A method for initial registration of global fuel tank point clouds based on structural feature descriptors

Abstract

This invention relates to an initial registration method for overall fuel tank point clouds based on structural feature descriptors. It includes an efficient algorithm for extracting fuel tank contour lines and fastener vertices, and designs a validity judgment strategy to select feature combinations that contribute significantly to subsequent descriptor construction. A novel structural feature descriptor is designed, based on which accurate registration between point clouds is achieved. A distance constraint condition replaces the iterative steps in the registration process, effectively improving algorithm efficiency. This invention addresses the problems of poor registration accuracy and low reliability of point cloud data in locally highly similar scenes, thereby improving the accuracy of digital 3D reconstruction of overall fuel tanks for large aircraft and providing high-quality 3D data for subsequent inspection and measurement tasks.

Description

Technical Field

[0001] This invention relates to the field of digital modeling technology for overall fuel tanks, and in particular to an initial registration method for point clouds of overall fuel tanks based on structural feature descriptors. Background Technology

[0002] Large aircraft fuel tanks have numerous sealed areas, poor openness, and internal obstructions and interference. Therefore, the sealant application process inevitably encounters issues such as missed areas, insufficient coating thickness, and surface defects. This necessitates the use of digital inspection technology to assess the sealing quality. However, conventional 3D model reconstruction techniques face several challenges in the context of large fuel tank scenarios (enclosed, confined, and highly structured). The primary challenge is ensuring the accuracy of model generation, which involves multiple techniques including initial registration, precise registration, and reconstruction error optimization. Precise registration and subsequent reconstruction error optimization heavily rely on a good initial position. Therefore, initial registration, as a crucial step, is particularly important for its ability to describe and distinguish similar features. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for initial registration of overall fuel tank point clouds based on structural feature descriptors. This method solves the problems of poor registration accuracy and low reliability of traditional methods in scenarios where the local height of the fuel tank wall panels is similar. This invention extracts reliable fuel tank structural features to construct novel structural feature descriptors and stores them in a hash table. Simultaneously, it employs a distance constraint condition to replace the iterative steps in the registration process, achieving accurate registration between fuel tank point clouds while effectively improving algorithm efficiency.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a method for initial registration of a global fuel tank point cloud based on structural feature descriptors, comprising the following steps:

[0005] S1. The method for extracting line features in the joint area extracts straight line features of the tank wall panel from the single frame point cloud data of the tank wall panel and divides them into sub-line segment features, and constructs a set of sub-line segment features.

[0006] S2. Fastener region vertex feature extraction method: Segment the point cloud data frame into a local coplanar point cloud, voxelize the local coplanar point cloud, and use the fastener reference model to perform template matching and filter noise to obtain the vertex features of the fastener and construct a vertex feature set.

[0007] S3. A point and line feature filtering method based on validity verification: Design a validity verification strategy, which constructs a combination of structural features, defines a validity verification scoring function, and filters the combination of structural features based on the scoring function to obtain a candidate set of structural features;

[0008] S4. A descriptor construction method based on structural features describes structured features using a candidate set of structural features to obtain structural feature descriptors.

[0009] S5. A point cloud correspondence search method based on distance constraints: For each point cloud frame, a hash table is constructed based on the structural feature descriptor. The hash table is used to obtain the initial correspondence set between adjacent point cloud frames, and the final correspondence is obtained by filtering.

[0010] S6. Based on the correspondence relationship, the initial registration method of point cloud frames is used to construct a world coordinate system. The rigid transformation matrix between adjacent frame point clouds is solved using the final correspondence relationship obtained in step S5. The rigid transformation matrix is ​​then used to transform the point cloud to the world coordinate system frame by frame.

[0011] Furthermore, in step S1, the specific process includes the following steps:

[0012] S11. Calculate the normals of the single-frame point cloud data of the tank wall panel to obtain the plane normal vector, and obtain the reference plane and coplanar region point cloud through the region growing method.

[0013] S12. Project the single-frame point cloud data of the tank wall panel onto the reference plane. Based on the centroid of the coplanar region point cloud, its projection point on the reference plane, and the plane normal vector, establish the coordinate system of the reference plane and obtain the conversion formula from three-dimensional point cloud coordinates to two-dimensional plane coordinates.

[0014] S13. Arrange the point cloud of the coplanar region in ascending order of scale, and take the first 95% of the points as the scale factor of the reference plane.

[0015] S14. Based on the scale factor, convert the two-dimensional planar coordinates on the reference plane into pixel coordinates of the binary image to obtain the binary image of the fuel tank wall panel and the corresponding conversion formula.

