Intelligent positioning method and system for hyperbolic plate processing

Abstract

The application relates to the field of image processing, in particular to a hyperbolic plate processing intelligent positioning method and system. The method comprises the following steps: obtaining a three-dimensional scanning model and a theoretical three-dimensional model of a hyperbolic plate, obtaining a scanning topological graph and a theoretical topological graph; obtaining scanning priority sequencing and theoretical priority sequencing; obtaining the optimal matching point of each vertex in the scanning topological graph in the theoretical topological graph, and determining the corresponding theoretical point of each optimal matching point; respectively determining the corresponding original point of each vertex in the scanning topological graph; combining the original point and the theoretical point corresponding to each vertex in each scanning topological graph to form a three-dimensional point corresponding sequence, obtaining the mapping relationship corresponding to the three-dimensional point corresponding sequence through a SIFT method; obtaining a fine registration transformation matrix; decomposing the fine registration transformation matrix to obtain a change amount, converting the change amount into a control instruction, and positioning and adjusting a defective link in a hyperbolic plate manufacturing process. The application can improve the production precision of the hyperbolic plate.

Description

TECHNICAL FIELD

[0001] The application relates to the field of image processing, in particular to an intelligent positioning method and system for hyperbolic plate processing. BACKGROUND

[0002] In high-precision intelligent positioning of hyperbolic plate processing, first, actual three-dimensional point cloud data of the hyperbolic plate is obtained through laser scanning, and an ICP (Iterative Closest Point) algorithm is used to accurately match and compare the scanning data with a theoretical CAD design model, so as to calculate accurate pose deviation for subsequent processing correction.

[0003] However, the smooth and continuous geometry structure, sparse features and gentle curvature variation of the hyperbolic plate result in a lack of clear corresponding targets in the "closest point search" step of the algorithm core, a large number of fuzzy and erroneous matching point pairs are generated on the curved surface; at the same time, the similar local geometric features between different regions make the iterative process extremely dependent on the initial pose, and it is extremely easy to fall into a local optimal solution and cannot converge to a globally correct registration position, which seriously restricts the final matching accuracy and reliability. SUMMARY

[0004] The application provides an intelligent positioning method and system for hyperbolic plate processing to solve the existing problems.

[0005] The intelligent positioning method for hyperbolic plate processing provided by the application adopts the following technical scheme:

[0006] One embodiment of the application provides an intelligent positioning method for hyperbolic plate processing, which comprises the following steps:

[0007] Obtain a three-dimensional scanning model and a theoretical three-dimensional model of the hyperbolic plate, and obtain a scanning topological graph and a theoretical topological graph according to the three-dimensional scanning model and the theoretical three-dimensional model;

[0008] Obtain the topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph, and sort them respectively to obtain a scanning priority order and a theoretical priority order;

[0009] According to the scanning priority order and the theoretical priority order, obtain the optimal matching point of each vertex in the scanning topological graph in the theoretical topological graph, and determine the corresponding theoretical point of each optimal matching point;

[0010] Determine the original point corresponding to each vertex in the scanning topological graph respectively;

[0011] Combine the original point and the theoretical point corresponding to each vertex in the scanning topological graph to form a three-dimensional point correspondence sequence, and obtain the mapping relationship corresponding to the three-dimensional point correspondence sequence by using a SIFT method;

[0012] According to the mapping relationship, the three-dimensional scanning model and the theoretical three-dimensional model, an accurate registration transformation matrix is obtained;

[0013] The accurate registration transformation matrix is decomposed to obtain a change amount, and the change amount is converted into a control instruction to position and adjust a defective link in a manufacturing process of the double-curved plate.

[0014] Optionally, the obtaining of the scanning topological graph and the theoretical topological graph according to the three-dimensional scanning model and the theoretical three-dimensional model specifically comprises:

[0015] The perspective graph and the depth graph are obtained according to the three-dimensional scanning model and the theoretical three-dimensional model respectively, wherein the obtaining parameters of the perspective graph and the depth graph are the same;

[0016] The scanning key points and the theoretical key points corresponding to the perspective graph and the depth graph are obtained respectively by using a sift algorithm;

[0017] The triangular nets are constructed by taking the scanning key points and the theoretical key points as the vertices respectively by using a Delaunay triangulation method, so that the scanning topological graph and the theoretical topological graph are obtained.

[0018] Optionally, the obtaining of the topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph specifically comprises:

[0019] The degree of each vertex in the topological graph is obtained, and the ratio of the degree of the Nth vertex to the maximum degree of the vertex is determined as the relative degree centrality of the Nth vertex in the topological graph, wherein the topological graph is the scanning topological graph or the theoretical topological graph;

[0020] The distance sum of the Nth vertex in the topological graph is obtained by summing up the distances between the Nth vertex and other vertices in the topological graph by using a Dijkstra shortest path algorithm, and the distance sum of the Nth vertex is taken as a negative number and taken as an exponential of a natural base to obtain the centrality of the Nth vertex;

[0021] The product of the relative degree centrality of the Nth vertex and the centrality of the Nth vertex is determined as the topological relationship strength of the Nth vertex;

[0022] The topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph are determined respectively.

[0023] Optionally, the obtaining of the optimal matching point of each vertex in the scanning topological graph in the theoretical topological graph according to the scanning priority order and the theoretical priority order specifically comprises:

[0024] The first element in the scanning priority order is determined as the first matching point corresponding to the vertex in the scanning topological graph;

[0025] determine a candidate matching point corresponding to the first matching point according to the first matching point from the theoretical priority order, wherein an absolute value of a difference between a degree of the first matching point and a degree of the candidate matching point is less than a preset degree threshold value;

[0026] calculate a similarity between the descriptor of each candidate matching point and the descriptor of the first matching point respectively, and determine the candidate matching point with a similarity greater than a preset similarity threshold value as a reserved candidate matching point;

[0027] determine a neighborhood point priority sequence of the first matching point according to the scan topology graph;

[0028] determine a neighborhood priority sequence of the reserved candidate matching point according to the theoretical topology graph;

[0029] obtain a cosine similarity of the scan topology graph and the theoretical topology graph;

[0030] calculate a matching degree of the first matching point and each reserved candidate matching point according to the cosine similarity, the neighborhood point priority sequence and the neighborhood priority sequence, and determine a reserved candidate matching point with a maximum matching degree as an optimal matching point of the first matching point;

[0031] determine a vertex corresponding to an A-th element in the scan priority order in the scan topology graph as an A-th matching point;

[0032] obtain an optimal matching point of the vertex corresponding to each element in the scan priority order in the scan topology graph.

