A method for three-dimensional point cloud error detection and visualization guidance of a structural member

Abstract

The application provides a kind of structural component three-dimensional point cloud error detection and visualization guiding method, belongs to computer three-dimensional data processing technical field, method includes: S1.variation quantum neural network algorithm is based on the global optimization solution of the number of measuring station position of three-dimensional laser scanner, generates optimal measuring station arrangement scheme;S2.according to optimal measuring station arrangement scheme, using fixed three-point group algorithm is spliced to the point cloud data of multiple measuring station, obtains the complete point cloud data after splicing;S3.using point cloud segmentation large model extracts the measured point cloud of pre-installed component, discrete three-dimensional design model is design point cloud, using fixed three-point group algorithm is point cloud alignment and analyzes the deviation of both, through virtual reality technology, deviation information and corresponding installation adjustment guide visualization, complete the guide of component.The application has the advantages of high efficiency, high precision and intelligent visualization, and can be used for virtual error detection and on-site installation guide before structural component installation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer three-dimensional data processing, and in particular to a structural component three-dimensional point cloud error detection and visualization guiding method. BACKGROUND

[0002] With the large-scale construction of large-scale transportation infrastructure and industrial buildings, the installation precision requirements of large prefabricated components (such as bridge girder and factory steel column) have been improved to millimeter level, and traditional technology cannot meet the needs of digital and high-precision quality control. Although the current three-dimensional point cloud technology has been applied in component error detection and precision control, there are still the following key technical defects:

[0003] Point cloud collection efficiency and coverage imbalance: the existing station arrangement lacks systematic optimization algorithm and relies on manual experience planning, which is prone to problems such as redundant number of stations (leading to high data collection cost and low efficiency) or incomplete coverage of key areas, and cannot balance full coverage and economy;

[0004] Insufficient robustness of multi-station point cloud splicing: traditional splicing methods use maximum group search to match point pairs, and the computational complexity increases exponentially, and are sensitive to noise, dust, light interference and local occlusion in the field, making it difficult to ensure the geometric consistency of the spliced point cloud, resulting in distorted basic data for subsequent error analysis;

[0005] Insufficient error visualization and decision support: the error analysis results of existing technologies are mostly presented in the form of two-dimensional reports or abstract data, lacking intuitive three-dimensional visualization means, and the point cloud deviation cannot be converted into intuitive information that can be directly used for precision judgment, making it difficult for technical personnel to quickly and accurately locate the deviation position and size, affecting the decision-making efficiency of subsequent operations;

[0006] Existing technical defects lead to disconnection between virtual and physical operations: error checking results in virtual space are mostly presented in the form of drawings or reports, and cannot be directly converted into installation guidance that can be intuitively understood by field construction personnel, leading to a technical gap between virtual verification and physical installation, affecting the efficiency of guided construction.

[0007] The application effect of three-dimensional point cloud technology in large-scale structural component precision control is limited, and a closed-loop technical solution integrating station optimization, data processing, error analysis and installation guidance is urgently needed to solve the efficiency, precision and readability problems in the whole process of point cloud processing. SUMMARY

[0008] The present application provides a structural component three-dimensional point cloud error detection and visualization guiding method, which has the advantages of high efficiency, high precision and intelligent visualization, and can be used for virtual error detection before installation of structural components and field installation guidance.

[0009] To achieve the above object, the application adopts the following technical solutions:

[0010] A three-dimensional point cloud error detection and visualization guiding method for structural members, comprising:

[0011] S1. Obtain a three-dimensional design model of a structure to be constructed and a site environment constraint condition of a construction area, and perform global optimization and solving on the station position and number of a three-dimensional laser scanner based on a variational quantum neural network algorithm to generate an optimal station arrangement scheme;

[0012] S2. According to the optimal station arrangement scheme, collect multi-station point cloud data of the construction area by using the three-dimensional laser scanner, and perform splicing processing on the multi-station point cloud data by using a fixed three-point cluster algorithm to obtain complete point cloud data after splicing;

[0013] S3. Extract the measured point cloud of the pre-installed member from the complete point cloud data after splicing by using a point cloud segmentation large model, discretize the three-dimensional design model of the structure to be constructed into design point cloud matching the density of the measured point cloud, align and analyze the deviation of the design point cloud and the measured point cloud by using the fixed three-point cluster algorithm in step S2, and visualize the deviation information and the corresponding installation adjustment guide by using virtual reality technology to complete the guiding of the member.

[0014] In the specification, the specific process of generating the optimal station arrangement scheme in step S1 includes: taking the effective scanning range of the three-dimensional laser scanner as the benchmark to construct a plurality of surrounding spaces, making the three-dimensional design model be completely covered and the adjacent surrounding spaces maintain a preset overlap degree by stacking the surrounding spaces, taking the center of each surrounding space as a candidate station position to form a candidate station position set; and optimizing the candidate station position set based on the variational quantum neural network algorithm, wherein the point cloud coverage cost, the scanning accuracy cost and the station economic cost are considered simultaneously in the optimization process, and the optimal station arrangement scheme satisfying the point cloud full coverage, the accuracy requirement and the least number of stations is output.