[0016] S15. The least squares method is used to fit the contour lines in the binary image of the tank wall panel. The obtained line features are inversely transformed to two-dimensional plane coordinates through the transformation formula in step S14. Then, the two-dimensional plane coordinates are inversely transformed to coordinates on the three-dimensional point cloud through the transformation formula in step S12 to obtain the straight line features of the tank wall panel on the three-dimensional point cloud frame.

[0017] S16. Divide the straight line feature of the fuel tank wall into multiple sub-segment features with a fixed length, and construct a set of sub-segment features.

[0018] Further, in step S1: the straight line features of the tank wall panel are sorted sequentially according to their length values, and discontinuous or low-discrimination line features are eliminated.

[0019] Furthermore, in step S2, the specific process includes the following steps:

[0020] S21. Input the point cloud data frame of the fuel tank wall panel and fit the main plane based on the plane fitting method;

[0021] S22. Calculate the vertical distance between each point cloud in a single frame of point cloud data and the main plane, and use this as a basis to divide the point cloud data frame into local coplanar point clouds corresponding to the main plane.

[0022] S23. Convert the local coplanar point cloud into several cubic grids, calculate the curvature of the point cloud in a single grid, and select candidate grids for fasteners based on the curvature.

[0023] S24. Merge adjacent fastener candidate grids into a large grid, perform template matching through the fastener reference model, and if the proportion of overlapping sampling points between the point cloud in the large grid and the fastener reference model reaches a threshold, then segment it as a candidate fastener point cloud.

[0024] S25. For a single candidate fastener point cloud, select its farthest point from the main plane and its neighboring points as candidate points. Calculate the median and average distances between the candidate points and the main plane as reference indicators. Filter out noisy point clouds based on the reference indicators. Use the point cloud at the median position that is retained as the vertex feature of the corresponding fastener. Repeat the above steps for all candidate fastener point clouds to construct a vertex feature set.

[0025] Furthermore, in step S3, the specific process includes the following steps:

[0026] S31. Calculate the distance between the starting point feature, ending point feature, and midpoint feature of each sub-segment feature and each vertex feature, and define each pair of point features with calculated distances as a structural feature combination.

[0027] S32. Define the validity verification scoring function. :

[0028] ;

[0029] in:

[0030] ;

[0031] ;

[0032] In the above formula, This represents the Euclidean distance between corresponding point features. For all The average value, Represents the total number of combinations of structural features; Standard deviation is used to measure all The degree of dispersion of the distribution Indicates all Standardization and;

[0033] S33. Calculate the score according to the validity verification scoring function, and select the structural feature combinations whose scores are within the set threshold range as the structural feature candidate set.

[0034] Furthermore, in step S4, the specific process includes: using the sub-segment features and vertex features in the candidate set of structural features to describe the structured features and construct a structural feature descriptor. :

[0035] ;

[0036] ;

[0037] In the formula, Represents a linear vector. Midpoint feature of sub-segment With vertex features The Euclidean distance between them; This refers to calculating the angle between two vectors. Specifically, Represents a linear vector and by and The angle between the vectors formed Indicates by and The vectors formed by and The angle between the vectors formed; and the angle and The range of values ​​for is .

[0038] Furthermore, in step S4: the structural feature descriptor can adaptively select the optimal combination to describe the corresponding feature points according to different numbers of principal planes. When the vertex features and line features are located on different principal planes, the proposed method uses the vertex features as a benchmark to continuously construct the corresponding set of structural feature descriptors with the sub-line segment features. When the vertex features and line features are located on the same principal plane, the proposed method first evaluates the contribution of constructing unique structural feature descriptors based on the above-mentioned validity verification scoring function in order to remove redundant or poorly descriptive structural feature combinations.

[0039] Furthermore, in step S5, the specific process includes the following steps:

[0040] S51, The hash table includes key values, structural feature descriptors, and three-dimensional coordinates of corresponding point features;

[0041] S52. Represent the structural feature descriptor as a two-dimensional vector consisting of distance and angle. That is, design the corresponding sampling step size for two different physical quantities, and then map the sampling results to the same vector as the key value in the hash table.

[0042] S53. Given two adjacent point cloud frames, defined as source point cloud frame and target point cloud frame, calculate their corresponding key values ​​based on the structural feature descriptor of the source point cloud frame and index them into the structural feature descriptor in the hash table constructed by the target point cloud frame. Repeat the above steps for all adjacent point cloud frames to obtain the initial correspondence between adjacent point cloud frames and construct the initial correspondence set.

