Multi-view point cloud data quality enhancement method for aviation assembly gap measurement

Abstract

The invention provides a multi-view point cloud data quality enhancement method for aviation assembly gap measurement, and the method comprises the steps: enabling a line laser sensor to be fixed to a multi-view collection device, and collecting the point cloud of a to-be-measured gap from a plurality of views; converting the plurality of slit point clouds collected under the plurality of visual angles into the same coordinate system, and preprocessing the data; curvature features of the gap point cloud are calculated, and edge points, namely critical points, of the skin surfaces on the two sides of the gap are extracted; performing linear fitting on the critical point by adopting a least square method, and determining a gap direction; projecting the three preprocessed slit point clouds to a plane perpendicular to the slit direction along the slit direction; and merging the three slit point clouds, and carrying out denoising and down-sampling processing to obtain complete slit point cloud data at a certain point to be measured. According to the method, the problems of incomplete data acquisition under a single view angle and large accumulative error of traditional multi-view angle data acquisition are solved, the measurement quality of the gap data is optimized, and the details and characteristics of the gap data are enhanced.

Description

Technical Field

[0001] The present invention relates to the fields of three-dimensional digital detection and machinery, and particularly to a method for enhancing the quality of multi-view point cloud data for measuring aviation assembly gaps. Background Art

[0002] In the field of aviation manufacturing, the assembly quality of an aircraft directly affects the aerodynamic performance, structural strength, and safety of the entire aircraft. During the assembly process, the gap size and consistency between various components of the airframe are one of the important parameters for measuring the assembly quality. To ensure precise control of the gaps, high-precision measurement and analysis are required. However, technologies such as contact measurement and single-view laser scanning are often limited by measurement perspectives, data acquisition accuracy, and environmental factors, making it difficult to comprehensively obtain complete data of the gaps. Especially in the presence of complex curved surfaces, local occlusion, or highly reflective materials, the measurement results are easily affected by noise, which affects the stability and reliability of the data. Therefore, in response to the high-precision measurement requirements for aviation assembly gaps, there is an urgent need for a measurement method that can obtain complete point cloud data from multiple perspectives and improve data quality to enhance assembly accuracy and detection efficiency.

[0003] Currently, research on assembly gap measurement mainly focuses on methods such as single-view laser scanning, structured light scanning, and three-dimensional vision measurement. For example, a single-line laser scanner combined with a high-precision mechanical displacement platform can scan the gap area point by point or line by line to obtain measurement data. However, this method is limited by the fixed measurement angle and is prone to data loss when facing complex structures or occlusions. In addition, although three-dimensional structured light scanning can obtain relatively complete three-dimensional data, the measurement accuracy is difficult to guarantee due to lighting conditions and surface materials. In recent years, the multi-view point cloud fusion method has been developed, which attempts to obtain gap data from different angles through multiple sensors or mobile measurement devices and fuse them through coordinate transformation and data registration. However, these methods are prone to introducing cumulative errors during the data conversion process, and the research on optimizing the quality of point clouds is still relatively limited, resulting in certain uncertainties in the final measurement results. Therefore, how to optimize the multi-view point cloud acquisition method and enhance the quality of point clouds in subsequent data processing to improve the accuracy and integrity of gap measurement is an urgent problem to be solved in current research. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for enhancing the quality of multi-view point cloud data for measuring aviation assembly gaps, which solves the problem of poor data quality caused by incomplete data acquisition in a single view, optimizes the quality of point cloud data collected from multiple views, and enhances the details and features of gap data.

[0005] To achieve the above technical objectives, the present invention provides the following technical solutions:

[0006] A multi - perspective point cloud data quality enhancement method for aviation assembly gap measurement, which specifically includes the following steps:

[0007] S1. Fix the line laser sensor on the multi - perspective acquisition device, measure the gap to be measured from multiple perspectives, and obtain multiple gap point clouds;

[0008] S2. Convert the multiple obtained gap point clouds to the same coordinate system and pre - process the data;

[0009] S3. Calculate the curvature features of each pre - processed gap point cloud, and extract the edge points of the skin surfaces on both sides of each gap point cloud, denoted as critical points;

[0010] S4. Use the least - squares method to perform linear fitting on the extracted critical points, and determine the gap direction according to the direction of the fitting line;

[0011] S5. Project the multiple pre - processed gap point clouds along the gap direction onto a plane perpendicular to this direction;

[0012] S6. Merge the multiple projected gap point clouds, and perform outlier detection and down - sampling to obtain a complete and accurate gap point cloud at a certain point to be measured.

