Structural slab hole site measurement method based on L2 distance similarity evaluation registration

Abstract

A structural plate hole site measurement method based on L2 distance similarity evaluation registration belongs to the technical field of high-precision automatic measurement, and comprises the following steps: S1, extracting three-dimensional coordinates of structural hole sites from a three-dimensional model of a structural plate to form a group of three-dimensional point clouds as reference point clouds; s2, building a structural plate hole site measurement environment, and measuring three-dimensional coordinates of a structural hole site through photogrammetry to serve as a measurement point cloud; s3, calculating L2 distance sets for the reference point cloud and the measurement point cloud respectively; s4, constructing a similarity distribution map based on the L2 distance according to the L2 distance set calculated in the S3; s5, searching the similarity distribution map in the step S4, and extracting an optimal matching point pair; s6, calculating a coordinate transformation function between the matching point pairs through SVD decomposition, and taking the coordinate transformation function as an initial value of an ICP algorithm; s7, iteratively calculating a coordinate transformation relation by using an ICP algorithm to realize high-precision registration; and S8, calculating a hole site measurement result.

Description

Technical Field

[0001] The invention relates to a structural plate hole position measurement method based on L2 distance similarity evaluation and registration, and belongs to the technical field of high-precision automatic measurement. Background Art

[0002] With the rapid development of computer image processing and visual inspection technology, people can easily obtain spatial point information on the surface of objects. This information is essentially a series of geometric three-dimensional coordinates of sampling points on the surface of objects, which is called three-dimensional point cloud data. Among them, three-dimensional point cloud registration technology is the core of many technologies such as pose estimation, data splicing, and multi-sensor calibration. It is one of the key technologies in point cloud data processing technology and has been applied in many fields such as industrial inspection, aerospace, navigation and positioning, and medical diagnosis. Especially in the field of aerospace structural plate inspection, the use of three-dimensional point cloud registration technology can greatly improve the efficiency of hole detection.

[0003] The connotation of 3D point cloud matching technology can be described as: for two groups of 3D point clouds with arbitrary postures and points, find a set of spatial coordinate transformation relationships so that the spatial positions of each point in one group of point clouds are as consistent as possible with those in the other group of points after the coordinate transformation. At present, the algorithms for 3D point cloud registration are mainly divided into two categories: registration algorithms based on probability models and registration algorithms based on least squares. For the first method, the mainstream method is the Coherent Point Drift (CPD) algorithm proposed by AndriyMyronenko and Xubo Song, which transforms the solution of the coordinate transformation matrix into the maximum likelihood estimation problem of fitting the centroid point cloud of the Gaussian mixture model to the observed data point cloud. It has strong stability to the noise and deviation of the points, but the registration accuracy is not high and it is easy to fall into the local optimum; for the second method, the most representative one is the Iterative Closest Points Algorithm (ICP) proposed by BeslPJ and Mckay ND. The registration accuracy is very high, but its results are highly dependent on the global optimality and accuracy of the initial value.

[0004] Photogrammetry is an important method for detecting the appearance size of spacecraft structural plates. The measurement of structural plate holes is characterized by high precision, large number, diverse shapes, and complex position distribution. The traditional method is based on physical contact measurement using a three-coordinate measuring instrument. This method has high precision but low detection efficiency and is not compatible with automated detection of large-size and side holes. Photogrammetry has the advantages of strong flexibility, simple operation, fast measurement speed, and high measurement accuracy. It has gradually become an important method for detecting the appearance size of spacecraft structural plates.

