Cabin section key assembly feature extraction method based on measured data

Abstract

The invention discloses a cabin section key assembly feature extraction method based on measured data, and the method comprises the steps: obtaining the point cloud data of a movable cabin section and the point cloud data of a fixed cabin section, and carrying out the preprocessing; large-scale axial features and end face features are extracted from the preprocessed cabin point cloud data, then serialized partitioning is carried out along the axis, and a pin hole feature data set is extracted; small-scale positioning pin hole features are extracted from the pin hole feature data set through a neighborhood point dynamic adjustment and generalized fraction flexible fitting method; and an improved multi-dimensional least square method is adopted to solve the assembly pose, and high-precision butt-joint assembly is achieved. The method provided by the invention solves the problem that the key assembly features of the small-scale pin hole are difficult to accurately extract in the traditional method, greatly improves the accuracy of extracting the key assembly features of the large cabin section, and further provides accurate guidance for actual assembly.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud extraction, and particularly to a method for extracting key assembly features of a cabin section based on measured data. Background Art

[0002] As the core structural framework of a spacecraft, large cabin sections widely adopt a segmented design and are assembled through high-precision positioning pin-hole fits to ensure structural integrity and mechanical stability. The high-precision docking of cabin sections is a key link in the overall assembly process of a spacecraft and is widely used in the manufacturing of complex aerospace equipment such as rockets, spacecraft bodies, and missiles. Its assembly quality directly affects the overall mechanical performance, structural stability, and operation safety of the spacecraft, and is crucial for the service life and mission reliability of the spacecraft. In addition, with the continuous improvement of the requirements of space missions, the improvement of the docking accuracy of cabin sections has become one of the core directions for optimizing the manufacturing process of spacecraft.

[0003] Currently, traditional methods for extracting assembly features have many limitations in terms of accuracy, robustness, and adaptability, and it is difficult to meet the stringent requirements of high-precision cabin section docking. The ability to identify small features is limited. When facing high-density point clouds, complex curved surfaces, and small pin-hole features, it is easily affected by noise interference, resulting in a decrease in recognition accuracy and difficulty in maintaining the integrity and consistency of feature extraction across scales. In addition, the uneven density distribution of point clouds leads to unstable extraction results of fixed neighborhood or global methods. There is an urgent need for a more intelligent, robust, and efficient feature extraction strategy to improve the accuracy and reliability of spacecraft assembly. Summary of the Invention

[0004] Aiming at the problems existing in the prior art, the present invention provides a method for extracting key assembly features of a cabin section based on measured data, aiming to solve the problem that it is difficult to accurately extract the key assembly features of small-scale pin holes by traditional methods, thereby improving the accuracy of extracting key assembly features of large cabin sections and providing precise guidance for actual assembly.

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

[0006] A method for extracting key assembly features of a cabin section based on measured data, characterized in that it specifically includes the following steps:

[0007] S1. Obtain the point cloud data M of the moving cabin section and the point cloud data F of the fixed cabin section, and preprocess the obtained cabin section point cloud data, including denoising and downsampling;

[0008] S2. Extract large-scale axial features and end face features from the preprocessed cabin section point cloud data, and then perform serialization partitioning along the axis to extract the pin hole feature dataset;

[0009] S3. Extract small-scale locating pin hole features from the pin hole feature dataset through neighborhood point dynamic adjustment and generalized fractional flexible fitting method;

[0010] S4. Solve the assembly pose by using the improved multi-dimensional least squares method. By calculating the pose transformation of the locating pin hole point cloud, accurately solve the rotation matrix and translation vector of the moving cabin section; then convert the rotation matrix and translation vector into the motion instructions of the assembly platform to achieve high-precision docking assembly.

[0011] Further, step S1 specifically includes:

[0012] S11. Measure the to-be-assembled moving cabin section and fixed cabin section through an automated measuring device to obtain the complete moving cabin section point cloud data M and fixed cabin section point cloud data F;

[0013] S12. For the noise existing in the acquisition process, use voxel consistency filtering for denoising. Divide the point cloud into voxels, calculate the point cloud features within each voxel, perform voxel consistency screening, and use the voxel grid for downsampling. Merge the point clouds of each voxel to reduce the number of point clouds and obtain the preprocessed cabin section point cloud data.

