A method and system for identifying a docking ring of a spatially failed satellite based on three-dimensional point cloud matching

Abstract

The application discloses a kind of space invalid satellite docking ring identification method and system based on three-dimensional point cloud matching, it is related to target identification technical field.The technical points of the present application include: the local feature description of the three-dimensional reconstruction point cloud corresponding to space invalid satellite is carried out, SAC-IA rough matching is carried out with docking ring standard point cloud, and the area point cloud containing docking ring component is obtained;With the pose calculated in rough matching as initial value, ICP fine matching is carried out with docking ring standard point cloud, and the docking ring component is accurately identified, and the point cloud coordinates are extracted using nearest neighbor search;And the principal component analysis method is used to project and calculate the center space position and diameter of docking ring component.The present application can accurately identify the docking ring component of space invalid satellite, and simultaneously determine its characteristic site size and position information, provide technical support for the implementation of subsequent approach, capture and other tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target recognition, and particularly relates to a space failure satellite docking ring recognition method and system based on three-dimensional point cloud matching. BACKGROUND

[0002] In recent years, with the increasing demand of human beings for space resource utilization and exploration, developing reliable autonomous on-orbit service and spacecraft autonomous operation control technology to perform autonomous proximity operation under the condition of limited resources on the satellite has become an important research direction of future space technology. Spacecrafts are installed with key components having certain characteristics in order to realize certain functions, and identifying these characteristics can provide information for autonomous proximity and navigation. For space non-cooperative targets, it is of great significance to utilize natural characteristics of target surfaces in space attack and defense confrontation. Docking rings are components installed on most spacecrafts, and play an important role in space failure satellite pose measurement, rendezvous and docking, and proximity and capture of target spacecrafts. Three-dimensional point clouds contain accurate spatial positions, and are more conducive to subsequent pose measurement. Therefore, space failure satellite docking ring recognition on three-dimensional point clouds has important value and significance.

[0003] Research shows that the prior art adopts a method of recognizing point cloud rectangular surface features to process the pose measurement problem of non-cooperative targets, and some people extract point curvature, normal, point cloud density and other geometric features in point clouds to recognize targets, and calculate the similarity of features of two adjacent frames of point clouds to track space failure satellites. However, most of the recognition researches on docking rings, which are key components of space failure satellites, are in two-dimensional images, such as detecting the satellite-rocket docking ring part in the image by using the least square fitting ellipse method or recognizing the docking ring component from the image with high precision by using the deep learning method. However, images lack depth information, which cannot provide accurate spatial positions of targets. SUMMARY

[0004] In view of the above problems, the present application provides a space failure satellite docking ring recognition method and system based on three-dimensional point cloud matching, so as to recognize the space failure satellite docking ring component in the three-dimensional reconstruction point cloud by matching with the standard point cloud, determine the diameter and center position coordinates thereof, and provide support for subsequent tasks.

[0005] According to an aspect of the present application, a space failure satellite docking ring recognition method based on three-dimensional point cloud matching is provided, which comprises the following steps:

[0006] Step one, obtaining a three-dimensional reconstruction point cloud containing a satellite;

[0007] Step two, using a sampling consistency initial registration algorithm to coarsely match the three-dimensional reconstruction point cloud with a docking ring standard point cloud, and obtaining a region point cloud containing a docking ring component;

[0008] Step three, using the iterative closest point algorithm to perform fine matching on the region point cloud containing the docking ring component and the docking ring standard point cloud, and using the nearest neighbor search to extract the three-dimensional point coordinates of the docking ring component;

[0009] Step four, according to the three-dimensional point coordinates of the docking ring component, using principal component analysis method projection to calculate the center coordinates and diameter of the docking ring component.

[0010] Further, the specific process of step two includes:

[0011] Step two one, uniformly down-sampling the three-dimensional reconstruction point cloud P and the docking ring standard point cloud Q respectively to obtain P' and Q';

[0012] Step two two, calculating the normal vector n P' , n Q' of each point in P' and Q' respectively;

[0013] Step two three, calculating the point fast feature histogram feature of each point in Q', and constructing a spatial index structure based on KDTree for the point fast feature histogram feature;

[0014] Step two four, selecting n sampling points p" i from P', calculating the point fast feature histogram feature of the sampling points; according to the feature similarity principle, using K nearest neighbor search in Q' to find the points with the closest feature distance to p" i , and selecting the point with the smallest normal vector difference from p" i as the corresponding point to obtain the corresponding point set {P", Q"};

[0015] Step two five, calculating the rigid transformation matrix between the corresponding points, and judging the performance of the current registration transformation by solving the distance error sum function after the transformation of the corresponding points;