[0013] Further, step S1 specifically includes:

[0014] S11. Drive a single - degree - of - freedom link structure through a servo motor to drive the sensor to always rotate around the center of the seam to be measured from multiple perspectives;

[0015] S12. Input the angle required for the sensor to rotate at the control end, and collect multiple gap point clouds from different perspectives at the seam to be measured; take the perspective perpendicular to the gap plane as the middle perspective, and divide the left and right sides by α° deviation into multiple pairs of left and right perspectives; and at least select a set of three gap point clouds under the left, middle, and right perspectives as the basis for subsequent analysis and processing.

[0016] Further, step S2 specifically includes:

[0017] S21. Manually introduce an offset in the y - axis direction to make the gap point clouds obtained from the left, middle, and right perspectives form a dislocation in the y - axis direction. Denote the gap point clouds from the left, middle, and right perspectives after introducing the offset as L0, M0, and R0 respectively;

[0018] S22. According to the relative position relationship between the gap point clouds, register the gap point clouds from the left and right perspectives to the coordinate system of the middle perspective through coordinate transformation. Denote the registered gap point clouds as L0′, M0′, and R0′ respectively;

[0019] S23. Apply the radius filtering method to the registered seam point clouds to filter out the free points and outliers in the seam point clouds, and obtain the three preprocessed seam point clouds L, M, and R.

[0020] Further, step S3 specifically includes:

[0021] S31. Apply the K-means clustering algorithm to the left and right regions of each seam point cloud under the left, middle, and right viewpoints after preprocessing for point cloud clustering segmentation to obtain six independent subsets, denoted as L1, L2, M1, M2, R1, and R2 respectively;

[0022] S32. Check that the arrangement order of each point in each subset is arranged from far away from the seam to close to the seam. If the order is incorrect, the data points need to be re-sorted;

[0023] S33. Take subset L1 as an example. For each point in it Select its first k / 2 and last k / 2, a total of k neighborhood points, calculate the vectors between adjacent points, and find the average value of the angles between the vectors as the curvature at this point to reflect the degree of curvature of the local area near this point;

[0024] S34. Set the curvature threshold C th , traverse each point of subset L1, and find the point where the curvature first exceeds this threshold, which is the critical point of this subset, denoted as P L1 ; the area in front of the critical point is considered the skin surface, and the area of the critical point is considered the seam;

[0025] S35. The remaining five subsets are used to extract critical points in the same way; finally, the critical points of the corresponding point cloud regions of each subset are obtained, denoted as P L1 、P L2 、P M1 、P M2 、P R1 、P R2 .

[0026] Further, step S4 specifically includes:

[0027] S41. For the critical points P L1 、P L2 、P M1 、P M2 、P R1 、P R2 of the left and right regions of the seam point clouds from the left, middle, and right viewpoints, pair them up in pairs according to the same viewpoint to obtain three pairs of critical point pairs; calculate the midpoint for each pair of critical point pairs, denoted as P L 、P M 、P R; Fit the three midpoints with a straight line using the least squares method, and constrain the fitted straight line to pass through point P M ; Assume the equation of the fitting line is as follows:

[0028] r(t)=B+md;

[0029] Among them, r(t)=(x(t),y(t),z(t)) is any point on the line, d=(d x ,d y ,d z ) is the direction vector of the line, and m is the parameter;

[0030] S42. Calculate the midpoints of the three sets of critical points and obtain: Construct the offset matrix O to describe P L , P R Relative to P M The offset of O is expressed as:

[0031]

[0032] Then singular value decomposition SVD is used to solve the straight line direction vector, and the formula is expressed as:

[0033] U,S,V T =SVD(O);

[0034] Among them, U and V are left and right orthogonal matrices, S is a singular value matrix; the first column of V corresponds to the direction vector d of the least squares fitting line, which is the gap direction.