[0014] Further, step S2 specifically includes:

[0015] S21. Analyze the point cloud data through the RANSAC algorithm, and extract the axial features and end face features of the cabin section as large-scale assembly features

[0016] S22. Serially partition the point cloud data along the axial direction of the cabin section to sequentially construct continuous point cloud datasets; for a given partition direction τ, the spatial set with n serially partitioned regions is expressed as:

[0017]

[0018] Taking the fixed cabin section point cloud data F as an example, its representation after serial partitioning is:

[0019]

[0020] Identify the pin hole features along the τ direction in each partition, and the final pin hole feature set obtained is expressed as:

[0021] The moving cabin section point cloud data M is serially partitioned and pin hole feature identified in the same way; the pin hole feature datasets in any partition of the fixed cabin section and the moving cabin section are respectively denoted as and and

[0022]

[0023] Furthermore, step S3 specifically includes:

[0024] S31, with fixed compartments in any partition Pin hole feature dataset in As an example, curvature-guided bidirectional feature aggregation is used to analyze The normal direction of the circumference point is used to classify pins and holes. The circumference point normal points to the cluster center for holes, and the circumference point normal points in the opposite direction to the center for pins.

[0025] S32, dynamically adjust the pin hole feature data set according to the local characteristics of the point cloud The number of neighboring points of any point in the sphere; and generalized fractional flexible fitting;

[0026] S33, for each point q in the neighborhood determined in step S32 i , using its normal direction n i Construct the equation of the line: L i :x=q i +tn i ; Find the intersection point x of each pair of normal lines ij And record the intersection set X = {x ij}; Perform weighted average on all intersection points, with the weight being the angle θ between the normal lines of the neighboring points ij Determine, specifically Then find the partition The center C and radius R of the center pin hole are:

[0027]

[0028] S34. The obtained center and radius are used as the final circle extraction result, and the extraction results of all serialized partitions are merged to obtain all small-scale positioning pin hole features, which are used together with large-scale axial features and end face features as key assembly features to obtain the assembly posture.

[0029] More specifically, the number of neighborhood points dynamically adjusted in step S32 is as follows:

[0030] The curvature-density collaborative adaptive method is adopted, and the curvature factor C(P i ), dynamically adjust the number of neighborhood points k according to the local characteristics of the point cloud; Pin hole feature dataset in For example, for any point P i , the calculation formula of the number of neighborhood points k is:

[0031]

[0032] Among them, Round(·) is the rounding function, γ is the density weight exponent, C(P i ) is the curvature factor at point P i , δ represents the overall scale of the point cloud, Res(P) is the point cloud resolution;

[0033] The local sampling density LocalDensity is calculated by the kernel density estimation method, and the formula is expressed as:

[0034]

[0035] Among them, P j is the neighborhood point of P i , N(P i ) is the neighborhood point set of point P i , is the square of the Euclidean distance between point P j and point P j , and σ is the density influence factor;

[0036] The global average point cloud density GlobalDensity is the mean value of the local sampling densities of all points, that is:

[0037]

[0038] More specifically, the generalized fractional flexible fitting in step S32 is specifically as follows:

[0039] Perform generalized fractional flexible fitting on the neighborhood of point P i after determining the number of neighborhood points, and propose a fractional order continuity control method to break through the traditional integer order derivative constraint; the mathematical model of the fitting curve is:

[0040]

[0041] Among them, C α (u) represents the fractional order continuity fitting curve, u is the parameter variable, is the fractional order basis function, w i is the weight coefficient, Ps i is the control point participating in the fractional flexible fitting, generated according to the points in the pinhole feature dataset , and n is the total number of control points;

[0042] The fractional order basis function is defined as:

[0043]

[0044] Among them, Γ(1 + α) is the Gamma function, used for normalizing the fractional order calculation, α is the fractional order, u i is the control point parameter, θ is the integration variable, p is the derivative order, Ni,p B-spline basis function is \(N_{i,p}(\theta)\), and \(p\) is the order of derivative.

