data stream at different spatiotemporal locations in the construction scene; Based on the spatiotemporal distribution of feature point sets, rigid transformation parameters between point clouds with adjacent timestamps are calculated to achieve spatiotemporal consistency registration of point cloud sequences. In the registered point cloud sequence, for the same three-dimensional spatial location, data points collected at different timestamps are compared, and the data points with the latest color intensity information and the highest geometric consistency are retained, while the rest of the redundant data points are filtered out. The retained data points are organized according to the collection timestamp and fused with color intensity information to construct a spatiotemporal fused 3D point cloud.

[0007] As a further aspect of the present invention: the output process of the component point cloud cluster is as follows: Based on the 3D coordinates and color intensity information of each point in the spatiotemporal fusion 3D point cloud, the normal vector distribution and curvature change of the points in the local neighborhood of the 3D coordinates and color intensity information of each point in the spatiotemporal fusion 3D point cloud are calculated to generate a local geometric topological descriptor representing the micro-geometric morphology. Based on local geometric topological descriptors, points that are spatially adjacent and have similar descriptors are aggregated to initially form geometric structural primitives. Combined with color intensity information, candidate component regions with continuous surfaces in the spatiotemporal fusion 3D point cloud are identified. The spatial connection relationships and relative positional constraints between the candidate component regions are analyzed, and global contextual reasoning is performed based on a predefined engineering semantic rule base, which defines the existence logic and interrelationships of different engineering semantic components in space. Based on the results of global context reasoning, the candidate component regions are ultimately determined as permanent components awaiting acceptance, temporary construction components, or unconstructed areas, and the corresponding component point cloud clusters are output.

[0008] As a further aspect of the present invention: the generation of local geometric topological descriptors characterizing microscopic geometric morphology specifically includes: Taking each point to be calculated in the spatiotemporal fusion 3D point cloud as the center, find and determine all neighboring points within a preset radius spatial range to form a local spatial neighborhood of the neighboring points. Based on the three-dimensional coordinates of all neighboring points in the local spatial neighborhood, the covariance matrix of the local spatial neighborhood is constructed, and the covariance matrix is ​​decomposed into eigenvalues ​​to obtain three eigenvalues ​​and their corresponding eigenvectors arranged in descending order. The ratio of the minimum eigenvalue to the eigenvalue adjacent to the minimum eigenvalue, as well as the difference relationship between the three eigenvalues, are combined to calculate the curvature metric value that characterizes the degree of change in local surface curvature for the point to be calculated. The direction vector corresponding to the smallest eigenvalue in the eigenvector is determined as the normal vector of the corresponding point to be calculated, and the curvature metric value is combined with the direction encoding of the normal vector to form the local geometric topological descriptor of the point to be calculated.

[0009] As a further aspect of the present invention: the specific construction stage states specifically include: By using the three-dimensional geometric shape and spatial distribution characteristics of the point cloud clusters corresponding to the permanent components to be inspected, the feature regions that characterize the physical connection relationship between the components are identified and extracted. The feature regions include at least the point cloud of bolt holes and the point cloud of weld surfaces. Based on the point cloud density of the feature region, the regularity of the three-dimensional contour, and the gap distance with the point cloud clusters of adjacent components, the tightness and continuity of the connection state between components are quantitatively evaluated, and a quantitative index of the connection state is generated. Based on the spatial occupancy, mutual combination relationship and connection status quantification indicators of all permanent component point cloud clusters to be accepted, and in accordance with the predefined construction progress logic rules that include process dependencies, the final process node that has been met in the current construction scenario is determined through state matching reasoning, and the final process node is marked as the specific construction stage state.

[0010] As a further aspect of the present invention: the generation of connection state quantification indicators specifically includes: The number of three-dimensional points contained in a unit volume within the statistical feature region is used to obtain the point cloud density of the feature region. The ratio of the point cloud density to the expected density value under the preset standard tightness is calculated as the first sub-index reflecting the filling density. Surface reconstruction is performed on the 3D contour point cloud of the feature region. The average distance and maximum distance between the actual surface of the feature region and the preset standard geometric shape surface are calculated. The combination of the average distance and the maximum distance is used as the second sub-index reflecting the regularity of the morphology. The minimum distances between the corresponding positions of the measurement feature area and the point cloud clusters of adjacent components are measured. The average and variance of these minimum distances are calculated. If the average value meets the design requirements and the variance is lower than the preset threshold, the gap is judged to be uniform, and a third sub-index reflecting the assembly alignment is generated. The first, second, and third sub-indicators are weighted and merged to generate a comprehensive connectivity state quantification index.