[0043] S54. Based on step S53, further filter the correspondences in the initial correspondence set based on distance constraints, remove the incorrectly matched correspondences, and obtain the final correspondences.

[0044] Furthermore, in step S6, the specific process includes the following steps:

[0045] S61. Use a scanner to acquire point cloud frames, take the first acquired point cloud frame as the reference frame, and construct a world coordinate system with the camera optical center of the scanner as the coordinate origin.

[0046] S62. Define the Euclidean distance between adjacent frame point clouds, assuming they satisfy their final correspondence, as the objective function:

[0047] ;

[0048] in, , For point clouds of adjacent frames, For rotation matrix, Let be the translation vector and the rigid transformation matrix. That is, the variable to be optimized in the objective function;

[0049] S63. Calculate the centroids of the point clouds in adjacent frames respectively, simplify the objective function using the centroids, and obtain the rigid transformation matrix between the point clouds in adjacent frames by using singular value decomposition (SVD). Using this rigid transformation matrix The point cloud is converted to the world coordinate system frame by frame to achieve initial registration between point cloud frames of adjacent oil tank wall panels.

[0050] By employing the above technical solution, the present invention provides a method for initial registration of the overall fuel tank point cloud based on structural feature descriptors, which has at least the following beneficial effects:

[0051] (1) This invention reduces the computational cost of invalid or low-discrimination feature combinations by designing a point and line feature screening method based on validity verification, and ensures the robustness of subsequent initial registration by constructing a combination of unique descriptors.

[0052] (2) This invention improves the search efficiency of corresponding descriptors and the accuracy of feature matching by constructing a hash table to store structural feature descriptor vectors;

[0053] (3) The present invention effectively improves the efficiency of the algorithm by using a distance constraint to replace the iterative steps in the registration process. Attached Figure Description

[0054] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0055] Figure 1 This is a framework diagram of an initial registration method for the overall fuel tank point cloud based on structural feature descriptors according to the present invention.

[0056] Figure 2 This is a schematic diagram of the linear feature discretization in the overall fuel tank point cloud initial registration method based on structural feature descriptors according to the present invention.

[0057] Figure 3 This serves as a validity verification strategy for the overall fuel tank point cloud initial registration method based on structural feature descriptors of the present invention.

[0058] Figure 4 This is a schematic diagram illustrating the construction of a structural feature descriptor for an initial registration method of a global fuel tank point cloud based on a structural feature descriptor according to the present invention.

[0059] Figure 5 This is a schematic diagram illustrating the combination of structural feature descriptors, hash table construction, and indexing of a method for initial registration of overall fuel tank point clouds based on structural feature descriptors according to the present invention.

[0060] Figure 6 This invention provides a method for initial registration of the overall fuel tank point cloud based on structural feature descriptors, which involves searching for point cloud correspondences based on distance constraints. Detailed Implementation

[0061] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0062] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] Please refer to Figures 1-6 This illustration shows a specific implementation of the present embodiment. This embodiment selects feature combinations that contribute significantly to the subsequent descriptor construction by designing a validity judgment strategy; it designs a novel structural feature descriptor to achieve accurate registration between point clouds; and it adopts a distance constraint condition to replace the iterative steps in the registration process, effectively improving the algorithm efficiency. This solves the problems of poor registration accuracy and low reliability of point cloud data in locally highly similar scenes, improves the accuracy of digital 3D reconstruction of the overall fuel tank of large aircraft, and provides high-quality 3D data for subsequent detection and measurement tasks.

[0064] Please refer to Figure 1 This embodiment proposes an initial registration method for the overall fuel tank point cloud based on structural feature descriptors. The method includes the following steps:

[0065] S1. The method for extracting line features in the joint area extracts straight line features of the tank wall panel from the single frame point cloud data of the tank wall panel and divides them into sub-line segment features, and constructs a set of sub-line segment features.

[0066] As a preferred embodiment of step S1, the specific process includes the following steps:

[0067] S11. Calculate the normal vectors from the single-frame point cloud data of the fuel tank wall panel to obtain the plane normal vector. The reference plane is obtained through the region growing method. Point cloud of the co-surface region This allows for the extraction of corresponding structural features.