[0033] Optionally, the determining of the neighborhood point priority sequence of the first matching point according to the scan topology graph specifically comprises:

[0034] determine a neighborhood point of the first matching point from the scan topology graph, wherein the neighborhood point of the first matching point and the first matching point are connected through an edge of the scan topology graph in the scan topology graph;

[0035] calculate a similarity between the descriptor of each neighborhood point of the first matching point and the descriptor of the first matching point respectively, and perform ascending order sorting to obtain a neighborhood point sequence of the first matching point;

[0036] determine a sequence number value of each element in the scan priority order as a priority order value of the vertex corresponding to the element in the scan topology graph to obtain a priority order value of each vertex in the scan topology graph;

[0037] replace a value of each element in the neighborhood point sequence of the first matching point with a priority order value of the vertex corresponding to the element in the scan topology graph to obtain the neighborhood point priority sequence.

[0038] Optionally, the determining of the neighborhood priority sequence of the reserved candidate matching point according to the theoretical topology graph specifically comprises:

[0039] determining neighborhood points of the reserved candidate matching point from the theoretical topology graph, wherein the neighborhood points of the reserved candidate matching point and the reserved candidate matching point are connected by edges of the theoretical topology graph in the theoretical topology graph;

[0040] calculating similarity between a descriptor of each neighborhood point of the reserved candidate matching point and a descriptor of the reserved candidate matching point respectively, and performing ascending order sorting to obtain a neighborhood point sequence of the reserved candidate matching point;

[0041] determining a sequence number value of each element in the theoretical priority sequence as a priority sequence value of a corresponding vertex in the theoretical topology graph, to obtain a priority sequence value of each vertex in the theoretical topology graph;

[0042] replacing a value of each element in the neighborhood point sequence of the reserved candidate matching point with a priority sequence value of a corresponding vertex of each element in the theoretical topology graph, to obtain a neighborhood priority sequence.

[0043] Optionally, the cosine similarity between the scan topology graph and the theoretical topology graph is obtained, and specifically includes:

[0044] calculating eigenvalues of a normalized Laplacian matrix of the scan topology graph, and arranging the eigenvalues in descending order to obtain a first eigenvalue sequence;

[0045] calculating eigenvalues of a normalized Laplacian matrix of the theoretical topology graph, and arranging the eigenvalues in descending order to obtain a second eigenvalue sequence;

[0046] obtaining elements greater than a preset first eigenvalue in the first eigenvalue sequence to obtain a first vector;

[0047] obtaining elements greater than a preset second eigenvalue in the second eigenvalue sequence to obtain a second vector;

[0048] calculating a cosine similarity between the first vector and the second vector to determine the cosine similarity between the scan topology graph and the theoretical topology graph.

[0049] Optionally, the matching degree between the first matching point and each reserved candidate matching point is calculated according to the cosine similarity, the neighborhood point priority sequence and the neighborhood priority sequence, and specifically includes:

[0050] determining a similarity between the descriptor of the Mth reserved candidate matching point and the descriptor of the first matching point as an Mth descriptor similarity;

[0051] multiplying the cosine similarity after taking a negative number and adding 1 by the Mth descriptor similarity to obtain a first matching degree of the Mth reserved candidate matching point;

[0052] calculating a similarity between the neighborhood point priority sequence of the first matching point and the neighborhood priority sequence of the Mth reserved candidate matching point to obtain an Mth neighborhood similarity;

[0053] The product of the Mth neighborhood similarity and the cosine similarity is determined as the second matching degree of the Mth reserved candidate matching point;

[0054] The sum of the first matching degree of the Mth reserved candidate matching point and the second matching degree of the Mth reserved candidate matching point is determined as the matching degree of the first matching point and the Mth reserved candidate matching point;

[0055] The matching degree of the first matching point and each reserved candidate matching point is obtained.

[0056] Optionally, the obtaining of the fine registration transformation matrix according to the mapping relationship, the three-dimensional scanning model and the theoretical three-dimensional model specifically comprises:

[0057] The mapping relationship, the preset minimum point set size, the inner point distance threshold and the maximum iteration number are taken as inputs of the RANSAC robust estimation algorithm, and the rigid body transformation matrix is output through the RANSAC robust estimation algorithm;

[0058] The rigid body transformation matrix, the three-dimensional scanning model and the theoretical three-dimensional model are input into the ICP algorithm, and the fine registration transformation matrix is output through the ICP algorithm.

[0059] The present application provides a kind of for hyperbolic plate processing intelligent positioning system, including memory, processor and stored in memory and can be run on the computer program of processor, the computer program is implemented when processor executes as the steps of the kind for hyperbolic plate processing intelligent positioning method.

