A method for identifying and classifying similar aircraft connectors based on three-dimensional measurement

Abstract

The present application belongs to the technical field of visual recognition of similar parts and components, and is a method for identifying and classifying similar aircraft connectors based on three-dimensional measurement. It proposes to establish a standard point cloud database and feature template library for similar aircraft connectors, and on this basis, innovatively divides the identification and classification of similar aircraft connectors into three steps: maximum outer contour screening, similarity calculation based on spatial positioning, and key feature comparison: based on the definition and creation of four-point sets and the point-to-point ICP algorithm as the theoretical basis, the point clouds are aligned to complete the spatial positioning of similar connectors; the feature point cloud data corresponding to each similar connector point cloud to be classified in the standard database is extracted, the corresponding feature parameters are calculated, and compared with the parameters in the feature template library to complete the final identification and classification of similar connectors, with high recognition accuracy and high recognition efficiency.

Description

Technical Field

[0001] The present application belongs to the technical field of visual recognition of similar parts, and in particular, relates to a method for identifying and classifying similar connecting parts of aircraft based on three-dimensional measurement. Background Art

[0002] At present, the aviation equipment chain is becoming more complete, with the production and manufacturing capabilities of the entire industry chain. Among them, aircraft connectors are an indispensable part of the research and development and production of aviation equipment. Their function is to transfer and distribute the loads between different components. The stress state is complex and they belong to "transitional elements". Moreover, there are many types of connectors, large quantities, and small differences in size and shape. Some connectors are highly similar, with the smallest size difference being only 0.1mm. Therefore, in the production process of aviation equipment, the identification and selection of connectors is very important. In traditional manufacturing, the identification and classification of similar connectors requires workers to use calipers, gauges and other tools for manual measurement, which has high technical requirements for workers, high labor intensity, and low identification efficiency.

[0003] Most of the research work on similar parts identification and classification uses template matching or deep learning methods to solve the problem, but in the identification and classification of highly similar parts, its accuracy cannot meet the requirements; among them, the template matching-based method has poor robustness, and the neural network algorithm also requires the production of data sets and a lot of training, which makes it difficult to take into account both the accuracy and efficiency of identification and classification. With the research and development of 3D laser scanning technology, especially in the aviation field, it has been widely studied and applied in the detection and identification of aviation parts due to its advantages of non-contact, fast speed, high accuracy, and strong anti-interference ability on site, with the highest accuracy reaching the micron level.

[0004] Therefore, how to ensure the accuracy and efficiency of similar connector identification at the same time is a problem that needs to be solved. Summary of the invention

[0005] The purpose of this application is to provide a method for identifying and classifying similar aircraft connectors based on three-dimensional measurement, so as to solve the problem in the prior art that it is difficult to strike a balance between classification accuracy and efficiency when identifying similar connectors using template matching or deep learning methods.

[0006] The technical solution of the present application is: a method for identifying and classifying similar aircraft connectors based on three-dimensional measurement, comprising:

[0007] Randomly sample the surface of the theoretical three-dimensional model of the standard similar connector to obtain standard point cloud data of each standard similar connector; extract features and classify and store each standard point cloud data to obtain a standard database;

[0008] The characteristic parameters of each standard similar connector are calculated according to the standard database. The characteristic parameters include the standard three-dimensional size of the minimum circumscribed rectangular box, the standard area of ​​the two-dimensional feature projection point cloud, and the standard length of the key geometric similar feature point cloud extracted by segmentation. Each characteristic parameter is integrated and classified to form a characteristic template library.

[0009] Use a 3D scanner to scan and obtain physical point cloud data of similar connectors to be identified, analyze and determine the type of noise in the physical point cloud data, denoise and downsample the physical point cloud data, and obtain point cloud data of parts to be classified;

[0010] Calculate the size data of the point cloud of each similar connector to be identified by scanning according to the part point cloud data, perform maximum outer contour screening on the similar connector to be identified, and obtain different categories of similar connectors to be identified;

[0011] The point cloud data of the parts of similar connectors to be identified that are classified into the same category are spatially positioned based on registration, and the numerical accuracy index after spatial positioning is calculated. Then, the point clouds of similar connectors to be identified with high similarity are sorted from high to low according to the numerical accuracy index, and the point clouds of similar connectors to be identified with high similarity are grouped together;

[0012] The key features of the point cloud of similar connectors to be identified in the same group are extracted, the feature parameters in the feature template library corresponding to the key features are calculated, and the similar connectors to be identified in each group are finally classified according to the feature parameters.

[0013] Preferably, the specific process of establishing the standard database is:

[0014] The theoretical digital model of each standard similar connector in a standard posture is converted into an STL format, and based on a random sampling algorithm program, the surface of the theoretical digital model is randomly sampled to generate standard point cloud data of each standard similar connector;

[0015] Construct a feature projection surface for each standard similar connector, project at least two sets of standard point cloud data of the same type containing similar features into a two-dimensional plane to form at least two sets of two-dimensional feature projection point clouds, calculate the normal vectors of the two-dimensional feature projection point clouds containing similar feature point clouds, and extract the key geometric similar feature point clouds of the standard similar connectors according to the normal vectors;

[0016] Randomly select two points in the key geometric similarity feature point cloud to create a class, fit the two selected points into a straight line, search in a round-robin manner for points in various key geometric similarity feature point clouds whose distances from the straight line meet a threshold and classify them into the same class, extract each key geometric similarity feature point cloud from the two-dimensional feature projection point cloud containing similar features by calculating the normal vector of the similar feature point cloud, and classify them by searching in a round-robin manner for points whose straight line distances meet a threshold, and obtain standard key geometric similarity feature point clouds of different categories;

[0017] The standard point cloud data, two-dimensional feature projection point cloud, and key geometric similarity feature point cloud of each standard similar connector are integrated, classified, and stored to form a standard database.