[0068] S12. Project the single-frame point cloud data of the fuel tank wall panel onto the reference plane. and point clouds of coplanar regions center of mass As the center of the plane, The first point in Projected onto reference plane Its projection point is denoted as ,definition for The axis, its positive direction is ,but Axial direction can be passed Confirm, at this point, click Coordinates on a two-dimensional plane can be represented as:

[0069] ;

[0070] S13. Define the scale factor of the reference plane as follows: ,in Indicates will A collection of midpoint clouds arranged in ascending order of scale; This means selecting the 95th percentile of points in ascending order to eliminate the influence of some noisy point clouds and outliers on feature extraction;

[0071] S14. Calculate the width and height values ​​of the binary image, i.e. and ,in and Let represent the maximum and minimum coordinate values ​​of the two-dimensional plane, respectively. Then the coordinates of the two-dimensional plane... The corresponding pixel coordinates of the binary image It can be calculated using the following formula:

[0072] ;

[0073] The two-dimensional planar coordinates on the reference plane are converted into pixel coordinates of a binary image to obtain a binary image of the fuel tank wall panel;

[0074] S15. The least squares method is used to fit the contour lines existing in the binary image of the fuel tank wall panel to obtain line features, and the transformation formula in step S14 is used to convert the features. The linear features of the position are inversely transformed into two-dimensional plane coordinates. Then, the two-dimensional plane coordinates are converted into coordinates on the three-dimensional point cloud using the conversion formula in step S12, resulting in the straight line features of the tank wall panel on the three-dimensional point cloud frame. ;

[0075] S16. Considering that the straight line features shared between adjacent point cloud frames may only partially overlap, it is necessary to fit the straight line features of the fuel tank wall. A set composed of multiple sub-segments of fixed length. The sub-segment feature is defined as follows: ,in Indicates the midpoint of the sub-segment. Indicates the direction of the sub-segment, such as Figure 2 As shown.

[0076] More specifically, the straight line features of the fuel tank wall can be sorted sequentially according to their length values, thereby eliminating discontinuous or low-discrimination line features.

[0077] In this embodiment, the process of constructing the sub-segment feature set can be regarded as discretizing the complete straight line feature of the tank wall into a set of equally spaced point features to ensure that the common straight line feature of the tank wall (between adjacent point cloud frames) can be described by the sub-segment feature in the overlapping area.

[0078] S2. Fastener region vertex feature extraction method: Segment the point cloud data frame into a local coplanar point cloud, voxelize the local coplanar point cloud, and use the fastener reference model to perform template matching and filter noise to obtain the vertex features of the fastener and construct a vertex feature set.

[0079] As a preferred embodiment of step S2, the specific process includes the following steps:

[0080] S21. Input the point cloud data frame of the fuel tank wall panel, and fit the main fuel tank wall panel plane, i.e., the main plane, from the input point cloud data frame using a plane fitting method. This serves as the reference plane for extracting the vertex features of the fastener;

[0081] S22. Calculate the vertical distance between each point cloud in a single frame of point cloud data and the principal plane, and use this distance as a basis to divide the point cloud data frame into locally coplanar point clouds corresponding to the principal plane. ;

[0082] S23. Divide the coplanar region point cloud into segments. Voxelization is performed for objects belonging to the reference plane. The point cloud is divided into several cubic grids with the fastener size as a reference. The curvature of the point cloud in a single grid is calculated. If the curvature is close to 0, it means that the point cloud at that point is a planar point cloud. If the curvature of the point cloud in the grid is not zero, it is added to the fastener candidate grid.

[0083] S24. For fastener candidate grids, adjacent fastener candidate grids are merged into a large grid, and template matching is performed through the fastener reference model. If the proportion of sampling points that overlap between the point cloud in the large grid and the fastener reference model reaches a threshold, it is segmented as a candidate fastener point cloud.

[0084] S25. For a single candidate fastener point cloud, select its farthest point from the main plane and its neighboring points as candidate points, and simultaneously calculate the median distance between each candidate point and the main plane. and average As a reference metric, the relative differences between reference metrics are calculated to assess the impact of noisy point cloud data on vertex feature extraction, as shown in the following formula:

[0085] ;

[0086] in This represents Euclidean distance, if there are differences between reference indicators. If the value is too large, it indicates that the point cloud data within the cubic grid has significant noise, and the vertex feature needs to be removed; if... Less than the set threshold If the point cloud noise data and outliers are considered to be within the error tolerance range, then the median is retained. The point cloud at the location is used as the vertex feature of the corresponding fastener. The mathematical description of this process is as follows:

[0087] ;

[0088] Repeating the above calculation steps for all candidate fastener point cloud data yields the vertex feature set. ,in , Represents the coordinates on a 3D point cloud.