[0060] The beneficial effects of the technical scheme of the present application are:

[0061] In the embodiment of the present application, by the SIFT matching method based on depth map, first, multi-view depth maps are generated from the hyperbolic plate scanning point cloud and CAD model and robust SIFT features invariant to rotation and scaling are extracted;Then, through high-dimensional descriptor matching and RANSAC robust estimation, a high-precision initial rigid body transformation matrix is calculated, thereby realizing global coarse registration of the scanning data and the design model. Further, the matching ambiguity problem caused by the smooth surface and sparse features of the hyperbolic plate can be effectively overcome, providing an almost ideal iteration starting point for subsequent ICP matching, so that it does not need to consume iteration resources in the wrong direction, can focus on small local deformation optimization, and further significantly improves the accuracy, convergence speed and overall robustness of the final ICP matching. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0063] Figure 1 A flow chart of a method for intelligent positioning of hyperbolic plate processing provided by an embodiment of the present application;

[0064] Figure 2 A scatter plot of key points;

[0065] Figure 3 A triangulation plot obtained with key points as vertices;

[0066] Figure 4 A structure diagram of a system for intelligent positioning of hyperbolic plate processing provided by an embodiment of the present application. DETAILED DESCRIPTION

[0067] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined purposes, the specific embodiments, structures, features and effects of a method for intelligent positioning of hyperbolic plate processing according to the present application are described in detail below in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0069] The specific scheme of the method for intelligent positioning of hyperbolic plate processing provided by the present application is specifically described below in combination with the drawings.

[0070] The embodiments of the present application provide a method and system for intelligent positioning of hyperbolic plate processing. Please refer to Figure 1 which shows a flow chart of a method for intelligent positioning of hyperbolic plate processing provided by an embodiment of the present application. The method comprises the following steps:

[0071] S101, obtaining a three-dimensional scanning model and a theoretical three-dimensional model of the hyperbolic plate, and obtaining a scanning topological graph and a theoretical topological graph according to the three-dimensional scanning model and the theoretical three-dimensional model.

[0072] In this embodiment, the scanning topological graph and the theoretical topological graph are obtained according to the three-dimensional scanning model and the theoretical three-dimensional model, which specifically comprises:

[0073] According to the three-dimensional scanning model and the theoretical three-dimensional model, a perspective map and a depth map are obtained respectively, wherein the acquisition parameters of the perspective map and the depth map are the same;

[0074] The scanning key points and the theoretical key points corresponding to the perspective map and the depth map are obtained respectively by using a sift algorithm.

[0075] The triangular nets are constructed by taking the scanning key points and the theoretical key points as the vertices respectively by using a Delaunay triangulation method, so as to obtain a scanning topological map and a theoretical topological map.

[0076] Exemplarily, the process aims to obtain the three-dimensional geometric information of the hyperbolic plate physical surface by a precision measuring device, and convert it into a digital model.

[0077] Data acquisition: A high-precision laser scanner is used to perform full-coverage scanning on the surface of the hyperbolic plate workpiece. By planning a scanning path and with the help of the high-density sampling capability of the device itself, it is ensured that the massive three-dimensional point coordinate data of the entire outer surface of the workpiece is obtained. During scanning, special reference marker points are usually arranged around the workpiece to provide a unified coordinate reference framework for the entire scanning process, ensuring the global consistency of the data.

[0078] Data fusion and generation: The original point cloud data obtained by the scanner is fused, denoised and packaged by professional processing software, and finally a three-dimensional scanning model (usually stored in the form of triangular mesh patches, such as.stl or.obj files) accurately reflecting the manufacturing topography of the hyperbolic plate physical object is generated. The model completely records the actual size and shape of the workpiece.

[0079] The process of obtaining the theoretical three-dimensional model aims to obtain the accurate theoretical design model of the hyperbolic plate in the ideal state:

[0080] Data source: The model is directly derived from the initial design stage of the product. Designers use computer-aided design (CAD) software (such as CATIA, Siemens NX, etc.) to directly create the theoretical three-dimensional model of the hyperbolic plate according to the performance requirements of the product in terms of aerodynamics, structure, etc.

[0081] Data preparation: The final version of the CAD design file of the part is extracted from the product data management (PDM) system. According to the needs of the subsequent matching process, it can be converted in format (such as exported as.stp or.iges format), but the geometric definition and size accuracy must be strictly consistent with the original design.

[0082] For the three-dimensional scanning model obtained by laser scanning, six main view directions (up, down, left, right, front and back, the specific angle of the main view, the user can add and adjust according to the actual shape of the hyperbolic plate) of the workpiece can be projected onto a virtual image plane respectively, and then six perspective views are obtained. The gray value of each pixel in each perspective view represents the distance from the corresponding point to the perspective view.

[0083] For the theoretical three-dimensional model, the CAD model is virtually rendered from the same perspective and parameters (focal length, imaging size, etc.) as the three-dimensional scanning model to generate a depth map for each perspective (each perspective of the six perspectives).

[0084] Up to now, the three-dimensional matching problem has been converted into a two-dimensional image matching problem between multiple depth maps.

[0085] For the perspective view and the depth map under the same perspective, the key points in the image are obtained by the sift algorithm (the key points in the perspective view are the perspective view, and the key points in the depth map are the theoretical key points). Each key point corresponds to a 128-dimensional vector, which is used to describe the position, scale and direction information of the key point.

[0086] The traditional method directly selects the key point with the maximum vector similarity as the key point pair, but due to the smooth and continuous, sparse feature and gentle curvature change of the hyperbolic plate, for each key point, multiple matching key points with high similarity and little difference can be obtained. At this time, if the matching key point with the maximum similarity is directly selected as the matching point pair of the key point, a large number of fuzzy and incorrect matching point pairs will be generated.

[0087] Based on this, the embodiment considers which key point on which image is preferentially matched by combining more global information in the selection of matching key points, and then obtains the optimal matching point by combining the global information of the preferential key point in the numerous matching points, instead of directly selecting the matching point with the maximum similarity as the matching point of the preferential key point on the other image.

[0088] For all key points on each image, a triangular mesh is constructed by taking these key points as vertices through the Delaunay triangulation method, and a topological graph (the topological graph includes a scanning topological graph and a theoretical topological graph) is obtained.

[0089] As shown in Figure 2 , the scatter plot of the key points is shown in Figure 2 . Figure 2 The scatter points in the scatter plot can be the key points obtained by using the sift algorithm according to any perspective view or any depth map. As shown in Figure 3 , the triangulation graph obtained by taking the key points as vertices is shown in Figure 3 .