[0011] As a further aspect of the present invention: the dynamic invocation of phased three-dimensional design specification point cloud data that conforms to the construction objectives of a specific construction phase specifically includes: Establish a semantic mapping relationship between specific construction stage status and multiple pre-stored specification entries in the design 3D specification library. Each specification entry is associated with a stage-specific 3D design specification point cloud data uniquely identified by the construction node, component scope, and design version. Based on the process dependencies implied by the specific construction stage, the set of identifiers of all target components that must be completed in the current stage and are subject to acceptance is parsed out. Based on the identifier set of the target component, all specification entries containing the identifier set are traversed and matched in the design 3D specification library. By comparing the semantic similarity between the specific construction stage status and the construction nodes of each entry, the unique target specification entry with the highest matching degree is selected. Retrieve the 3D point cloud data with design versions associated with the target specification entry from the 3D design specification library, and output it as the phased 3D design specification point cloud data.

[0012] As a further aspect of the present invention: the parsing process of the identifier set is as follows: Based on the specific construction stage status, locate the corresponding current process node in the predefined process dependency table; Extract all direct predecessor process nodes of the current process node from the process dependency table to form a predecessor process set, and query the construction record database to obtain the completion status marker of each process in the predecessor process set; Based on the completion status flag, verify whether all direct preceding processes have been completed; if they have been completed, extract the unique identifiers of all components associated with the current process node, which are only allowed to be accepted after the current process is completed, according to the component-process association mapping table. The unique identifiers of all extracted components are aggregated to generate a set of identifiers for all target components.

[0013] As a further aspect of the present invention: the quality compliance determination result defined in the output stage specifically includes: Based on the design coordinates of components in the point cloud data of the phased 3D design specifications, the point cloud clusters of permanent components to be accepted are initially spatially aligned to obtain the initial registration results. The positional deviation distribution between corresponding component point clouds in the initial registration result is analyzed. The constraint weights in different spatial directions are dynamically adjusted according to the positional deviation distribution in the subsequent registration process. Iterative optimization registration is performed to obtain the stage-adaptive registration result. From the stage adaptation registration results, the three-dimensional coordinate deviation between the actual position and the design position of each component is extracted. At the same time, according to the specific construction stage status and component type, the corresponding acceptance tolerance threshold is obtained from the preset multidimensional tolerance table. The three-dimensional coordinate deviation of each component is compared with the corresponding acceptance tolerance threshold to generate a conformity matrix composed of Boolean values, which serves as the output of the stage-limited quality conformity judgment result.

[0014] As a further aspect of the present invention: the generation of the conformity matrix composed of Boolean values ​​specifically includes: For each component to be accepted in the phase adaptation registration results, calculate the deviation vector between the three-dimensional coordinates of all point clouds of the component to be accepted and the three-dimensional coordinates of the corresponding design position in the phase three-dimensional design specification point cloud data. Based on the specific construction stage and the engineering type of the components to be inspected, the multidimensional tolerance table is queried to obtain the independent tolerance thresholds and overall comprehensive tolerance thresholds of the corresponding components in three spatial directions. The absolute value of the components of the deviation vector in each spatial direction is compared with the corresponding independent tolerance threshold, and the magnitude of the deviation vector is calculated and compared with the overall comprehensive tolerance threshold. The corresponding component is assigned a value of true if and only if the component comparisons in all directions and the overall comparison both meet the requirements; otherwise, it is assigned a value of false. Finally, the Boolean judgment values ​​of all components are arranged in a preset spatial or logical order to generate a conformity matrix composed of Boolean values.

[0015] The beneficial effects of this invention are: (1) This invention accurately isolates non-acceptance targets through semantic segmentation of "construction status perception"; and automatically infers the current construction stage based on connection status analysis, dynamically calling the strictly matching stage design specifications for comparison. This technical solution ensures the consistency between the comparison logic and the actual construction from two fundamental levels: data source (filtering out interference) and evaluation benchmark (dynamic adaptation), thereby minimizing invalid alarms caused by information asynchrony or contextual misunderstanding, and making the automated detection results highly reliable and practically valuable in engineering.

[0016] (2) This invention constructs a fully closed-loop automated process from data fusion, semantic understanding, state inference to intelligent comparison. It can not only automatically identify defects, but also understand the engineering context of the defects (which construction stage they are in) and make objective quantitative judgments based on staged tolerance standards. This avoids the problems of premature or misaligned acceptance that are common in traditional methods, allowing quality management work to focus on the most critical and compliant acceptance objects at present, saving manpower and time costs, and providing clear, consistent, and traceable data basis for engineering decisions. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, the present invention is a vision-based building engineering quality inspection and evaluation system, comprising: The multi-source spatiotemporal point cloud fusion module is used to collect three-dimensional point cloud data streams from different spatiotemporal locations in the construction scene and construct a spatiotemporally fused three-dimensional point cloud that integrates color intensity information and collection timestamps. The engineering semantic perception and segmentation module, based on the local geometric topology and global context relationship of points in the spatiotemporal fusion 3D point cloud, segments the point cloud into component point cloud clusters with different engineering semantics through semantic understanding. The engineering semantics include permanent components to be accepted, temporary construction components, and unconstructed areas. The construction phase status inference module analyzes the connection status and assembly integrity between components represented by the component point cloud clusters corresponding to the permanent components to be accepted, and deduces the specific construction phase status of the construction scenario at the current time point based on the predefined construction progress logic rules. The phased design specification dynamic retrieval module dynamically retrieves phased 3D design specification point cloud data that matches the construction objectives of the specific construction stage from the 3D design specification library organized according to the construction progress sequence, based on the specific construction stage status. The phase adaptability quality comparison and evaluation module performs phase adaptability registration between the point cloud cluster of permanent components to be accepted and the point cloud data of the phased 3D design specifications, and quantifies the differences based on the acceptance tolerance threshold applicable to the current construction phase, outputting the phase-limited quality compliance judgment result.

