A method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurement

Abstract

The application discloses a kind of anti-shielding three-dimensional scanning measurement field construction methods for large size measurement, belong to optical three-dimensional measurement technical field.The method is first based on the adaptive sampling of workpiece geometric characteristics and generates candidate marker point set;Secondly, a virtual measurement scene is constructed, and the visibility and observation quality of each point under multiple viewpoints are calculated by simulating the dynamic scanning process using ray tracing technology;Then, the measurement accuracy is quantified using the Fisher information matrix, and a layout optimization model is established to maximize the information gain while satisfying the coverage rate and engineering constraints;Finally, the optimal marker point subset is obtained by solving the model, and verification and output are performed.The application solves the problem of measurement interruption, low dot placement efficiency and difficulty in ensuring accuracy caused by occlusion during scanning of large-size complex components, and realizes the automated construction of high-robustness and high-precision measurement field.

Description

Technical Field

[0001] This invention belongs to the field of optical three-dimensional measurement technology, specifically relating to a method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements. Background Technology

[0002] In industrial inspection and manufacturing, high-precision 3D measurement of large-sized components is crucial for ensuring product quality and assembly accuracy. Tracking laser scanning systems are widely used in such tasks due to their combination of high precision and efficiency. These systems typically consist of a binocular vision tracker and a laser scanning head. Their working principle is as follows: reflective markers are first placed on the surface of the component under test. The tracker identifies these markers in real time to determine the relative pose of the scanning head and the component, thereby achieving accurate stitching and reconstruction of 3D data. Therefore, the markers, as the basic framework of the measurement field, directly determine the reliability, efficiency, and accuracy of the measurement task based on their placement.

[0003] However, existing marker placement methods have significant shortcomings when dealing with large components with complex structures and auxiliary tooling. Current research and practice mainly focus on the placement planning of markers in static measurement fields, or only optimize visibility coverage under a single viewpoint, failing to systematically solve the problem of temporal occlusion and accuracy coupling in moving scans. This is mainly reflected in the following aspects:

[0004] First, large-scale scanning involves long paths and complex fixtures, making dynamic occlusion issues particularly prominent. Current practices generally rely on static viewpoints for coverage analysis, lacking analysis of the workpiece's own geometric features and the obstruction of the optical path by the fixtures during the actual scanning process. This results in the tracking system being unable to simultaneously observe a sufficient number of marker points on critical scanning paths, thus failing to calculate pose, causing measurement interruptions, and even requiring manual shutdown to add markers, severely reducing measurement efficiency.

[0005] Secondly, there is a lack of adaptive sampling mechanisms for large-sized objects. Existing marker generation methods mostly adopt a uniform sampling strategy with a fixed density (such as Poisson disk sampling). For large-sized components, maintaining high-density sampling will lead to a surge in the number of markers, which will severely slow down the computational efficiency and increase the implementation cost; if the density is reduced, feature points in areas with drastic curvature changes may be lost, affecting the positioning accuracy.

[0006] Furthermore, existing marker placement methods have a singular optimization objective, focusing only on static visibility coverage while neglecting the accuracy requirements of large-scale measurement fields. An unreasonable spatial distribution of markers can lead to ill-conditioned observation equations, resulting in significant pose estimation errors. Existing methods typically fail to incorporate key indicators characterizing the lower bound of estimation accuracy into their optimization objectives, making it theoretically difficult to guarantee optimal measurement accuracy from their planning results.

[0007] Therefore, developing an anti-occlusion 3D scanning measurement field construction method that can comprehensively consider large-size adaptive sampling, dynamic multi-viewpoint visibility constraints, and maximization of measurement information is of great engineering application value for improving the stability and accuracy of automated inspection of large and complex components. Summary of the Invention

[0008] The technical problem to be solved:

[0009] To overcome the shortcomings of existing technologies, this invention provides an occlusion-resistant 3D scanning measurement field construction method for large-scale measurements. It generates a high-quality candidate set through multi-level adaptive sampling, performs dynamic visibility and quality pre-simulation using ray tracing, and finally achieves accuracy-oriented layout optimization through a Fisher information matrix model. This invention overcomes the shortcomings in the construction of large-scale 3D scanning measurement fields, such as the imbalance between the number of sampling points and computational efficiency, loss of line of sight due to dynamic occlusion, and the difficulty in ensuring global measurement accuracy with a single geometrically covered target.

[0010] The technical solution of this invention is: a method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements, comprising the following steps:

[0011] Step 1. Data Acquisition and Model Building: Acquire the 3D mesh model of the large workpiece under test and the preset scanning pose sequence, and determine the camera imaging model under each scanning pose based on the scanning pose sequence and the camera parameters of the binocular measurement system.

[0012] Step 2. Adaptive generation of candidate marker points: Based on the geometric features of the 3D mesh model, multi-level adaptive sampling is performed to generate a set of candidate marker points that cover the workpiece surface and whose density distribution matches the geometric complexity;

[0013] Step 3. Dynamic Visibility Simulation: Based on the camera imaging model, in a virtual measurement scene containing the workpiece and tooling fixtures, the dynamic scanning process is simulated using ray tracing technology. The visibility status and observation quality of each candidate marker point under all scanning poses are calculated, and a visibility matrix and a quality weight matrix are constructed.

[0014] Step 4. Construction of layout optimization model: Based on the visibility matrix, quality weight matrix and camera imaging model, a measurement accuracy evaluation model represented by Fisher information matrix is ​​constructed, and a marker layout optimization model is established with maximizing overall information gain as the core objective while satisfying coverage and engineering constraints.

[0015] Step 5. Solving and Verification: Solve the marker layout optimization model, select the optimal marker subset, verify and analyze the coverage and theoretical accuracy of the optimal marker subset, and generate the final anti-occlusion 3D scanning measurement field layout scheme.

[0016] A further technical solution of the present invention is: in step 2, the multi-level adaptive sampling specifically includes:

[0017] Calculate the diagonal length and total surface area of ​​the bounding box of a 3D mesh model;

[0018] Calculate the size scaling factor based on the diagonal length of the bounding box and the preset reference length, and calculate the area compensation factor based on the total surface area and the preset reference area.

[0019] The global adaptive sampling density is calculated by combining the preset baseline sampling density, size scaling factor, and area compensation factor.

[0020] Identify high curvature regions in a 3D mesh model whose curvature values ​​exceed a preset threshold;

[0021] Based on the global adaptive sampling density, a curvature weighting factor is superimposed on the high curvature region to form a local density field;

[0022] Based on the local density field, a set of candidate marker points that satisfy the minimum spacing constraint is generated using a weighted Poisson disk sampling algorithm.