[0090] In this embodiment, the stronger the topological relationship of the points on the topological graph is, the more the points are regarded as the priority matching points. Such points are closely related to other key points, like the core hub in the network, which can provide stronger constraints and higher reliability for the subsequent matching process, thereby effectively reducing the propagation of false matching and improving the accuracy and robustness of the overall registration.

[0091] S102, the topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph are obtained and sorted respectively, and scanning priority sorting and theoretical priority sorting are obtained respectively.

[0092] In this embodiment, the topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph are obtained, which specifically includes:

[0093] The degree of each vertex in the topological graph is obtained, and the ratio of the degree of the Nth vertex to the maximum degree of the vertex is determined as the relative degree centrality of the Nth vertex in the topological graph, wherein the topological graph is a scanning topological graph or a theoretical topological graph;

[0094] The distance between the Nth vertex and other vertices in the topological graph is obtained by Dijkstra's shortest path algorithm and summed up to obtain the distance sum of the Nth vertex. The distance sum of the Nth vertex is taken as a negative number, and the Nth vertex is taken as the exponent of the natural base to obtain the centrality of the Nth vertex;

[0095] The product of the relative degree centrality of the Nth vertex and the centrality of the Nth vertex is determined as the topological relationship strength of the Nth vertex;

[0096] The topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph are determined respectively.

[0097] Exemplarily, taking the topological graph corresponding to any one of the perspective graph or the depth graph as an example, for each vertex in the topological graph, the more the number of edges directly connected to the vertex, the more "direct neighbors" it has. It is like a busy traffic hub, which is directly connected to many places and is at least the center of a local area; the smaller the "total distance" from it to all other points in the graph, i.e. it does not need to pass through many intermediate nodes to reach anywhere. Such vertices are closer to the true geometric center or information transmission center. Further, the degree of each vertex and the reciprocal of the average shortest path length of each vertex to all other vertices in the graph are used to quantify the strength of the topological relationship of each vertex.

[0098] For each vertex in each topology graph, the ratio of the degree of the vertex to the maximum degree among all vertices is denoted as n1 (i.e., relative degree centrality); the distance from the vertex to each other vertex is obtained by the Dijkstra shortest path algorithm, and the distance sum x is obtained; the product of n1 and z of x is taken as the topology relationship strength of the vertex; the stronger the topology relationship, the closer the contact with other key points, and the stronger the constraint and the higher the reliability provided for the subsequent matching process, thereby effectively reducing the propagation of false matching. As the centrality z of the vertex.

[0099] For each vertex in each topology graph, the product of n1 and z of the vertex is taken as the topology relationship strength of the vertex; the stronger the topology relationship, the closer the contact with other key points, and the stronger the constraint and the higher the reliability provided for the subsequent matching process, thereby effectively reducing the propagation of false matching.

[0100] For the view graph and the depth graph under the same view angle, the topology relationship strength of each key point is obtained, and then the priority order of the key points is obtained according to the order of the topology relationship strength of the key points from large to small (the scanning priority order is obtained by sorting the topology relationship strength of each vertex in the scanning topology graph; the theoretical priority order is obtained by sorting the topology relationship strength of each vertex in the theoretical topology graph), and the matching priority of the key point in the sequence is greater.

[0101] The sorting manner in the embodiment can be in sequence or in reverse sequence, the topology relationship strength is sorted from large to small, and the matching priority of the key point in the sequence is greater. The topology relationship strength is sorted from small to large, and the matching priority of the key point in the sequence is greater.

[0102] S103, according to the scanning priority order and the theoretical priority order, the optimal matching point of each vertex in the scanning topology graph in the theoretical topology graph is obtained, and each optimal matching point corresponding to the theoretical point is determined.

[0103] In the embodiment, according to the scanning priority order and the theoretical priority order, the optimal matching point of each vertex in the scanning topology graph in the theoretical topology graph is obtained, and the optimal matching point of each vertex in the scanning topology graph in the theoretical topology graph is obtained, and the optimal matching point of each vertex in the scanning topology graph in the theoretical topology graph is obtained.

[0104] The first element in the scanning priority order is determined as the first matching point in the scanning topology graph;

[0105] According to the first matching point, the candidate matching point corresponding to the first matching point is determined from the theoretical priority order, wherein the absolute value of the difference between the degree of the first matching point and the degree of the candidate matching point is less than a preset degree threshold;

[0106] The similarity between the descriptor of each candidate matching point and the descriptor of the first matching point is calculated, and the candidate matching point with a similarity greater than a preset similarity threshold is determined as a reserved candidate matching point;

[0107] According to the scanning topology graph, the neighborhood point priority sequence of the first matching point is determined;

[0108] determining a neighborhood priority sequence of the reserved candidate matching point according to the theoretical topology graph;

[0109] obtaining a cosine similarity of the scan topology graph and the theoretical topology graph;

[0110] calculating a matching degree of the first matching point and each reserved candidate matching point according to the cosine similarity, the neighborhood point priority sequence and the neighborhood priority sequence, and determining a reserved candidate matching point with the largest matching degree as an optimal matching point of the first matching point;

[0111] determining a vertex corresponding to an A-th element in the scan priority sequence in the scan topology graph as an A-th matching point;

[0112] obtaining an optimal matching point of a vertex corresponding to each element in the scan priority sequence in the scan topology graph.

[0113] determining a neighborhood point priority sequence of the first matching point according to the scan topology graph, and the method specifically comprises:

[0114] determining a neighborhood point of the first matching point from the scan topology graph, wherein the neighborhood point of the first matching point and the first matching point are connected through an edge of the scan topology graph in the scan topology graph;

[0115] calculating a similarity between a descriptor of each neighborhood point of the first matching point and a descriptor of the first matching point, and performing ascending sequence sorting to obtain a neighborhood point sequence of the first matching point;

[0116] determining a sequence number value of each element in the scan priority sequence as a priority sequence value of a vertex corresponding to the element in the scan topology graph to obtain a priority sequence value of each vertex in the scan topology graph;

[0117] replacing a value of each element in the neighborhood point sequence of the first matching point with a priority sequence value of a vertex corresponding to the element in the scan topology graph to obtain a neighborhood priority sequence.