[0021] In the multi-source spatiotemporal point cloud fusion module, a set of feature points with stable geometric characteristics is extracted from the 3D point cloud data streams at different spatiotemporal locations in the construction scene. Specifically, for each frame of input point cloud, the curvature value of each point is calculated. The curvature value is calculated by finding the rate of change of the normal vector on the local surface formed by the point's nearest neighbors. Points with curvature values ​​lower than a preset first curvature threshold are initially selected as candidate feature points. Next, within the spatial neighborhood of each candidate feature point, the consistency of the direction vector distribution of the lines connecting the point to all its neighbors is calculated. If the standard deviation of the direction vector distribution is lower than a preset direction consistency threshold, the geometric features of the point are determined to be stable and significant in the local environment, and it is ultimately included in the feature point set.

[0022] Secondly, based on the spatiotemporal distribution of feature point sets, rigid transformation parameters are calculated between point clouds with adjacent timestamps to achieve spatiotemporal consistency registration of the point cloud sequence. Specifically, for two consecutive frames of point clouds acquired in time, matching feature point pairs are found in their respective feature point sets by comparing the local geometric feature descriptors of the points (which consist of the point's curvature, normal vector, and histogram of its neighborhood point distribution). Using at least three non-collinear matching feature point pairs, a spatial transformation is solved using the least squares method. This transformation optimally coincides the coordinates of one set of matching points with the coordinates of another set of matching points through rotation and translation operations. The obtained rotation matrix and translation vector are the rigid transformation parameters. These parameters are applied to uniformly transform the 3D coordinates of all points in the subsequent frame of the point cloud, aligning them spatially with the previous frame. This process is repeated sequentially for all point cloud frames in the sequence.

[0023] Subsequently, in the point cloud sequence after spatiotemporal consistency registration, for the same 3D spatial location, all data points collected at different timestamps are compared to filter out redundant data points. All points whose spatial coordinates lie within the same pre-defined microcube volume are considered to correspond to the same spatial location. Among these points, their color intensity information (RGB values) is compared, and the point with the most recent recording time is retained. If multiple points with the most recent times exist, the deviation distance between each point and the average coordinates of other points in its spatial neighborhood is further calculated. The point with the smallest deviation distance, i.e., the point with the highest geometric consistency, is retained, and the remaining points are filtered out.

[0024] Finally, all the data points retained in the above steps are organized according to the order of their collection timestamps. Each data point contains its three-dimensional spatial coordinates, color intensity information, and its corresponding timestamp. In this way, all valid data points are integrated into a unified data structure, that is, a spatiotemporally fused 3D point cloud is constructed. This point cloud fully records the three-dimensional geometric shape, surface color texture, and information on the evolution of the construction scene over time within a specific period.

[0025] In the engineering semantic perception segmentation module, firstly, the local spatial neighborhood of each point to be calculated in the spatiotemporally fused 3D point cloud is determined. Specifically, a preset radius value is established, determined based on the overall density of the point cloud and the typical feature dimensions of the component to be identified, and is set to 5 to 10 times the average point spacing of the point cloud. In 3D space, the Euclidean distance between the point to be calculated and all other points in the point cloud is calculated. Points with distances less than or equal to the preset radius value are selected to form the local spatial neighborhood point set of the point to be calculated.

[0026] Secondly, based on the three-dimensional coordinates of all neighboring points within the local spatial neighborhood, a covariance matrix for that neighborhood is constructed. The specific process is as follows: First, the arithmetic mean of the three-dimensional coordinates of all neighboring points within the local spatial neighborhood is calculated, resulting in a three-dimensional average coordinate point. Then, for each neighboring point within the neighborhood, the difference vector between its three-dimensional coordinates and the coordinates of the three-dimensional average coordinate point is calculated. Next, for each pair of spatial coordinate dimensions (X, Y, Z), the arithmetic mean of the product of the differences between all neighboring points in those two dimensions is calculated. For example, for the X and Y dimensions, this average is equal to the sum of the differences in X coordinates multiplied by the differences in Y coordinates of all neighboring points, divided by the total number of neighboring points. The calculation results of all dimensions are arranged in a specific order to form a 3x3 symmetric matrix, which is the covariance matrix. Subsequently, eigenvalue decomposition is performed on the covariance matrix. This decomposition process is accomplished by solving an eigenvalue equation, which consists of the covariance matrix, a scalar (eigenvalue) to be determined, and a corresponding non-zero vector (eigenvector). After solving, we obtain three eigenvalues ​​and their corresponding orthogonal three-dimensional unit eigenvectors. The three eigenvalues ​​are then arranged in descending order of their values, and the order of their corresponding eigenvectors is also determined accordingly.