[0018] Preferably, the specific method for establishing the feature template library is:

[0019] Based on the standard point cloud data of each standard similar connector in the standard database, the minimum circumscribed rectangular box of the corresponding standard similar connector is calculated, and the three-dimensional size of the minimum circumscribed rectangular box of each standard similar connector is recorded respectively;

[0020] Based on the two-dimensional feature projection point cloud and key geometric similarity feature point cloud of each standard similar connector in the standard database, the area of ​​the two-dimensional feature projection point cloud and the length of the key geometric similarity feature point cloud extracted by segmentation are calculated respectively;

[0021] The calculated feature parameters are integrated and classified. The feature parameters include the three-dimensional size of the minimum circumscribed rectangular box, the area of ​​the two-dimensional feature projection point cloud, and the length of the key geometric similarity feature point cloud extracted by segmentation. The feature parameters of each independent standard similar connector are constructed and stored separately to form a feature template library.

[0022] Preferably, the specific method for acquiring the part point cloud data is:

[0023] First, the 3D scanner is spatially calibrated, and similar connectors to be identified and classified are scanned to obtain physical point cloud data;

[0024] The point cloud data of similar connectors to be identified are downsampled using a voxel downsampling method that retains the original features, and the point closest to the centroid of the voxel grid is selected as the point after downsampling to form a noise point cloud and obtain the noise type;

[0025] The noise point cloud in the physical point cloud data of similar connectors to be identified is spatially located, and the three-dimensional straight-through filtering method is used to remove the noise to obtain the point cloud data of the parts to be classified.

[0026] Preferably, the specific method for screening the maximum outer contour of similar connectors to be identified is:

[0027] The main direction vector of the physical point cloud data is calculated using the main direction eigenvalue analysis method;

[0028] According to the main direction vector, the transformation matrix of the main direction vector of the similar connection to be identified is calculated to align with the positive direction of the Z axis, and the physical point cloud data translated to the center of mass is applied for rigid transformation to obtain the physical point cloud data of the similar connection to be identified after rigid transformation. ,calculate:

[0029]

[0030] In the above formula, is the principal direction vector Alignment matrix with the positive Z axis, The matrix constructed by the three-dimensional coordinates of the point cloud after removing the centroid, It is the physical point cloud data of similar connectors to be identified after rigid transformation;

[0031] Physical point cloud data of similar connectors to be identified after rigid transformation Traverse each point in the to-be-identified similar connector to find its maximum and minimum values ​​on the three coordinate axes, and then calculate the minimum circumscribed rectangular box size of each similar connector point cloud to be identified, and obtain the maximum outer contour of the similar connector to be identified;

[0032] The minimum bounding rectangular box size of the point cloud of similar connectors to be identified is compared with the standard three-dimensional size of the minimum bounding rectangular box in pairs, and the similar connectors to be identified that meet the threshold standard are classified into one category.

[0033] Preferably, the calculation process of the main direction vector is as follows:

[0034] Let the point cloud data set be Q, where Q is a data set containing n points, and each point By three-dimensional coordinates Representation, let the unit vector is the main direction vector of the point cloud data; a 3×n matrix X is constructed through the point cloud data Q, where each column corresponds to the coordinates of a point;

[0035] Calculate the centroid coordinates of point cloud data , the centroid coordinates are the row means of X, Y and Z respectively, and the specific calculation formula is as follows:

[0036] The physical point cloud data of similar connectors to be identified are de-centred to obtain the matrix constructed by the three-dimensional coordinates of the point cloud after de-centring. The change formula is as follows:

[0037] Calculate the covariance matrix C of the point cloud data after removing the centroid. The calculation formula is as follows:

[0038] Perform eigenvalue decomposition on the covariance matrix C, obtain three eigenvalues ​​and set the three eigenvalues and the corresponding eigenvector matrix V; where the column vectors of V are the eigenvectors , the main direction vector is selected as the eigenvector corresponding to the maximum eigenvalue, that is, .

[0039] Preferably, a specific method for classifying similar connectors to be identified that meet the threshold standard into one category is:

[0040] Calculate the absolute value of the three-dimensional size of the minimum bounding rectangle:

[0041]

[0042]

[0043]

[0044] In the formula, , , Point cloud data for physical objects The maximum value on the three coordinate axes, , , It is the maximum value of the standard three-dimensional size of the smallest circumscribed rectangular box in the standard point cloud data on the three coordinate axes. , , is the absolute value of the three-dimensional size of the minimum bounding rectangular box;

[0045] Setting the Threshold , , , All less than , the two similar connectors to be identified are placed in the same category, denoted as Similar connectors to be identified.

[0046] Preferably, the specific steps for calculating the numerical accuracy index are as follows:

[0047] First, the 4PCS registration algorithm based on the classic four-point set theory that defines invariants and invariant ratios is used to perform rough registration of the point cloud of similar connectors to be identified. Then, the point-to-point ICP-based registration algorithm is used to achieve fine registration of the point cloud of similar connectors to be identified and complete spatial positioning.