[0023] A further technical solution of the present invention is: in step 3, the dynamic visibility simulation specifically includes:

[0024] Construct a hierarchical bounding box acceleration structure for the triangular mesh model of the workpiece and tooling fixture;

[0025] For each scan pose, traverse the candidate marker set, and for each candidate marker, calculate the line-of-sight vector from the optical centers of the left and right cameras to that candidate marker.

[0026] Use the hierarchical bounding box to perform ray intersection detection on each view vector. If the ray intersects with any face in the scene before reaching the candidate marker point, it is determined to be occluded.

[0027] If the candidate marker is not obstructed by the line of sight of the left and right cameras in a scanning pose, and its working distance to the camera is within the preset depth of field and the angle of incidence of the line of sight is less than the preset maximum value, it is determined to be validly visible.

[0028] Based on the incident angle, observation distance, and binocular intersection angle of the candidate marker, calculate its comprehensive observation quality weight, and summarize the visibility status and quality weight of all scan poses and candidate markers to construct a visibility matrix and a quality weight matrix.

[0029] A further technical solution of the present invention is: in step 4, constructing the measurement accuracy evaluation model represented by the Fisher information matrix specifically involves:

[0030] In response to the The visible first scan pose For each candidate marker point, calculate its image reprojection error with respect to the six degrees of freedom parameters of that pose using the Jacobian matrix. The expression is as follows:

[0031]

[0032] In the formula, For the camera's six-DOF pose perturbation parameters, For the first The three-dimensional spatial coordinates of the candidate marker points in the world coordinate system For the first The camera's external homogeneous transformation matrix for each scan pose. Represents a binocular perspective projection function based on a camera model, which projects from three-dimensional spatial coordinates to two-dimensional image plane pixel coordinates.

[0033] Based on the observation quality weight of this point Construct its observation noise covariance matrix The expression is as follows:

[0034]

[0035] In the formula, Measure the noise standard deviation for the reference image. It is the identity matrix;

[0036] For any given subset of candidate markers In the Fisher information matrix under each scanning pose It is obtained by summing the information contributions of all visible points, and the expression is as follows:

[0037]

[0038] In the formula, For candidate labeled subsets In the middle of A subset of locally visible marker points that satisfy visibility constraints under a given scan pose; For candidate labeled subsets In the Fisher information matrix under each scan pose; This is a regularization term used to prevent matrix singularities due to insufficient visible points, ensuring the stability of numerical computation. is the regularization coefficient.

[0039] A further technical solution of the present invention is: in step 4, the established marker layout optimization model is as follows:

[0040] The decision variable for optimization is defined as an optimally selected subset of entities. The optimization objective is to maximize the sum of the natural logarithmic determinants of the Fisher information matrix under all scan poses, as expressed below:

[0041]

[0042] In the formula, This represents the mathematical objective of the optimization process, namely, solving for the objective function. The optimal selected subset of entities that achieves the maximum value ; For the optimal selected entity subset The global information content evaluation function; This represents the total number of scan poses in the preset scan path; The determinant operator of a matrix; To select the optimal entity subset In the Fisher information matrix under each scan pose;

[0043] And satisfy the following constraints:

[0044] (a) Coverage constraint: For each scan pose, at least one The selected points are visible;

[0045]

[0046] (b) Physical distance constraint: The Euclidean distance between any two selected points is not less than [value missing]. ;

[0047]

[0048] (c) Total number constraint: The total number of selected points does not exceed ;

[0049]

[0050] In the formula, The minimum number of visible points for a single scan pose is set to [value] for solving the six-DOF spatial pose. ; For the first Under each scan pose, the global visibility matrix A subset of locally valid visible marker points that satisfy the constraint. ; To ensure a safe distance for physical application, the distance should be set to be slightly larger than the physical diameter of a single marker point. Representing a subset The Middle The and the first Euclidean distance between the marked points; For the first The three-dimensional spatial coordinates of the candidate marker points in the world coordinate system; The maximum allowable number of measurement points for the entire measurement field is set by combining component size and algorithm performance.

[0051] A further technical solution of the present invention is: in step 5, a greedy algorithm is used to solve the marker point layout optimization model, including:

[0052] Initialize the selected point set to empty;

[0053] Iterative execution: Calculate the information gain increment brought about by adding each candidate marker point to the current point set;

[0054] Candidate markers that maximize information gain increment and do not violate conflict constraints are preferentially selected to be added to the point set;

[0055] Once all coverage constraints are met, continue selecting the point with the largest information gain until the total amount constraint is reached or the gain falls below the threshold.

[0056] A further technical solution of the present invention is: in step 5, the theoretical accuracy verification specifically includes:

[0057] Based on the optimal subset of marked points, calculate the local Fisher information matrix for each scan pose on the scan path;

[0058] Calculate the inverse of each local Fisher information matrix as the theoretical covariance matrix of the corresponding scan pose;

[0059] The position estimation variance and attitude estimation variance corresponding to the scanning pose are extracted from each theoretical covariance matrix and used as the lower bound of the theoretical accuracy of the measurement field.

[0060] A further technical solution of the present invention is that step 5 further includes performing Monte Carlo simulation verification:

[0061] Add Gaussian random noise to the simulated image observation coordinates;

[0062] A certain proportion of visible marker points are randomly blocked to simulate sudden occlusion or marker point failure in actual measurements.

[0063] Multiple pose calculations were performed using a Gaussian probability distribution model with zero mean, and the error distribution between the calculation results and the true pose was statistically analyzed.

[0064] The robustness of the measurement field layout scheme under actual disturbances is evaluated based on the error distribution.

[0065] An occlusion-resistant 3D scanning measurement field construction device for large-size measurements includes:

[0066] One or more processors;

[0067] Memory, which stores computer program instructions;

[0068] When computer program instructions are executed by the processor, the device performs a method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements.

[0069] An occlusion-resistant 3D scanning measurement field construction system for large-size measurements includes:

[0070] The data acquisition and model building module is used to acquire the three-dimensional mesh model of the large workpiece under test and the preset scanning pose sequence, and to determine the camera imaging model under each scanning pose based on the scanning pose sequence and the camera parameters of the binocular measurement system.

[0071] The candidate marker adaptive generation module, connected to the data acquisition and model building module, is used to perform multi-level adaptive sampling based on the geometric features of the 3D mesh model to generate a set of candidate markers that cover the workpiece surface and whose density distribution matches the geometric complexity.

[0072] The dynamic visibility simulation module is connected to the data acquisition and model building module and the candidate marker adaptive generation module, respectively. It is used to simulate the dynamic scanning process in a virtual measurement scene containing workpieces and tooling fixtures, based on the camera imaging model and using ray tracing technology, to calculate the visibility status and observation quality of each candidate marker in all scanning poses, and to construct the visibility matrix and quality weight matrix.