[0027] Subsequently, a curvature metric, characterizing the degree of curvature change at the local surface of the point to be calculated, is calculated. Specifically, the smallest eigenvalue obtained from the decomposition is divided by its adjacent, second-smallest eigenvalue, to obtain a ratio. Simultaneously, the sum of the three eigenvalues ​​and the ratio of the smallest eigenvalue to the sum of the three eigenvalues ​​are calculated. The first ratio and the second ratio are then weighted and summed, with weighting coefficients preset based on experience. The result of this weighted summation is the curvature metric. A smaller value indicates a flatter local surface; a larger value indicates a more drastic change in curvature of the local surface.

[0028] Finally, the local geometric topological descriptor is constructed. The eigenvector corresponding to the smallest eigenvalue obtained from eigenvalue decomposition is determined as the normal vector of the point to be calculated. This normal vector is then oriented: the three components of the normal vector are converted into their corresponding signs (positive or negative) and the proportion of the absolute value of that component in the sum of the absolute values ​​of the three components. The encoded direction information and the calculated curvature metric are combined in a predetermined order to form a multidimensional data vector. This data vector is the local geometric topological descriptor of the point to be calculated, which comprehensively represents the microscopic geometric morphology and orientation characteristics near that point.

[0029] In the construction phase state inference module, firstly, the first sub-index reflecting the filling density is calculated. For the feature region, such as a bolt hole or a section of weld, the volume of its three-dimensional bounding box is calculated. The total number of all three-dimensional points within the feature region is counted. The point cloud density of the region is obtained by dividing the total number of points by the volume of the bounding box. The preset expected density value under the standard fastening state is the average density value obtained in advance through statistical analysis of a large number of accepted samples of the same type of connection components. The calculated actual point cloud density is divided by this expected density value to obtain a ratio, which is the first sub-index. The closer this index is to 1, the denser the filling; the further it deviates from 1, the more likely there are holes, missing parts, or incomplete scanning.

[0030] Secondly, a second sub-index reflecting the regularity of the morphology is calculated. The 3D contour point cloud of the feature region is reconstructed using a Poisson surface reconstruction algorithm to generate a continuous triangular mesh surface as the actual surface model. The preset standard geometric surface refers to an ideal geometric model generated according to the design drawings, such as a standard cylindrical surface (corresponding to a bolt) or a smooth transition surface (corresponding to a weld). A series of points are uniformly sampled on the actual surface model, and the shortest distance from each sampling point to the standard geometric surface is calculated. The arithmetic mean of all these distances is calculated to obtain the average distance; and the maximum value is found to obtain the maximum distance. The average distance and the maximum distance are combined as a two-dimensional vector, and this combination is the second sub-index. The average distance reflects the overall morphological deviation, while the maximum distance reflects the degree of the maximum local defect.

[0031] Subsequently, a third sub-index reflecting assembly alignment is calculated. Reference positions are selected at regular intervals along the contact or mating interface between the feature region and the adjacent component point cloud clusters. At each reference position, the closest point pair is found from the two component point cloud clusters, and the Euclidean distance between the point pair is calculated to obtain a series of minimum distance values. The arithmetic mean of this series of minimum distance values ​​is calculated to evaluate the average assembly clearance; simultaneously, their variance is calculated to evaluate the uniformity of the clearance. The average value meeting design requirements means that the calculated average clearance value falls within the tolerance zone specified in the design drawings (e.g., 0.5 mm to 2.0 mm). The preset threshold is an upper limit of variance set based on engineering experience to determine uniformity, for example, 0.3 square millimeters. When the average value meets design requirements and the variance is lower than the preset threshold, the clearance is determined to be uniform, and a value representing excellent alignment (e.g., 1.0) is set as the third sub-index; otherwise, depending on the degree of deviation of the average value or the excessive variance, it is mapped to a lower value (e.g., between 0.2 and 0.8) according to preset rules.

[0032] Finally, a comprehensive quantitative index of connection status is generated. The first sub-index, the second sub-index (whose two components are combined into a single scalar value using a linear function), and the third sub-index obtained above are weighted and fused. The weights are determined based on the importance of different indices to connection performance. For example, for bolted connections primarily bearing shear force, assembly alignment may have the highest weight; for welds primarily ensuring sealing, morphological regularity may have the highest weight. The weight coefficients are pre-set based on domain expert experience or historical data analysis, and satisfy the condition that the sum of the three is 1. The fusion calculation uses a weighted summation method, multiplying each sub-index by its corresponding weight coefficient, and then adding the three products to obtain the final quantitative index of connection status. This index is a comprehensive scalar value, ranging from 0 to 1, with higher values ​​indicating better connection status.