[0118] determining a neighborhood priority sequence of the reserved candidate matching point according to the theoretical topology graph, and the method specifically comprises:

[0119] determining a neighborhood point of the reserved candidate matching point from the theoretical topology graph, wherein the neighborhood point of the reserved candidate matching point and the reserved candidate matching point are connected through an edge of the theoretical topology graph in the theoretical topology graph;

[0120] calculating a similarity between a descriptor of each neighborhood point of the reserved candidate matching point and a descriptor of the reserved candidate matching point, and performing ascending sequence sorting to obtain a neighborhood point sequence of the reserved candidate matching point;

[0121] Determine the sequence number value of each element in the theoretical priority ranking as the priority ranking value of the corresponding vertex in the theoretical topology graph, to obtain the priority ranking value of each vertex in the theoretical topology graph;

[0122] Replace the value of each element in the sequence of neighborhood points of the reserved candidate matching point with the priority ranking value of the corresponding vertex of each element in the theoretical topology graph to obtain the neighborhood priority sequence.

[0123] Obtain the cosine similarity of the scan topology graph and the theoretical topology graph, specifically including:

[0124] Calculate the eigenvalues of the normalized Laplacian matrix of the scan topology graph, and arrange the eigenvalues in descending order to obtain a first eigenvalue sequence;

[0125] Calculate the eigenvalues of the normalized Laplacian matrix of the theoretical topology graph, and arrange the eigenvalues in descending order to obtain a second eigenvalue sequence;

[0126] Obtain the elements greater than a preset first eigenvalue in the first eigenvalue sequence to obtain a first vector;

[0127] Obtain the elements greater than a preset second eigenvalue in the second eigenvalue sequence to obtain a second vector;

[0128] Calculate the cosine similarity of the first vector and the second vector, and determine it as the cosine similarity of the scan topology graph and the theoretical topology graph.

[0129] According to the cosine similarity, the neighborhood point priority sequence, and the neighborhood priority sequence, calculate the matching degree of the first matching point and each reserved candidate matching point, specifically including:

[0130] Determine the similarity of the Mth reserved candidate matching point descriptor and the first matching point descriptor as the Mth descriptor similarity;

[0131] Take the negative of the cosine similarity and add 1, and multiply it by the Mth descriptor similarity to obtain the first matching degree of the Mth reserved candidate matching point;

[0132] Calculate the similarity of the neighborhood point priority sequence of the first matching point and the neighborhood priority sequence of the Mth reserved candidate matching point to obtain the Mth neighborhood similarity;

[0133] Determine the product of the Mth neighborhood similarity and the cosine similarity as the second matching degree of the Mth reserved candidate matching point;

[0134] Determine the sum of the first matching degree of the Mth reserved candidate matching point and the second matching degree of the Mth reserved candidate matching point as the matching degree of the first matching point and the Mth reserved candidate matching point;

[0135] Obtain the matching degree of the first matching point and each reserved candidate matching point.

[0136] Exemplarily, taking the scanning priority ranking and the scanning priority ranking as an example, both of which are ranked from large to small. The first key point in the scanning priority ranking, i.e., the first matching point, is obtained, and the matching point is recorded as the first matching point. The degree of the first matching point is denoted by i. The degree of each key point in the theoretical priority ranking is obtained. The key point with an absolute value of the difference from i less than 3 is taken as a candidate matching point. For the candidate matching point, firstly, the similarity of the descriptor of each candidate matching point and the first matching point is calculated. The candidate matching point with a similarity greater than 0.7 is reserved and recorded as a reserved candidate matching point.

[0137] Optionally, the preset degree threshold and the preset similarity threshold are 3 and 0.7 respectively. Here, only an example is taken, and the user can adjust according to the actual production demand, and no specific numerical limitation is made here.

[0138] In the embodiment, the cooperative consistency of the matching point and its neighborhood points on the feature is taken as an important index for measuring the similarity of the matching point and the candidate point, that is, if a candidate point has a high similarity with the feature of the matching point itself, and the neighborhood relationship reflected by the surrounding topological structure and the feature distribution are highly consistent with the relationship of the matching point and its neighborhood, the matching priority of the candidate point is higher.

[0139] For the first matching point, the key points directly connected to the point on the scanning topological graph are recorded as the neighborhood points of the point. The similarity of the descriptor of each neighborhood point and the point is calculated. The similarity is ranked in order from small to large to obtain the corresponding neighborhood point sequence.

[0140] The topological relationship strength of each vertex on the scanning topological graph where the first matching point is located is obtained. In order from large to small, the priority ranking value of each vertex is obtained, such as: the priority ranking value of the vertex with the strongest topological relationship strength is 1, the priority ranking value of the vertex with the second strongest topological relationship strength is 2, and so on. The priority ranking value of each vertex is obtained.

[0141] For the first matching point, the label of each neighborhood point in the neighborhood point sequence is represented by the priority ranking value of the corresponding point to obtain the neighborhood point priority sequence.

[0142] The construction of the neighborhood point priority sequence contains two levels of information representation: firstly, the sequence is based on the similarity ranking of the descriptor of the neighborhood point and the first matching point, which reflects the feature similarity relationship between the neighborhood points and the first matching point; further, by replacing the similarity value with the topological priority ranking value of the corresponding point, the sequence further expresses the relative importance distribution of the neighborhood points based on the topological connection strength. Therefore, the neighborhood point priority sequence not only depicts the correlation properties between the first matching point and its neighborhood points, but also quantitatively reveals the structural saliency of different position nodes in the local neighborhood.