[0073] The layout optimization model building module, connected to the dynamic visibility simulation module, is used to build a measurement accuracy evaluation model represented by the Fisher information matrix based on the visibility matrix, the quality weight matrix, and the camera imaging model, and to establish a marker layout combination optimization model with the core objective of maximizing overall information gain while satisfying coverage and engineering constraints.

[0074] The solution and verification module is connected to the layout optimization model construction module. It is used to solve the layout optimization model to select the optimal subset of marker points, and to verify and analyze the coverage and theoretical accuracy of the optimal subset of marker points, and generate the final anti-shading measurement field layout scheme.

[0075] Beneficial effects

[0076] The beneficial effects of this invention are as follows: By constructing an adaptive sampling model based on geometric features and dynamically adjusting the sampling density according to the model size and curvature, this invention improves the problem of uneven distribution and redundancy of marker points on the surface of large-sized workpieces; by combining ray tracing technology for multi-viewpoint visibility calculation, it effectively eliminates invalid observation points caused by workpiece self-occlusion and mutual occlusion of tooling fixtures during dynamic scanning, thus improving the integrity of the measurement field; simultaneously, by using the Fisher information matrix to optimize the marker point layout, it enhances the geometric constraints on camera scanning pose while satisfying coverage constraints, thereby improving the pose calculation accuracy and anti-interference capability of the 3D scanning system, such as... Figure 6 and Figure 7 As shown. The specific effects are reflected in the following aspects:

[0077] First, this invention employs a multi-level adaptive sampling mechanism based on model geometric features (size, surface area, curvature) to dynamically adjust the generation density of candidate marker points according to the actual shape complexity of the workpiece. This not only overcomes the computational explosion or feature loss problems caused by uniform sampling on large-sized workpieces, but also significantly reduces the number of invalid candidate marker points.

[0078] Secondly, this invention constructs a virtual scene containing the workpiece and tooling fixtures, and utilizes ray tracing technology to accurately determine the visibility and quality of all candidate marker points at every pose along the preset scanning path. This allows for the early identification and elimination of points that are invisible due to self-occlusion or tooling fixture obstruction. This ensures that the final selected set of marker points provides stable and sufficient visual observation throughout the dynamic scanning process, fundamentally avoiding measurement interruptions caused by loss of line of sight and reducing reliance on manual intervention.

[0079] Third, this invention innovatively introduces the Fisher information matrix as a quantitative evaluation tool for the accuracy of the measurement field, and constructs an optimization model based on this matrix with the goal of maximizing information gain. This ensures that the selection of marker points not only meets the basic requirement of "visibility coverage," but also optimizes the geometric constraint strength of the spatial distribution of marker points on camera pose parameters from the perspective of estimation theory.

[0080] Fourth, this invention not only outputs the optimal point layout scheme, but also quantitatively verifies the coverage performance and disturbance resistance robustness of the scheme through theoretical accuracy lower bound calculation and Monte Carlo simulation experiments. This provides reliable theoretical and data support for the practical application of the scheme, enabling users to predict and evaluate the performance of the measurement field in advance and reducing the risks of on-site implementation. Attached Figure Description

[0081] Figure 1 This is the main flowchart of the anti-occlusion three-dimensional scanning measurement field construction method for large-size measurement according to an embodiment of the present invention;

[0082] Figure 2 This is a schematic diagram of the adaptive sampling effect based on geometric features in an embodiment of the present invention;

[0083] Figure 3 This is a schematic diagram of a virtual assembly scene in an embodiment of the present invention;

[0084] Figure 4 This is a schematic diagram illustrating the principle of visibility calculation and occlusion culling based on ray tracing in an embodiment of the present invention;

[0085] Figure 5 This is a heatmap verifying the coverage of marker points under the multi-view scanning path in this embodiment of the invention;

[0086] Figure 6 This is an analysis result diagram of the selected marker layout scheme in the embodiments of the present invention;

[0087] Figure 7 This is a graph showing the analysis results of the comparative experiment in the embodiments of the present invention.

[0088] Explanation of the attached labels: 1. Workpiece to be measured, 2. Camera pose, 3. Fixture, 4. Marker point to be observed. Detailed Implementation

[0089] The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention.

[0090] When dealing with large components with complex structures and auxiliary tooling, existing technologies can be mainly categorized into the following three types:

[0091] First, current research focuses on the layout planning of static or quasi-static measurement fields (such as CN113536496A), or only optimizes visibility under a single viewpoint (such as CN117804377A). It fails to systematically simulate and optimize the temporal visibility of marker points on the entire preset scanning path sequence, resulting in poor robustness of the planning scheme in actual dynamic scanning.

[0092] Second, for workpieces with complex structures and large dimensions, using uniform sampling (such as uniform point cloud discretization in CN116341327A) or simple random sampling often leads to two drawbacks: first, generating a massive number of candidate marker points to cover the entire surface results in a heavy computational burden; second, ignoring surface curvature variations leads to insufficient sampling in key feature-rich areas, affecting local positioning accuracy. Existing methods fail to dynamically adjust the sampling density based on the model's macroscopic dimensions, surface area, and local curvature characteristics, making it difficult to achieve a balance between computational efficiency and feature fidelity.

[0093] Third, most methods limit their optimization objectives to maximizing geometric coverage or minimizing the number of devices (e.g., CN104154859A, CN120651141A), neglecting the impact of the spatial distribution of marker points on the accuracy of subsequent visual pose estimation algorithms. An unreasonable marker point layout can lead to ill-conditioned observation equations, potentially resulting in significant pose estimation errors even if coverage requirements are met. Furthermore, existing technologies generally do not incorporate indices characterizing the lower bound of theoretical accuracy in parameter estimation into their optimization models, making it difficult to theoretically guarantee optimal measurement accuracy from the planning results.

[0094] In summary, although existing technologies have made progress in areas such as measurement equipment networking, scanning path planning, and viewpoint optimization, a systematic solution has yet to be developed for addressing the challenges of anti-occlusion, high precision, and adaptive marker placement during dynamic scanning of large and complex components. Therefore, there is an urgent need to propose a measurement field construction method that integrates adaptive sampling of geometric features, dynamic multi-viewpoint occlusion analysis, and accuracy optimization guided by information theory, in order to improve the robustness, accuracy, and efficiency of automated 3D measurement of large components.