[0035] Furthermore, step S5 specifically includes:

[0036] S51, normalize the direction vector d to a unit vector d'=(d' x ,d′ y ,d′ z );

[0037] S52. Take the gap point cloud r from the right perspective as an example, find any point R i =(x i ,y i ,z i ) is the projection R in the direction of the unit vector d′ i,proj , the formula is:

[0038] R i,proj =(R i ·d′)d′;

[0039] Among them, R i ·d′=x i d′ x +yi d′ y +z i d′ z ;

[0040] Then the projection point on the plane perpendicular to the slit direction is R′ i =R i -R i,proj ;

[0041] S53, calculate the projection points of all points in R, and obtain the gap point cloud R′ projected onto the plane perpendicular to the gap direction; the projections under the middle and left perspectives are calculated in the same way, and finally the projected gap point clouds are L′, M′, and R′ respectively.

[0042] Furthermore, step S6 specifically includes:

[0043] S61, merge the projected gap point clouds L′, M′, and R′ from the left, middle, and right perspectives to obtain the point cloud P fusion , the formula is:

[0044] P fusion =L′∪M′∪R′;

[0045] S62, for P fusion Any point in Find its n nearest neighbor point set Calculate the mean Euclidean distance from this point to each point in the neighborhood The formula is:

[0046]

[0047] S63, recalculation The standard deviation σ d , the formula is:

[0048]

[0049] in, for The mean of

[0050] S64. According to the 3σ principle in statistics, remove abnormal outliers that exceed the mean threshold of the Euclidean distance;

[0051] Euclidean distance mean threshold d threshold The formula is:

[0052]

[0053] Among them, α is the control parameter; if P fusion Any point in satisfy Then it is considered that this point is an abnormal outlier and is removed from the point cloud P fusion Finally, the point cloud P after removing the abnormal outliers is obtained filtered ;

[0054] S65. Downsample the point cloud P filtered by the voxel grid method; Divide the point cloud P filtered into regular three-dimensional voxel grids according to the voxel size v; For each voxel, find the set P V ={p1, p2,..., p l} of all points inside it, and then calculate the centroid of the points inside this voxel:

[0055]

[0056] Finally, integrate the centroid points of all voxels to obtain the downsampled point cloud P downsampled , that is, the complete and accurate gap data at a certain point to be measured is obtained.

[0057] Based on the above technical solutions, the present invention has at least the following beneficial effects:

[0058] Through the multi-view acquisition device and the introduction of the offset, the present invention converts the two-dimensional data collected by the line laser sensor into a three-dimensional point cloud, and completes the information through precise point cloud processing and projection methods, obtaining the complete data of the gap; realizing the effective improvement of the gap measurement accuracy and data integrity, and being able to effectively avoid the single-view measurement blind area and eliminate the errors existing in the traditional multi-view measurement, thereby ensuring the precise reconstruction of the gap. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0060] Figure 1 is a flowchart of a multi-view point cloud data quality enhancement method for aircraft assembly gap measurement proposed by the present invention;

[0061] Figure 2 is a design diagram of a connecting rod structure for multi-view point cloud data acquisition in the present invention;

[0062] Figure 3 is a schematic diagram of multi-view point cloud data acquisition for gap measurement in the present invention;

[0063] Figure 4 is a point cloud diagram after the y-axis offset of the two-dimensional line laser data in the present invention;

[0064] Figure 5Schematic diagram for extracting critical points on both sides of the gap in the present invention;

[0065] Figure 6 Schematic diagram of the spatial distribution of the gap direction in the present invention;

[0066] Figure 7 Schematic diagram of the process for optimizing data quality by multi - perspective point cloud projection in the present invention. Detailed implementation manners

[0067] To make the above - mentioned objects, features, and advantages of the present invention more obvious and understandable, the following further describes the present invention in detail with reference to the attached Figure 1-7 drawings and embodiments. Thereby, the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0068] Those of ordinary skill in the art can understand that all or part of the steps in the methods of the above - mentioned embodiments can be completed by instructing relevant hardware through a program. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer - usable storage media (including but not limited to disk storage, CD - ROM, optical storage, etc.) containing computer - usable program code.

[0069] Please refer to Figure 1 - Figure 7 , which shows a specific implementation manner of the present invention. In this embodiment, the two - dimensional data collected by the line laser sensor is converted into three - dimensional point clouds through a multi - perspective acquisition device, and the information is complemented through precise point cloud processing and projection methods, obtaining complete data of the gap, effectively improving the measurement accuracy and data integrity, being able to effectively eliminate blind areas and errors, and ensuring the accurate reconstruction of the gap.