[0143] For each reserved candidate matching point of the first matching point, the neighborhood point sequence of each reserved candidate matching point is obtained by the same method, the topological relationship strength of each key point on the theoretical topological graph where the reserved candidate matching point is located is obtained, and then the neighborhood priority sequence is obtained.

[0144] For the scan topological graph of the image where the first matching point is located, the eigenvalues of the normalized Laplace matrix of the scan topological graph are calculated, and a first eigenvalue sequence is obtained in descending order. The second eigenvalue sequence of the Laplace matrix of the corresponding theoretical topological graph is also calculated for each reserved candidate matching point by the same method, and the first seven (the specific number of the first few can be set by the user as needed) largest eigenvalues of each eigenvalue sequence are extracted to form two seven-dimensional vectors. The cosine similarity s of the two seven-dimensional vectors is calculated, and the greater s is, the greater the similarity of the structure of the scan topological graph and the theoretical topological graph is, and then the greater the calculation weight of the neighborhood priority sequence is.

[0145] Optionally, in the embodiment, the preset first eigenvalue and the preset second eigenvalue are obtained according to the first seven obtained according to the sorting result, and the user can set it according to the actual situation, and no numerical limit is made here, which is only a preferred embodiment.

[0146] For the first matching point and each reserved candidate matching point, the similarity of the corresponding descriptor is denoted as x1, the similarity of the neighborhood priority sequence is denoted as x2, (1-s) is taken as the weight of x1, and s is taken as the weight of x2. The matching degree of the first matching point and each reserved candidate matching point is obtained by weighted summation of x1 and x2. The reserved candidate matching point with the maximum matching degree is taken as the matching point of the first matching point in the other image.

[0147] For the second matching point, the third matching point,..., the matching points are obtained by the same method, respectively. That is, a plurality of matching pairs of the depth map and the perspective view under the same perspective are obtained.

[0148] And for each optimal matching point in each matching pair, the corresponding 3D point in the original three-dimensional data (theoretical three-dimensional model) is found by using the projection mapping relationship when the depth map is generated, and the theoretical point is obtained.

[0149] S104, respectively, determine the corresponding original point of each vertex in the scan topological graph.

[0150] In the embodiment, the corresponding 3D point in the three-dimensional scan model obtained by laser scanning is found by using the projection mapping relationship when the perspective view is generated, and the original point is obtained.

[0151] S105, the original point and the theoretical point corresponding to each vertex in the scanning topological graph are combined to form a three-dimensional point correspondence sequence, and a mapping relationship corresponding to the three-dimensional point correspondence sequence is obtained by using the SIFT method.

[0152] In this embodiment, the original point and the theoretical point corresponding to each vertex in the scanning topological graph are combined to form a three-dimensional point correspondence sequence, and a set of three-dimensional point correspondence sequences is obtained: { (3D point A1 corresponding to the three-dimensional scanning model obtained by laser scanning, 3D point B1 in the theoretical three-dimensional model), (A2, B2)... (An, Bn)}. And a mapping relationship corresponding to the three-dimensional point correspondence sequence is obtained by using the SIFT method.

[0153] S106, obtaining a fine registration transformation matrix according to the mapping relationship, the three-dimensional scanning model and the theoretical three-dimensional model.

[0154] In this embodiment, a fine registration transformation matrix is obtained according to the mapping relationship, the three-dimensional scanning model and the theoretical three-dimensional model, and the method specifically comprises:

[0155] The mapping relationship, the preset minimum point set size, the inlier distance threshold and the maximum iteration number are taken as inputs of the RANSAC robust estimation algorithm, and a rigid body transformation matrix is output by the RANSAC robust estimation algorithm;

[0156] The rigid body transformation matrix, the three-dimensional scanning model and the theoretical three-dimensional model are input into the ICP algorithm, and a fine registration transformation matrix is output by the ICP algorithm.

[0157] Exemplarily, due to the information loss of three-dimensional to two-dimensional projection mapping, the similarity ambiguity of feature descriptors on smooth surfaces, and the machining errors and noise interference between the real workpiece and the ideal model, there are still some false matches in the corresponding point pairs in the three-dimensional space. Further, the robust estimation algorithm (such as RANSAC) is used to solve the most robust rigid body transformation matrix.

[0158] The mapping relationship corresponding to the three-dimensional point correspondence sequence obtained by the SIFT method, the preset minimum point set size (usually 3), the inlier distance threshold and the maximum iteration number are taken as inputs, and the RANSAC (random sample consensus) robust estimation algorithm is iteratively calculated. The algorithm randomly extracts the minimum point set to calculate the transformation hypothesis, and verifies all the point pairs, and finally obtains the rigid body transformation matrix. The rigid body transformation matrix represents the optimal rotation and translation parameters required for the coarse alignment of the scanning point cloud and the CAD model, and the inlier set corresponding to the rigid body transformation matrix represents the high-confidence correct matching point pairs selected from the initial matching, which provides a pure and reliable data basis for the subsequent ICP fine matching.

[0159] The rigid body transformation matrix, the three-dimensional scanning model and the theoretical three-dimensional model are input into the ICP algorithm, and a fine registration transformation matrix is output through the ICP algorithm.

[0160] In S107, the fine registration transformation matrix is decomposed to obtain a change amount, and the change amount is converted into a control instruction to position and adjust a defective link in the hyperbolic plate manufacturing process.

[0161] Exemplarily, the change amount of the 6 degrees of freedom (translation amounts ΔX, ΔY and ΔZ along the X, Y and Z axes and rotation amounts θx, θy and θz around the X, Y and Z axes) of the workpiece that need to be adjusted are decomposed from the rigid body transformation matrix obtained in the last step.

[0162] The change amounts are converted into motion control instructions of each positioner or robot joint by considering the kinematic model of the actuator (such as a robot gripper).

[0163] The control system drives the positioning actuator (such as the gripper at the end of the robot) to fine-tune the workpiece according to the calculated adjustment amount. After the adjustment is completed, the three-dimensional scanner is started again to perform re-measurement, and new point cloud data is obtained. The new point cloud and the CAD model are matched again to verify whether the current pose meets the tolerance requirement.