[0033] In the phased design specification dynamic retrieval module, firstly, a semantic mapping relationship is established between the specific construction phase status and multiple pre-stored specification entries in the design 3D specification library. The design 3D specification library is a structured database, where each record is defined as a specification entry. Each specification entry contains three key identifier fields: construction node, component scope, and design version. The construction node field uses standardized engineering terminology to describe a specific construction milestone, such as "main keel installation completed." The component scope field is a list recording the unique identifiers of all components expected to be installed at that construction node. The design version field records the version number of the design drawings corresponding to that group of components. Furthermore, each specification entry is associated with a phased 3D design specification point cloud data file, which is pre-generated by 3D digital sampling of the corresponding version of the design drawings, representing the ideal spatial location, geometry, and dimensions of all relevant components at that specific construction node. The semantic mapping relationship is implemented through an index table that associates and matches the descriptive text keywords of the specific construction phase status with the construction node fields of each specification entry.

[0034] Secondly, the process involves parsing the set of identifiers for all target components that must be completed and are subject to acceptance in the current stage. Specifically, based on the input description of the specific construction stage, a semantically matching entry is searched in a predefined process dependency table. This table stores the dependencies between various construction processes in a graph structure or adjacency list, thus locating the corresponding current process node. Next, all process nodes that are direct successors to the current process node are retrieved from this process dependency table, forming a set of preceding processes. Then, the construction record database is queried. This database records the actual completion status of each process node in a time series. By retrieving the latest record corresponding to each preceding process node, its completion status marker (usually "completed" or "not completed") is obtained. Then, logical verification is performed based on these completion status markers: if all markers are "completed," it indicates that the current process has met the prerequisites for commencement and acceptance. After successful verification, the component-process association mapping table is queried. This table defines which process is responsible for installing each component. From this table, unique identifiers of all components whose associated processes are exactly equal to the current process node are extracted. Finally, all extracted component unique identifiers are compiled into a list to generate a set of identifiers for all target components.

[0035] Next, based on the identifier set of the target component, a unique target specification entry is matched and filtered. Using the identifier set obtained in the previous step, all specification entries in the design 3D specification library are traversed. The matching logic is: check whether the identifier list recorded in the component scope field of each specification entry completely contains all elements in the identifier set. All specification entries that satisfy the inclusion relationship are initially filtered out. Then, these candidate entries are finely filtered using semantic similarity: the description text of the specific construction stage status is compared with the description text of the construction node field of each candidate entry. The comparison is achieved by calculating the overlap of text keywords, word order similarity, or by calculating cosine similarity using pre-trained semantic vectors. The candidate entry with the highest semantic similarity calculation result is selected as the target specification entry.

[0036] Finally, the staged 3D design specification point cloud data is retrieved and output. Based on the unique identifier of the target specification item determined in the previous step (composed of its construction node, component scope, and design version), a precise query is performed in the 3D design specification library to locate the data file associated with that item. This file is read to obtain the staged 3D design specification point cloud data, which is then output as the benchmark data for subsequent quality comparison. This point cloud data accurately represents the ideal 3D state that all components that should be accepted should present under the current specific construction stage.

[0037] In the phased adaptability quality comparison and evaluation module, firstly, initial spatial alignment is performed to obtain initial registration results. Based on the design coordinates of each component in the phased 3D design specification point cloud data, i.e., its theoretical center point coordinates in the global coordinate system, the point cloud cluster of the permanent components to be accepted is initially located. Specifically, for each component to be accepted, the arithmetic mean of the 3D coordinates of all points in its component point cloud cluster is calculated to obtain the measured center point coordinates of the component. Next, a displacement vector is constructed between the measured center point coordinates of the component and its corresponding theoretical center point coordinates in the design data. The arithmetic mean of the displacement vectors calculated for all components to be accepted is calculated to obtain an overall average displacement vector. This average displacement vector is then subtracted from the 3D coordinates of each point in the point cloud cluster of the permanent components to be accepted, thereby completing the overall translational alignment. The point cloud position state obtained at this point is the initial registration result.