[0070] As Figure 1 shown, it shows a method for enhancing the quality of multi - perspective point cloud data for measuring gaps in aircraft assembly proposed by the present invention. It specifically includes the following steps:

[0071] S1. Fix the line laser sensor on the multi - perspective acquisition device, measure the gap to be measured at multiple perspectives, and obtain multiple pieces of gap point clouds;

[0072] As a preferred implementation manner, step S1 specifically includes:

[0073] S11. Drive a single - degree - of - freedom link structure through a servo motor to drive the sensor to always rotate around the center of the butting joint to be measured at multiple perspectives; specifically, as Figure 2As shown in the figure, the present invention designs a single-degree-of-freedom linkage mechanism. The servo motor is installed at the blue rotor, and the white rotor is used to fix the line laser sensor. By driving the linkage mechanism with the servo motor, multi-view rotation of the sensor can be achieved, and it always maintains its pointing to the center of the seam to be measured. This design not only ensures the integrity of data acquisition but also effectively improves the measurement accuracy.

[0074] S12. Input the angle required for the sensor to rotate at the control end, and collect multiple sets of seam point clouds at different perspectives at the seam to be measured; take the perspective perpendicular to the seam plane as the middle perspective, and divide the left and right by α° deviation into multiple pairs of left and right perspectives; and at least select three sets of seam point clouds at the left, middle, and right perspectives as the basis for subsequent analysis and processing. It should be noted here that in practical applications, considering the measurement accuracy and acquisition efficiency comprehensively, at least collecting a set of point cloud data at the left, middle, and right three perspectives ensures that the seam information is basically complete, and then according to the result of the seam measurement (such as whether there is still a blind area), it is decided whether to add more perspective point cloud data; the added multi-perspective point cloud also includes the seam point cloud at the middle perspective and the seam point clouds at the left and right perspectives with different deviation values from the middle perspective.

[0075] In this embodiment, as Figure 3 , the deviations of the left and right perspectives from the middle perspective are set to -30° and 30° respectively. Through the multi-perspective point cloud data, the blind area situation that occurs during single-perspective acquisition can be avoided. At the same time, in the prior art, obtaining seam data at different angles through multiple sensors or mobile measurement devices is prone to introducing cumulative errors during the data conversion process, and the research on optimizing the point cloud quality is still relatively limited; therefore, the single-degree-of-freedom linkage structure designed by the present invention drives the sensor to alleviate the cumulative errors caused by multiple sensors and mobile measurement devices, and at the same time further optimizes the point cloud quality in the subsequent steps.

[0076] S2. Convert the obtained multiple sets of seam point clouds into the same coordinate system and preprocess the data;

[0077] As a preferred implementation manner, step S2 specifically includes:

[0078] S21. Artificially introduce an offset in the y-axis direction to make the seam point clouds obtained at the left, middle, and right perspectives form a dislocation in the y-axis direction. Denote the seam point clouds at the left, middle, and right perspectives after introducing the offset as L0, M0, and R0 respectively; the line laser sensor usually can only collect point cloud data on the x-axis and z-axis, and in this application, by introducing a y-axis offset, the two-dimensional data is converted into three-dimensional point cloud, realizing preliminary enhancement and complementation. The finally obtained three sets of seam point clouds are as Figure 4 shown.

[0079] S22. According to the relative position relationship between the gap point clouds, through coordinate transformation, the gap point clouds in the left view and the right view are registered to the coordinate system in the middle view, and the registered gap point clouds are denoted as L0′, M0′, and R0′ respectively;

[0080] Specifically, the rigid body transformation of the three-dimensional point cloud consists of a rotation matrix R and a translation vector T, and its mathematical expression is as follows:

[0081] P′ = RP + t;

[0082] where P represents the point cloud coordinates in the original coordinate system, is the rotation matrix, is the translation vector, and P' is the coordinates of the point cloud after transformation in the target coordinate system. The rotation matrix R is calculated from the rotation angles θ x , θ y , θ z around the x, y, and z axes:

[0083] R = R z (θ z )R y (θ y )R x (θ x );

[0084] where the rotation matrices for each axis are defined as follows:

[0085]

[0086] In this embodiment, it is known that the three gap point clouds do not involve translation transformation and only rotate around the y-axis. Therefore, the translation vector t = 0, and only the point clouds in the left and right views need to be converted to the coordinate system in the middle view. Specifically, the middle view point cloud M0 remains unchanged, i.e., M0 = M0'; the left view point cloud L0 needs to rotate 30° around the y-axis to obtain L0'; the right view point cloud R0 needs to rotate -30° around the y-axis to obtain R0'.