[0015] In the specification, the setting standard of the preset overlap degree is that the overlapping area of the adjacent surrounding spaces can meet the geometric consistency checking requirement in subsequent multi-station point cloud splicing.

[0016] In the specification, in the optimization process of the variational quantum neural network algorithm, the priority of the point cloud coverage cost, the scanning accuracy cost and the station economic cost is adjusted by setting the weight coefficients, wherein the weight coefficient of the station economic cost is higher than the weight coefficients of the other two costs, so as to preferentially realize the minimization of the number of stations.

[0017] In the specification, the specific process of splicing the multi-station point cloud data in step S2 by using the fixed three-point group algorithm includes: for any two groups of point cloud data to be spliced, establishing a spatial compatibility relationship of point pairs in the point cloud, judging the consistency of the point pairs in the spatial structure; constructing a second-order compatibility graph based on the spatial compatibility relationship, quantifying the neighborhood consistency of the point pairs; selecting a number of point pairs with the highest score from the second-order compatibility graph as pivots, screening a third point from the common neighbors of each pivot to form a candidate three-point group, calculating the weight of each candidate three-point group and retaining the candidate three-point groups with high weight as fixed three-point groups; calculating the optimal rigid transformation of the point cloud data based on the fixed three-point groups, completing the splicing of the two groups of point cloud data by maximizing the number of inliers of the transformed point cloud, and repeating the process until all multi-station point cloud data is spliced into complete point cloud data.

[0018] In the specification, when judging the spatial structure consistency of the point pairs, the consistency judgment threshold is determined based on the density distribution characteristics of the point cloud, and when the spatial structure difference of the point pairs is less than the threshold, it is determined that the point pairs have spatial compatibility.

[0019] In the specification, when selecting the pivots, the point pairs ranked in the top K 1 of the second-order compatibility graph are selected as the pivots, K 1 is determined according to the size of the point cloud data; when screening the candidate three-point groups, the candidate three-point groups ranked in the top K 2 of the weight are retained for each pivot, K 2 is determined according to the noise level of the point cloud data, and the number of fixed three-point groups finally generated is the product of K 1 and K 2.

[0020] In the specification, the point cloud segmentation large model is constructed based on the PointNet++ framework, which obtains local point sets in the point cloud through hierarchical sampling, aggregates features of the local point sets to obtain multi-scale geometric information, and obtains structure features from local to global through feature transmission, and realizes segmentation and extraction of the pre-installed component measured point cloud based on the structure features.

[0021] In the specification, when the three-dimensional design model is discretized into a design point cloud, the discretization density is consistent with the density of the measured point cloud, ensuring the comparison accuracy of the design point cloud and the measured point cloud in subsequent alignment analysis; when aligning the design point cloud and the measured point cloud by using the fixed three-point group algorithm, the goal is to align the point cloud deviation to be less than a preset accuracy threshold, and the preset accuracy threshold is determined according to the accuracy requirement of the component installation.

[0022] In the specification, in step S3, the deviation of the design point cloud and the measured point cloud is converted into a three-dimensional deviation vector when visualized by the virtual reality technology, the direction and size of the deviation vector are displayed in an intuitive graphic mark in the virtual reality device, and a step-by-step installation adjustment guide is generated according to the deviation vector to guide the construction personnel to complete the installation adjustment of the component according to the guide.

[0023] In summary, the present application has at least the following beneficial effects:

[0024] 1. The measurement station arrangement is optimized by the variational quantum neural network algorithm, relying on the global search capability, the number of measurement stations is minimized under the premise of ensuring full coverage and precision, the data acquisition cost is reduced, and the operation efficiency is improved.

[0025] 2. The high-precision and robust splicing of multi-station point clouds is realized by using the fixed three-point group algorithm, and the problems of sensitivity to initial values and easy to fall into local optimum of traditional methods are overcome.

[0026] 3. Real-time visualization guidance of the installation process is realized by virtual reality fusion technology, the installation precision is controlled within millimeter level, and the construction quality and efficiency are significantly improved.

[0027] 4. The whole method forms a complete data acquisition-processing-analysis-guiding technology closed loop, and provides a complete solution for digital installation of structural components.

[0028] The present application is suitable for virtual inspection and installation guidance of complex structural components in large-scale traffic facility construction, and has important engineering application value. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0030] Figure 1 It is a schematic diagram of the three-dimensional point cloud error detection and visualization guidance method of the structural component involved in the present application.

[0031] Figure 2 It is a flowchart of the variational quantum neural network algorithm for generating an optimal measurement station arrangement scheme involved in the present application.

[0032] Figure 3 It is a schematic diagram of the multi-measurement station point cloud data splicing process based on the fixed three-point group algorithm involved in the present application.

[0033] Figure 4A schematic diagram of the component error analysis and guided installation process involved in the virtual-real alignment in the present application. DETAILED DESCRIPTION

[0034] In the following, only certain exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present application. Therefore, the drawings and the description are considered to be exemplary in nature rather than limiting.