[0038] Secondly, the positional deviation distribution in the initial registration results is analyzed, and iterative optimization registration is performed to obtain the stage-adaptive registration results. Based on the initial registration results, for each component to be accepted, its point cloud is matched with the design point cloud in a more refined manner. The matching point pairs of the component are calculated in three spatial directions (e.g., ...). , , Positional deviation components in the axial direction. Statistical analysis of all components in... The absolute value of the deviation in direction is calculated, and its average value is denoted as . And calculate its standard deviation, denoted as . Similarly, calculate direction and Average deviation in direction , and standard deviation , Based on these statistics, the constraint weights for different directions in subsequent registration iterations are dynamically adjusted. The adjustment of these constraint weights aims to reduce the influence of directions with larger standard deviations (i.e., more dispersed bias distributions) on registration optimization, making registration more focused on directions with better bias consistency. Weight coefficients are defined. (in Represents direction, with a value of , , The formula for calculating ) is: ; in, Represents direction The standard deviation. This formula ensures that the weighting coefficients are inversely proportional to the standard deviation, and that the sum of the weights in the three directions is 1. This represents a loop index variable used for iterating over sets. All directions within the range. Subsequently, the iterative nearest-point algorithm is used for optimized registration, but in each iteration, the spatial transformation parameters (rotation matrix) are solved. Translation vector When , direction weights are incorporated into the objective function. The optimization process aims to find the optimal... and This process minimizes the weighted sum of squared distances in three directions between all corresponding point pairs in the transformed point cloud of the component to be inspected and the design point cloud. After multiple iterations until convergence, the final optimized point cloud position is obtained, which is the stage adaptation registration result.

[0039] Subsequently, the 3D coordinate deviation is extracted from the phased adaptation registration results, and the corresponding acceptance tolerance threshold is queried. For each component to be accepted, after its point cloud is accurately registered with the design point cloud, the difference between the 3D coordinates of each point cloud and the corresponding design position coordinates is calculated to obtain the deviation vector of that point. The arithmetic mean of the deviation vectors of all points of the component is calculated to obtain the average deviation vector representing the overall positional deviation of the component. ;in For component indexing, Indicates the first The average positional deviation vector of the components to be inspected Represents the deviation vector In the global coordinate system Components in the axial direction, Represents the deviation vector In the global coordinate system Components in the axial direction, Represents the deviation vector In the global coordinate system The component in the axial direction; based on the specific construction stage (e.g., "steel structure main installation stage") and the engineering type of the component (e.g., "main beam", "secondary beam", "connecting plate"), a pre-set multidimensional tolerance table is consulted. This table is indexed by construction stage and component type, defining its tolerance in each case. , , Independent positional tolerance thresholds in three directions And a comprehensive spatial location tolerance threshold The overall tolerance threshold is used to control the total displacement of the component.

[0040] Finally, a conformity matrix consisting of Boolean values ​​is generated. For each component to be accepted, multiple conditions need to be considered. First, the magnitude of its average deviation vector δ_i is calculated, which represents its overall displacement. The calculation formula is: ; Then, perform a comparison and judgment on the four parallel items: judge Is the absolute value less than or equal to the independent tolerance threshold? .

[0041] judge Is the absolute value less than or equal to the independent tolerance threshold? .

[0042] judge Is the absolute value less than or equal to the independent tolerance threshold? .

[0043] judge Is it less than or equal to the overall composite tolerance threshold? .

[0044] Only when all of the above four comparison and judgment results are "yes", it is determined that the quality compliance of the component in the current construction stage is qualified, and it is given the Boolean value "true" (True or 1). As long as any one of the judgment results is "no", it is determined to be unqualified and given the Boolean value "false" (False or 0). After all the components to be accepted are judged, these Boolean judgment values are arranged in the order of the spatial position of the components (such as from top to bottom, from left to right) or in the order of their logical numbers in the design, forming a one-dimensional array or a two-dimensional table. This ordered set of Boolean values is the compliance matrix. This matrix intuitively and unambiguously expresses whether each permanent component to be accepted meets the stage quality requirements in the current construction stage, and is output as the final stage-limited quality compliance judgment result. The entire calculation and judgment process is based on clear mathematical rules and predefined thresholds, ensuring the objectivity and repeatability of the evaluation results.

[0045] The working principle of the present invention: First, collect the multi-temporal and multi-spatial three-dimensional point cloud data stream of the construction scene, and construct a spatio-temporal fusion three-dimensional point cloud that combines color and timestamp through feature point extraction and spatio-temporal registration; Second, perform semantic segmentation based on the local geometric topology and global context relationship of the points in the point cloud to obtain component point cloud clusters with different engineering semantics such as permanent components to be accepted, construction temporary components, and unconstructed areas; Then, analyze the point cloud density, morphology regularity, and assembly gap of the connection areas (such as bolt holes, welds) in the point cloud cluster of the permanent components to be accepted, generate connection state quantification indicators, and deduce the current specific construction stage state based on the predefined process dependency relationship logic; Then, dynamically match and retrieve the corresponding stage three-dimensional design specification point cloud data from the design three-dimensional specification library organized by the progress sequence according to this stage state; Finally, perform stage adaptability weighted registration on the point cloud cluster of the permanent components to be accepted and the retrieved design specification point cloud data, quantify the deviation component by component according to the tolerance threshold specific to the current stage and component type, generate a Boolean value compliance matrix through multi-condition comparison, and output the stage-limited quality compliance judgment result.

[0046] The above has described a detailed description of an embodiment of the present invention, but the content described is only the preferred embodiment of the present invention, and cannot be considered as used to limit the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the present invention application should still fall within the scope covered by the patent of the present invention.