[0087] S23. Apply the radius filtering method to the registered gap point clouds. The main purpose of this step is to remove the useless points on the concave plane inside the gap; these points not only have no substantial effect in subsequent processing but may also affect the accuracy of data processing. Therefore, this application uses the radius filtering method to effectively remove the free points and outliers in the point cloud, which is suitable for the filtering process of line laser acquisition data. For each point in the point cloud, with this point as the center, use the KNN algorithm to construct the neighborhood of this point, calculate the average distance between the center point and the neighbor points in the neighborhood, and based on this distance information, judge whether the center point is an outlier; let the center point be p0, the number of neighborhood points be k, and the radius range be r, then the determination of the noise points by radius filtering can be expressed by the following formula:

[0088]

[0089] Calculate the distances between adjacent points in the acquired point cloud, and set a radius threshold slightly higher than the maximum distance. When implementing the present invention, the optimal threshold is experimentally determined to be r = 0.1. After filtering, three denoised slit point clouds L, M, and R are obtained.

[0090] S3. Calculate the curvature features of each preprocessed slit point cloud, and extract the edge points of the skin surfaces on both sides of each slit point cloud, denoted as critical points;

[0091] As a preferred implementation manner, step S3 specifically includes:

[0092] S31. Perform point cloud clustering segmentation on the left and right regions of each preprocessed slit point cloud in the left, middle, and right views using the K-means clustering algorithm to obtain six independent subsets, denoted as L1, L2, M1, M2, R1, and R2 respectively;

[0093] The specific process is as follows: Randomly select 6 data points as the initial cluster centers, and calculate the Euclidean distance d of each remaining data point from each cluster center euclidean , and assign it to the nearest cluster; The calculation formula of d euclidean is

[0094]

[0095] where X is the data point, C k is the center of the Kth cluster, x i is the i-th dimensional coordinate feature of the point cloud, and c k,i is the i-th dimensional coordinate feature of the cluster center;

[0096] For each cluster, calculate the mean value of all its data points, and use this mean value as the new cluster center; Repeat the assignment and update steps until the cluster centers no longer change significantly or reach a predetermined number of iterations; Through continuous iteration, the cluster centers will gradually stabilize, and finally form clusters with similar characteristics to complete the clustering process and obtain six subsets L1, L2, M1, M2, R1, and R2.

[0097] S32. Check that the arrangement order of each point in each subset is arranged from far away from the slit to close to the slit. If the order is incorrect, the data points need to be reordered; In this embodiment, as Figure 5 shown, the points in all point sets should be arranged along the arrow direction to ensure the correctness of subsequent curvature calculations.

[0098] Taking subset L1 as an example, for each point in it, select its first k / 2 and last k / 2, a total of k neighborhood points, denoted as

[0099] Calculate the vectors between adjacent points:

[0100] And find the average value of the angles between the vectors as the curvature at this point The formula is expressed as:

[0101]

[0102] Among them, v j and v j+1 are two adjacent vectors respectively, ||v j || represents the Euclidean norm of the vector, and C i represents the curvature at the point reflecting the degree of curvature of the local area near this point;

[0103] S34. Set the curvature threshold C th , traverse each point in the subset L1, and find the point where the curvature first exceeds this threshold, which is the critical point of this subset, denoted as P L1 ; The area in front of the critical point is considered the skin surface, and the area of the critical point is considered the gap;

[0104] S35. The other five subsets are used to extract critical points in the same way; finally, the critical points of the corresponding point cloud regions of each subset are obtained, denoted as P L1 , P L2 , P M1 , P M2 , P R1 , P R2 ; In this embodiment, through multiple experiments, it is found that when the curvature threshold C th = 0.01, the screening effect is the best, that is, the straight-line segment regions on both sides of the gap can be accurately identified; for the final result, please refer to Figure 5 , the green points are the finally extracted critical points, the red area is the gap area, and the other areas are the skin surfaces; at this time, the obtained gap area is only a fuzzy area, and in this embodiment, the direction of the gap and the complete and accurate point cloud will be determined in the subsequent steps.