[0035] The following disclosure provides many different embodiments, or examples, for implementing different structures of the embodiments of the present application. For the purpose of simplifying the disclosure of the embodiments of the present application, the components and settings of specific examples are described in the following. Of course, they are only examples, and the purpose is not to limit the embodiments of the present application. In addition, the embodiments of the present application can repeatedly refer to numerals and / or reference letters in different examples, and such repetition is for the purpose of simplification and clarity, which itself does not indicate the relationship between the various embodiments and / or settings discussed.

[0036] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0037] As Figure 1 shown, the present embodiment provides a structure component three-dimensional point cloud error detection and visualization guidance method, comprising:

[0038] S1. Obtain the three-dimensional design model of the structure to be constructed and the site environment constraint condition of the construction area, and based on the variational quantum neural network algorithm, globally optimize and solve the station position and quantity of the three-dimensional laser scanner, to generate an optimal station arrangement scheme;

[0039] S2. According to the optimal station arrangement scheme, use a three-dimensional laser scanner to collect multi-station point cloud data of the construction area, and use a three-point group algorithm to splice and process the multi-station point cloud data to obtain complete point cloud data after splicing;

[0040] S3. Use a point cloud segmentation large model to extract the measured point cloud of the pre-installed component from the complete point cloud data after splicing, discretize the three-dimensional design model of the structure to be constructed into design point cloud matching the density of the measured point cloud, use the three-point group algorithm in step S2 to align and analyze the deviation between the design point cloud and the measured point cloud, visualize the deviation information and the corresponding installation adjustment guidance through virtual reality technology, and complete the guidance of the component.

[0041] In some embodiments, the specific process of generating the optimal station arrangement scheme in step S1 includes: constructing a plurality of surrounding spaces based on the effective scanning range of the three-dimensional laser scanner, stacking the surrounding spaces so that the three-dimensional design model is completely covered and the adjacent surrounding spaces maintain a preset overlap degree, taking the center of each surrounding space as a candidate station position, and forming a candidate station position set; and optimizing the candidate station position set based on a variational quantum neural network algorithm, considering the point cloud coverage cost, scanning accuracy cost and station economic cost during the optimization process, and outputting an optimal station arrangement scheme that meets the point cloud full coverage, accuracy requirements and has the least number of stations.

[0042] In some embodiments, the preset overlap degree is set according to the requirement of geometric consistency check during subsequent multi-station point cloud data splicing.

[0043] In some embodiments, during the optimization process of the variational quantum neural network algorithm, the priority of the point cloud coverage cost, the scanning accuracy cost and the station economic cost is adjusted by setting weight coefficients, and the weight coefficient of the station economic cost is higher than that of the other two costs, so as to prioritize the minimization of the number of stations.

[0044] In some embodiments, the specific process of splicing multi-station point cloud data by using the fixed three-point group algorithm in step S2 includes: establishing a spatial compatibility relationship of point pairs in the point cloud for any two groups of point cloud data to be spliced, and judging the consistency of the point pairs in the spatial structure; constructing a second-order compatibility graph based on the spatial compatibility relationship to quantify the neighborhood consistency of the point pairs; selecting a number of point pairs with the highest scores from the second-order compatibility graph as pivots, and selecting a third point from the common neighbors of each pivot to form a candidate three-point group, calculating the weight of each candidate three-point group and retaining the candidate three-point groups with high weights as fixed three-point groups; calculating the optimal rigid transformation of the point cloud data based on the fixed three-point groups, maximizing the number of inliers of the transformed point cloud to complete the splicing of the two groups of point cloud data, and repeating the process until all multi-station point cloud data is spliced into complete point cloud data.

[0045] In some embodiments, when judging the spatial structure consistency of the point pairs, the consistency judgment threshold is determined based on an empirical adjustment coefficient and point cloud density distribution characteristics. When the spatial structure difference of the point pairs is less than the threshold, it is determined that the point pairs have spatial compatibility. The empirical adjustment coefficient is a multiplier factor for tightening the matching conditions, and its core purpose is to balance and adjust between the recall rate of matching and the accuracy rate of matching, so that the algorithm can adapt to point cloud data of different qualities and characteristics. It is usually not a fixed value, but a hyperparameter that needs to be manually debugged according to specific data conditions such as noise level, overlap rate and point cloud density.

[0046] In some embodiments, when selecting the pivots, the top N point pairs with the highest scores are selected from the second-order compatibility graph, and N is a positive integer.K 1 group of point pairs as pivots, K 1 is determined according to the size of the point cloud data; when screening the candidate three-point groups, the top K 2 candidate three-point groups are reserved for each pivot, K 2 is determined according to the noise degree of the point cloud data, and the number of final generated three-point groups is the product of K 1 and K 2.

[0047] In some embodiments, the point cloud segmentation large model is constructed based on the PointNet++ framework, which obtains local point sets in the point cloud through hierarchical sampling, aggregates features of the local point sets to obtain multi-scale geometric information, and obtains structure features from local to global through feature transmission, and realizes the segmentation and extraction of the measured point cloud of the pre-installed component based on the structure features.