[0045] Furthermore, step S4 specifically includes:

[0046] S41. First, calculate the centroid of each pair of locating pin hole point clouds. Taking the fixed cabin section pin point cloud data \(A = \{a_1, a_2, \cdots, a_n\}\) n and the moving cabin section hole point cloud data \(B = \{b_1, b_2, \cdots, b_n\}\) as an example, the calculation formula is: n

[0047]

[0048] Then, de - center to obtain point sets \(A'\) and \(B'\), which is expressed by the formula:

[0049] \(A'=\{a_i' = a_i - C_A|i = 1, 2, \cdots, n\}\), \(B'=\{b_i' = b_i - C_B|i = 1, 2, \cdots, n\}\); i where \(a_i' = a_i - C_A\) i -C_A A |i = 1, 2, \cdots, n\}, B'=\{b i ′ = b i -C B |i = 1, 2, \cdots, n\};

[0050] where \(C_A\) A and \(C_B\) B are the centroids of the fixed cabin section pin point cloud data and the moving cabin section hole point cloud data respectively;

[0051] S42. Define the covariance matrix to describe the correlation between the corresponding pin hole point clouds. Denote the moving cabin section hole point cloud data \(B = \{b_1, b_2, \cdots, b_n\}\), n add weights to optimize and construct the covariance matrix \(H\), which is expressed by the formula:

[0052]

[0053] where the weight \(m_i\) i is adaptively adjusted by the normal vector angle and point density, and is expressed by the formula: \(\theta_i\) i is the normal vector angle of the corresponding points of the pin holes, \(j\) is the point density; then perform singular value decomposition \(H = USV\) T , where \(U\) and \(V\) are orthogonal matrices, \(S\) is a diagonal matrix of singular values, and the pose parameters of the moving cabin section are obtained. The pose parameters include the rotation matrix \(R\) Cabin =VU^T T and the translation vector \(T\) Cabin =C_B - R C_A^T; B -R Cabin C A ;

[0054] ​S43. Convert the obtained pose parameters into the motion parameters (x, y, z, α, β, γ) of the six-degree-of-freedom driving platform through Euler angle decomposition, and convert them into the change amount of the driving rod length of the driving platform. x, y, and z are translation parameters, representing the linear displacements of the mobile cabin section in the x, y, and z directions. α, β, and γ are rotation parameters, representing the rotation angles of the mobile cabin section around the x, y, and z axes. Drive the mobile cabin section to perform the docking action to achieve high-precision docking and assembly of the cabin sections.

[0055] Based on the above technical solutions, the present invention has at least the following beneficial effects:

[0056] The present invention solves the problem that it is difficult to accurately extract the key assembly features of small-scale pin holes in traditional methods. By dynamically adjusting the number of neighboring points of the point cloud, the traditional method uses a fixed neighborhood scale, which is prone to losing details in high-curvature regions and may introduce redundant noise in low-curvature regions. The present invention adaptively adjusts the number of neighboring points according to features such as the local density and curvature distribution of the point cloud to ensure that the feature point set can maintain integrity and reduce the interference of redundant points. The introduction of generalized fractional-order flexible fitting improves the fitting accuracy of the point cloud features, greatly improves the accuracy of extracting key assembly features of different scales of large cabin sections, and finally converts the extracted key assembly features into pose parameters through the multi-dimensional least squares method improved by weight optimization to provide precise guidance for actual assembly. Description of the Drawings

[0057] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0058] Figure 1 is a flowchart of a method for extracting key assembly features of a cabin section based on measured data proposed by the present invention;

[0059] Figure 2 is a schematic diagram of the point clouds of the fixed cabin section and the mobile cabin section to be assembled after preprocessing;

[0060] Figure 3 is a schematic diagram of the extraction results of multi-scale key assembly features by the method proposed by the present invention. Detailed Embodiments

[0061] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following combines the attached Figures 1-3 drawings and embodiments to further elaborate on the present invention. Thereby, a full understanding of how the present application applies technical means to solve technical problems and achieve the realization process of technical effects can be obtained and implemented accordingly.