[0048] The RMSE value of the point cloud of similar connectors to be identified after fine registration is calculated, and based on this, the similarity of similar connectors to be identified after rough registration and fine registration is characterized. The calculation formula is as follows:

[0049] In the above formula, m represents the number of matching point pairs after the target point cloud and the source point cloud are registered. and Respectively represent the spatial three-dimensional coordinate values ​​of the corresponding points of the target point cloud and the source point cloud after point cloud registration;

[0050] Setting Thresholds , the similarity value is less than the threshold The similar connectors to be identified and have an associated relationship are grouped into the same group again, which is recorded as Group.

[0051] Preferably, the specific method of finally classifying the similar connectors to be identified in each group by using the characteristic parameters is as follows:

[0052] Construct a two-dimensional feature projection surface, project the scanned similar connector point cloud to be identified on the two-dimensional feature projection surface to obtain a two-dimensional projection point cloud, and then remove duplicate points in the two-dimensional projection point cloud by creating a two-dimensional surface grid to obtain boundary points, and obtain boundary point cloud data B by statistics;

[0053] Calculate the total number of points S in the two-dimensional projection point cloud after removing duplicate points, and compare it with the total number of points of similar connectors corresponding to the standard in the standard template library. Compare and get the absolute value of the difference :

[0054]

[0055] Calculate the normal vector of the two-dimensional projection point cloud and extract the contour points in the two-dimensional projection point cloud;

[0056] According to the extracted boundary points, a small sample is randomly selected and the feature model is fitted. Then, the error from the contour point to the model is calculated to select the correct point set. Through continuous iteration, the fitting segmentation result is finally obtained, and the final classification of similar connectors to be identified is completed, which is recorded as .

[0057] Preferably, the specific method of iteratively screening the correct point set is:

[0058] Input boundary point cloud data B, and randomly select two points to create a spatial line;

[0059] Calculate the distance from the contour point to the spatial straight line and set the distance screening threshold , filter out the inliers according to the distance screening threshold and fit the feature model;

[0060] Given the number of iterations n, repeatedly screen the inliers and count the number of inliers, and extract the data points corresponding to the feature model with the most inliers;

[0061] Repeatedly extract the data points in the feature model with the most internal points until all the feature line point clouds in B are segmented;

[0062] Based on the segmented feature line point cloud, the corresponding length parameter in the standard feature template library is calculated. The characteristic length of similar connectors to be classified and identified is designed to be L. The length parameter in the standard feature template library is , the absolute value of the difference is :

[0063]

[0064] Setting Thresholds and , when the following conditions are met:

[0065] That is, the final classification of similar connectors to be identified is considered completed.

[0066] The present application discloses a method for identifying and classifying similar aircraft connectors based on three-dimensional measurement, and proposes to establish a standard point cloud database and feature template library for similar aircraft connectors. Based on this, the method innovatively divides the identification and classification of similar aircraft connectors into three steps: maximum outer contour screening, similarity calculation based on spatial positioning, and key feature comparison. The method uses the definition and creation of a four-point set and a point-to-point ICP algorithm as a theoretical basis to align the point cloud and complete the spatial positioning of similar connectors. The method extracts the feature point cloud data corresponding to each similar connector point cloud to be classified in the standard database, calculates the corresponding feature parameters, and compares them with the parameters in the feature template library to complete the final identification and classification of similar connectors, with high recognition accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solution provided by the present application, the following is a brief introduction to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present application.

[0068] Figure 1 This is the overall flow chart of this application. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] A method for identifying and classifying similar aircraft connectors based on three-dimensional measurement, such as Figure 1 As shown, the following steps are included:

[0071] Step S100, establishing a standard database based on standard similar connector point clouds:

[0072] Taking the theoretical three-dimensional model of each standard similar connector as the data basis, the surface of the theoretical three-dimensional model is randomly sampled to obtain the standard point cloud data of each standard similar connector; feature extraction is performed on each standard point cloud data to obtain the key geometric similarity feature point cloud of each standard similar connector, and the key geometric similarity feature point cloud is classified by calculating the straight-line distance to obtain the key geometric similarity feature point clouds of different categories of standards; and a standard database is obtained through classification and storage.

[0073] The specific process of establishing the standard database is as follows:

[0074] Step S110, converting the theoretical digital model of each standard similar connector in a standard posture into an STL format, and randomly sampling the surface of the theoretical digital model based on a random sampling algorithm program to generate standard point cloud data of each standard similar connector;

[0075] Step S120, constructing a feature projection surface for each standard similar connector, projecting at least two groups of standard point cloud data of the same type containing similar features into a two-dimensional plane to form at least two groups of two-dimensional feature projection point clouds, calculating normal vectors of the two-dimensional feature projection point clouds containing similar feature point clouds, and extracting key geometric similar feature point clouds of the standard similar connectors according to the normal vectors;

[0076] The normal vectors in the two-dimensional feature projection point clouds formed by different types of standard similar connectors are extracted respectively to obtain key geometric similarity feature point clouds of different types of standard similar connectors;

[0077] Step S130, randomly select two points in the key geometric similarity feature point cloud to create a class, fit the two selected points into a straight line, search in a round-robin manner for points in various key geometric similarity feature point clouds whose distances from the straight line meet a threshold and classify them into the same class, extract each key geometric similarity feature point cloud from a two-dimensional feature projection point cloud containing similar features by calculating a normal vector of the similar feature point cloud, and classify them by searching in a round-robin manner for points whose straight line distances meet a threshold, thereby obtaining standard key geometric similarity feature point clouds of different categories, and realizing fitting segmentation and extraction of multiple similar local contour features;

[0078] Step S140, integrate, classify and store the standard point cloud data, two-dimensional feature projection point cloud, and key geometric similarity feature point cloud before and after fitting segmentation of each standard similar connector to form a standard database for subsequent indexing and calling.