[0016] Step two six, iteratively repeating steps two four to two five until the distance error sum function reaches a preset threshold or reaches a maximum number of iterations, at which time the corresponding rigid transformation matrix is the final registration transformation matrix {R0, T0};

[0017] Step two seven, performing pose conversion on the three-dimensional reconstruction point cloud P by the registration transformation matrix {R0, T0} to obtain the point cloud P0=R0P+T0;

[0018] Step two eight, using the BIRCH clustering algorithm to perform segmentation processing on the point cloud P0 to obtain a plurality of partial point sets {P 0i};

[0019] Step two nine, calculating the distance between Q' and each partial point set P 0iThe overlap rate of the three-dimensional point coordinates is used to select the point set with the largest overlap rate, which is the point cloud of the region containing the docking ring component.

[0020] Furthermore, the distance error sum function described in step two-five is represented using the Huber penalty function, denoted as... in:

[0021]

[0022] In the formula, m l For a pre-defined value, l i Let be the distance difference between the corresponding points in the i-th group after transformation.

[0023] Furthermore, the specific steps of step three include:

[0024] Step 31: Using the region point cloud P containing the docking ring component... part Using the registration transformation matrix {R0, T0} as initial values, calculate the point cloud P. part For each point in the standard point cloud Q of the docking loop, the nearest point in the standard point cloud Q is obtained, resulting in the corresponding point set (P). part Q j );

[0025] Step 3.2: Obtain the set of corresponding points (P) through singular value decomposition. part Q j Average distance in The minimum rigid body transformation {R', T'}, where n is the number of points, R' is the rotation matrix, and T' is the translation vector, then the new transformed point set P' part =R'P part +T';

[0026] Step 3: Calculate the transformed point set P p ' art Average distance between the docking ring and the standard point cloud Q The calculation stops when the average distance is less than a given threshold; otherwise, the above steps are repeated to calculate a new registration transformation matrix using the current {R', T'} as the initial value until... The iterative calculation stops when the threshold requirement is met or the number of iterations reaches a given value; at this point, the final registration transformation matrix {R, T} and the transformed point cloud are obtained.

[0027] Steps 3 and 4: Based on the spatial index structure of the KDTree constructed in Steps 2 and 3, transform the point cloud in the standard point cloud Q of the docking loop. The nearest neighbor search is used to find the point closest to a point in Q, and the coordinates P of the nearest point are extracted. roll This means accurately identifying the point cloud of the docking ring component.

[0028] Further, the specific steps of step four include:

[0029] Step four one, calculating the mean value of each dimension of point cloud coordinate P roll (X roll ,Y roll ,Z roll ) to zero-mean the three-dimensional point coordinate, and obtaining point cloud coordinate matrix P' roll ;

[0030] Step four two, solving the covariance matrix of P' roll ;

[0031] Step four three, performing eigenvalue decomposition V -1 C V = D, where D is a diagonal matrix composed of eigenvalues of the covariance matrix, and V is an eigenvector matrix, also known as a principal component coefficient matrix; taking the first two rows of the principal component coefficient matrix V to form a matrix E, and projecting P" roll = E * P roll to the XY plane;

[0032] Step four four, calculating the diameter of the docking ring where X max , X min , Y max , Y min represent the maximum and minimum values of the coordinates, and let the spatial position coordinates of the center of the circle be (X0, Y0, Z0), then Z0 takes the maximum value of the third dimension data.

[0033] According to another aspect of the present application, a space failure satellite docking ring identification system based on three-dimensional point cloud matching is provided, which comprises:

[0034] A reconstructed point cloud acquisition module configured to acquire a three-dimensional reconstructed point cloud containing a satellite;

[0035] A coarse matching module configured to perform coarse matching of the three-dimensional reconstructed point cloud and a docking ring standard point cloud by using a sampling consistency initial registration algorithm, and to acquire a regional point cloud containing a docking ring component;

[0036] A fine matching module configured to perform fine matching of the regional point cloud containing the docking ring component and the docking ring standard point cloud by using an iterative closest point algorithm, and to extract three-dimensional point coordinates of the docking ring component by using nearest neighbor search;

[0037] A docking ring identification module configured to calculate the center coordinates and the diameter of the docking ring component by using principal component analysis projection according to the three-dimensional point coordinates of the docking ring component.