[0048] In some embodiments, when the three-dimensional design model is discretized into a design point cloud, the discretization density is consistent with the density of the measured point cloud, ensuring the comparison accuracy of the design point cloud and the measured point cloud during subsequent alignment analysis; when aligning the design point cloud and the measured point cloud using the three-point group algorithm, the deviation of the aligned point cloud is less than a preset accuracy threshold, and the preset accuracy threshold is determined according to the accuracy requirement of the component installation.

[0049] In some embodiments, when visualizing in step S3 through virtual reality technology, the deviation of the design point cloud and the measured point cloud is converted into a three-dimensional deviation vector, the direction and size of the deviation vector are displayed in a intuitive graphical mark in the virtual reality device, and step-by-step installation adjustment instructions are generated according to the deviation vector to guide the construction personnel to complete the installation adjustment of the component according to the instructions.

[0050] The technical concept of the application is as follows:

[0051] S1. Generating an optimal station layout scheme based on a variational quantum neural network algorithm

[0052] The station optimization module constructs a candidate station position set with the three-dimensional design model and the field environment constraint conditions (construction site boundary, obstacles, and areas where stations cannot be arranged) as inputs; global optimization is solved using a quantum neural network to achieve full coverage and high-precision collection of point cloud data with the least number of stations; and an optimal station position and quantity scheme is output. The specific implementation includes:

[0053] The range of the three-dimensional laser scanner is used as the size of the enclosing space, and the field environment constraint is used as the limiting condition, multiple enclosing spaces are stacked to completely surround the three-dimensional design model, and a certain degree of overlap is ensured between each enclosing space. The center of each enclosing space is defined as a point in the candidate station position set ​, a set of candidate station locations is constructed The optimal subset is found by a variational quantum neural network algorithm So that under the constraints of full coverage, high accuracy, etc., the efficiency is the highest. The objective function is constructed :

[0054] ;

[0055] ;

[0056] Wherein represents a set of candidate station locations; represents the position of the th candidate station in the set i ; represents a vector of a station arrangement scheme; represents the decision variable, , represents setting up a station at position ; represents not setting up a station; represents the total number of candidate stations. represents the coverage cost function, which calculates the coverage degree of the scheme to the target area;

[0057] ;

[0058] Wherein: wherein is the area of the th surface element, is the total number of elements; is an indicator function:

[0059] ;

[0060] represents the accuracy cost function, which evaluates the overall scanning accuracy that the scheme can achieve;

[0061] ;

[0062] Wherein: The number of active stations; Weight coefficient, reflecting the importance of the region, set according to the specific structure and the object of attention; is the degree of overlap between two stations;

[0063] ;

[0064] Wherein: represents the station and The degree of overlap; The actual overlap ratio; The optimal overlap ratio, recommended value 20%-30%; The tolerance parameter of the degree of overlap, recommended value 1.5.

[0065] The economic cost function, directly related to the total number of stations:

[0066] ;

[0067] Wherein: The number of activated stations; The total number of stations;

[0068] Weighting coefficients. The three positive real numbers are used to adjust the importance and priority of the three sub-costs in the total target. It is recommended to set Between 0.4-0.5 to ensure the hard constraint of full coverage, Between 0.3-0.4 to strongly drive the minimization of the number of stations, and Between 0.1-0.25 to optimize the scanning accuracy; typical initial values can be , , This configuration efficiently pursues economy on the basis of ensuring no-missing scanning, and takes accuracy as an important optimization item. The final value needs to be determined after iterative debugging according to the constraint strength of the specific scene.

[0069] Use direct encoding, each candidate station location is represented by a quantum bit:

[0070] ;

[0071] Indicates the quantum state, which is used to encode a station arrangement scheme. The quantum state |0> represents not setting a station here, and |1> represents setting a station here; Indicates the quantum state composed of Quantum bits, where each Is a quantum bit. An N-qubit state can represent 2 N Possible station arrangement schemes at the same time. Map the objective function to the energy operator of the quantum system . The eigenvalue on the computational basis vector is the cost of the scheme:

[0072] ;

[0073] Wherein H, which is the mapping of the Hamiltonian operator, is the objective function in the quantum system, and its expectation value represents the scheme cost. , , correspond to the coverage, accuracy and economy penalty terms, respectively.

[0074] where is the most important subterm, used to punish excessive number of stations:

[0075] ;

[0076] ;

[0077] ;

[0078] where, is the operator acting on the i th qubit; is the identity operator; the total number of voxels after discretization of the target area; represents the importance weight of voxel , and the recommended value is 0.3; the set of stations that can cover the voxel; ∏ represents the continuous multiplication of operators; represents the expansion of all operators in the Pauli operator basis, which is convenient for quantum computing. Its expectation value represents the probability of setting up a station at the i th position. Then introduce the parameterized quantum circuit , whose structure is defined as:

[0079] ;

[0080] where represents the parameter vector, represents the parameterized quantum circuit; , , represent the entanglement operation layer, quantum bit rotation layer, and repetition layer of the structure, respectively.