[0079] Step S200, establishing a feature template library of standard similar connector point clouds:

[0080] The feature parameters in the standard database are obtained. The feature parameters include standard point cloud data, two-dimensional feature projection point cloud and key geometric similarity feature point cloud. The standard three-dimensional size of the minimum circumscribed rectangular box of each standard similar connector, the standard area of ​​the two-dimensional feature projection point cloud and the standard length of the key geometric similarity feature point cloud extracted by segmentation are calculated respectively. Each feature parameter is integrated and classified to form a feature template library for index matching.

[0081] The specific method of establishing the feature template library is as follows:

[0082] Step S210, based on the standard point cloud data of each standard similar connector in the standard database, calculate the minimum circumscribed rectangular box of the corresponding standard similar connector, and record the three-dimensional size of the minimum circumscribed rectangular box of each standard similar connector.

[0083] Step S220, using the two-dimensional feature projection point cloud and the key geometric similarity feature point cloud of each standard similar connector in the standard database as data basis, respectively calculate the area of ​​the two-dimensional feature projection point cloud and the length of the key geometric similarity feature point cloud extracted by segmentation, and replace the area of ​​the two-dimensional feature projection point cloud with the number of points after downsampling processing by the voxel downsampling method that retains the original features, and replace the length of the key geometric similarity feature point cloud with the distance between any two largest points in the point cloud.

[0084] In step S230, the feature parameters calculated in steps S210 and S220 are integrated and classified. The feature parameters include the three-dimensional size of the minimum circumscribed rectangular box, the area of ​​the two-dimensional feature projection point cloud, and the length of the key geometric similarity feature point cloud extracted by segmentation. The feature parameters of each independent standard similar connector are constructed and stored separately to form a feature template library for subsequent index query. Then, an information transmission interface of the feature template library of standard similar connectors is constructed to match the mapping relationship between the feature parameters and the corresponding geometric quantities, laying a data foundation for the final classification.

[0085] Step S300, preprocessing the physical point cloud data of similar connectors to be identified:

[0086] Use a 3D scanner to obtain the physical point cloud data of similar connectors to be identified, analyze and determine the noise type in the physical point cloud data, and use an adapted 3D straight-through filtering method to remove the noise in the physical point cloud data of similar connectors to be identified obtained by scanning after downsampling the noise points, so as to improve the accuracy of subsequent recognition and classification; downsampling uses a voxel downsampling algorithm that retains the original features, and sets appropriate parameters to reduce the redundant data of the overall physical point cloud data, improve the processing speed of subsequent point clouds, and finally obtain the point cloud data of parts to be classified.

[0087] The specific method for obtaining part point cloud data is as follows:

[0088] In step S310, the three-dimensional scanner is first spatially calibrated, and then similar connecting parts to be identified and classified are scanned to obtain physical point cloud data.

[0089] Step S320, use the voxel downsampling method that retains the original features to downsample the point cloud data of the similar connectors to be identified, select the point closest to the voxel grid centroid as the point after downsampling, form a noise point cloud and obtain the noise type; this design can reduce the amount of data and improve the subsequent processing speed while maintaining the detail size of the similar features of the point cloud of the similar connectors to be identified and homogenize the main point cloud.

[0090] Step S330, spatially locate the noise point cloud in the physical point cloud data of similar connectors to be identified, select a suitable filtering method to remove the noise, preferably use a straight-through filtering algorithm with specified three coordinates to remove the located noise point cloud, and obtain the point cloud data of the parts to be classified to improve the subsequent recognition and classification accuracy.

[0091] Step S400, screening the similar connectors to be identified for their maximum outer contours:

[0092] Based on the scanned part point cloud data obtained in step S300, the length, width, and height dimension data of the point cloud of each similar connector to be identified are calculated. Specifically, the minimum circumscribed rectangular box of the point cloud is calculated to obtain the maximum outer contour of the similar connector to be identified; screening is performed according to the different three-dimensional sizes of the minimum circumscribed rectangular box to obtain different categories of similar connectors to be identified; specifically, a threshold is set for comparison. If the three-dimensional size difference is large, they are classified as different types of similar connectors to be identified; if the three-dimensional size difference is small, they are classified as the same type of similar connector to be identified, to be further subdivided.

[0093] The specific method for screening the maximum outer contour of similar connectors to be identified is:

[0094] Step S410, using the main direction eigenvalue analysis method to calculate the main direction vector of the physical point cloud data, the calculation process is as follows:

[0095] ① Let the point cloud data set be Q, where Q is a data set containing n points, and each point By three-dimensional coordinates Representation, let the unit vector is the main direction vector of the point cloud data; a 3×n matrix X is constructed through the point cloud data Q, where each column corresponds to the coordinates of a point;

[0096]

[0097] ②Calculate the centroid coordinates of the point cloud data , the centroid coordinates are the row means of X, Y and Z respectively, and the specific calculation formula is as follows:

[0098]

[0099] ③ The physical point cloud data of the similar connector to be identified is de-centred to obtain the matrix constructed by the three-dimensional coordinates of the point cloud after de-centring. The change formula is as follows:

[0100]

[0101] In the above formula, The matrix constructed for the three-dimensional coordinates of the point cloud after removing the centroid;

[0102] ④ Calculate the covariance matrix C of the point cloud data after removing the centroid. The calculation formula is as follows:

[0103] ;

[0104] ⑤ Perform eigenvalue decomposition on the covariance matrix C, obtain three eigenvalues ​​and set three eigenvalues and the corresponding eigenvector matrix V; where the column vectors of V are the eigenvectors , the main direction vector is selected as the eigenvector corresponding to the maximum eigenvalue, that is, .