[0005] In the photogrammetry of the holes in the structural plate, there are high requirements for the accuracy and precision of the registration of the landmark points. The characteristics of the point cloud are as follows: there is only a rigid body transformation between the two groups of point clouds to be registered, and there is a strict correspondence between the points; the noise is small, but it contains some outliers and missing points; there is a certain symmetry, and in addition to the real registration posture, it may also contain multiple different good registration postures. In this case, the CPD algorithm is prone to fall into the local optimum, resulting in wrong registration. Relatively speaking, the ICP algorithm can obtain higher accuracy and has a broader application prospect, but it requires a good estimate of the matching relationship between each point in the point cloud. Achieving highly robust global point cloud registration is a key issue in the application of the ICP algorithm. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and solve the problem that the ICP algorithm requires a high-precision initial point cloud registration relationship in the current high-precision automatic measurement process of the hole position of aerospace structural plates. A high-stability, high-efficiency and high-precision structural plate hole position measurement method is proposed.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A method for measuring hole positions of structural plates based on L2 distance similarity evaluation and registration, comprising:

[0009] S1. Extract the three-dimensional coordinates of the structural holes from the three-dimensional model of the structural plate to form a set of three-dimensional point clouds as reference point clouds;

[0010] S2. Build a structural plate hole measurement environment, and measure the three-dimensional coordinates of the structural hole through photogrammetry as a measurement point cloud;

[0011] S3, respectively calculate the L2 distance set of the reference point cloud and the measured point cloud;

[0012] S4, constructing a similarity distribution graph based on the L2 distance according to the L2 distance set calculated in S3;

[0013] S5, searching the similarity distribution graph in S4, extracting the best matching point pair;

[0014] S6, calculating the coordinate transformation function between the matching point pairs through SVD decomposition as the initial value of the ICP algorithm;

[0015] S7, use ICP algorithm to iteratively calculate the coordinate transformation relationship to achieve high-precision registration;

[0016] S8. Calculation of hole position measurement results.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] (1) High-stability point cloud global matching. The point cloud automatic registration method based on L2 distance similarity evaluation proposed in the present invention comprehensively analyzes all registration situations between the reference point cloud and the measured point cloud. It is a global registration method that does not require an initial value. It overcomes the problem of local erroneous convergence in registration due to the high symmetry of hole positions in the measurement of aerospace structural plates, and can stably obtain the global optimal solution that meets the least squares condition.

[0019] (2) High-precision matching parameter calculation. The point cloud automatic registration method based on L2 distance similarity evaluation proposed in the present invention is based on the strict correspondence between the aerospace structural plate theory and the measured hole position. Through the registration calculation based on the high-precision ICP algorithm, high-precision coordinate transformation parameters can be calculated based on global stable registration.

[0020] (3) Efficient identification and extraction of hole positions. Aiming at the structural characteristics of aerospace structural plates with planes, multiple sides and multiple holes, the present invention can automatically identify the front and sides of structural plates of different shapes and sizes through feature recognition algorithms, extract feature parameters and hole position information, and improve efficiency.

[0021] (4) Automated and efficient calculation of hole position measurement results. Aiming at the multi-parameter measurement requirements of aerospace structural plate holes, the present invention can automatically perform post-processing analysis on the point cloud registration results, and directly output the result parameters such as the hole position, hole spacing, straightness, and flatness of the front and side panels, thereby improving measurement efficiency.

[0022] (5) Highly stable, efficient and high-precision structural plate hole position measurement. The present invention ensures the stability and high precision of the results through L2 distance similarity evaluation and registration, and realizes the automation of the entire process from input model hole position identification and extraction, point cloud registration and calculation, to data post-processing output, greatly improving the calculation efficiency, thereby achieving highly stable, efficient and high-precision structural plate hole position measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The hole position measurement process of the structural plate is evaluated based on the L2 distance similarity registration;

[0024] Figure 2 Schematic diagram of the calculation process of the L2 distance similarity distribution graph C(u,v)

[0025] Figure 3 Collect point clouds for point clouds extracted from 3D models and for devices that are not registered;

[0026] Figure 4 This is the final effect after registering the device-collected point cloud with the model point cloud. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0028] A structural plate hole position measurement method based on L2 distance similarity evaluation registration, such as Figure 1 As shown, the following steps are included:

[0029] S1, extract the three-dimensional coordinates of the structural holes from the three-dimensional model of the structural plate to form a reference point cloud Y = (y1, y2, ..., y m );

[0030] Step S1 specifically includes:

[0031] S101: Preset area threshold s and distance threshold d

[0032] S102, front panel identification. Read the model surface features and their areas, segment the surface features with an area greater than s as the front and back sides of the structural plate, and manually select one surface feature from the front and back sides through the visual interface as the front panel to be measured;

[0033] S103, extracting the hole positions of the front panel. Read the circle and its center coordinates in the model line feature, as well as the spatial mathematical expression of the front panel, calculate the distance from the center of the circle to the front panel, and extract the center coordinates of the circle whose distance is less than d as the hole positions of the front panel;

[0034] S104, side panel identification. Extract the normal vector and the three-dimensional coordinates of the vertices of several sides of the front panel from the front panel; traverse the surface features of the model, extract the normal vector and spatial expression of each, if the normal of the surface feature is perpendicular to the normal of the front panel, and the distance between the vertex of the side edge and the surface feature is less than d, then identify the surface feature as a side panel and save it;

[0035] S105, extracting the hole positions of the side panel. Read the circle in the model line feature and the spatial mathematical expression of the side panel, extract the three-dimensional coordinates of the center of the circle, calculate the spatial distance from the center of the circle to the side panel, and extract the coordinates of the center of the circle whose distance is less than d as the hole position of the side panel;

[0036] S2, build the structural plate hole measurement environment, and measure the three-dimensional coordinates of the structural hole through photogrammetry as the measurement point cloud X = (x1, x2, ..., x n). The structural plate hole position measurement environment consists of a measuring platform, an industrial camera, a planar light source array, a light shield, a high-precision positioning target and a host computer. The structural plate is placed on the measuring platform, and the planar light source array is arranged around the measuring platform to provide uniform lighting; the high-precision positioning target is placed in the hole to be detected on the structural plate, and the light shield covers the above facilities to avoid interference from ambient light. The high-precision positioning target is a special target that matches the aperture and thread of the structural plate hole. It consists of a screw, a nut and a photogrammetry target point. The photogrammetry target point is flat on the nut, and the position offset between the center of the photogrammetry target point and the center of the nut is less than 0.01mm. The photogrammetry area fully covers all high-precision positioning targets on the structural plate, and the center of the target point on the positioning target is extracted from the image. The position of this center point is used as the position of the structural plate hole, such as Figure 3 shown.

[0037] S3, traverse all points in the reference point cloud and the measured point cloud respectively, and calculate the L2 distance set d X (i,j),d Y (i,j)

[0038] d X (i,j)=||x i -x j ||,(i,j=0,1,...,n,i≠j)

[0039] d Y (i,j)=||y i -y j ||,(i,j=0,1,...,m,i≠j)

[0040] Among them, || || represents modulus, i, j are the serial numbers of the two points involved in the calculation in the corresponding point cloud;

[0041] S4, based on the L2 distance set d calculated in S3 X (i,j),d Y (i, j), construct a similarity distribution graph C(u, v) based on L2 distance; the calculation process is shown in the following figure Figure 2

[0042] Step S4 specifically includes:

[0043] S401, for the L2 distance set d X (i,j), calculate its maximum value max d X 、Minimum valuemin d X ;

[0044] S402, given an estimated value σ of the upper limit of the absolute value of the measurement error of the measurement point cloud; define a container D X (t), used to store d according to certain rulesX (i,j), where t is the index of the container, D X (t) represents the tth item of the container, which is the d stored X The set of (i,j),

[0045] S403, traverse d X (,i), calculate the container subscript index corresponding to each item in means round down and d X (i,j) is stored in the corresponding item D of the container X (t), so D X (t) will contain multiple d X (i,j), and D X All d in (t) X The value range of (i,j) is mind X +tσ~mind X +(t+1)σ, container D X The capacity of (t) is (maxd X -mind X ) / σ;