[0105] S4. Use the least squares method to perform linear fitting on the extracted critical points, and determine the gap direction according to the direction of the fitted line;

[0106] As a preferred implementation method, step S4 specifically includes:

[0107] S41. For the critical points P L1 , P L2 , P M1 , P M2 , P R1 , P R2, combine them in pairs according to the same perspective to obtain three pairs of critical point pairs; calculate the midpoint for each pair of critical point pairs and denote it as P L 、P M 、P R ; perform linear fitting on the three midpoints by the least squares method, and constrain the fitted line to pass through point P M ; This ensures the fitting accuracy and reduces the error caused by the misalignment of the gap point cloud in the y-axis direction; it should be noted here that in this application, when the three gap point clouds are misaligned in the y-axis direction, the gap point cloud in the middle actually does not change, but is the accurate point cloud measured in reality, and it is also used as the reference for subsequent registration. Therefore, using P M as a constraint condition can optimize the fitting result and make the subsequent point cloud projection more accurate.

[0108] Assume the linear fitting equation is as follows:

[0109] r(t) = B + md;

[0110] where r(t) = (x(t), y(t), z(t)) is any point on the line, d = (d x , d y , d z ) is the direction vector of the line, and m is a parameter;

[0111] S42. After calculating the midpoints of the three groups of critical points, we get: Construct an offset matrix O to describe the offsets of P L 、P R relative to P M ; The formula of O is expressed as:

[0112]

[0113] Subsequently, use singular value decomposition SVD to solve the line direction vector, and the formula is expressed as:

[0114] U, S, V T = SVD(O);

[0115] where U and V are left and right orthogonal matrices, and S is the singular value matrix; the first column of V corresponds to the direction vector d of the least squares fitted line, which is the gap direction;

[0116] In the ideal state, the fitted line should be parallel to the y-axis. In the actual experimental process, when the distribution of the gap direction in the three-dimensional space is almost perpendicular to the XOZ plane and the deviation is only a few micrometers, the error can be ignored. In this embodiment, as shown in Figure 6, in the figure, the x - coordinate of each grid is 0.00002, which is almost 0, that is, it can be considered perpendicular to the XOZ plane; therefore, the spatial direction of the fitting line is highly consistent with the ideal gap direction, further verifying the reliability and accuracy of this method.

[0117] S5. Project the pre - processed multiple gap point clouds along the gap direction onto a plane perpendicular to this direction;

[0118] As a preferred embodiment, step S5 specifically includes:

[0119] S51. Normalize the direction vector d to a unit vector d′=(d′ x ,d′ y ,d′ z ); The formula is expressed as:

[0120]

[0121] S52. Taking the gap point cloud R in the right - view as an example, find the projection R i =(x i ,y i ,z i ) of any point in the unit vector d′ direction. The formula is expressed as: i,proj

[0122] R i,proj =(R i ·d′)d′;

[0123] Among them, R i ·d′ = x i d′ x +y i d′ y +z i d′ z ;

[0124] Then the projection point on the plane perpendicular to the gap direction is R′ i =R i -R i,proj ;

[0125] S53. Calculate the projection points of all points in R to obtain the gap point cloud R′ projected onto the plane perpendicular to the gap direction; The projections in the middle - view and left - view adopt the same calculation method, and finally the projected gap point clouds are L′, M′, and R′ respectively.

[0126] S6. Merge the projected multiple gap point clouds, and perform outlier detection and down - sampling to obtain a complete and accurate gap point cloud at a certain point to be measured;

[0127] Figure 7 As a preferred embodiment, such as Figure 7As shown, step S6 specifically includes:

[0128] S61, merge the projected gap point clouds L′, M′, and R′ from the left, middle, and right perspectives to obtain the point cloud P fusion , the formula is:

[0129] P fusion = L′∪M′∪R′;

[0130] S62, for P fusion Any point in Find its n nearest neighbor point sets Calculate the mean Euclidean distance from this point to each point in the neighborhood The formula is:

[0131]

[0132] S63, recalculation The standard deviation σ d , the formula is:

[0133]

[0134] in, for The mean of

[0135] S64. According to the 3σ principle in statistics, remove abnormal outliers that exceed the mean threshold of the Euclidean distance;

[0136] Euclidean distance mean threshold d threshold The formula is expressed as:

[0137]

[0138] Wherein, α is a control parameter, which is usually 1.5 to 2.0, and is 1.5 in the embodiment of the present invention; if P fusion Any point in satisfy The point is considered to be an abnormal outlier. fusion Finally, we get the point cloud P after the abnormal outliers are removed. filtered ;