[0089] S3. A point and line feature filtering method based on validity verification: Design a validity verification strategy, which constructs a combination of structural features, defines a validity verification scoring function, and filters the combination of structural features based on the scoring function to obtain a candidate set of structural features;

[0090] As a preferred embodiment of step S3, the specific process includes the following steps:

[0091] S31. Calculate the distance between the starting point feature, ending point feature, and midpoint feature of each sub-segment feature and each vertex feature, and define each pair of point features with calculated distances as a structural feature combination.

[0092] S32, such as Figure 3 As shown, and They belong to different sub-segment characteristics. Sub-segment characteristics The starting point, the ending point, and the midpoint. Representing the midpoint characteristics of the sub-segments respectively Vertex features in the fastener feature set The distance between them; extending the above feature point distance calculation to and Then the validity verification scoring function The following formula can be used for calculation:

[0093] ;

[0094] in:

[0095] ;

[0096] ;

[0097] In the above formula, This represents the Euclidean distance between corresponding point features. For all The average value, Represents the total number of combinations of structural features; Standard deviation is used to measure all The degree of dispersion of the distribution Indicates all Standardization and validity verification scoring function This function measures the contribution of different combinations of structural features to constructing a unique structural feature descriptor. Mapped to the (0,1) interval, the closer the output value is to 1, the better (based on...) The higher the validity of the constructed structural feature descriptor, the closer it is to 0, indicating lower validity.

[0098] S33. Based on the validity verification scoring function Calculate the score, select the structural feature combinations whose scores are within the set threshold range as the structural feature candidate set, and participate in the subsequent construction of structural feature descriptors.

[0099] In this embodiment, the designed point-line feature filtering method based on validity verification can reduce the computational cost of invalid or low-discrimination feature combinations. Furthermore, it ensures the robustness of subsequent initial registration by constructing combinations of unique descriptors. For example, for... Figure 3 Sub-segment features and In other words, due to the different angles between the lines connecting the positions and features of the two relative to the fastener area, the former's The score should be slightly higher than the latter, indicating that the construction of an unambiguous descriptor is successful. The contribution is less than Then, structural features within the threshold range are included in the candidate set. This is used to construct the corresponding feature descriptors in the future.

[0100] S4. A descriptor construction method based on structural features describes structured features using a candidate set of structural features to obtain structural feature descriptors.

[0101] As a preferred implementation of step S4, the specific process includes: using the sub-segment features and vertex features in the candidate set of structural features to describe the structured features, and constructing a structural feature descriptor. :

[0102] ;

[0103] ;

[0104] In the formula, Represents a linear vector. Midpoint feature of sub-segment With vertex features The Euclidean distance between them; This refers to calculating the angle between two vectors. Specifically, Represents a linear vector and by and The angle between the vectors formed Indicates by and The vectors formed by and The angle between the vectors formed; and the angle and The range of values ​​for is .

[0105] More specifically, the structural feature descriptor described in step S4 can adaptively select the optimal combination to describe the corresponding feature points based on different numbers of principal planes, such as... Figure 4 As shown in (a), when the vertex features and line features are located in different principal planes (i.e., regions with rich structural features, such as stiffener regions), the proposed method uses the fastener vertex features as a benchmark and continuously constructs corresponding descriptor subsets with the discretized line features; the corresponding feature combinations are as follows: Figure 4 As shown in (b), when all structural feature sets are located in the same principal plane (the wall panel region with fewer structural features), the proposed method first evaluates the contribution of constructing unique structural feature descriptors based on the above-mentioned validity verification scoring function, so as to remove redundant or poorly descriptive structural feature combinations.

[0106] In this embodiment, the structural feature descriptor is intended to describe the geometric shape of a single frame of fuel tank wall point cloud data. The structural feature descriptor vector will be stored by establishing a hash table, thereby improving the search efficiency of the corresponding descriptor and the accuracy of feature matching.

[0107] S5. A point cloud correspondence search method based on distance constraints: For each point cloud frame, a hash table is constructed based on the structural feature descriptor. The hash table is used to obtain the initial correspondence set between adjacent point cloud frames, and the final correspondence is obtained by filtering.