Claims

1. A visual-based construction engineering quality detection and evaluation system, characterized in that, include: The multi-source spatiotemporal point cloud fusion module is used to collect three-dimensional point cloud data streams from different spatiotemporal locations in the construction scene and construct a spatiotemporally fused three-dimensional point cloud that integrates color intensity information and collection timestamps. The engineering semantic perception and segmentation module, based on the local geometric topology and global context relationship of points in the spatiotemporal fusion 3D point cloud, segments the point cloud into component point cloud clusters with different engineering semantics through semantic understanding. The engineering semantics include permanent components to be accepted, temporary construction components, and unconstructed areas. The construction phase status inference module analyzes the connection status and assembly integrity between components represented by the component point cloud clusters corresponding to the permanent components to be accepted, and deduces the specific construction phase status of the construction scenario at the current time point based on the predefined construction progress logic rules. The phased design specification dynamic retrieval module dynamically retrieves phased 3D design specification point cloud data that matches the construction objectives of the specific construction stage from the 3D design specification library organized according to the construction progress sequence, based on the specific construction stage status. The phase adaptability quality comparison and evaluation module performs phase adaptability registration between the point cloud cluster of permanent components to be accepted and the point cloud data of the phased 3D design specifications, and quantifies the differences based on the acceptance tolerance threshold applicable to the current construction phase, outputting the phase-limited quality compliance judgment result.

2. The visual-based construction quality inspection and assessment system according to claim 1, wherein, The process of constructing the spatiotemporal fusion 3D point cloud is as follows: Extract a set of feature points with stable geometric features from the 3D point cloud data stream at different spatiotemporal locations in the construction scene; Based on the spatiotemporal distribution of feature point sets, rigid transformation parameters between point clouds with adjacent timestamps are calculated to achieve spatiotemporal consistency registration of point cloud sequences. In the registered point cloud sequence, for the same three-dimensional spatial location, data points collected at different timestamps are compared, and the data points with the latest color intensity information and the highest geometric consistency are retained, while the rest of the redundant data points are filtered out. The retained data points are organized according to the collection timestamp and fused with color intensity information to construct a spatiotemporal fused 3D point cloud.

3. The visual based construction quality inspection and assessment system as claimed in claim 1, wherein, The output process of the component point cloud cluster is as follows: Based on the 3D coordinates and color intensity information of each point in the spatiotemporal fusion 3D point cloud, the normal vector distribution and curvature change of the points in the local neighborhood of the 3D coordinates and color intensity information of each point in the spatiotemporal fusion 3D point cloud are calculated to generate a local geometric topological descriptor representing the micro-geometric morphology. Based on local geometric topological descriptors, points that are spatially adjacent and have similar descriptors are aggregated to initially form geometric structural primitives. Combined with color intensity information, candidate component regions with continuous surfaces in the spatiotemporal fusion 3D point cloud are identified. The spatial connection relationships and relative positional constraints between the candidate component regions are analyzed, and global contextual reasoning is performed based on a predefined engineering semantic rule base, which defines the existence logic and interrelationships of different engineering semantic components in space. Based on the results of global context reasoning, the candidate component regions are ultimately determined as permanent components awaiting acceptance, temporary construction components, or unconstructed areas, and the corresponding component point cloud clusters are output.

4. The visual based construction quality inspection and assessment system as claimed in claim 3, wherein, The generation of local geometric topological descriptors representing microscopic geometric morphology specifically includes: Taking each to-be-calculated point in the spatio-temporal fused three-dimensional point cloud as the center, all neighboring points in a preset radius space range of the to-be-calculated point are searched and determined to form a local spatial neighborhood of the neighboring points; Based on the three-dimensional coordinates of all the neighboring points in the local spatial neighborhood, a covariance matrix of the local spatial neighborhood is constructed, and eigenvalue decomposition is performed on the covariance matrix to obtain three eigenvalues arranged in descending order and corresponding eigenvectors; The ratio of the minimum eigenvalue to the adjacent eigenvalue of the minimum eigenvalue and the difference relationship among the three eigenvalues are combined to obtain a curvature measure value representing the degree of change in the local surface curvature of the to-be-calculated point; The direction vector corresponding to the minimum eigenvalue in the eigenvector is determined as the normal vector of the to-be-calculated point, and the curvature measure value and the direction of the normal vector are combined to form a local geometric topological descriptor of the to-be-calculated point.

5. The visual based construction quality inspection and assessment system as claimed in claim 1, wherein, The specific construction phase state specifically includes: By the three-dimensional geometric shape and spatial distribution characteristics of the component point cloud cluster corresponding to the to-be-inspected permanent component, a feature area representing the physical connection relationship between components is identified and extracted, and the feature area at least includes bolt hole point cloud and weld surface point cloud; Based on the point cloud density, three-dimensional contour regularity and gap distance from the adjacent component point cloud cluster of the feature area, the tightness and continuity of the connection state between components are quantitatively evaluated, and a connection state quantitative index is generated; According to the spatial occupation range, mutual combination relationship and connection state quantitative index of all to-be-inspected permanent component point cloud clusters, the construction progress logical rules including process dependency relationship are matched by state matching reasoning according to the pre-defined final process node which has been satisfied by the current construction scene, and the final process node is identified as the specific construction phase state.