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

[0063] As Figure 1 shown, the present invention proposes a flowchart of a method for extracting key assembly features of a cabin section based on measured data, which specifically includes the following steps:

[0064] S1. Obtain the point cloud data M of the moving cabin section and the point cloud data F of the fixed cabin section, and preprocess the obtained cabin section point cloud data, including denoising and downsampling;

[0065] As a preferred implementation manner, step S1 specifically includes:

[0066] S11. Measure the to-be-assembled moving cabin section and fixed cabin section through an automated measuring device to obtain the complete point cloud data M of the moving cabin section and the point cloud data F of the fixed cabin section;

[0067] S12. For the noise existing in the acquisition process, use voxel consistency filtering for denoising, divide the point cloud into voxels, calculate the point cloud features within each voxel, perform voxel consistency screening, and use the voxel grid for downsampling, and merge the point clouds of each voxel to reduce the number of point clouds, obtaining the preprocessed cabin section point cloud data, as Figure 2 shown;

[0068] S2. Extract the large-scale axial features and end face features from the preprocessed cabin section point cloud data, and then perform serial partitioning along the axis to extract the pin hole feature dataset;

[0069] As a preferred implementation manner, step S2 specifically includes:

[0070] S21. Analyze the point cloud data through the RANSAC algorithm to extract the axial features and end face features of the cabin section as large-scale assembly features

[0071] S22. Perform serial partitioning on the point cloud data along the axis of the cabin section to sequentially construct continuous point cloud datasets; for a given partitioning direction τ, the spatial set with n serial partitions is expressed as:

[0072]

[0073] Taking the point cloud data F of the fixed cabin section as an example, its representation after serial partitioning is:

[0074]

[0075] If pin hole feature recognition is performed along the τ direction in each partition, the resulting final set of pin hole features is represented as:

[0076] The point cloud data M of the moving cabin section is serialized and partitioned and pin hole feature recognition is performed in the same way; the pin hole feature data sets in any partition of the fixed cabin section and the moving cabin section are respectively denoted as in and

[0077]

[0078] S3. Extract small-scale locating pin hole features from the pin hole feature data set through the neighborhood point dynamic adjustment and the generalized fractional flexible fitting method;

[0079] As a preferred implementation manner, step S3 specifically includes:

[0080] S31. Taking the pin hole feature data set in any partition of the fixed cabin section as an example, adopt curvature-guided bidirectional feature aggregation, analyze the normal direction of the circumferential points in , classify the pins and holes, the circumferential points whose normals point to the clustering center are holes, and the circumferential points whose normals point in the opposite direction to the center are pins;

[0081] S32. Dynamically adjust the number of neighborhood points of any point in the pin hole feature data set according to the local characteristics of the point cloud; and perform generalized fractional flexible fitting;

[0082] In this application, the number of neighborhood points is first dynamically adjusted to enhance the adaptability to complex structures; the complexity of point cloud data is reflected in aspects such as curvature change, density change, and different scales of local geometric features. When traditional methods perform feature extraction, they usually select a fixed-size neighborhood (K-nearest neighbor or radius neighborhood); when the result of point cloud data is complex, a fixed-size neighborhood may lead to:

[0083] ① Local density change, redundant point cloud in high-density areas and sparse in low-density areas;

[0084] ② Difficulty in adapting to complex curved surfaces: it is difficult to obtain all the feature details in places with large curvature;

[0085] ③ When there are cross-scale problems, small-scale features will not be overwhelmed by large-scale features;

[0086] ​Therefore, this application designs a method for dynamically adjusting the number of neighborhood points to enhance local feature perception in high-curvature regions and reduce redundant data in low-curvature regions, so as to adapt to the complex shapes with different curvature sizes of the docking section (such as regions with sudden curvature changes, uneven distribution regions, and fine assembly features). As a preferred embodiment, the dynamic adjustment of the number of neighborhood points in step S32 is specifically as follows:

[0087] Adopt a curvature-density collaborative adaptive method, introduce a curvature factor C(P i ), and dynamically adjust the number of neighborhood points k according to the local characteristics of the point cloud; take the pin hole feature data set in any partition of the fixed section as an example, for any point P i in it, the calculation formula for the number of its neighborhood points k is:

[0088]

[0089] where, Round(·) is a rounding function, γ is a density weight exponent, C(P i ) is the curvature factor at point P i , δ represents the overall scale of the point cloud, and Res(P) is the point cloud resolution;

[0090] The local sampling density LocalDensity is calculated by the kernel density estimation method, and the formula is expressed as:

[0091]