[0038] Further, the specific process of acquiring the regional point cloud containing the docking ring component in the coarse matching module includes:

[0039] Step two one, uniformly down-sampling the three-dimensional reconstructed point cloud P and the standard point cloud Q of the butt joint ring respectively to obtain P' and Q';

[0040] Step two two, calculating the normal vector n of each point in P' and Q' respectively P' Q' ;

[0041] Step two three, calculating the point fast feature histogram feature of each point in Q', and constructing a KDTree-based spatial index structure based on the point fast feature histogram feature;

[0042] Step two four, selecting n sampling points p" from P' i , calculating the point fast feature histogram feature of the sampling points; according to the feature similarity principle, using K-nearest neighbor search in Q' to find the points closest to p" i in feature distance, and selecting the point with the smallest difference from p" i in normal vector as the corresponding point to obtain the corresponding point set {P", Q"};

[0043] Step two five, calculating the rigid transformation matrix between the corresponding points, and judging the performance of the current registration transformation by solving the distance error sum function after the transformation of the corresponding points;

[0044] Step two six, iteratively repeating steps two four to two five until the distance error sum function reaches a preset threshold or reaches a maximum number of iterations, at which time the corresponding rigid transformation matrix is the final registration transformation matrix {R0, T0};

[0045] Step two seven, performing pose conversion on the three-dimensional reconstructed point cloud P by the registration transformation matrix {R0, T0} to obtain the point cloud P0=R0P+T0;

[0046] Step two eight, performing segmentation processing on the point cloud P0 by using the BIRCH clustering algorithm to obtain a plurality of partial point sets {P 0i};

[0047] Step two nine, calculating the three-dimensional point coordinate overlap rate of Q' and each partial point set P 0i , and selecting the point set with the largest overlap rate, that is, the area point cloud containing the butt joint ring component.

[0048] Further, the distance error sum function in the coarse matching module is expressed using a Huber penalty function, denoted as wherein:

[0049]

[0050] In the formula, m l is a pre-given value, and l i ​Let be the distance difference between the corresponding points in the i-th group after transformation.

[0051] Furthermore, the specific steps in the fine matching module for performing fine matching between the region point cloud containing the docking ring component and the standard point cloud of the docking ring using the iterative nearest neighbor algorithm, and for extracting the three-dimensional point coordinates of the docking ring component using nearest neighbor search, include:

[0052] Step 31: Using the point cloud P of the region containing the docking ring component... part Using the registration transformation matrix {R0, T0} as initial values, calculate the point cloud P. part For each point in the loop, the nearest point in the standard point cloud Q of the docking ring is obtained, resulting in the corresponding point set (P). part Q j );

[0053] Step 3.2: Obtain the set of corresponding points (P) through singular value decomposition. part Q j Average distance in The minimum rigid body transformation {R', T'}, where n is the number of points, R' is the rotation matrix, and T' is the translation vector, then the new transformed point set P p ' art =R'P part +T';

[0054] Step 3: Calculate the transformed point set P p ' art Average distance between the docking ring and the standard point cloud Q The calculation stops when the average distance is less than a given threshold; otherwise, the above steps are repeated to calculate a new registration transformation matrix, using the current {R', T'} as the initial value, until... The iterative calculation stops when the threshold requirement is met or the number of iterations reaches a given value; at this point, the final registration transformation matrix {R, T} and the transformed point cloud are obtained.

[0055] Steps 3 and 4: Based on the spatial index structure of the KDTree constructed in Steps 2 and 3, transform the point cloud in the standard point cloud Q of the docking loop. The nearest neighbor search is used to find the point closest to a point in Q, and the coordinates P of the nearest point are extracted. roll This means accurately identifying the point cloud of the docking ring component.

[0056] Furthermore, the specific steps in the docking ring identification module for calculating the center coordinates and diameter of the docking ring component using principal component analysis projection based on the three-dimensional point coordinates of the docking ring component include:

[0057] Step 41: Calculate the point cloud coordinates P roll (X roll ,Y rollZ roll )Mean of each dimension, zero-mean of three-dimensional point coordinates, get point cloud coordinate matrix P' roll ;

[0058] Step four two, solve the covariance matrix of P' roll

[0059] Step four three, eigenvalue decomposition of the covariance matrix V -1 CV=D, where D is the diagonal matrix of eigenvalues of the covariance matrix, V is the eigenvector matrix, also called principal component coefficient matrix; Take the first two rows of the principal component coefficient matrix V to form a matrix E, and use P" roll =E*P roll Project the point cloud to the XY plane;

[0060] Step four four, calculate the diameter of the docking ring Where X max , X min , Y max , Y min represent the maximum value of the coordinates, let the spatial position coordinates of the center of the circle be (X0, Y0, Z0), then Z0 take the maximum value of the third dimension data.