[0108] As a preferred embodiment of step S5, the specific process includes the following steps:

[0109] S51. Construct a corresponding hash table based on the candidate set of structural features. The hash table includes key values, structural feature descriptors, and the three-dimensional coordinates of the corresponding point features, such as... Figure 5 As shown;

[0110] S52. Represent the structural feature descriptor as a two-dimensional vector consisting of distance and angle. That is, design the corresponding sampling step size for two different physical quantities. Then map the sampling result to the same vector as the key value in the hash table. Perform the above calculation on all candidate sets of structural feature descriptors to complete the construction of the hash table for the point cloud of the structured oil tank wall.

[0111] S53. Given two adjacent point cloud frames, define them as source point cloud frames respectively. and target point cloud frame First, target point cloud frames Constructing a hash table from a subset of structural feature descriptions Then, structural feature descriptors are calculated from the candidate structural feature set in the source point cloud frame. and the corresponding key-value pairs ,like Figure 5 As shown in (a), if in Constructed hash table There exists a key with the same value. Then it can be passed Index to hash table Corresponding structural feature descriptor It is important to note that here... and This refers to the correspondence between one of the descriptors in two point cloud frames, where... and Let be the set of coordinates of the corresponding structural feature descriptors in the source point cloud frame and the target point cloud frame, respectively. This initial correspondence is denoted as . Repeat the above steps to calculate the structural feature descriptor key values ​​in the candidate set to obtain the initial correspondence between the two frame point clouds, thus constructing the initial correspondence set. The structural feature descriptor correspondence search method is as follows: Figure 5 As shown in (b) of the image, the source point cloud frame A structural feature descriptor For example, the above process can be briefly described as follows: Calculate the corresponding key value And indexed to the target point cloud frame Constructed hash table Structural feature descriptors in The process.

[0112] S54. Based on step S53, further filter the correspondences in the initial correspondence set based on distance constraints, remove incorrectly matched correspondences, and obtain the final correspondences. Specifically, from the initial correspondences... Select three points from the set, and denote them as follows: Then, determine whether the Euclidean distance between any two adjacent points is equal. The mathematical description of this process can be expressed as:

[0113] ;

[0114] in, , Represents the distance constraint threshold and candidate initial correspondence. Points that meet the above conditions will be retained as the final correspondence, denoted as . Otherwise, it will be eliminated. Figure 6 Taking the correspondence shown as an example, where Figure 6 (a) in the image represents the input source point cloud frame. and ; Figure 6 (b) in Figure 6 The calculation process in (e) reflects the distance calculation and judgment results of three sets of adjacent point pairs. Specifically, green edges indicate that adjacent point pairs meet the requirements of the above formula, while red edges indicate that adjacent point pairs do not meet the correspondence. In this example, only Figure 6 The three point pairs selected in (b) that satisfy the corresponding distance constraints can be used as point clouds. and The final correspondence between them; Appendix Figure 6 The numbers 1, 2, 3, and 4 in the middle represent the test points 1, 2, 3, and 4 extracted from the point cloud frame, which are used to form adjacent point pairs.

[0115] S6. The point cloud frame initial registration method based on the correspondence relationship is to construct a world coordinate system, use the final correspondence relationship obtained in step S5 to solve the rigid transformation matrix between adjacent frame point clouds, and use the rigid transformation matrix to transform the point cloud to the world coordinate system frame by frame.

[0116] As a preferred embodiment of step S6, the specific process includes the following steps:

[0117] S61, Adjacent point cloud frames acquired by a surface structured light 3D scanner , For example, a single-frame point cloud uses the optical center of the scanner's camera as the origin of the coordinate system, and the first frame... Construct a world coordinate system for the reference frame;

[0118] S62. Define the point clouds of adjacent frames in a way that satisfies their final correspondence. The Euclidean distance, given this premise, is the objective function:

[0119] ;

[0120] in, , For point clouds of adjacent frames, For rotation matrix, Let be the translation vector and the rigid transformation matrix. This refers to the variable to be optimized in the objective function, and the transformation matrix can be solved by minimizing the objective function;

[0121] S63. The essence of the above transformation matrix solution process is to obtain the optimal solution through least squares. Therefore, singular value decomposition (SVD) can be used for the solution. Specifically, the centroids of the source point cloud and the target point cloud need to be calculated separately, denoted as... , Secondly, assuming a source point cloud. With target point cloud Correspondence If the number of midpoints is N, then:

[0122] ;

[0123] center of mass , By incorporating the objective function and further expanding and simplifying the above equation, the objective function can be expressed as:

[0124] ;

[0125] in and These represent the centroid-free spatial locations of the source and target point clouds, respectively. At this point, the objective function is minimized, i.e., minimizing the squared term in the above formula. Observation reveals that the squared term... Since only R is involved, we prioritize minimizing it, and the objective function can be simplified to:

[0126] ;

[0127] Singular value decomposition (SVD) of the above equation yields:

[0128] ;

[0129] Because the coefficients of the first two terms in the above formula Since all values ​​are independent of R, the objective function can be further simplified to:

[0130] ;

[0131] in, This represents the trace operation of a matrix, while express The matrix can be obtained through eigenvalue decomposition:

[0132] ;

[0133] in, Let represent a diagonal matrix composed of eigenvalues, and , Let each represent a diagonal matrix, when When the rank is full, the rotation matrix R can be calculated:

[0134] ;

[0135] The centroids of point clouds in adjacent frames are calculated, and the objective function is simplified using these centroids. Singular Value Decomposition (SVD) is then used to solve the objective function, yielding a rigid transformation matrix T between adjacent point clouds. This rigid transformation matrix T is then used to transform the point clouds frame by frame to the world coordinate system, achieving initial registration between adjacent fuel tank wall point cloud frames. Finally, the corresponding translation vectors are calculated. The rigid transformation matrix T is obtained by combining the rotation matrix R, and the initial registration between point cloud frames of adjacent tank wall panels can be achieved through this transformation matrix.

[0136] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0137] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0138] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for initial registration of overall fuel tank point clouds based on structural feature descriptors, characterized in that, The method comprises the following steps: S1, a junction surface area line feature extraction method, extracting a straight line feature of an oil tank wall plate from single frame point cloud data of the oil tank wall plate and dividing the straight line feature into sub-line segment features to construct a sub-line segment feature set; S2, a fastener area vertex feature extraction method, segmenting a point cloud data frame into local coplanar point clouds, voxelizing the local coplanar point clouds, and performing template matching and noise filtering by using a fastener reference model to obtain vertex features of the fastener and construct a vertex feature set; The specific process comprises the following steps: S21, inputting a point cloud data frame of the oil tank wall plate, fitting a main plane based on a plane fitting method; S22, calculating a vertical distance between each point cloud in a single frame point cloud data and the main plane, and segmenting the point cloud data frame into local coplanar point clouds corresponding to the main plane according to the vertical distance; S23, voxelizing the local coplanar point clouds into a plurality of cubic grids, calculating the curvature of the point cloud in a single grid, and screening fastener candidate grids according to the curvature; S24, merging adjacent fastener candidate grids into large grids, performing template matching by using a fastener reference model, and if the proportion of the sampling points between the point cloud in the large grid and the fastener reference model reaches a threshold value, segmenting the large grid as a candidate fastener point cloud; S25, for a single candidate fastener point cloud, selecting the farthest point from the main plane and the neighborhood points thereof as candidate points, calculating the median and average of the distance between the candidate points and the main plane as reference indexes, filtering the point cloud with large noise according to the reference indexes, and retaining the point cloud at the median position as the vertex feature of the corresponding fastener, and repeating the above steps for all candidate fastener point clouds to construct a vertex feature set; S3, a point-line feature screening method based on effectiveness verification, designing an effectiveness verification strategy, constructing a structure feature combination, defining an effectiveness verification scoring function, and screening the structure feature combination based on the scoring function to obtain a structure feature candidate set; The specific process comprises the following steps: S31, calculating the distance between the starting point feature, the terminal point feature and the midpoint feature of each sub-line segment feature and each vertex feature, and defining each pair of point features with calculated distance as a structure feature combination; S32, define validity verification score function : ； Wherein: ； ； In the above formula, denotes the Euclidean distance between the corresponding point features, is the average of all , denotes the total number of structure feature combinations; denotes the standard deviation, which is used to measure the dispersion degree of all distribution, denotes the standardized sum of all . S33, calculating the score according to the effectiveness verification scoring function, and selecting the structure feature combination with the score within a set threshold range as the structure feature candidate set; S4, a structural feature descriptor construction method, which uses a structural feature candidate set to describe a structured feature to obtain a structural feature descriptor; the specific process includes: using the sub-line segment features and vertex features in the structural feature candidate set to describe the structured feature to construct the structural feature descriptor : ； ； wherein, represents a straight line vector, represents the Euclidean distance between the midpoint feature of the sub-segment feature and the vertex feature ; represents the angle between two vectors, in particular, represents a straight line vector and the angle between the vectors composed of and ; represents the angle between the vectors composed of and and the vectors composed of and ; and the angle between ranges from ;​ S5, a point cloud correspondence relationship searching method based on distance constraint, constructing a hash table according to the structure feature descriptor for each point cloud frame, obtaining an initial correspondence relationship set between adjacent point cloud frames by using the hash table, and screening to obtain a final correspondence relationship; S6, a point cloud frame initial registration method based on the correspondence relationship, constructing a world coordinate system, solving a rigid transformation matrix between adjacent frame point clouds by using the final correspondence relationship obtained in step S5, and converting the point clouds frame by frame to the world coordinate system by using the rigid transformation matrix.