[0092] where, P j is the neighborhood point of P i , N(P i ) is the neighborhood point set of point P i , is the square of the Euclidean distance between point P j and point P j , and σ is a density influence factor;

[0093] The global average point cloud density GlobalDensity is the mean value of the local sampling densities of all points, that is:

[0094]

[0095] On the basis of dynamically adjusting the neighborhood points, this application further processes the pin hole feature data set through generalized fractional flexible fitting to fit the curve more smoothly, more flexibly, and more finely, so as to better adapt to complex structures and improve the expression ability of the point cloud for complex curved surfaces; the specific process of generalized fractional flexible fitting is as follows:

[0096] For point P iThe neighborhood uses generalized fractional flexible fitting; under the traditional integer-order constraint, the smoothness of the curve is determined by the integer-order derivative. Once a certain order is selected, it is impossible to flexibly adjust the smoothness of the curve in different regions. Therefore, a fractional-order continuity control method is proposed to break through the traditional integer-order derivative constraint. The mathematical model of the fitting curve is:

[0097]

[0098] where C α (u) represents the fractional-order continuity fitting curve, u is the parameter variable, is the fractional-order basis function, w i is the weight coefficient, Ps i are the control points participating in the fractional flexible fitting, generated according to the points in the pinhole feature dataset , and n is the total number of control points;

[0099] The fractional-order basis function is defined as:

[0100]

[0101] where Γ(1 + α) is the Gamma function, used for normalizing the fractional-order calculation, α is the fractional-order, u i is the control point parameter, θ is the integration variable, p is the derivative order, N i,p (θ) is the B-spline basis function, is the Pth derivative.

[0102] S33. After improving the adaptability to complex point cloud structures in step S32, step S33 will solve the two-dimensional circular features of the pinholes in each partition. Specifically: for each point q i in the fitted neighborhood, use its normal direction n i to construct the straight line equation: L i : x = q i + tn i ; find the intersection points x ij of the normals pairwise and record the intersection point set X = {x ij}; perform weighted averaging on all intersection points, and the weights are determined by the normal angle θ ij of the neighborhood points. Specifically and then find the center C and radius R of the pinhole in the partition as follows:

[0103]

[0104] S34. Take the obtained center and radius as the final circular extraction result, and merge the extraction results of all serialized partitions to obtain all the positioning pin hole features at a small scale. Together with the axial features and end face features at a large scale, they are used as key assembly features, as Figure 3 shown. It is used to obtain the assembly pose.

[0105] S4. Use the improved multi-dimensional least squares method to solve the assembly pose. By calculating the pose transformation of the positioning pin hole point cloud, accurately solve the rotation matrix and translation vector of the moving cabin section; then convert the rotation matrix and translation vector into the motion instructions of the assembly platform to achieve high-precision docking assembly;

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

[0107] S41. First calculate the centroid of each pair of positioning pin hole point clouds; take the fixed cabin section pin point cloud data A = {a1, a2,..., a n} and the moving cabin section hole point cloud data B = {b1, b2,..., b n} as an example. The calculation formula is:

[0108]

[0109] Then de-center to obtain point sets A' and B', which are expressed by the formula:

[0110] A′ = {a i ′ = a i - C A ∣i = 1, 2,..., n}, B' = {b i ′ = b i - C B ∣i = 1, 2,..., n};

[0111] Among them, C A 、C B are the centroids of the fixed cabin section pin point cloud data and the moving cabin section hole point cloud data;

[0112] S42. Define the covariance matrix to describe the correlation between the corresponding pin hole point clouds; denote the moving cabin section hole point cloud data B = {b1, b2,..., b n}, add weights to optimize and construct the covariance matrix H, which is expressed by the formula:

[0113]

[0114] Among them, the weight m i is adaptively adjusted by the normal vector angle and point density, and is expressed by the formula: θ i is the normal vector angle of the corresponding points of the pin holes, and j is the point density; then perform singular value decomposition H = USVT , U and V are orthogonal matrices, S is a singular value diagonal matrix, and the posture parameters of the mobile cabin are obtained. The posture parameters include the rotation matrix R Cabin =VU T and the translation vector T Cabin =C B -R Cabin C A ;

[0115] S43. Convert the obtained posture parameters into motion parameters (x, y, z, α, β, γ) of a six-degree-of-freedom driving platform through Euler angle decomposition, and convert them into length changes of driving rods of the driving platform. x, y, z are translation parameters, representing the linear displacement of the mobile cabin in the x, y, z directions; α, β, γ are rotation parameters, representing the rotation angle of the mobile cabin around the x, y, z axes, and driving the mobile cabin to perform docking action to achieve high-precision docking assembly of the cabin. The specific docking assembly process is: first, perform rough axial alignment according to large-scale axial features, and then perform pin hole alignment and end face fitting alignment according to the posture parameters solved based on the end face features and small-scale positioning pin hole features.