[0046] For d X (i,j),d Y (i,j) calculate the container D respectively according to the above method X (t), D Y (t);

[0047] S404, constructing a similarity distribution graph C(u,v), where u and v are the serial numbers of the reference point cloud and the measured point cloud respectively, the image size is n×m, and all values ​​are set to 0;

[0048] S405, for D X (t) X (i X ,j X ), whose index is t X ; From container D Y (t), take out D Y (t X -1) D Y (t X )、D Y (t X +1) All elements contained in the three items d Y (i Y ,j Y ), respectively with d X (i X ,j X ) Calculate the similarity coefficient, the calculation formula is as follows

[0049]

[0050] The calculated correlation coefficient value c(i X ,j X ,i Y ,j Y ), accumulated to C(i X ,i Y )、C(i X ,j Y )、C(j X ,i Y )、C(j X ,j Y )

[0051] S406, repeat S405, traverse container D X For all elements in (t), calculate the correlation coefficient c and add them up to obtain the similarity distribution graph C(u,v);

[0052] S5, search the similarity distribution map C(u,v) in S4 and extract the best matching point pair; the similarity distribution map C(u,v) can be regarded as a two-dimensional matrix, traverse each element in C(u,v), for point number p,q, if C(p,q) is the maximum value in the row and column direction of the position at the same time, then the point x is considered p ,y q The highest similarity is saved as a matching point pair. After the traversal is completed, the reference point cloud and the measured point cloud are rearranged according to the matching relationship to obtain the matching point cloud (x1, x2, ..., x s ),(y1,y2,...,y s ), between two point clouds, the points with the same subscript are matching point pairs;

[0053] S6, calculate the coordinate transformation relationship between the matching point pairs through SVD decomposition, which is used as the initial value of the ICP algorithm;

[0054] Step S6 specifically includes:

[0055] S601, calculate the centroid positions x and y of the two groups of point clouds respectively, and then calculate the centroid coordinates x of each point i ',y i ':

[0056]

[0057] x i '=x i -x,y i '=y i -y

[0058] S602, define and calculate the matrix W,

[0059]

[0060] S603, perform SVD decomposition on the matrix to obtain

[0061] W=UΣV T

[0062] Among them, Σ is a diagonal matrix composed of singular values, the diagonal elements are arranged from large to small, U and V are orthogonal matrices, let

[0063] R=VU T

[0064] t=y-Rx

[0065] Get the coordinate transformation parameters R,t so that the point cloud is optimally registered, that is, satisfying the following least squares optimization function

[0066]

[0067] S7, uses the ICP algorithm to iteratively calculate the coordinate transformation relationship to achieve high-precision registration.

[0068] Step S7 specifically includes:

[0069] S701, set the maximum deviation threshold e, traverse the point x in the measurement point cloud i , according to R,t calculated in S6, re-search the matching point y in the reference point cloud j , to satisfy

[0070]

[0071] S702, the matching point pairs are used to form a new matching point cloud, and the new R,t is calculated again according to the step of S6; and the step of S7 is repeated and iterated. Given an iteration threshold ε, if If the iteration is successful, output R,t; if the number of iterations is greater than 10, exit the iteration. Figure 4 This is the final effect after registering the device-collected point cloud with the model point cloud.

[0072] S8, calculation of hole position measurement results;

[0073] Step S8 specifically includes:

[0074] S801: Subtract the measured point cloud from the matched reference point cloud after coordinate transformation to obtain the hole position measurement deviation.

[0075] dx i =Rx i +ty i

[0076] S802, according to the attribution relationship of the reference point cloud in step S1, the hole position measurement deviation data is divided into the front panel and each side panel; according to the measurement requirements, for the front panel holes, the hole position deviation values ​​are projected along the normal of the front panel, and then the square root is taken and the average is taken to obtain the flatness of the front panel; for the side panel holes (single row), the hole position deviation values ​​are projected along the normal of the side panel, and then the square root is taken and the average is taken to obtain the parallelism of the side panel; the hole position deviation values ​​are projected onto the plane of the front panel or side panel to which they belong, respectively, to obtain the hole position deviation; for the side panel, the hole spacing of the single row of holes is calculated in sequence.