6. The visual based construction quality inspection and assessment system as claimed in claim 5, wherein, The generation of the connection state quantitative index specifically includes: The number of three-dimensional points contained in a unit volume in the feature area is counted to obtain the point cloud density of the feature area, and the ratio of the point cloud density to the expected density value in the preset standard tightness state is calculated as a first sub-index reflecting the filling density; The three-dimensional contour point cloud of the feature area is surface reconstructed, the average distance and the maximum distance between the actual surface of the feature area and the preset standard geometric shape surface are calculated, and the combination of the average distance and the maximum distance is taken as a second sub-index reflecting the topographic regularity; The average value and the variance of all minimum distances between the feature area and the corresponding positions of the adjacent component point cloud cluster are calculated, and the average value meets the design requirements and the variance is lower than the preset threshold value, which is determined as a gap uniformity, and a third sub-index reflecting the assembly alignment degree is generated; The first sub-index, the second sub-index and the third sub-index are weighted and fused to generate a comprehensive connection state quantitative index.

7. The visual based construction quality inspection and assessment system as claimed in claim 1, wherein, The dynamic calling of the phased three-dimensional design specification point cloud data corresponding to the specific construction phase construction target specifically includes: A semantic mapping relationship between the specific construction phase state and a plurality of specification items pre-stored in the design three-dimensional specification library is established, and each specification item is associated with a phased three-dimensional design specification point cloud data uniquely identified by a construction node, a component range and a design version; According to the process dependency relationship implied by the specific construction phase state, the identification set of all target components that must be completed in the current phase as acceptance objects is parsed; Based on the identification set of the target components, all specification entries containing the identification set are matched in the design three-dimensional specification library, and the unique target specification entry with the highest matching degree is selected by comparing the semantic similarity between the specific construction phase state and the construction node of each entry; The three-dimensional point cloud data associated with the target specification entry and having a design version is retrieved from the design three-dimensional specification library as the output of the three-dimensional design specification point cloud data of the phase.

8. The visual based construction quality inspection and assessment system as claimed in claim 7, wherein, The parsing process of the identification set is as follows: According to the specific construction phase state, the corresponding current process node is located in the pre-defined process dependency relationship table; All direct pre-process nodes of the current process node are extracted from the process dependency relationship table to form a pre-process set, and the completion state marks of each process in the pre-process set are obtained by querying the construction record database; According to the completion state marks, it is verified whether all direct pre-processes have been completed; if so, according to the association mapping table of components and processes, the unique identification of all components associated with the current process node and only allowed to be accepted after the current process is completed is extracted; All unique identifications of the extracted components are summarized to generate the identification set of all target components.

9. The visual based construction quality inspection and assessment system as claimed in claim 1, wherein, The output of the phase-limited quality compliance judgment result specifically includes: Based on the design coordinates of the components in the three-dimensional design specification point cloud data, the initial spatial alignment of the permanent component point cloud cluster to be accepted is performed to obtain an initial registration result; The position deviation distribution between the corresponding component point clouds in the initial registration result is analyzed, the constraint weights in different spatial directions in the subsequent registration process are dynamically adjusted according to the position deviation distribution, the iterative optimization registration is performed, and the phase adaptation registration result is obtained; From the phase adaptation registration result, the three-dimensional coordinate deviation between the actual position and the design position of each component is extracted, and the corresponding acceptance tolerance threshold is obtained from the pre-set multi-dimensional tolerance table according to the specific construction phase state and the component type; The three-dimensional coordinate deviation of each component is compared with the corresponding acceptance tolerance threshold to generate a compliance matrix composed of Boolean values as the output of the phase-limited quality compliance judgment result.

10. The visual based construction quality inspection and assessment system as claimed in claim 9, wherein, The generation of the compliance matrix composed of Boolean values specifically includes: For each acceptance component in the phase adaptation registration result, the deviation vector between the three-dimensional coordinates of all point clouds of the acceptance component and the three-dimensional coordinates of the corresponding design position in the three-dimensional design specification point cloud data of the phase is calculated; According to the specific construction phase state and the engineering type of the acceptance component, the multi-dimensional tolerance table is queried to obtain the independent tolerance threshold and the overall comprehensive tolerance threshold of the corresponding component in three spatial directions; The absolute value of the component of the deviation vector in each spatial direction is compared with the corresponding independent tolerance threshold, and the module length of the deviation vector is calculated and compared with the overall comprehensive tolerance threshold; The value of the corresponding component is true only when the component comparison and the overall comparison in all directions meet the requirements, otherwise it is false. Finally, the Boolean judgment values of all components are arranged in a predetermined spatial or logical order to generate a compliance matrix composed of Boolean values.

Images