[0116] At this point, the present invention solves the problem that small-scale pin hole key assembly features are difficult to accurately extract using traditional methods. By dynamically adjusting the number of point cloud neighborhood points, it ensures that the feature point set maintains integrity and reduces redundant point interference. The generalized fractional-order flexible fitting is introduced to improve the fitting accuracy of point cloud features, greatly improving the accuracy of key assembly feature extraction at different scales. Finally, the extracted key assembly features are converted into pose parameters through the improved multidimensional least squares method, which can provide precise guidance for the actual assembly of the cabin.

[0117] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0118] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for extracting key assembly features of a cabin section based on measured data, characterized in that: The specific steps include: S1, obtaining the mobile compartment point cloud data M and the fixed compartment point cloud data F, and preprocessing the obtained compartment point cloud data, including denoising and downsampling; S2, extracting large-scale axial features and end surface features from the pre-processed cabin point cloud data, and then performing serialization partitioning along the axis to extract the pin hole feature data set; S3, extracting small-scale positioning pin hole features from the pin hole feature dataset through dynamic adjustment of neighborhood points and generalized fractional flexible fitting method; S4. Use the improved multi-dimensional least squares method to solve the assembly posture. By calculating the posture transformation of the positioning pin hole point cloud, the rotation matrix and translation vector of the mobile cabin are accurately solved; then the rotation matrix and translation vector are converted into motion instructions for the assembly platform to achieve high-precision docking assembly.

2. The method for extracting key assembly features of a cabin section based on measured data according to claim 1 is characterized in that: Step S1 specifically includes the following steps: S11. Using an automated measuring device, measure the mobile cabin and the fixed cabin to be assembled to obtain complete mobile cabin point cloud data M and fixed cabin point cloud data F; S12. In order to deal with the noise existing in the acquisition process, voxel consistency filtering is used for denoising, the point cloud is divided into voxels, the point cloud features in each voxel are obtained, voxel consistency screening is performed, and down-sampling is performed using the voxel grid. The point cloud of each voxel is merged to reduce the number of point clouds and obtain the pre-processed cabin point cloud data.

3. The method for extracting key assembly features of a cabin section based on measured data according to claim 1 is characterized in that: Step S2 specifically includes: S21. Analyze the point cloud data through the RANSAC algorithm to extract the axial features and end surface features of the cabin as large-scale assembly features S22, serialize and partition the point cloud data along the cabin axis, and construct a continuous point cloud data set in sequence; for a given partition direction τ, the spatial set with n serialized partitions is expressed as: Taking the fixed compartment point cloud data F as an example, its serialized partitioning is represented as follows: Pin hole feature recognition is performed along the τ direction in each partition, and the final pin hole feature set is expressed as: The mobile cabin point cloud data M is serialized and partitioned and pin hole feature recognized in the same way; any partition of the fixed cabin and the mobile cabin The pin hole feature data sets in and 4. The method for extracting key assembly features of a cabin section based on measured data according to claim 3 is characterized in that: Step S3 specifically includes: S31, with fixed compartments in any partition Pin hole feature dataset in As an example, curvature-guided bidirectional feature aggregation is used to analyze The normal direction of the circumference point is used to classify pins and holes. The circumference point normal points to the cluster center for holes, and the circumference point normal points in the opposite direction to the center for pins. S32, dynamically adjust the pin hole feature data set according to the local characteristics of the point cloud The number of neighboring points of any point in the sphere; and generalized fractional flexible fitting; S33, for each point q in the fitted neighborhood i , using its normal direction n i Construct the equation of the line: L i :x=q i +tn i ; Find the intersection point x of each pair of normal lines ij And record the intersection set X = {x ij }; Perform weighted average on all intersection points, with the weight being the angle θ between the normal lines of the neighboring points ij Determine, specifically Then find the partition The center C and radius R of the center pin hole are: S34. The obtained center and radius are used as the final circle extraction result, and the extraction results of all serialized partitions are merged to obtain all small-scale positioning pin hole features, which are used together with large-scale axial features and end face features as key assembly features to obtain the assembly posture.