[0095] Based on the above problem analysis, this invention provides a method for constructing an occlusion-resistant 3D scanning measurement field for large-size measurements. This method performs multi-level adaptive sampling based on the geometric features of the object being measured, utilizes ray tracing technology to solve the visibility calculation problem under dynamic multi-viewpoint conditions, and introduces a Fisher information matrix to construct an information gain-oriented optimization model. This achieves automated, high-precision, and occlusion-resistant design of the measurement field marker layout. The specific steps are as follows:

[0096] Step 1. Data Acquisition and Model Building: Acquire the 3D mesh model of the large workpiece under test and the preset scanning pose sequence, and determine the camera imaging model under each scanning pose based on the scanning pose sequence and the camera parameters of the binocular measurement system.

[0097] Step 2. Adaptive generation of candidate marker points: Based on the geometric features of the 3D mesh model, multi-level adaptive sampling is performed to generate a set of candidate marker points that cover the workpiece surface and whose density distribution matches the geometric complexity;

[0098] Step 3. Dynamic Visibility Simulation: Based on the camera imaging model, in a virtual measurement scene containing the workpiece and tooling fixtures, the dynamic scanning process is simulated using ray tracing technology. The visibility status and observation quality of each candidate marker point under all scanning poses are calculated, and a visibility matrix and a quality weight matrix are constructed.

[0099] Step 4. Construction of layout optimization model: Based on the visibility matrix, quality weight matrix and camera imaging model, a measurement accuracy evaluation model represented by Fisher information matrix is ​​constructed, and a marker layout optimization model is established with maximizing overall information gain as the core objective while satisfying coverage and engineering constraints.

[0100] Step 5. Solving and Verification: Solve the marker layout optimization model, select the optimal marker subset, verify and analyze the coverage and theoretical accuracy of the optimal marker subset, and generate the final anti-occlusion 3D scanning measurement field layout scheme.

[0101] This invention also proposes an anti-occlusion three-dimensional scanning measurement field construction device for large-size measurements, comprising:

[0102] One or more processors;

[0103] Memory, which stores computer program instructions;

[0104] When computer program instructions are executed by the processor, the device performs a method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements.

[0105] This invention also proposes an anti-occlusion three-dimensional scanning measurement field construction system for large-size measurements, comprising:

[0106] The data acquisition and model building module is used to acquire the three-dimensional mesh model of the large workpiece under test and the preset scanning pose sequence, and to determine the camera imaging model under each scanning pose based on the scanning pose sequence and the camera parameters of the binocular measurement system.

[0107] The candidate marker adaptive generation module, connected to the data acquisition and model building module, is used to perform multi-level adaptive sampling based on the geometric features of the 3D mesh model to generate a set of candidate markers that cover the workpiece surface and whose density distribution matches the geometric complexity.

[0108] The dynamic visibility simulation module is connected to the data acquisition and model building module and the candidate marker adaptive generation module, respectively. It is used to simulate the dynamic scanning process in a virtual measurement scene containing workpieces and tooling fixtures, based on the camera imaging model and using ray tracing technology, to calculate the visibility status and observation quality of each candidate marker in all scanning poses, and to construct the visibility matrix and quality weight matrix.

[0109] The layout optimization model building module, connected to the dynamic visibility simulation module, is used to build a measurement accuracy evaluation model represented by the Fisher information matrix based on the visibility matrix, the quality weight matrix, and the camera imaging model, and to establish a marker layout combination optimization model with the core objective of maximizing overall information gain while satisfying coverage and engineering constraints.

[0110] The solution and verification module is connected to the layout optimization model construction module. It is used to solve the layout optimization model to select the optimal subset of marker points, and to verify and analyze the coverage and theoretical accuracy of the optimal subset of marker points, and generate the final anti-shading measurement field layout scheme.

[0111] The following is in conjunction with the appendix Figure 1-7 The technical solution of the present invention will be further described in detail below.

[0112] In one embodiment, refer to Figure 1 As shown in the figure, this embodiment presents a method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements. The construction method includes the following steps:

[0113] Step 1: Obtain the 3D mesh model of the large workpiece to be measured and the preset scanning pose sequence, and perform data preprocessing;

[0114] The specific steps of step 1 are as follows:

[0115] S101: Obtain the three-dimensional mesh model of the workpiece being tested.

[0116] In this embodiment, the read 3D mesh model is in STL format.

[0117] S102: Through bounding box size analysis, the vertex coordinates of the 3D mesh model are uniformly converted into standard units (millimeters mm).

[0118] In this embodiment, the input model is a cylindrical part, and its volume frame is... The tooling fixture is a simplified grid plate with a shield.

[0119] S103: Remove duplicate vertices and degenerate patches, and use the Laplacian smoothing algorithm to remove scan noise, ensuring the smoothness of the model surface and avoiding subsequent curvature calculation errors due to poor mesh quality.

[0120] S104: Detects whether the mesh is watertight. For non-watertight meshes, it automatically fills in tiny holes and flips erroneous normal vectors to ensure that the normal vectors of all faces point uniformly to the outside of the model.

[0121] S105: Calculate the axis-aligned bounding box of the 3D mesh model of the workpiece under test, and extract the diagonal length of the bounding box. and the total surface area of ​​the three-dimensional mesh model of the workpiece being measured. .

[0122] In this embodiment, the calculation is obtained , .

[0123] S106: Obtain and parse the scan pose sequence. The scan pose sequence contains... Each scan pose is a discrete scan pose. Each scan pose includes the position of the scanner's optical center in the world coordinate system. and attitude quaternions The system converts the input scan pose sequence into a representation of the first... Each scan pose Camera external homogeneous transformation matrix By combining the intrinsic parameter matrix of the binocular camera and the baseline distance between the left and right cameras, a complete camera imaging model is constructed, providing optical center coordinates and line-of-sight direction for subsequent ray tracing visibility calculations.

[0124] In this embodiment, the trajectory data is stored in NumPy format. The imported scan path file contains 42 key scan poses, covering the top surface, side surface and main feature areas of the workpiece, simulating the process of the part being placed horizontally on the ground and the measuring device collecting marker points along the central axis.

[0125] Step 2: Based on the geometric features of the three-dimensional mesh model, perform adaptive sampling to generate a set of candidate marker points covering the workpiece surface.

[0126] The specific steps of step 2 are as follows:

[0127] S201: Calculate the scaling factor based on the diagonal length of the bounding box and a preset reference length. The area compensation factor is calculated based on the total surface area and the preset reference area. Combined with the preset benchmark density Calculate the global adaptive sampling density The calculation formula is as follows:

[0128]

[0129] Among them, the size scaling factor It is negatively correlated with the model scale and is used to prevent overloading of sampling points in large-sized models; area compensation factor Used to compensate for insufficient sampling in small-area models.

[0130] In this embodiment, the preset reference density for That is, approximately 500 points per square meter.