[0105] Step S420, based on the main direction vector, calculate the transformation matrix of the main direction vector of the similar connection to be identified to align with the positive direction of the Z axis, and apply the physical point cloud data translated to the centroid to perform a rigid transformation to obtain the physical point cloud data of the similar connection to be identified after the rigid transformation. ,calculate;

[0106]

[0107] In the above formula, is the principal direction vector Alignment matrix with the positive Z axis, It is the physical point cloud data of similar connectors to be identified after rigid transformation.

[0108] Step S430: the physical point cloud data of the similar connector to be identified after the rigid transformation Traverse each point in the , find its maximum and minimum values ​​on the three coordinate axes, and then calculate the minimum circumscribed rectangular box size of each similar connector point cloud to be identified, and obtain the maximum outer contour of the similar connector to be identified. The calculation process is as follows:

[0109]

[0110]

[0111]

[0112] In the formula, , , They are The maximum value on the three coordinate axes, , , They are The minimum value on the three coordinate axes, , , is the minimum size of the bounding rectangle of the point cloud data of similar connectors to be identified;

[0113] Step S440, calculate the minimum bounding rectangular box size of each pre-processed similar connection point cloud to be identified, compare them with the standard three-dimensional size of the minimum bounding rectangular box, and classify the similar connection to be identified that meets the threshold standard into one category. The specific method is as follows:

[0114] ① Calculate the absolute value of the three-dimensional size of the minimum bounding rectangle:

[0115]

[0116]

[0117]

[0118] In the formula, , , Point cloud data for physical objects The maximum value on the three coordinate axes, , , It is the maximum value of the standard three-dimensional size of the smallest circumscribed rectangular box in the standard point cloud data on the three coordinate axes. , , is the absolute value of the three-dimensional size of the minimum bounding rectangular box;

[0119] ②Set threshold , , , All less than , the two similar connectors to be identified are placed in the same category, denoted as Similar connectors to be identified. The setting is adjusted according to the size magnitude of similar connectors to be identified.

[0120] Step S500, similarity calculation of the point cloud of the similar connectors to be identified based on spatial positioning:

[0121] For the similar connectors to be identified that are classified into the same category in the maximum outer contour screening in step S400, the similar connectors to be identified are further subdivided; the part point cloud data of the similar connectors to be identified that are classified into the same category are spatially positioned based on the registration, and the numerical accuracy index after the spatial positioning is calculated to characterize the point cloud similarity between the similar connectors to be identified in this category, and then the similar connectors to be identified in the same category are sorted from high to low according to the numerical accuracy index, and the point clouds of the similar connectors to be identified with high similarity are grouped into one group for further subdivision.

[0122] The specific steps for calculating numerical accuracy indicators are as follows:

[0123] Step S510, first use the 4PCS registration algorithm of the classical four-point set theory that defines invariants and invariant ratios to perform coarse registration of the point cloud of similar connectors to be identified, and then use the point-to-point ICP-based registration algorithm to achieve fine registration of the point cloud of similar connectors to be identified, and complete spatial positioning;

[0124] Step S520, calculating the RMSE value of the point cloud of the similar connector to be identified after fine registration, and based on this, characterizing the similarity of the similar connector to be identified after rough registration and fine registration, the calculation formula is as follows:

[0125] In the above formula, m represents the number of matching point pairs after the target point cloud and the source point cloud are registered. and Respectively represent the spatial three-dimensional coordinate values ​​of the corresponding points of the target point cloud and the source point cloud after point cloud registration;

[0126] Step S530, setting a threshold , such as 5%, the similarity value is less than the threshold The similar connectors to be identified and have an associated relationship are grouped into the same group again, which is recorded as Group, thereby completing the screening of similar connectors to be identified in the same group.

[0127] Step S600, compare key features of similar connectors to be identified and perform final classification:

[0128] For the similar connectors to be identified that are classified into the same group in the identification in step S500, there is only a difference of 0.1mm-0.2mm in size and shape; the point cloud of the similar connectors to be identified includes a feature surface point cloud and a feature line point cloud, the feature surface point cloud includes the feature surface shape and the feature surface area, the feature line point cloud includes the line point cloud length, the angle between feature line segments and the distance between parallel line segments, and by extracting the key features of the points, lines and surfaces of the point cloud of the similar connectors to be identified in the same group, and calculating the feature parameters in the feature template library corresponding to the key features, the similar connectors to be identified in each group are finally classified based on this.