5. The method for extracting key assembly features of a cabin section based on measured data according to claim 4 is characterized in that: The specific method of dynamically adjusting the number of neighborhood points in step S32 is as follows: The curvature-density collaborative adaptive method is adopted, and the curvature factor C(P i ), dynamically adjust the number of neighborhood points k according to the local characteristics of the point cloud; Pin hole feature dataset in For example, for any point P i , the calculation formula of the number of neighborhood points k is: Among them, Round(·) is the rounding function, γ is the density weight index, C(P i ) is point P i The curvature factor at , δ represents the overall scale of the point cloud, and Res(P) is the point cloud resolution; The local sampling density LocalDensity is calculated by the kernel density estimation method, and the formula is expressed as: Among them, P j P i Neighborhood points, N(P i ) is point P i The neighborhood point set of Point P j With point P j The square of the Euclidean distance between them, σ is the density influence factor; The global average point cloud density GlobalDensity is the mean of the local sampling density of all points, that is:

6. The method for extracting key assembly features of a cabin section based on measured data according to claim 4 is characterized in that: The generalized fractional flexible fitting in step S32 is specifically: After determining the number of neighboring points, point P i The neighborhood of is fitted with generalized fractional flexible fitting, and a fractional-order continuity control method is proposed to break through the traditional integer-order derivative constraint; the mathematical model of the fitting curve is: Among them, C α (u) represents the fractional order continuity fitting curve, u is the parameter variable, is a fractional basis function, w i is the weight coefficient, Ps i For the control points involved in fractional flexible fitting, according to the pin hole feature data set The points in are generated, n is the total number of control points; The fractional order basis function is defined as: Among them, Γ(1+α) is the Gamma function, which is used for normalized fractional order calculation, α is the fractional order, and u i is the control point parameter, θ is the integral variable, p is the derivative order, N i,p (θ) is the B-spline basis function, is the P-order derivative.

7. The method for extracting key assembly features of a cabin section based on measured data according to claim 1 is characterized in that: Step S4 specifically includes the following steps: S41, first calculate the centroid of each pair of positioning pin hole point cloud; select the fixed cabin pin point cloud data A = {a1, a2, ..., a n } and mobile compartment hole point cloud data B = {b1, b2, ..., b n } as an example, the calculation formula is: Then decentralize to get point sets A′ and B′, the formula is expressed as: A′={a i ′=a i -C A ∣i=1,2,...,n},B'={b i ′=b i -C B ∣i=1,2,...,n}; Among them, C A , C B It is the centroid of the fixed cabin pin point cloud data and the mobile cabin hole point cloud data; S42, define the covariance matrix to describe the correlation between the corresponding pin hole point clouds; record the mobile cabin hole point cloud data B = {b1, b2, ..., b n }, add weight optimization to construct the covariance matrix H, the formula is expressed as: Among them, the weight m i The formula for adaptive adjustment of normal vector angle and point density is expressed as: θ i is the normal vector angle of the pin hole corresponding point, j is the point density; then perform singular value decomposition H = USV T , U and V are orthogonal matrices, S is a singular value diagonal matrix, and the posture parameters of the mobile cabin are obtained. The posture parameters include the rotation matrix R Cabin =VU T and the translation vector T Cabin =C B -R Cabin C A ; S43. The obtained posture parameters are converted into motion parameters (x, y, z, α, β, γ) of the six-degree-of-freedom driving platform through Euler angle decomposition, and then converted into the length change of the driving rod of the driving platform. x, y, z are translation parameters, indicating the linear displacement of the mobile cabin in the x, y, z directions. α, β, γ are rotation parameters, indicating the rotation angle of the mobile cabin around the x, y, z axes, which drive the mobile cabin to perform the docking action and realize the high-precision docking assembly of the cabin.

Images