[0077] The contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

[0078] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for measuring the hole position of a structural plate based on L2 distance similarity evaluation and registration, characterized in that: include: S1. Extract the three-dimensional coordinates of the structural holes from the three-dimensional model of the structural plate to form a set of three-dimensional point clouds as reference point clouds; S2. Build a structural plate hole measurement environment, and measure the three-dimensional coordinates of the structural hole through photogrammetry as a measurement point cloud; S3, respectively calculate the L2 distance set of the reference point cloud and the measured point cloud; S4, constructing a similarity distribution graph based on the L2 distance according to the L2 distance set calculated in S3; S5, searching the similarity distribution graph in S4, extracting the best matching point pair; S6, calculating the coordinate transformation function between the matching point pairs through SVD decomposition as the initial value of the ICP algorithm; S7, use ICP algorithm to iteratively calculate the coordinate transformation relationship to achieve high-precision registration; S8. Calculate the hole position measurement results.

2. The method for measuring the hole position of a structural plate according to claim 1, characterized in that: Step S1 specifically includes: S101, preset an area threshold s and a distance threshold d; S102, front panel identification: read the model surface features and their areas, segment the surface features with an area greater than s as the front and back sides of the structural plate, manually select through the visual interface, and select a surface feature from the front and back sides as the front panel to be measured; S103, extracting the hole positions of the front panel: reading the circle and its center coordinates in the model line feature and the spatial mathematical expression of the front panel, calculating the distance from the center of the circle to the front panel, and extracting the center coordinates of the circle whose distance is less than d as the hole positions of the front panel; S104, side panel identification: extract the normal vector and the three-dimensional coordinates of the vertices of several sides of the front panel from the front panel; traverse the surface features of the model, extract the normal vector and the spatial expression of each, if the normal of the surface feature is perpendicular to the normal of the front panel, and the distance between the side edge vertex and the surface feature is less than d, then identify the surface feature as a side panel and save it; S105, side panel hole position extraction: read the circle in the model line feature and the spatial mathematical expression of the side panel, extract the three-dimensional coordinates of the center of the circle, calculate the spatial distance from the center of the circle to the side panel, and extract the coordinates of the center of the circle with a distance less than d as the side panel hole position.

3. The method for measuring the hole position of a structural plate according to claim 1, characterized in that: In step S2, a structural plate hole position measurement environment is built, including a measurement platform, an industrial camera, a planar light source array, a light shield, a high-precision positioning target and a host; the structural plate is placed on the measurement platform, and the planar light source array is arranged around the measurement platform to provide uniform lighting; the high-precision positioning target is placed in the hole position to be detected on the structural plate, and the light shield covers the above facilities to avoid interference from ambient light; the high-precision positioning target is a special target that matches the aperture and thread of the structural plate hole position, and is composed of a screw, a nut and a photogrammetry target point. The photogrammetry target point is flat on the nut, and the position offset between the center of the photogrammetry target point and the center of the nut is less than 0.01mm. The photogrammetry area fully covers all high-precision positioning targets on the structural plate, and the center of the target point on the positioning target is extracted from the image, and the center point position is used as the position of the structural plate hole.