[0131] In this embodiment, the size scaling factor Based on the diagonal length of the model bounding box With preset reference length The ratio is determined. The calculation formula is:

[0132]

[0133] In the formula, These are the maximum and minimum value operators, used to calculate the size scaling factor. Limit the values ​​to the range of [0.5, 1.5] to prevent excessive sparsity or density.

[0134] In this embodiment, the area compensation factor Based on the total surface area of ​​the model Compared with the preset reference area The ratio is determined and used for density compensation of models with abnormally small surface areas but complex features. The calculation formula is:

[0135]

[0136] In the formula, For adjustment coefficients, To prevent small quantities with a denominator of zero, This is the lower bound of the compensation factor.

[0137] In this embodiment, the adjustment coefficient Take 0.1;

[0138] S202: Construct a local density field based on curvature weights, superimposing local curvature weights on top of the global density. Traverse the vertices of the 3D mesh model, for any mesh vertex... Curvature weight function Defined as:

[0139]

[0140] In the formula, For grid vertices The local curvature weights after normalization. For curvature enhancement ratio, To activate the curvature threshold of the weights, As a transition smoothing factor, This represents an exponential function with the natural constant as its base. This formula makes the weights in flat regions approach 1.0, while the weights in high-curvature regions smoothly transition to... .

[0141] In this embodiment, Set to 2.0. Set it to 0.15.

[0142] S203: Based on the constructed local density field, a candidate marker point set is generated using a weighted Poisson disk sampling algorithm. The generation method is as follows: initialize an empty candidate marker point set. Subsequently, an initial seed point is randomly generated in the region with a high density field on the mesh surface. Add it In the middle. And based on the local density at that point, a minimum distance constraint radius is set. The calculation formula is:

[0143]

[0144] In the formula, The distribution coefficient is used to control the compactness of the point cloud distribution. for The continuous interpolation curvature weights at the point are due to the newly generated For a triangle located inside a 3D mesh, interpolation is performed using the curvature weights of the three vertices of that triangle. The calculation formula is as follows:

[0145]

[0146] In the formula, They represent enclosed areas respectively. The three vertices of the current triangular face. Representing points respectively The three centroid coordinate components within the triangular facet satisfy nonnegativity and normalization constraints.

[0147] New seed points are continuously generated on the mesh surface. And calculate its corresponding ,like and If the Euclidean distance of all existing sample points is strictly greater than the constraint radius, then add it to the list. Repeat the above process until no new sample points satisfying the constraints can be added to the model surface.

[0148] In this embodiment, the distribution coefficient Take 0.8.

[0149] S204: Generate the candidate marker point set data structure.

[0150] In this embodiment, refer to Figure 2 As shown, the final generated candidate marker set It contains 7353 points, and the three-dimensional coordinates of these points are... and its corresponding normal vector Store it as a KD-Tree structure to facilitate rapid nearest neighbor search and distance constraint determination in subsequent steps.

[0151] Step 3: Construct a virtual measurement scene containing the workpiece and tooling fixtures, use ray tracing algorithm to calculate the visibility status and observation quality of the candidate marker point set under each scanning pose, and construct a visibility matrix.

[0152] The specific steps of step 3 are as follows:

[0153] S301: Construct an accelerated ray intersection structure based on hierarchical bounding boxes. Import the workpiece mesh model processed in step 1 and the tooling fixture model into the same coordinate system, such as... Figure 3 As shown, using the Open3D engine, hierarchical bounding boxes are constructed for all triangles in the scene. This structure recursively divides the space into hierarchical bounding box nodes, reducing the time complexity of ray intersection from... Reduce to Based on the parameters of the binocular structured light scanner, the optical center positions and field of view of the left and right cameras are defined.

[0154] In this embodiment, the parameters of the binocular structured light scanner are: camera baseline distance. Optimal working distance Effective depth of field range ,in This represents the scanner's near-field depth of field limit. This represents the depth-of-field limit for the scanner.

[0155] S302: Performs bidirectional ray tracing and occlusion culling. Traverses each discrete pose in the scan path. For each candidate marker point Perform visibility determination. i The index of the candidate marker, such as Figure 4 As shown. First, calculate the absolute 3D coordinates of the optical centers of the left and right cameras respectively. Then, calculate the points pointing from the left and right optical centers to the candidate marker points respectively. Direction vector and Then, a ray is emitted from the camera's optical center along the line-of-sight vector. A hierarchical bounding box structure is used to detect whether the ray intersects with the scene model. If the distance between the ray and any facet intersection point is less than the distance from the optical center to each candidate marker point... If the distance is less than a certain value, the line of sight is considered obstructed. Only when the lines of sight of both the left and right cameras are not obstructed is the point marked as geometrically visible in the current pose.

[0156] In this embodiment, a tolerance for line-of-sight occlusion detection is set. .

[0157] S303: Verify whether the candidate markers meet the physical imaging constraints of the scanner. The physical imaging constraints include working distance constraints, incident angle constraints, and field of view constraints.

[0158] In this embodiment, the working distance constraint is set as follows: calculate candidate marker points. Projection distance to the camera plane .like If it is not visible (outside the depth of field), then it is determined to be invisible.

[0159] In this embodiment, the incident angle constraint is set as follows: calculate the direction vectors respectively. , With candidate markers normal vector The included angles are denoted as the left angle of incidence. and right angle of incidence .like or If it is not visible, then it is determined to be invisible.

[0160] In this embodiment, the maximum incident angle threshold Set as .

[0161] In this embodiment, the field of view constraint is set as follows: candidate marker points If the coordinates are projected onto the camera image plane and exceed the image resolution range, then the image is determined to be invisible.

[0162] S304: For the current... Candidate markers identified as visible under each scan pose Calculate its observation quality weight Construct a quality weight matrix The formula for calculating the observation quality weight is:

[0163]

[0164] in, For weights based on the incident angle, Distance weights are based on a Gaussian distribution. This refers to the quality weighting of triangulation based on binocular intersection angle.

[0165] In this embodiment, the formula for calculating the incident angle weight is:

[0166]

[0167] in, For the current number The average incident angle of binoculars under each scanning pose, i.e. The smaller the incident angle, the higher the reflection intensity of the marker point, and the greater its weight.

[0168] In this embodiment, the distance weight calculation formula is as follows:

[0169]

[0170] in, Candidate markers Up to the current number Spatial observation distance of the scanner measurement center (i.e., the midpoint of the left and right camera baselines) under each scanning pose. The standard deviation constant is used to control the penalty sensitivity. The closer the observation distance is to the optimal working distance... The clearer the image, the greater the weight.