[0081] Prepare the quantum state , measure the expectation value of under the current quantum state on the quantum processor, which represents the average cost of the current station layout scheme. Optimize and update the parameters based on the parameter displacement rule. First, calculate the gradient of the objective function with respect to each parameter. For each parameter , prepare the quantum state and after the parameter offset, and measure the corresponding expectation values and , by the gradient formula:

[0082] ;

[0083] where is the unit vector with the th component being 1; is the fixed displacement amount; the gradient of each component is accurately calculated, and the complete gradient vector is finally obtained. The adaptive moment estimation optimization algorithm is used for parameter updating. The first moment estimation and the second moment estimation are maintained, which respectively accumulate the mean information and variance information of the gradient. At each iteration, first update the momentum term: , where is the first moment estimation of the current time step; is the first moment decay rate, usually taking 0.9; is the first moment estimation of the last time step; is the gradient of the current time step. Then update the adaptive learning rate term: , where is the second moment estimation of the current time step; is the second moment decay rate, usually taking 0.999; is the second moment estimation of the last time step; is the square of the gradient of each component. To avoid initial bias, bias correction is performed on the moment estimation: , . The final parameter update amount is , where is the base learning rate, is the numerical stability constant. During the entire optimization process, the learning rate adopts an adaptive scheduling mechanism. A larger learning rate (0.1) is used in the early stage of optimization to promote rapid convergence, a medium learning rate (0.01) is smoothly decayed according to the cosine annealing strategy in the middle stage, and a small learning rate (0.001) is used for fine tuning in the later stage. The convergence judgment condition is set: when the change of the objective function in continuous multiple iterations is less than the threshold value δ (recommended value 10 -6 ) or the parameter change is less than , it is considered to be converged; otherwise, continue to iterate and optimize until the maximum iteration number is reached. Through this systematic gradient calculation and parameter updating mechanism, the quantum neural network can effectively search for the optimal station layout scheme that meets the requirements of full coverage, high precision, and economy.

[0084] where the number of quantum bits directly corresponds to the number of candidate station positions. That is, if M candidate station points are generated by the surrounding space method, then the number of quantum bits N = M;

[0085] The depth of the circuit is defined by the number of repeated layers. Deeper circuits (larger L) have stronger expressiveness and entanglement, and can represent more complex function relationships, thus have the potential to find better solutions. L is recommended to be 4~6;

[0086] Composition of the dataset: The dataset of the model is not a static, pre-labeled sample set in traditional machine learning, but a digital environment composed of multi-dimensional information for dynamically evaluating the feasibility of the solution. It includes four parts: 1) target geometric model (high-precision BIM or CAD model), which defines all the surface point clouds that need to be scanned and is the ultimate object of coverage evaluation; 2) candidate station set, which is a set of all possible station arrangement points generated in three-dimensional space in advance through algorithms, directly determining the size of the solution space and the number of qubits of the optimization problem; 3) scanning device parameters, including maximum / minimum scanning distance, field of view angle, and physical model of accuracy decay with distance, which provide objective basis for converting abstract coverage and accuracy targets into calculable mathematical functions; 4) site environment constraints, usually in the form of a three-dimensional obstacle model, used to determine whether there is an obstruction in the line of sight from the station to the target point, ensuring the site feasibility of the solution. In the training process, each station arrangement scheme generated by the VQNN will be dynamically simulated and evaluated in this digital environment: the system will calculate the loss function of the scheme for the target point, estimate its comprehensive scanning accuracy, and count the number of stations. Therefore, the entire dataset is a simulated verification system that integrates physical rules and geometric constraints.

[0087] S2. Splicing multi-station point cloud data based on multi-scale feature fusion optimization fixed three-point group algorithm

[0088] The multi-station data splicing module uses a fixed-size three-point group (fixed three-point group) instead of the traditional maximum group, optimizes through multi-scale feature fusion, ensures geometric consistency with the help of height-constrained compatibility maps, and realizes efficient and stable splicing of multi-station point cloud data using the pivot matching pair guiding strategy. The fixed three-point group algorithm avoids the exponential complexity brought by global group search when constructing matching relationships, significantly reducing computational complexity and time overhead. At the same time, the fixed three-point group algorithm can improve the matching stability of noisy data and local occlusion conditions while maintaining the accuracy of geometric constraints, thereby improving the robustness and real-time performance of multi-station point cloud splicing. Specifically, it includes:

[0089] Establishing a spatial compatibility relationship between matching pairs for the input point cloud data to be spliced

[0090] ;

[0091] where is the compatibility threshold,​ is a key point in the source point cloud; is a key point in the target point cloud; denotes spatial distance. Compatibility threshold is used to judge the consistency of point pairs in spatial structure. τ should be set to about 0.25 times the resolution of the point cloud, and the recommended value is 0.01 m to 0.08 m, which has stable performance in this range.