4. The method for measuring the hole position of a structural plate according to claim 1, characterized in that: Step S4 specifically includes: S401, for the L2 distance set d X (i,j), calculate its maximum value max d X 、Minimum valuemin d X ; S402, given an estimated value σ of the upper limit of the absolute value of the measurement error of the measurement point cloud; define a container D X (t), used to store d according to certain rules X (i,j), where t is the index of the container, D X (t) represents the tth item of the container, which is the d stored X The set of (i,j); S403, traverse d X (,i), calculate the container subscript index corresponding to each item in Indicates rounding down and d X (i,j) is stored in the corresponding item D of the container X (t), so D X (t) will contain multiple d X (i,j), and D X All d in (t) X The value range of (i,j) is min d X +tσ~min d X +(t+1)σ, container D X The capacity of (t) is (max d X -min d X ) / σ; For d X (i,j),d Y (i,j) calculate the container D respectively according to the above method X (t), D Y (t); S404, constructing a similarity distribution graph C(u,v), where u and v are the serial numbers of the reference point cloud and the measured point cloud respectively, the image size is n×m, and all values ​​are set to 0; S405, for D X (t) X (i X ,j X ), whose index is t X ; From container D Y (t), take out D Y (t X -1) D Y (t X )、D Y (t X +1) All elements contained in the three items d Y (i Y ,j Y ), respectively with d X (i X ,j X ) Calculate the similarity coefficient, the calculation formula is as follows The calculated correlation coefficient value c(i X ,j X ,i Y ,j Y ), accumulated to C(i X ,i Y )、C(i X ,j Y )、C(j X ,i Y )、C(j X ,j Y ) S406, repeat S405, traverse container D X For all elements in (t), the correlation coefficient c is calculated and accumulated to obtain the similarity distribution graph C(u,v).

5. The method for measuring the hole position of a structural plate according to claim 1, characterized in that: Step S5 specifically includes: treating the similarity distribution graph C(u,v) as a two-dimensional matrix, traversing each element in C(u,v), for point numbers p, q, if C(p,q) is the maximum value in the row and column directions of the position at the same time, then the point x is considered p ,y q The highest similarity is saved as a matching point pair; after the traversal is completed, the reference point cloud and the measured point cloud are rearranged according to the matching relationship to obtain the matching point cloud (x1, x2, ..., x s ),(y1,y2,...,y s ), between two point clouds, the points with the same subscript are matching point pairs.

6. The method for measuring the hole position of a structural plate according to claim 1, characterized in that: Step S6 specifically includes: S601, calculate the centroid positions x and y of the two groups of point clouds respectively, and then calculate the centroid coordinates x of each point i ',y i ': x i '=x i -x,y i '=and i -and S602, define and calculate the matrix W: S603, perform SVD decomposition on the matrix to obtain W=UΣV T Among them, Σ is a diagonal matrix composed of singular values, the diagonal elements are arranged from large to small, U and V are orthogonal matrices, let R=VU T t=y-Rx The coordinate transformation parameters R,t are obtained to achieve optimal registration of the point cloud.

7. The method for measuring the hole position of a structural plate according to claim 6, characterized in that: Step S7 specifically includes: S701, set the maximum deviation threshold e, traverse the point x in the measurement point cloud i , according to R,t calculated in S6, re-search the matching point y in the reference point cloud j , to satisfy S702, the matching point pairs are used to form a new matching point cloud, and the new R,t is calculated again according to the step of S6; and the step of S7 is repeated and iterated. Given an iteration threshold ε, if If the iteration is successful, output R,t; if the number of iterations is greater than 10, exit the iteration.

8. The method for measuring the hole position of a structural plate according to claim 1, characterized in that: Step S8 specifically includes: S801: Subtract the measured point cloud from the matched reference point cloud after coordinate transformation to obtain the hole position measurement deviation. dx i =Rx i +t-y i S802, according to the attribution relationship of the reference point cloud in step S1, the hole position measurement deviation data is divided into the front panel and each side panel; according to the measurement requirements, for the front panel hole, the hole position deviation value is projected along the normal of the front panel and then the square root is taken and the average is taken to obtain the flatness of the front panel; for the side panel hole, the hole position deviation value is projected along the normal of the side panel and then the square root is taken and the average is taken to obtain the parallelism of the side panel; the hole position deviation value is projected onto the plane of the front panel or side panel to which it belongs, respectively, to obtain the hole position deviation; For the side panels, calculate the hole spacing of a single row of holes one by one.

Images