[0139] S65. In order to prevent the merged point cloud from being too dense and to improve the computational efficiency of subsequent processing, in this embodiment, the point cloud is downsampled by a voxel grid method; specifically, a voxel size v is set, and the point cloud P filtered The space is divided into a regular 3D grid of cubes, with each voxel defined as:

[0140] V(i,j,k)=[iv x ,(i+1)vx ×[jv y ,(j + 1)v y ×[kv z ,(k + 1)v z

[0141] where v x ,v y ,v z is the side length of the voxel grid, usually set to 2 - 3 times the average spacing of the point cloud. In the embodiments of the present invention, it is set to 0.12, and i, j, k are grid indices; for each voxel V(i, j, k), find the set P V ={p1, p2,..., p l} of all the points inside it, and then calculate the centroid of the points inside the voxel:

[0142]

[0143] Finally, integrate the centroid points of all voxels to obtain the downsampled point cloud P downsampled , that is, obtain the complete and accurate gap data at a certain point to be measured.

[0144] So far, through the specially designed multi - view acquisition device and the introduction of the y - axis offset in the present invention, the two - dimensional data collected by the line laser sensor is converted into a three - dimensional point cloud, and the information is complemented through precise point cloud processing and projection methods, obtaining the complete data of the gap; effectively improving the measurement accuracy and data integrity of the gap, while being able to effectively avoid the single - view measurement blind area and eliminate the errors existing in traditional multi - view measurements, thus ensuring the precise reconstruction of the gap.

[0145] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above - mentioned exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non - restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0146] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.​

Claims

1. A multi-viewpoint cloud data quality enhancement method for measuring gaps in aircraft assembly, characterized in that Specifically, it includes the following steps: S1. Fix the line laser sensor on the multi-view acquisition device, measure the to-be-detected gap at multiple views, and obtain multiple pieces of gap point clouds; S2. Convert the obtained multiple pieces of gap point clouds to the same coordinate system and perform data preprocessing; S3. Calculate the curvature features of each piece of preprocessed gap point cloud, and extract the edge points of the skin surfaces on both sides of each piece of gap point cloud, denoted as critical points; S4. Use the least squares method to perform linear fitting on the extracted critical points, and determine the gap direction according to the direction of the fitting line; S5. Project the multiple pieces of preprocessed gap point clouds along the gap direction onto a plane perpendicular to this direction; S6. Merge the multiple pieces of projected gap point clouds, and perform outlier removal and downsampling processing to obtain a complete and accurate gap point cloud at a certain to-be-detected point.

2. A multi-viewpoint cloud data quality enhancement method for measuring the gaps in aircraft assembly, characterized in that, Step S1 specifically includes: S11. Drive a single-degree-of-freedom link structure through a servo motor to drive the sensor to always rotate around the center of the to-be-detected butting joint at multiple views; S12. Input the rotation angle required by the sensor at the control end, and collect multiple pieces of gap point clouds at different views at the to-be-detected butting joint; take the view perpendicular to the gap plane as the middle view, and divide the left and right by a deviation of α° on each side into multiple pairs of left and right views; and at least select a set of three pieces of gap point clouds at the left, middle, and right views as the basis for subsequent analysis and processing.

3. A multi - perspective point cloud data quality enhancement method for aircraft assembly gap measurement according to claim 2, characterized in that, Step S2 specifically includes: S21. Manually introduce an offset in the y-axis direction to make the gap point clouds obtained at the left, middle, and right views form a dislocation in the y-axis direction. Denote the gap point clouds at the left, middle, and right views after introducing the offset as L0, M0, and R0 respectively; S22. According to the relative position relationship between the gap point clouds, register the gap point clouds at the left and right views to the coordinate system at the middle view through coordinate transformation. Denote the registered gap point clouds as L0′, M0′, and R0′ respectively; S23. Use the radius filtering method for the registered gap point clouds to filter out the free points and outlier points in the gap point clouds, and obtain three pieces of preprocessed gap point clouds L, M, and R.