[0092] Construct multi-scale feature fusion compatibility graph, in order to improve the robustness of multi-station data under noise and occlusion conditions, multi-scale feature fusion optimization is introduced. By hierarchical feature extraction, geometric information of different scales is captured, and the discrimination ability of the compatibility graph is enhanced. The spatial compatibility relationship between matching pairs of multi-scale feature fusion is established :

[0093] ;

[0094] wherein is the weight of controlling geometric compatibility, and the recommended value is 0.7, which highlights the dominant role of geometric constraints; is the weight of controlling multi-scale features, and the recommended value is 0.3; is the number of hierarchical multi-scale feature extraction, usually 3, representing three different scales of local (radius 1~5m, capturing surface geometry), medium (radius about 5-8m, capturing object parts), and global (>8m, capturing the entire scene), The value can also be adjusted according to the resolution of the point cloud and the application scenario; is the number of hierarchical multi-scale feature extraction weight, used to adjust the contribution of different scale features, the weight is usually set adaptively according to the importance of the scale, such as high-resolution scale (rich in details) may be given a greater weight. The weight satisfies the normalization condition ; is a similarity function, used to calculate the similarity between two feature vectors, ; is a feature extraction function, which extracts the neighborhood point set through spherical query with different radius , and uses multi-layer perception (MLP) for feature extraction, ;

[0095] Quantify the neighborhood consistency of matching pairs: The higher the value, the more consistent the matching pair in space; denotes the edge weight between node and node in the second-order compatibility graph; and are the node With , node With compatibility indication value. By pivot-guided search algorithm, search the paired point clouds efficiently. Guided by the top K 1 matching pair (called pivot) in the score of the second-order compatibility graph. For each pivot , find the third point in its common neighbor set :

[0096] ;

[0097] represents the index of the third matching pair, whose value range is 1 to the total number of matching pairs, but and must be excluded. Each corresponds to a matching pair, and if it meets the compatibility condition, it can form a triplet. Each candidate triplet formed in this way will be scored with an aggregate weight :

[0098] ;

[0099] , is the edge weight of node and , and in the second-order compatibility graph. Finally, for each pivot, the top triple-point group with the highest weight is retained, and a total of hypotheses are generated. =1000 balances between performance and speed, fixed value 2. For each triple-point group , , , respectively represent the matching pairs between the triplets . Calculate its rigid transformation and select the final best transformation by maximizing the number of inliers:

[0100] ;

[0101] where the candidate transformation set ; is the zth candidate transformation, , is the rotation matrix calculated by the rigid transformation, is the translation vector calculated by rigid transformation; The scoring function of the inlier is:

[0102]

[0103] where is an indicator function (returns 1 if the condition is met, otherwise returns 0); is a threshold value, which needs to be set according to the accuracy requirement.

[0104] S3, component error analysis and guided installation based on virtual-real alignment

[0105] The inspection and guidance module uses a point cloud segmentation large model to segment the pre-installed component, discretizes the design model into a point cloud, and accurately aligns it with the measured point cloud. Through difference analysis, the component error is analyzed, and the virtual reality technology is used to realize the visual guidance of the installation process. Specifically, the point cloud segmentation large model uses PointNet++ as the backbone network, and the core is hierarchical feature learning. First, the input point cloud is multi-level down-sampled through the set abstraction module: PointNet is used to extract features in each local area, and the receptive field is gradually expanded to capture multi-level geometric structures from details to the whole. In order to perform segmentation, the model needs to generate point-by-point labels, so a feature propagation module is connected as a decoder after the encoder. The feature propagation module uses reverse distance weighted interpolation to up-sample sparse high-level features to the original point cloud density, and fuse with the corresponding layer jump connection features of the encoder to recover the detail information. Finally, a multi-layer perceptron with shared weights outputs the classification score of each point.

[0106] In some embodiments, through the semantic features output by the model, the pre-installed component is segmented, the design model is discretized into a point cloud, and the measured point cloud is aligned and subtracted using the method in S2; through the deviation analysis of the aligned point cloud, the component size error, geometric deviation and other quality problems are identified according to the three-way projection plane; through the virtual reality device, the current position of the component is displayed in real time, and the system calculates the three-dimensional coordinate difference of the corresponding points to generate a deviation vector through the registration of the design model point cloud and the measured point cloud, and marks the deviation direction and size of the target position. The deviation vector provides intuitive installation adjustment guidance for on-site operators.

[0107] ​In some embodiments, to improve robustness in complex environments, laser point cloud, infrared image and motion compensated multi-modal data are fused to enhance the stability of segmentation. Subsequently, the model is discretized into point cloud and accurately registered with the measured point cloud, and the deviation vector is generated by calculating the three-dimensional coordinate difference to analyze the component size and geometric error. On this basis, the system displays the component position and deviation vector in real time through virtual reality equipment, providing intuitive visual guidance for operators. The closed-loop mechanism of "real-time scanning-deviation updating-guidance iteration" is introduced, so that the guidance information can be dynamically refreshed with the component adjustment until the error converges. In addition, the system introduces human-computer collaborative decision-making function, provides multi-scheme auxiliary selection in complex scenes, and expands the multi-component collaborative installation and dynamic interference detection capability, and real-time collision risk warning in the guidance process ensures the overall safety and efficiency of the installation process.

[0108] In some embodiments, a design model in the design and construction of a bridge and field collected data are used to prove the effectiveness of the method.