2. The method of claim 1, wherein: The specific process of step S1 comprises the following steps: S11, calculating the normal of the single frame point cloud data of the oil tank wall plate to obtain a plane normal vector, and obtaining a reference plane and coplanar area point cloud by a region growing method; S12, project the single-frame point cloud data of the oil tank wall plate to a reference plane, establish a coordinate system of the reference plane according to the centroid of the coplanar region point cloud, the projection point of the centroid on the reference plane and the plane normal vector, and obtain a conversion formula from three-dimensional point cloud coordinates to two-dimensional plane coordinates; S13, arrange the coplanar region point cloud in ascending order of scale, and take the points at the front 95% as the scale factor of the reference plane; S14, convert the two-dimensional plane coordinates on the reference plane into pixel coordinates of a binary image according to the scale factor, to obtain an oil tank wall plate binary image and a corresponding conversion formula; S15, fit the contour line existing in the oil tank wall plate binary image by using the least square method, convert the obtained line feature to two-dimensional plane coordinates through the conversion formula in step S14 in reverse, and then convert the two-dimensional plane coordinates to coordinates on the three-dimensional point cloud through the conversion formula in step S12 in reverse, to obtain the straight line feature of the oil tank wall plate on the three-dimensional point cloud frame; S16, divide the straight line feature of the oil tank wall plate into a plurality of sub-line segment features with a fixed length, and construct a sub-line segment feature set.

3. The method of claim 2, wherein: The straight line feature of the oil tank wall plate is sequentially sorted according to length values, and discontinuous or low-recognizability line features are removed.

4. The method of claim 1, wherein: The structure feature descriptor in step S4 can adaptively select the optimal combination to describe the corresponding feature points according to different main plane numbers. In the case that the vertex feature and the straight line feature are located in different main planes, the proposed method takes the vertex feature as the reference, and continuously constructs the corresponding structure feature descriptor set with the sub-line segment feature. In the case that the vertex feature and the straight line feature are both in the same main plane, the proposed method first evaluates the contribution of the structure feature descriptor with uniqueness to remove redundancy or poor feature description based on the above effectiveness verification scoring function.

5. The method of claim 1, wherein: The specific process of step S5 includes the following steps: S51, the hash table includes key values, structure feature descriptors and three-dimensional coordinates of corresponding point features; S52, the structure feature descriptor is represented as a two-dimensional vector composed of distance and angle, that is, a sampling step length is designed for two different physical quantities, and then the sampling results are mapped to the same vector as the key value in the hash table; S52, given two adjacent point cloud frames, defined as a source point cloud frame and a target point cloud frame, calculate the corresponding key values of the structure feature descriptors of the source point cloud frame, and index them to the structure feature descriptors in the hash table constructed by the target point cloud frame, repeat the above steps for all adjacent point cloud frames to obtain the initial correspondence relationship between the adjacent point cloud frames, and construct an initial correspondence relationship set; S53, on the basis of step S52, further screen the correspondence relationship in the initial correspondence relationship set based on distance constraint, remove the wrong matching correspondence relationship, and obtain the final correspondence relationship.

6. The method of claim 1, wherein: The specific process of step S6 includes the following steps: S61, use a scanner to obtain a point cloud frame, take the first obtained point cloud frame as a reference frame, and construct a world coordinate system with the camera optical center of the scanner as the coordinate origin; S62, define the Euclidean distance of the adjacent frame point cloud under the premise of meeting its final correspondence relationship as an objective function: ； wherein, , is a neighboring frame point cloud, is a rotation matrix, is a translation vector, a rigid transformation matrix are the variables to be optimized in the objective function; S63, respectively calculate the center of mass of the adjacent frame point cloud, simplify the above objective function by using the center of mass, and obtain the rigid transformation matrix between the adjacent frame point clouds by singular value decomposition (SVD) solution , using the rigid transformation matrix convert the point cloud frame by frame to the world coordinate system, and realize the initial registration between the adjacent tank wall plate point cloud frames.

Images