4. A multi - perspective point cloud data quality enhancement method for aircraft assembly gap measurement according to claim 2, characterized in that Step S3 specifically includes: S31. Use the K-means clustering algorithm to perform point cloud clustering and segmentation on the left and right regions of each piece of gap point cloud at the preprocessed left, middle, and right views to obtain six independent subsets, denoted as L1, L2, M1, M2, R1, and R2 respectively; S32. Check that the arrangement order of each point in each subset is arranged from far away from the gap to close to the gap. If the order is incorrect, the data points need to be re-sorted. S33. Taking subset L1 as an example, for each point in it Select its first k / 2 and last k / 2, a total of k neighboring points, calculate the vectors between adjacent points, and find the average value of the angles between the vectors as the curvature at this point to reflect the degree of curvature of the local area near this point; S34. Set the curvature threshold C th , traverse each point in subset L1, and find the point where the curvature first exceeds this threshold, which is the critical point of this subset, denoted as P L1 ; the area in front of the critical point is considered the skin surface, and the area of the critical point is considered the gap; S35. The remaining five subsets extract critical points in the same way; finally, the critical points of the corresponding point cloud regions of each subset are obtained, denoted as P L1 , P L2 , P M1 , P M2 , P R1 , P R2 respectively.

5. A multi-viewpoint cloud data quality enhancement method for measuring the gaps in aircraft assembly, as claimed in claim 2, wherein Step S4 specifically includes: S41. For the critical points P of the left and right regions of the gap point cloud from the left, middle, and right perspectives L1 , P L2 , P M1 , P M2 , P R1 , P R2 , pair them up according to the same perspective to obtain three pairs of critical point pairs; calculate the midpoint for each pair of critical point pairs, denoted as P L , P M , P R ; perform linear fitting on the three midpoints by the least squares method, and constrain the fitted line to pass through the point P M ; assume the fitted line equation is as follows: r(t) = B + md; where r(t) = (x(t), y(t), z(t)) is an arbitrary point on the line, d = (d x , d y , d z ) is the direction vector of the line, and m is a parameter; S42. After calculating the midpoints of the three groups of critical points, we obtain: Construct an offset matrix O to describe P L , P R with respect to P M offset; The formula for O is expressed as: Subsequently, use the singular value decomposition SVD to solve the line direction vector, and the formula expression is: U, S, V T = SVD(O); where U and V are left and right orthogonal matrices, and S is the singular value matrix; the first column of V corresponds to the direction vector d of the fitting line, which is the gap direction.

6. A multi - perspective point cloud data quality enhancement method for aircraft assembly gap measurement according to claim 5, characterized in that, Step S5 specifically includes: S51. Normalize the direction vector d to a unit vector d’=(d′ x , d' y , d' z ); Taking the gap point cloud R from the right perspective as an example, find any point R i =(x i , y i , z i )'s projection R i,proj in the direction of the unit vector d', which is expressed by the formula as: R i,proj = (R i · d')d'; wherein, R i ·d' = x i d′ x + y i d′ y + z i d′ z ; Then the projection point on the plane perpendicular to the gap direction is R' i = R i - R i,proj ; S53. Calculate the projection points of all points in R to obtain the gap point cloud R′ projected onto a plane perpendicular to the gap direction; use the same calculation method for the projections at the middle view and the left view. Finally, the projected gap point clouds are L′, M′, and R′ respectively.

7. A multi-viewpoint cloud data quality enhancement method for measuring gaps in aircraft assembly according to claim 2, characterized in that Step S6 specifically includes: S61. First, merge the projected gap point clouds L′, M′, and R′ under the left, middle, and right viewpoints to obtain the point cloud P fusion , which is expressed by the formula as: P fusion = L' ∪ M' ∪ R'; S62. For P fusion For any point Find its set of n nearest neighbor points Calculate the mean Euclidean distance from this point to each point within the neighborhood Expressed by the formula as: S63. Then calculate the standard deviation σ d using the formula: Among them, is the mean value of; S64. According to the 3σ principle in statistics, abnormal outliers beyond the threshold of the mean Euclidean distance are removed; Euclidean distance mean threshold d threshold is expressed by the formula as follows: where α is a control parameter; if any point fusion in satisfies then this point is considered an abnormal outlier and is removed from the point cloud P fusion to finally obtain the point cloud P filtered after the abnormal outliers are removed; S65. Downsample the point cloud P by the voxel grid method filtered ; Divide the point cloud P filtered into regular three-dimensional voxel grids according to the voxel size v; for each voxel, find the set P V ={p1, p2,..., p l} of all points inside it, and then calculate the centroid of the points within the voxel: Finally, the centroid points of all voxels are integrated to obtain the downsampled point cloud P downsampled , that is, the complete and accurate gap data at a certain point to be measured is obtained.

Images