[0109] 1) Discretize the design model and the collected image data into point cloud data (obtain low-precision global point cloud data by unmanned aerial photography), and segment them into bounding boxes of the same size as the three-dimensional laser scanning range to ensure a certain degree of overlap, and build a candidate station point set . After obtaining the initial optimized number of stations as 31 stations, the candidate station point set is input into the variational quantum neural network for optimization, and the output optimized number of stations is 17 stations. The specific process is shown in Figure 2 .

[0110] 2) According to the generated optimized station scheme, use a three-dimensional laser scanning instrument to collect data. Take the local four station data as input, replace the traditional maximum group with a fixed size three-point group, combine the height constraint compatibility map to ensure geometric consistency, and use the pivot matching pair to realize the splicing of multi-station point cloud data. The specific process is shown in Figure 3 .

[0111] 3) After the spliced point cloud data is used, the point cloud segmentation large model extracts the pre-installed component from the data, and the design model is discretized into point cloud and aligned with the measured point cloud for deviation analysis. In the virtual space, the rotation and translation matrix obtained by solving is used for construction installation guidance, and the error between the actual component and the design model is identified to analyze whether the component meets the construction requirements. Realize the virtual space error checking and guidance installation of the structure. The specific process is shown in Figure 4 .

[0112] The above-described embodiments are used to illustrate the present application and are not intended to limit the present application, so the change of example values or the substitution of equivalent elements should still belong to the scope of the present application.

[0113] From the above detailed description, one skilled in the art can clearly understand that the present application can achieve the above-mentioned purposes, and thus meets the requirements of the Patent Law.

[0114] Although the preferred embodiments of the application have been described, those skilled in the art will readily understand that there are numerous modifications and adaptations of the application that are within its scope. Therefore, the following claims are intended to include within their scope all such modifications and adaptations of the preferred embodiments as come within the scope of the present application.

[0115] It should be noted that the above description of the flow is merely for example and illustration, and does not limit the scope of the present specification. Those skilled in the art can make various modifications and changes to the flow under the guidance of the present specification. However, these modifications and changes are still within the scope of the present specification.

[0116] The above has described the basic concept, and it is obvious that the above-mentioned application disclosure is only as an example and does not constitute a limitation on the present application for those skilled in the art after reading this application. Although it is not explicitly stated here, those skilled in the art can make various modifications, improvements and modifications to the present application. Such modifications, improvements and modifications are suggested in the present application, so such modifications, improvements and modifications are still within the spirit and scope of the exemplary embodiments of the present application.

[0117] Meanwhile, specific words are used in the present application to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" means a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that the "an embodiment" or "one embodiment" or "an alternative embodiment" mentioned in different positions in the present specification does not necessarily refer to the same embodiment. In addition, certain features, structures or characteristics in one or more embodiments of the present application can be properly combined.

[0118] Moreover, as will be appreciated by persons skilled in the art, the present application is capable of being embodied with several different types of categories or circumstances of patentable subject matter including any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof. Accordingly, the various aspects of the present application can be embodied in hardware and / or in software (including firmware, resident software, micro-code, etc.) that runs on a processor or controller to cause a computer or computer- based device to perform various computer-implemented processes. The computer program instructions can be embodied in a computer program product, which can include a computer-readable medium, such as a floppy disk, a CD-ROM, a DVD, a Blu-ray Disc, a flash drive, a memory stick, a magnetic tape, a magnetic disk, a memory chip, a semiconductor, or the like, which can be accessed by a computer or computer- based device. Accordingly, the various aspects of the present application can be embodied in a number of different forms, all of which have been contemplated to be within the scope of the present application as disclosed herein. In addition, each feature or combination of features of the present application can be expressed as a method, device, program, or computer readable medium, or any combination of these.

[0119] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, conventional procedural programming languages, such as the C programming language, Visual Basic, Fortran 2103, Perl, COBOL 2102, PHP, ABAP, dynamic programming languages, such as Python, Ruby and Groovy, or another programming language. The program code can execute entirely on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any network, such as a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet) or within a cloud computing environment, or as a service, such as Software as a Service (SaaS).

[0120] In addition, the order of execution or sequence of processing elements and sequences, unless otherwise specifically indicated, can vary from that described. For example, although the description above refers to a particular arrangement of process steps implemented in a particular order, other arrangements could be implemented involving other orders or sequences of steps. For example, although the above disclosure discusses some presently preferred embodiments of the application by way of various examples, it is to be understood that this disclosure is not to be limited to the specifics of these examples, as these are intended to be illustrative only. Additional claims can be presented below which are in no way intended to be limited to the specific embodiments recited in this disclosure. For example, although the implementation of various components described above can be in the form of hardware devices, it can also be implemented as a pure software solution, for example, as an installation on an existing server or mobile device.

[0121] For similar reasons, it is to be appreciated that the teachings of the present application provided herein can be applied to any embodiment of the present application, and that actual claims applied for or patent granted can be broader than any single, featured embodiment. Accordingly, a patent applicant has constructed and filed examples to particularly point out and distinctly claim those aspects which are regarded as novel and those aspects specifically shown.