[0164] If it is qualified: the process is ended, the tooling is locked, and subsequent processing (such as drilling and edge milling) is performed; if it is not qualified: the fine registration transformation matrix, the new point cloud data and the theoretical three-dimensional model are input into the ICP algorithm, a new fine registration transformation matrix is output through the ICP algorithm, the control instruction is recalculated according to the new fine registration transformation matrix, and the point cloud data is obtained again, and the operation is repeated to perform secondary fine-tuning, forming a closed-loop feedback system, until the accuracy requirement is met. Usually, 1-2 iterations can achieve very high accuracy.

[0165] Thus, the application is completed.

[0166] In summary, in the embodiment of the application, the SIFT matching method based on the depth map is used, first, multi-view depth maps are generated from the hyperbolic plate scanning point cloud and the CAD model, and the rotation and scale invariant robust SIFT features are extracted; then, through high-dimensional descriptor matching and RANSAC robust estimation, a high-precision initial rigid body transformation matrix is calculated, so that the global coarse registration of the scanning data and the design model is realized. Furthermore, the matching ambiguity problem caused by the smooth surface and the sparse features of the hyperbolic plate can be effectively overcome, an almost ideal iteration starting point is provided for the subsequent ICP matching, so that the iteration resources are not consumed in the wrong direction, the optimization of the small local deformation can be focused on, and the accuracy, convergence speed and overall robustness of the final ICP matching are significantly improved.

[0167] The application further provides a double-curved plate processing intelligent positioning system. Figure 4 The application further provides a double-curved plate processing intelligent positioning system.

[0168] The data acquisition module 101 is configured to acquire a three-dimensional scanning model and a theoretical three-dimensional model of the double-curved plate, and acquire a scanning topological graph and a theoretical topological graph according to the three-dimensional scanning model and the theoretical three-dimensional model.

[0169] The data processing module 102 is configured to acquire a topological relationship strength of each vertex in the scanning topological graph and a topological relationship strength of each vertex in the theoretical topological graph, and sort the topological relationship strengths respectively to obtain a scanning priority sequence and a theoretical priority sequence respectively; acquire an optimal matching point of each vertex in the scanning topological graph in the theoretical topological graph according to the scanning priority sequence and the theoretical priority sequence, and determine a theoretical point corresponding to each optimal matching point; determine an original point corresponding to each vertex in the scanning topological graph respectively; form a three-dimensional point correspondence sequence by combining the original point and the theoretical point corresponding to each vertex in the scanning topological graph, acquire a mapping relationship corresponding to the three-dimensional point correspondence sequence by using a SIFT method; and acquire a fine registration transformation matrix according to the mapping relationship, the three-dimensional scanning model and the theoretical three-dimensional model.

[0170] The defect positioning module 103 is configured to decompose the fine registration transformation matrix to obtain a variation, convert the variation into a control instruction, and position and adjust a defective link in a double-curved plate manufacturing process.

[0171] It should be noted that the system provided in the above examples is only exemplified by the division of the above functional modules, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the double-curved plate processing intelligent positioning system and the double-curved plate processing intelligent positioning method provided in the above examples belong to the same concept, and the specific implementation process is described in the method embodiment, which will not be repeated here.

[0172] It should be noted that the above-mentioned order of the embodiments of the application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0173] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

[0174] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for intelligent positioning of hyperbolic plate processing, characterized in that, The method comprises the following steps: obtain a three-dimensional scanning model and a theoretical three-dimensional model of a hyperbolic plate, and obtain a scanning topological graph and a theoretical topological graph according to the three-dimensional scanning model and the theoretical three-dimensional model; obtain the topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph, and sort them respectively to obtain a scanning priority sequence and a theoretical priority sequence respectively; determine the vertex corresponding to the first element in the scanning priority sequence in the scanning topological graph as a first matching point; determine a candidate matching point corresponding to the first matching point from the theoretical priority sequence, wherein the absolute value of the difference between the degree of the first matching point and the degree of the candidate matching point is less than a preset degree threshold; calculate the similarity between the descriptor of each candidate matching point and the descriptor of the first matching point, and determine the candidate matching point with a similarity greater than a preset similarity threshold as a reserved candidate matching point; determine a neighborhood point priority sequence of the first matching point according to the scanning topological graph; determine a neighborhood priority sequence of the reserved candidate matching point according to the theoretical topological graph; obtain the cosine similarity of the scanning topological graph and the theoretical topological graph; calculate the matching degree of the first matching point and each reserved candidate matching point according to the cosine similarity, the neighborhood point priority sequence and the neighborhood priority sequence, and determine the reserved candidate matching point with the maximum matching degree as the optimal matching point of the first matching point; determine the vertex corresponding to the A-th element in the scanning priority sequence in the scanning topological graph as the A-th matching point; obtain the optimal matching point of each vertex corresponding to each element in the scanning priority sequence in the scanning topological graph, and determine the theoretical point corresponding to each optimal matching point; determine the original point corresponding to each vertex in the scanning topological graph respectively; compose the original point and the theoretical point corresponding to each vertex in each scanning topological graph into a three-dimensional point correspondence sequence, and obtain the mapping relationship corresponding to the three-dimensional point correspondence sequence by using the SIFT method; obtain a fine registration transformation matrix according to the mapping relationship, the three-dimensional scanning model and the theoretical three-dimensional model; decompose the fine registration transformation matrix to obtain a change amount, convert the change amount into a control instruction, and position and adjust the defective link in the manufacturing process of the hyperbolic plate.