[0129] The specific steps for classification by feature parameters are as follows:

[0130] Step S610, constructing a two-dimensional feature projection surface, projecting the scanned similar connector point cloud to be identified on the two-dimensional feature projection surface to obtain a two-dimensional projection point cloud, and then removing duplicate points in the two-dimensional projection point cloud by creating a two-dimensional surface grid to obtain boundary points, and obtaining boundary point cloud data B by statistics, which makes the total number of points in the point cloud a fixed parameter representation;

[0131] Step S620: calculate the total number of points S in the two-dimensional projection point cloud after removing duplicate points, and compare it with the total number of points of similar connectors corresponding to the standard in the standard template library. Compare and get the absolute value of the difference :

[0132]

[0133] Step S630, calculate the normal vector of the two-dimensional projection point cloud and extract the contour points in the two-dimensional projection point cloud. The calculation process is as follows:

[0134] ① Select any point P in the point cloud i , set the search radius r of the neighboring point, and calculate the normal vector n of the point based on this;

[0135] ②According to the selected point P i Make a tangent plane for the point with the calculated normal vector, and project all points in the neighborhood onto the tangent plane;

[0136] ③ Choose a neighboring point P after projection j ,by is the x-axis, the normal vector n is the y-axis, x×y is the z-axis, and P i Establish a local coordinate system for the coordinate center;

[0137] ④ Calculate the clockwise angle θ between the vectors from other neighborhood projection points to the coordinate origin and the coordinate axis x, and then subtract the angles of adjacent vectors to obtain the angle θ' between adjacent vectors;

[0138] ⑤ Set the threshold, sort θ' in descending order, and take the maximum value θ' max , if the value is greater than the given threshold, the point is considered a contour point;

[0139] Step S640, using the random sampling consistency algorithm to randomly extract a small sample of the extracted boundary points in step S610 and fit the feature model, then calculate the error from the contour point to the model and select the correct point set, and finally obtain the fitting segmentation result through continuous iteration. The specific steps of iteratively selecting the correct point set are as follows:

[0140] Step S641, input the boundary point cloud data B, and randomly select two points therein to create a spatial straight line;

[0141] Step S642, calculate the distance from the contour point to the space straight line and set the distance screening threshold , filter out the inliers according to the distance screening threshold and fit the feature model;

[0142] Step S643, given the number of iterations n, continuously iterate steps S641-S642, repeatedly screen inliers and count the number of inliers, and extract data points corresponding to the feature model with the most inliers;

[0143] Step S644, repeating steps S641-S643, repeatedly extracting data points in the feature model with the most inliers until all feature line point clouds in B are obtained by segmentation;

[0144] Based on the segmented feature line point cloud, the corresponding length parameter in the standard feature template library is calculated. The characteristic length of similar connectors to be classified and identified is designed to be L. The length parameter in the standard feature template library is , the absolute value of the difference is :

[0145]

[0146] Step S645, setting threshold and , when the following conditions are met:

[0147] That is to say, the final classification of similar connectors to be identified is completed, which is recorded as .

[0148] In summary, there are 6 steps in total, including the establishment of a standard database of similar connectors, the establishment of a feature template library of similar connectors, the preprocessing of point cloud data of similar connectors, the maximum outer contour screening of similar connectors, the similarity calculation of similar connectors based on spatial positioning, and the key feature comparison of similar connectors. The present invention proposes to establish a standard point cloud database and feature template library for similar connectors of aircraft, and on this basis, innovatively classifies the identification of similar connectors of aircraft into three steps: maximum outer contour screening, similarity calculation based on spatial positioning, and key feature comparison: Calculate the minimum circumscribed rectangular box of each similar connector and its length, width, and height dimensions , compare the three-dimensional size of the minimum circumscribed rectangular box, and make a preliminary classification of similar connectors; based on the definition and creation of four-point sets and the point-to-point ICP algorithm, align the point cloud to complete the spatial positioning of similar connectors, and calculate the four types of parameters after point cloud alignment, and weight them to calculate the final similarity value, so as to complete the further classification of similar connectors; extract the feature point cloud data corresponding to each similar connector point cloud to be classified in the standard database, calculate the corresponding feature parameters, and compare them with the parameters in the feature template library to complete the final recognition and classification of similar connectors, with high recognition accuracy and high recognition efficiency.

[0149] In order to verify the feasibility and effectiveness of the method of the present invention, an experimental verification is carried out on a typical scenario (identification and classification of canard bushings and similar connectors), and the accuracy of the method is evaluated.

[0150] The digital three-dimensional model and point cloud data of the canard bushing and its similar connector variants are obtained, wherein the A1 type uses the canard bushing as the original model, and the other types of canard bushings are variants created using the A1 type canard bushing model as a template, in order to detect the feasibility and effectiveness of the method of the present invention.

[0151] By calculating the minimum circumscribed rectangular box of the canard bushing and its similar connectors, the differences between Class A (A1, A2, A3), Class C and Class B parts can be easily distinguished; as shown in Table 1, the spatial registration of Class A1 parts with themselves and other variants is completed by defining the coarse spatial registration of four-point sets and the fine spatial registration of the ICP algorithm, and then the RMSE value in each registration case is calculated to characterize the similarity of the registered parts. It can be clearly seen that the RMSE value of Class A and Class C parts after registration is high, which means the similarity is low; the RMSE value of Class A parts after registration is low, indicating that the similarity is extremely high; as shown in Table 2, the heights of the three Class A similar connectors of the canard bushing are obtained by two-dimensional plane projection, similar contour extraction, and point cloud maximum length calculation. The height parameter is used as a characteristic parameter of the Class A canard bushing similar connector, which can easily distinguish the specific types of Class A canard bushing similar connectors, and the recognition and classification accuracy reaches the order of 0.1 mm.