[0171] In this embodiment, the triangulation mass weighting Based on binocular intersection angle ,in Defined as the spatial angle formed by the optical center of the left camera, the candidate marker point, and the optical center of the right camera, based on the error propagation model of binocular triangulation, the optimal range of the intersection angle is related to the camera baseline distance. The working distance is dynamically determined. The optimal reference intersection angle under the current measurement system configuration is set as... ,structure The calculation formula is:

[0172]

[0173] In the formula, The actual intersection angle. To control the standard deviation parameter of the intersection angle sensitivity, Determined by system calibration parameters.

[0174] S305: Generate the visibility matrix data structure, namely: binary visibility matrix Record the visibility state of all candidate marker points, and its matrix elements Defined as: if candidate marker point In the Effectively visible under each scan pose, then ,otherwise Floating-point mass weight matrix Record the corresponding observation quality weights, whose matrix elements are defined as follows: for valid visible points, the value is the observation quality weight calculated in step S304. For invisible points, its value is zero.

[0175] Step 4: Based on the visibility matrix and camera imaging model, construct the Fisher information matrix for evaluating measurement accuracy, and establish a marker layout optimization model with the goal of maximizing information gain and satisfying coverage constraints.

[0176] The specific steps of step 4 are as follows:

[0177] S401: Based on the camera model, define a binocular perspective projection function that transforms 3D spatial coordinates to 2D image plane pixel coordinates. For any candidate marker point that is determined to be visible in the k-th scan pose... Calculate the reprojection error with respect to the camera's six-DOF pose perturbation parameters. Jacobian matrix The calculation formula is:

[0178]

[0179] In the formula, For the first The three-dimensional spatial coordinates of the candidate marker points in the world coordinate system For the first The camera's external homogeneous transformation matrix for each scanning pose;

[0180] This matrix describes how small pose perturbations cause changes in image observations and is the basis for evaluating the strength of geometric constraints.

[0181] In this embodiment, the matrix is ​​calculated using an analytical method.

[0182] S402: Observation quality weights calculated in step 3 Construct a non-uniform observation noise covariance matrix The calculation formula is:

[0183]

[0184] In the formula, the standard deviation of the measurement noise of the reference image is: , This is the identity matrix. The formula indicates the observation quality weights. The higher the value, the lower the corresponding measurement uncertainty, and the greater the contribution weight of that point in subsequent optimization.

[0185] In this embodiment, the standard deviation of the reference image measurement noise is... Set to 0.05 pixels.

[0186] S403: For any given subset of candidate markers Calculate its in the first Fisher information matrix under each scanning pose According to the principle of information additivity, this matrix is ​​the sum of the information matrices of all visible points, and the calculation formula is:

[0187]

[0188] In the formula, For candidate labeled subsets In the middle of A subset of locally visible marker points that satisfy visibility constraints under a given scan pose. For candidate labeled subsets In the Fisher information matrix under each scan pose This is a regularization term used to prevent matrix singularities due to insufficient visible points, thus ensuring the stability of numerical computation.

[0189] In this embodiment, take .

[0190] S404: Establish a marker point layout combination optimization model with the goal of maximizing the overall positioning accuracy of the measurement field and meeting engineering implementation constraints. The following mathematical model is constructed:

[0191] Design variable: An optimal selected subset of entities , representing the final combination of physical marker points selected by the algorithm that satisfies .

[0192] Objective function: The D-optimal criterion is adopted, which is to maximize the sum of the logarithms of the determinants of the Fisher information matrix under all scan poses. This index corresponds to minimizing the uncertainty ellipsoid volume of the pose estimation, and the calculation formula is as follows:

[0193]

[0194] In the formula, This represents the mathematical objective of the optimization process, namely, solving for the objective function. The optimal selected subset of entities that achieves the maximum value , For the optimal selected entity subset The global information content evaluation function, This represents the total number of scan poses in the preset scan path. The determinant operator of a matrix.

[0195] Constraints:

[0196] Coverage constraint: Ensure that at least one coverage point is present in each pose. A number of effective observation points are established to avoid tracking loss.

[0197]

[0198] Physical distance constraint: The Euclidean distance between any two selected points is not less than [value missing]. To avoid overlapping markers that make placement difficult, the calculation formula is as follows:

[0199]

[0200] Total Limitation: To control implementation costs, the total number of selected points shall not exceed [a certain limit]. The calculation formula is:

[0201]

[0202] In the formula, The minimum number of visible points for a single scan pose is set to [value] for solving the six-DOF spatial pose. ; For the first Under each scan pose, the global visibility matrix A subset of locally valid visible marker points that satisfy the constraint. ; To ensure a safe distance for physical application, the distance should be set to be slightly larger than the physical diameter of a single marker point. Representing a subset The Middle The and the first Euclidean distance between the marked points For the first The three-dimensional spatial coordinates of the candidate marker points in the world coordinate system; The maximum allowable number of measurement points for the entire measurement field is set by combining component size and algorithm performance.

[0203] In this embodiment, , , .

[0204] Step 5: Solve the marker layout optimization model, select the optimal subset of marker points, verify the measurement field coverage and accuracy, and generate the final marker layout scheme.

[0205] The specific steps of step 5 are as follows:

[0206] S501: An improved greedy algorithm is used to solve the marker layout optimization model. An optimal selected entity subset is defined. Candidate marker set For all points generated in step 2, initialize the current information matrix. ;for Each candidate marker point Calculate and add it Increment of the post-objective function Based on the matrix determinant lemma, the incremental calculation can be simplified to:

[0207]

[0208] Calculate the initial gain of all candidate markers. This is used to construct a descending priority queue. Each time, the candidate marker with the maximum gain at the head of the queue is popped. Verify its compatibility with the optimal selected entity subset. Any distance within the range. If the physical anti-interference constraint is satisfied and the gain is higher than the threshold, then add the candidate marker point. Then update the information matrix corresponding to the pose, and repeat until the number of selected points reaches the maximum number of points. Or the loop is empty.

[0209] Traversing the binary visibility matrix In response to insufficient number of effective observation points The pose, while satisfying the anti-interference distance Under the premise, from Supplementing visible points to Continue until all poses satisfy the global coverage constraint, then output the finalized marker layout scheme. .

[0210] S502: Subset of final marked points obtained from the optimization solution Perform coverage performance analysis. Traverse all scan poses and count the number of valid visible points for each pose. Map the observed frequency of each marker point to the surface of the workpiece's 3D model to generate a coverage heatmap, such as... Figure 5 As shown.

[0211] This embodiment was ultimately selected. The results showed that the average coverage rate was 97.6%.

[0212] S503: Utilizing the final local information matrix Based on the Cramer-Robben lower bound theory, the final labeled subset is... The lower bound of the theoretical pose estimation accuracy is verified for each independent scan pose:

[0213] First calculate the... The theoretical covariance matrix corresponding to each scan pose The formula is:

[0214] .