Claims

1. A method for three-dimensional point cloud error detection and visualization guidance of structural components, characterized in that, include: S1. Obtain the three-dimensional design model of the structure to be constructed and the on-site environmental constraints of the construction area, and perform global optimization of the location and number of stations of the three-dimensional laser scanner based on the variable quantum neural network algorithm to generate the optimal station layout scheme; S2. According to the optimal station layout scheme, use a three-dimensional laser scanner to collect multi-station cloud data of the construction area, and use the fixed three-point cluster algorithm to stitch the multi-station cloud data to obtain the stitched complete point cloud data. S3. Using the point cloud segmentation large model, extract the measured point cloud of the pre-installed component from the spliced ​​complete point cloud data. Discretize the three-dimensional design model of the structure to be constructed into a design point cloud that matches the density of the measured point cloud. Use the fixed three-point cluster algorithm in step S2 to align the design point cloud with the measured point cloud and analyze the deviation between the two. Visualize the deviation information and the corresponding installation adjustment guide through virtual reality technology to complete the component guidance. The specific process of stitching multi-measurement point cloud data using the fixed three-point clique algorithm in step S2 includes: establishing spatial compatibility relationships between point pairs in any two sets of point cloud data to be stitched, and determining the consistency of point pairs in spatial structure; constructing a second-order compatibility graph based on the spatial compatibility relationships, and quantifying the neighborhood consistency of point pairs; selecting several sets of point pairs with the highest scores from the second-order compatibility graph as pivots, selecting a third point from the common neighbors of each pivot to form a candidate three-point clique, calculating the weight of each candidate three-point clique, and retaining the candidate three-point clique with high weight as the fixed three-point clique; calculating the optimal rigid transformation of the point cloud data based on the fixed three-point clique, and completing the stitching of two sets of point cloud data by maximizing the number of interior points in the transformed point cloud, and repeating this process until all multi-measurement point cloud data are stitched into complete point cloud data.

2. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 1, characterized in that, The specific process of generating the optimal station layout scheme in step S1 includes: constructing multiple enclosing spaces based on the effective scanning range of the 3D laser scanner; stacking the enclosing spaces to ensure that the 3D design model is completely covered and that adjacent enclosing spaces maintain a preset overlap; taking the center of each enclosing space as the candidate station location to form a set of candidate station locations; optimizing the set of candidate station locations based on the variable quantum neural network algorithm, taking into account the point cloud coverage cost, scanning accuracy cost, and station economic cost during the optimization process, and outputting the optimal station layout scheme that meets the requirements of full point cloud coverage, accuracy, and the minimum number of stations.

3. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 2, characterized in that, The standard for setting the preset overlap is: the overlapping area of ​​adjacent enclosing spaces can meet the geometric consistency verification requirements when splicing cloud data from multiple measurement stations.

4. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 2, characterized in that, In the optimization process of the variable quantum neural network algorithm, the priority of point cloud coverage cost, scanning accuracy cost and station economic cost is adjusted by setting weight coefficients. The weight coefficient of station economic cost is higher than that of the other two costs, so as to prioritize minimizing the number of stations.

5. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 1, characterized in that, When determining the spatial structure consistency of point pairs, a consistency judgment threshold is determined based on the point cloud density distribution characteristics. When the spatial structure difference of a point pair is less than this threshold, the point pair is determined to have spatial compatibility.

6. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 1, characterized in that, When selecting a pivot, choose the top-ranked score from the second-order compatibility graph. K One pair of points serves as the pivot. K The value of 1 is determined based on the scale of the point cloud data; when filtering candidate triclusters, the top weights are retained for each pivot. K Two candidate three-point groups, K The value of 2 is determined based on the noise level of the point cloud data, and the final number of fixed triangular clusters generated is [number missing]. K 1 and K The product of 2.

7. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 1, characterized in that, The large point cloud segmentation model is built on the PointNet++ framework. This model obtains local point sets in the point cloud through hierarchical sampling, performs feature aggregation on the local point sets to obtain multi-scale geometric information, and then obtains structural features from local to global through feature step-by-step transfer. Based on the structural features, the model realizes the segmentation and extraction of the measured point cloud of the pre-installed component.

8. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 1, characterized in that, When discretizing the 3D design model into a design point cloud, the discretization density is kept consistent with the density of the measured point cloud to ensure the accuracy of the comparison between the design point cloud and the measured point cloud during subsequent alignment analysis. When aligning the design point cloud and the measured point cloud using the fixed three-point cluster algorithm, the goal is to make the deviation of the aligned point cloud less than a preset accuracy threshold. The preset accuracy threshold is determined according to the accuracy requirements of component installation.

9. The method for three-dimensional point cloud error detection and visualization guidance of structural components according to claim 1, characterized in that, In step S3, when visualizing using virtual reality technology, the deviation between the design point cloud and the measured point cloud is converted into a three-dimensional deviation vector. The direction and magnitude of the deviation vector are displayed in the virtual reality device with intuitive graphic markers. Step-by-step installation and adjustment guidelines are generated based on the deviation vector to guide construction personnel to complete the installation and adjustment of components according to the guidelines.

Images