2. The intelligent positioning method for hyperbolic plate processing according to claim 1, characterized in that, The method of obtaining the scanning topological graph and the theoretical topological graph according to the three-dimensional scanning model and the theoretical three-dimensional model comprises the following steps: obtain a perspective graph and a depth graph according to the three-dimensional scanning model and the theoretical three-dimensional model respectively, wherein the acquisition parameters of the perspective graph and the depth graph are the same; obtain scanning key points and theoretical key points corresponding to the perspective graph and the depth graph respectively by using the SIFT algorithm; construct a triangular mesh by taking the scanning key points and the theoretical key points as vertices respectively by using the Delaunay triangulation method to obtain the scanning topological graph and the theoretical topological graph.

3. The intelligent positioning method for hyperbolic plate processing according to claim 1, characterized in that, The method of obtaining the topological relationship strength of each vertex in the scanning topological graph and the topological relationship strength of each vertex in the theoretical topological graph comprises the following steps: obtain the degree of each vertex in the topological graph, and determine the relative degree centrality of the N-th vertex in the topological graph as the ratio of the degree of the N-th vertex to the maximum degree of the vertex, wherein the topological graph is the scanning topological graph or the theoretical topological graph. The distance sum of the Nth vertex in the topology graph is obtained by Dijkstra shortest path algorithm, the distance sum of the Nth vertex is taken as a negative number, and the center of the Nth vertex is obtained by taking the negative number as an exponent of a natural base; The product of the relative degree centrality of the Nth vertex and the center of the Nth vertex is determined as the topology relationship strength of the Nth vertex; The topology relationship strength of each vertex in the scanning topology graph and the topology relationship strength of each vertex in the theoretical topology graph are determined respectively.

4. The intelligent positioning method for hyperbolic plate processing according to claim 1, characterized in that, The method for determining the priority sequence of the neighbor points of the first matching point according to the scanning topology graph comprises the following steps: The neighbor points of the first matching point are determined from the scanning topology graph, wherein the neighbor points of the first matching point and the first matching point are connected through edges of the scanning topology graph; The similarity between the descriptor of each neighbor point of the first matching point and the descriptor of the first matching point is calculated, and an ascending order sorting is performed to obtain a neighbor point sequence of the first matching point; The sequence number value of each element in the scanning priority sequence is determined as the priority sequence value of the corresponding vertex in the scanning topology graph, and the priority sequence value of each vertex in the scanning topology graph is obtained; The value of each element in the neighbor point sequence of the first matching point is replaced by the priority sequence value of the corresponding vertex in the scanning topology graph, and a neighbor priority sequence is obtained.

5. The intelligent positioning method for hyperbolic plate processing according to claim 1, characterized in that, The method for determining the neighbor priority sequence of the reserved candidate matching point according to the theoretical topology graph comprises the following steps: The neighbor points of the reserved candidate matching point are determined from the theoretical topology graph, wherein the neighbor points of the reserved candidate matching point and the reserved candidate matching point are connected through edges of the theoretical topology graph; The similarity between the descriptor of each neighbor point of the reserved candidate matching point and the descriptor of the reserved candidate matching point is calculated, and an ascending order sorting is performed to obtain a neighbor point sequence of the reserved candidate matching point; The sequence number value of each element in the theoretical priority sequence is determined as the priority sequence value of the corresponding vertex in the theoretical topology graph, and the priority sequence value of each vertex in the theoretical topology graph is obtained; The value of each element in the neighbor point sequence of the reserved candidate matching point is replaced by the priority sequence value of the corresponding vertex in the theoretical topology graph, and a neighbor priority sequence is obtained.

6. The intelligent positioning method for hyperbolic plate processing according to claim 1, characterized in that, The method for obtaining the cosine similarity of the scanning topology graph and the theoretical topology graph comprises the following steps: The eigenvalues of the normalized Laplacian matrix of the scanning topology graph are calculated, the eigenvalues are arranged in descending order, and a first eigenvalue sequence is obtained; The eigenvalues of the normalized Laplacian matrix of the theoretical topology graph are calculated, the eigenvalues are arranged in descending order, and a second eigenvalue sequence is obtained; An element greater than a preset first eigenvalue in the first eigenvalue sequence is obtained, and a first vector is obtained; An element greater than a preset second eigenvalue in the second eigenvalue sequence is obtained, and a second vector is obtained; The cosine similarity of the first vector and the second vector is calculated, and the cosine similarity of the scanning topology graph and the theoretical topology graph is determined.

7. The intelligent positioning method for hyperbolic plate processing according to claim 1, characterized in that, The method for calculating the matching degree between the first matching point and each reserved candidate matching point according to the cosine similarity, the neighbor priority sequence and the neighbor priority sequence comprises the following steps: Determine the similarity between the Mth reserved candidate matching point descriptor and the first matching point descriptor as the Mth descriptor similarity; Take the cosine similarity as negative and add 1, and multiply the Mth descriptor similarity to obtain the first matching degree of the Mth reserved candidate matching point; Calculate the similarity between the first matching point neighborhood point priority sequence and the Mth reserved candidate matching point neighborhood priority sequence to obtain the Mth neighborhood similarity; Determine the product of the Mth neighborhood similarity and the cosine similarity as the second matching degree of the Mth reserved candidate matching point; Determine the sum of the first matching degree of the Mth reserved candidate matching point and the second matching degree of the Mth reserved candidate matching point as the matching degree between the first matching point and the Mth reserved candidate matching point; Obtain the matching degree between the first matching point and each reserved candidate matching point.

8. The intelligent positioning method for hyperbolic plate processing according to claim 1, wherein, The method comprises the following steps: Input the mapping relationship, the preset minimum point set size, the inner point distance threshold value and the maximum iteration number into the RANSAC robust estimation algorithm, and output the rigid body transformation matrix through the RANSAC robust estimation algorithm; Input the rigid body transformation matrix, the three-dimensional scanning model and the theoretical three-dimensional model into the ICP algorithm, and output the fine registration transformation matrix through the ICP algorithm.

9. An intelligent positioning system for hyperbolic plate processing, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is executed by the processor to realize the steps of the method for intelligent positioning of hyperbolic plate processing according to any one of claims 1-8.

Images