[0215] Then, for this covariance matrix Divide the data into blocks, extract the translation covariance submatrices and rotation covariance submatrices on their diagonals, and calculate the first... The root mean square error of the position uncertainty of each scan pose and attitude uncertainty root mean square error The formula is:

[0216]

[0217] In the formula, The trace operator for a matrix; Represents the covariance matrix The translation block matrix in the top left corner, This represents the rotation block matrix in the lower right corner.

[0218] In this embodiment, the theoretical position error of all scanning poses is better than that of the target position. The attitude error is better than .

[0219] S504: Subset of final marked points To verify and evaluate the robustness of the measurement field under actual physical disturbances through Monte Carlo simulation, the specific approach involves conducting multiple independent simulation experiments. In each run, a uniform random dropout rate is mapped into the global binary visibility matrix to simulate sudden occlusion or marker contamination. Simultaneously, measurement noise following a Gaussian distribution is injected into the system. The position and attitude errors for each simulation are calculated based on a zero-mean Gaussian probability distribution model, where the standard deviation of this Gaussian distribution is proportional to the current measurement noise scale. Finally, Monte Carlo error statistical features, including the mean, standard deviation, and 95th / 99th quantiles, are extracted. For example... Figure 6 As shown, after introducing simulated random dropout and Gaussian measurement noise interference, the visible point count chart shows that the measurement field can still maintain a sufficient average effective number of observations; the Monte Carlo position error chart shows that the mean and standard deviation of the errors of multiple pose calculations remain at a low level (on the order of approximately 0.1 mm), and the upper limits of the 95th and 99th quantile errors, which represent severe working conditions, do not show serious divergence and are effectively enveloped within approximately 0.3 mm. This indicates that under the interference conditions set in this embodiment, the marker point layout scheme can suppress the error divergence caused by occlusion and noise to a certain extent, keeping the calculation results within a reasonable engineering error range, demonstrating a certain anti-interference ability and robustness.

[0220] To further verify the effectiveness of this invention in measuring field construction, this embodiment is based on the same large-scale model, with a limited upper limit on the total number of final selected marker points. Minimum visible point from a single viewpoint Under the same boundary conditions, the method of this invention was compared with two existing conventional deployment techniques in a simulation experiment. Comparative Example 1 only randomly and uniformly deployed points based on the geometric features of the workpiece surface, without introducing visual occlusion analysis. Comparative Example 2 introduced ray tracing analysis but did not combine it with Fisher information matrix to evaluate pose accuracy. The Monte Carlo simulation model described in S504 was used uniformly, and a dynamic random drop rate of 15% was applied to simulate on-site occlusion and contamination interference. The comparison results are as follows. Figure 7 As shown, under the same boundary conditions and interference environment, compared with Comparative Example 1 without occlusion analysis and Comparative Example 2 without Fisher information matrix, the candidate point size of the present invention is significantly reduced (from 50,000 points to 8,866 points), which reduces the optimization computation burden and shortens the computation time (about 41 seconds). Secondly, after introducing interference, the Fisher condition number of the system is significantly improved, and the Monte Carlo solution failure rate remains at 0, indicating that the spatial geometric constraints are more stable.

[0221] S505: The system uses a weighted scoring system based on coverage, accuracy, and Monte Carlo robustness characteristics to generate a final comprehensive evaluation and verification report. Upon successful verification, it automatically exports a detailed table of optimal marker point 3D coordinates (CSV format) and visual charts of coverage and accuracy analysis, assisting on-site personnel in quickly and accurately completing the standardized placement of marker points on the surface of large-sized workpieces.

[0222] In summary, this invention improves the problem of uneven distribution and redundancy of marker points on the surface of large workpieces by constructing an adaptive sampling model based on geometric features and dynamically adjusting the sampling density according to the model size and curvature. Combined with ray tracing technology for multi-view visibility calculation, it effectively eliminates invalid observation points caused by workpiece self-occlusion and mutual occlusion of tooling fixtures during dynamic scanning, thus improving the integrity of the measurement field. Simultaneously, by using the Fisher information matrix to optimize the marker point layout, it enhances the geometric constraints on camera pose while satisfying coverage constraints, thereby improving the pose calculation accuracy and anti-interference capability of the 3D scanning system.

[0223] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention.

Claims

1. A method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements, characterized in that, Includes the following steps: Step 1. Data Acquisition and Model Building: Acquire the 3D mesh model of the large workpiece under test and the preset scanning pose sequence, and determine the camera imaging model under each scanning pose based on the scanning pose sequence and the camera parameters of the binocular measurement system. Step 2. Adaptive generation of candidate marker points: Based on the geometric features of the 3D mesh model, multi-level adaptive sampling is performed to generate a set of candidate marker points that cover the workpiece surface and whose density distribution matches the geometric complexity; Step 3. Dynamic Visibility Simulation: Based on the camera imaging model, in a virtual measurement scene containing the workpiece and tooling fixtures, the dynamic scanning process is simulated using ray tracing technology. The visibility status and observation quality of each candidate marker point under all scanning poses are calculated, and a visibility matrix and a quality weight matrix are constructed. Step 4. Construction of layout optimization model: Based on the visibility matrix, quality weight matrix and camera imaging model, a measurement accuracy evaluation model represented by Fisher information matrix is ​​constructed, and a marker layout optimization model is established with maximizing overall information gain as the core objective while satisfying coverage and engineering constraints. Step 5. Solving and Verification: Solve the marker layout optimization model, select the optimal marker subset, verify and analyze the coverage and theoretical accuracy of the optimal marker subset, and generate the final anti-occlusion 3D scanning measurement field layout scheme.

2. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 1, characterized in that: Step 2, specifically, includes: Calculate the diagonal length and total surface area of ​​the bounding box of a 3D mesh model; Calculate the size scaling factor based on the diagonal length of the bounding box and the preset reference length, and calculate the area compensation factor based on the total surface area and the preset reference area. The global adaptive sampling density is calculated by combining the preset baseline sampling density, size scaling factor, and area compensation factor. Identify high curvature regions in a 3D mesh model whose curvature values ​​exceed a preset threshold; Based on the global adaptive sampling density, a curvature weighting factor is superimposed on the high curvature region to form a local density field; Based on the local density field, a set of candidate marker points that satisfy the minimum spacing constraint is generated using a weighted Poisson disk sampling algorithm.

3. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 2, characterized in that: Step 3, specifically, includes the dynamic visibility simulation: Construct a hierarchical bounding box acceleration structure for the triangular mesh model of the workpiece and tooling fixture; For each scan pose, traverse the candidate marker set, and for each candidate marker, calculate the line-of-sight vector from the optical centers of the left and right cameras to that candidate marker. Use the hierarchical bounding box to perform ray intersection detection on each view vector. If the ray intersects with any face in the scene before reaching the candidate marker point, it is determined to be occluded. If the candidate marker is not obstructed by the line of sight of the left and right cameras in a scanning pose, and its working distance to the camera is within the preset depth of field and the angle of incidence of the line of sight is less than the preset maximum value, it is determined to be validly visible. Based on the incident angle, observation distance, and binocular intersection angle of the candidate marker, calculate its comprehensive observation quality weight, and summarize the visibility status and quality weight of all scan poses and candidate markers to construct a visibility matrix and a quality weight matrix.

4. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 3, characterized in that: In step 4, the specific steps for constructing the measurement accuracy evaluation model represented by the Fisher information matrix are as follows: In response to the The visible first scan pose For each candidate marker point, calculate its image reprojection error with respect to the six degrees of freedom parameters of that pose using the Jacobian matrix. The expression is as follows: In the formula, For the camera's six-DOF pose perturbation parameters, For the first The three-dimensional spatial coordinates of the candidate marker points in the world coordinate system For the first The camera's external homogeneous transformation matrix for each scan pose. Represents a binocular perspective projection function based on a camera model, which projects from three-dimensional spatial coordinates to two-dimensional image plane pixel coordinates. Based on the observation quality weight of this point Construct its observation noise covariance matrix The expression is as follows: In the formula, Measure the noise standard deviation for the reference image. It is the identity matrix; For any given subset of candidate markers In the Fisher information matrix under each scanning pose It is obtained by summing the information contributions of all visible points, and the expression is as follows: In the formula, For candidate labeled subsets In the middle of A subset of locally visible marker points that satisfy visibility constraints under a given scan pose; For candidate labeled subsets In the Fisher information matrix under each scan pose; This is a regularization term used to prevent matrix singularities due to insufficient visible points, ensuring the stability of numerical computation. is the regularization coefficient.

5. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 4, characterized in that: In step 4, the established marker layout optimization model is as follows: The decision variable for optimization is defined as an optimally selected subset of entities. The optimization objective is to maximize the sum of the natural logarithmic determinants of the Fisher information matrix under all scan poses, as expressed below: In the formula, This represents the mathematical objective of the optimization process, namely, solving for the objective function. The optimal selected subset of entities that achieves the maximum value ; For the optimal selected entity subset The global information content evaluation function; This represents the total number of scan poses in the preset scan path; The determinant operator of a matrix; To select the optimal entity subset In the first Fisher information matrix under scanning pose; And satisfy the following constraints: (a) Coverage constraint: For each scan pose, at least one The selected points are visible; (b) Physical distance constraint: The Euclidean distance between any two selected points is not less than [amount missing]. ; (c) Total number constraint: The total number of selected points does not exceed ; In the formula, The minimum number of visible points for a single scan pose is set to [value] for solving the six-DOF spatial pose. ; For the first Under each scan pose, the global visibility matrix A subset of locally valid visible marker points that satisfy the constraint. ; To ensure a safe distance for physical application, the distance should be set to be slightly larger than the physical diameter of a single marker point. Representing a subset The Middle The and the first Euclidean distance between the marked points; For the first The three-dimensional spatial coordinates of the candidate marker points in the world coordinate system; The maximum allowable number of measurement points for the entire measurement field is set by combining component size and algorithm performance.

6. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 5, characterized in that: In step 5, a greedy algorithm is used to solve the marker layout optimization model, including: Initialize the selected point set to empty; Iterative execution: Calculate the information gain increment brought about by adding each candidate marker point to the current point set; Candidate markers that maximize information gain increment and do not violate conflict constraints are preferentially selected to be added to the point set; Once all coverage constraints are met, continue selecting the point with the largest information gain until the total amount constraint is reached or the gain falls below the threshold.

7. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 6, characterized in that: In step 5, the theoretical accuracy verification specifically involves: Based on the optimal subset of marked points, calculate the local Fisher information matrix for each scan pose on the scan path; Calculate the inverse of each local Fisher information matrix as the theoretical covariance matrix of the corresponding scan pose; The position estimation variance and attitude estimation variance corresponding to the scanning pose are extracted from each theoretical covariance matrix and used as the lower bound of the theoretical accuracy of the measurement field.

8. The method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements according to claim 7, characterized in that: Step 5 also includes performing Monte Carlo simulation verification: Add Gaussian random noise to the simulated image observation coordinates; A certain proportion of visible marker points are randomly blocked to simulate sudden occlusion or marker point failure in actual measurements. Multiple pose calculations were performed using a Gaussian probability distribution model with zero mean, and the error distribution between the calculation results and the true pose was statistically analyzed. The robustness of the measurement field layout scheme under actual disturbances is evaluated based on the error distribution.

9. A device for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements, characterized in that, include: One or more processors; Memory, which stores computer program instructions; When the computer program instructions are executed by the processor, the device performs the method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements as described in any one of claims 1-8.

10. A system for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements, used to implement the method for constructing an anti-occlusion three-dimensional scanning measurement field for large-size measurements as described in any one of claims 1-8; characterized in that, include: The data acquisition and model building module is used to acquire the three-dimensional mesh model of the large workpiece under test and the preset scanning pose sequence, and to determine the camera imaging model under each scanning pose based on the scanning pose sequence and the camera parameters of the binocular measurement system. The candidate marker adaptive generation module, connected to the data acquisition and model building module, is used to perform multi-level adaptive sampling based on the geometric features of the 3D mesh model to generate a set of candidate markers that cover the workpiece surface and whose density distribution matches the geometric complexity. The dynamic visibility simulation module is connected to the data acquisition and model building module and the candidate marker adaptive generation module, respectively. It is used to simulate the dynamic scanning process in a virtual measurement scene containing workpieces and tooling fixtures, based on the camera imaging model and using ray tracing technology, to calculate the visibility status and observation quality of each candidate marker in all scanning poses, and to construct the visibility matrix and quality weight matrix. The layout optimization model building module, connected to the dynamic visibility simulation module, is used to build a measurement accuracy evaluation model represented by the Fisher information matrix based on the visibility matrix, the quality weight matrix, and the camera imaging model, and to establish a marker layout combination optimization model with the core objective of maximizing overall information gain while satisfying coverage and engineering constraints. The solution and verification module is connected to the layout optimization model construction module. It is used to solve the layout optimization model to select the optimal subset of marker points, and to verify and analyze the coverage and theoretical accuracy of the optimal subset of marker points, and generate the final anti-shading measurement field layout scheme.

Images