[0152] Table 1 Comparative analysis of classification data based on similarity calculation of spatial positioning

[0153]

[0154] Table 2 Comparison results of key characteristic parameters

[0155]

[0156] Finally, it should be noted that: the drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other;

[0157] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying and classifying similar aircraft connectors based on three-dimensional measurement, characterized in that: include: Randomly sampling the surface of the theoretical three-dimensional model of the standard similar connector to obtain standard point cloud data of each standard similar connector; Extract features and classify and store each type of standard point cloud data to obtain a standard database; The characteristic parameters of each standard similar connector are calculated according to the standard database. The characteristic parameters include the standard three-dimensional size of the minimum circumscribed rectangular box, the standard area of ​​the two-dimensional feature projection point cloud, and the standard length of the key geometric similar feature point cloud extracted by segmentation. Each characteristic parameter is integrated and classified to form a characteristic template library. Use a 3D scanner to scan and obtain physical point cloud data of similar connectors to be identified, analyze and determine the type of noise in the physical point cloud data, denoise and downsample the physical point cloud data, and obtain point cloud data of parts to be classified; Calculate the size data of the point cloud of each similar connector to be identified by scanning according to the part point cloud data, perform maximum outer contour screening on the similar connector to be identified, and obtain different categories of similar connectors to be identified; The part point cloud data of similar connectors to be identified that are classified into the same category are spatially positioned based on registration, and the numerical accuracy index after spatial positioning is calculated. Then, the similar connectors to be identified in the same category are sorted from high to low according to the numerical accuracy index. The numerical accuracy index represents the point cloud similarity between the similar connectors to be identified in this category, and the point clouds of similar connectors to be identified with high similarity are grouped into one group. The key features of the point cloud of similar connectors to be identified in the same group are extracted, the feature parameters in the feature template library corresponding to the key features are calculated, and the similar connectors to be identified in each group are finally classified according to the feature parameters.

2. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 1, characterized in that: The specific process of establishing the standard database is as follows: The theoretical digital model of each standard similar connector in a standard posture is converted into an STL format, and based on a random sampling algorithm program, the surface of the theoretical digital model is randomly sampled to generate standard point cloud data of each standard similar connector; Construct a feature projection surface for each standard similar connector, project at least two sets of standard point cloud data of the same type containing similar features into a two-dimensional plane to form at least two sets of two-dimensional feature projection point clouds, calculate the normal vectors of the two-dimensional feature projection point clouds containing similar feature point clouds, and extract the key geometric similar feature point clouds of the standard similar connectors according to the normal vectors; Randomly select two points in the key geometric similarity feature point cloud to create a class, fit the two selected points into a straight line, search in a round-robin manner for points in various key geometric similarity feature point clouds whose distances from the straight line meet a threshold and classify them into the same class, extract each key geometric similarity feature point cloud from the two-dimensional feature projection point cloud containing similar features by calculating the normal vector of the similar feature point cloud, and classify them by searching in a round-robin manner for points whose straight line distances meet a threshold, and obtain standard key geometric similarity feature point clouds of different categories; The standard point cloud data, two-dimensional feature projection point cloud, and key geometric similarity feature point cloud of each standard similar connector are integrated, classified, and stored to form a standard database.

3. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 1, characterized in that: The specific method for establishing the feature template library is as follows: Based on the standard point cloud data of each standard similar connector in the standard database, the minimum circumscribed rectangular box of the corresponding standard similar connector is calculated, and the three-dimensional size of the minimum circumscribed rectangular box of each standard similar connector is recorded respectively; Based on the two-dimensional feature projection point cloud and key geometric similarity feature point cloud of each standard similar connector in the standard database, the area of ​​the two-dimensional feature projection point cloud and the length of the key geometric similarity feature point cloud extracted by segmentation are calculated respectively; The calculated feature parameters are integrated and classified. The feature parameters include the three-dimensional size of the minimum circumscribed rectangular box, the area of ​​the two-dimensional feature projection point cloud, and the length of the key geometric similarity feature point cloud extracted by segmentation. The feature parameters of each independent standard similar connector are constructed and stored separately to form a feature template library.

4. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 1, characterized in that: The specific method for obtaining the part point cloud data is as follows: First, the 3D scanner is spatially calibrated, and similar connectors to be identified and classified are scanned to obtain physical point cloud data; The point cloud data of similar connectors to be identified are downsampled using a voxel downsampling method that retains the original features, and the point closest to the centroid of the voxel grid is selected as the point after downsampling to form a noise point cloud and obtain the noise type; The noise point cloud in the physical point cloud data of similar connectors to be identified is spatially located, and the three-dimensional straight-through filtering method is used to remove the noise to obtain the point cloud data of the parts to be classified.

5. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 1, characterized in that: The specific method for screening the maximum outer contour of similar connectors to be identified is: The main direction vector of the physical point cloud data is calculated using the main direction eigenvalue analysis method; According to the main direction vector, the transformation matrix of the main direction vector of the similar connection to be identified is calculated to align with the positive direction of the Z axis, and the physical point cloud data translated to the center of mass is applied for rigid transformation to obtain the physical point cloud data X′ of the similar connection to be identified after rigid transformation. centered ,calculate: X centered =M0×X centered In the above formula, M0 is the main direction vector Alignment matrix with the positive Z axis, X centered X′ is the matrix constructed by the three-dimensional coordinates of the point cloud after removing the centroid. centered It is the physical point cloud data of similar connectors to be identified after rigid transformation; For the physical point cloud data X′ of the similar connector to be identified after rigid transformation centered Traverse each point in the cloud to find its maximum and minimum values ​​on the three coordinate axes, and then calculate the minimum circumscribed rectangular box size of each similar connector point cloud to be identified, and obtain the maximum outer contour of the similar connector to be identified; The minimum bounding rectangular box size of the point cloud of similar connectors to be identified is compared with the standard three-dimensional size of the minimum bounding rectangular box in pairs, and the similar connectors to be identified that meet the threshold standard are classified into one category.

6. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 5, characterized in that: The calculation process of the main direction vector is as follows: Let the point cloud dataset be Q, where Q is a data set containing n points, and each point q i From the three-dimensional coordinates (x i ,y i , z i ) represents, let the unit vector is the main direction vector of the point cloud data; it is constructed into a 3×n matrix X through the point cloud data set Q, where each column corresponds to the coordinates of a point; Calculate the centroid coordinates of point cloud data The centroid coordinates are the row means of X, Y, and Z respectively, and the specific calculation formula is as follows: The physical point cloud data of similar connectors to be identified are de-centred to obtain the matrix constructed by the three-dimensional coordinates of the point cloud after de-centring. The change formula is as follows: Calculate the covariance matrix C of the point cloud data after removing the centroid. The calculation formula is as follows: Perform eigenvalue decomposition on the covariance matrix C to obtain three eigenvalues ​​and set the three eigenvalues ​​λ1≥λ2≥λ3 and the corresponding eigenvector matrix V; the column vector of V is the eigenvector The principal direction vector is chosen to be the eigenvector corresponding to the maximum eigenvalue, i.e.

7. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 5, characterized in that: The specific method of classifying similar connectors to be identified that meet the threshold standard into one category is: Calculate the absolute value of the three-dimensional size of the minimum bounding rectangle: X dv =|X max1 -X max2 | AND dv =|And max1 -AND max2 | WITH dv =|From max1 -WITH max2 | Where, X max1 , Y max2 , Z max1 is the physical point cloud data X′ centered The maximum value on the three coordinate axes, X max2 , Y max2 , Z max2 is the maximum value of the standard three-dimensional size of the smallest circumscribed rectangular box in the standard point cloud data on the three coordinate axes, X DV , Y dv , Z dv is the absolute value of the three-dimensional size of the minimum bounding rectangular box; Set the threshold δ1, when X dv , Y dv , Z dv When both are less than δ1, the two similar connectors to be identified are placed in the same category, denoted as P i (i=1,2,3……)P i (i=1, 2, 3...) categories of similar connectors to be identified.

8. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 1, characterized in that: The specific steps for calculating numerical accuracy indicators are as follows: First, the 4PCS registration algorithm based on the classic four-point set theory that defines invariants and invariant ratios is used to perform rough registration of the point cloud of similar connectors to be identified. Then, the point-to-point ICP-based registration algorithm is used to achieve fine registration of the point cloud of similar connectors to be identified and complete spatial positioning. The RMSE value of the point cloud of similar connectors to be identified after fine registration is calculated, and based on this, the similarity of similar connectors to be identified after rough registration and fine registration is characterized. The calculation formula is as follows: In the above formula, m represents the number of matching point pairs after the target point cloud and the source point cloud are registered, X pi With X qi Respectively represent the spatial three-dimensional coordinate values ​​of the corresponding points of the target point cloud and the source point cloud after point cloud registration; Set a threshold δ2, and group similar connectors with similarity values ​​less than the threshold δ2 and with associated relationships into the same group again, denoted as P ij (i, j = 1, 2, 3...)P ij (i, j=1, 2, 3...) groups.

9. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 1, characterized in that: The specific method for finally classifying similar connectors to be identified in each group by feature parameters is as follows: Construct a two-dimensional feature projection surface, project the scanned similar connector point cloud to be identified on the two-dimensional feature projection surface to obtain a two-dimensional projection point cloud, and then remove duplicate points in the two-dimensional projection point cloud by creating a two-dimensional surface grid to obtain boundary points, and obtain boundary point cloud data B by statistics; Calculate the total number of points S in the two-dimensional projection point cloud after removing duplicate points, and compare it with the total number of points S′ of the corresponding standard similar connectors in the standard template library to obtain the absolute value of the difference Δ1: Δ1=|S′-S| Calculate the normal vector of the two-dimensional projection point cloud and extract the contour points in the two-dimensional projection point cloud; According to the extracted boundary points, a small sample is randomly selected and the feature model is fitted. Then, the error from the contour point to the model is calculated to select the correct point set. Through continuous iteration, the fitting segmentation result is finally obtained, and the final classification of similar connectors to be identified is completed, which is recorded as P ijk (i,j,k=1,2,3...)P ijk (i, j, k=1, 2, 3...).

10. The method for identifying and classifying similar aircraft connectors based on three-dimensional measurement according to claim 9, characterized in that: The specific method of iteratively screening the correct point set is: Input boundary point cloud data B, and randomly select two points to create a spatial line; Calculate the distance from the contour point to the spatial straight line and set the distance screening threshold δ boundary , filter out the inliers according to the distance screening threshold and fit the feature model; Given the number of iterations n, repeatedly screen the inliers and count the number of inliers, and extract the data points corresponding to the feature model with the most inliers; Repeatedly extract the data points in the feature model with the most internal points until all the feature line point clouds in B are segmented; Based on the segmented feature line point cloud, the corresponding length parameter in the standard feature template library is calculated. The characteristic length of similar connectors to be classified and identified is designed to be L, and the length parameter in the standard feature template library is L′. The absolute value of the difference is Δ2: Δ2=|L′-L| Thresholds δ3 and δ4 are set when the following conditions are met: Δ1<δ3,Δ2<δ4 That is, the final classification of similar connectors to be identified is considered completed.

Images