[0061] The beneficial technical effects of the present application are:

[0062] The present application provides a kind of based on three-dimensional point cloud matching's space failure satellite docking ring identification method and system, by clustering segmentation point cloud and point cloud rough matching combination, reduce the identification area of the key components docking ring of reconstructed point cloud, and provide the initial transformation pose with template point cloud;Further utilize accurate docking ring part of fine matching identification, based on the nearest neighbor search of Euclidean distance extraction three-dimensional position coordinates of docking ring;Further utilize principal component analysis method and project point cloud, calculate the center position coordinates and diameter size of docking ring.The present application can provide information for the approach, capture and other tasks of target spacecraft. BRIEF DESCRIPTION OF DRAWINGS

[0063] The present application can be better understood by referring to the description given below in conjunction with the accompanying drawings, which are included in the specification and form a part of the specification, and serve to further illustrate preferred embodiments of the present application and explain the principles and advantages of the present application.

[0064] Figure 1 A flow chart of a kind of based on three-dimensional point cloud matching's space failure satellite docking ring identification method described in the present application;

[0065] Figure 2 It is satellite body coordinate system establishment mode example graph in the embodiment of the present application.​

[0066] Figure 3 A result example diagram of establishing a satellite body coordinate system in an embodiment of the present application is shown in FIG. 1.

[0067] Figure 4 A result example diagram of satellite point cloud segmentation in an embodiment of the present application is shown in FIG. 2.

[0068] Figure 5 A result example diagram of coarse matching of a reconstructed point cloud and a docking ring standard point cloud SAC-IA in an embodiment of the present application is shown in FIG. 3.

[0069] Figure 6 A result example diagram of fine matching of a reconstructed region point cloud containing a docking ring region and a docking ring standard point cloud ICP in an embodiment of the present application is shown in FIG. 4.

[0070] Figure 7 A result example diagram of a reconstructed docking ring point cloud extracted based on a Euclidean distance nearest neighbor search in an embodiment of the present application is shown in FIG. 5. DETAILED DESCRIPTION

[0071] In order to make the person skilled in the art better understand the present application scheme, in the following will be combined with the drawings of the exemplary embodiments or embodiments of the present application will be described. Obviously, the described embodiments or embodiments are only a part of the embodiments or embodiments of the present application, rather than all. Based on the embodiments or embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the scope of protection of the present application.

[0072] The embodiment of the present application provides a space invalid satellite docking ring identification method based on three-dimensional point cloud matching, as shown in FIG. 6, the method comprises the following steps: Figure 1

[0073] Step one, obtaining a three-dimensional reconstructed point cloud containing a satellite;

[0074] Step two, using a sample consistency initial registration algorithm (Sample Consensus Initial Alignment, SAC-IA) to coarsely match the three-dimensional reconstructed point cloud with a docking ring standard point cloud, and obtaining a region point cloud containing a docking ring component; wherein the docking ring standard point cloud is obtained by using a laser radar to scan a docking ring component of a specific size;

[0075] Step three, using an iterative closest point algorithm (Iterative Closest Point, ICP) to finely match the region point cloud containing the docking ring component with the docking ring standard point cloud, and using a nearest neighbor search to extract a three-dimensional point coordinate of the docking ring component;

[0076] ​Step four, according to the three-dimensional point coordinates of the docking ring component, the principal component analysis method is used to project and calculate the center coordinates and diameter of the docking ring component.

[0077] In step one, for the point cloud data containing satellites obtained by laser radar scanning, a dense three-dimensional reconstruction point cloud is obtained by using an existing three-dimensional reconstruction algorithm.

[0078] In step two, the SAC-IA algorithm is used to coarsely match the three-dimensional reconstruction point cloud with the docking ring standard point cloud, and the region point cloud containing the docking ring component is obtained; according to the embodiment of the application, the three-dimensional reconstruction point cloud is segmented into multiple regions based on the BRICH clustering algorithm (Balanced Iterative Reducing and Clustering Using Hierarchies), key points are extracted from the reconstruction point cloud and the docking ring standard point cloud respectively, and the points are described using the Fast Point Feature Histogram (FPFH) feature, the SAC-IA algorithm is used to coarsely register the two groups of point clouds, the region point cloud on the docking ring standard point cloud registration is retained, and the point cloud component recognition range is reduced. The specific steps include:

[0079] Step two one, the reconstruction point cloud P and the docking ring template point cloud Q are uniformly down-sampled to obtain P' and Q', the number of three-dimensional points is reduced, and the operation speed is improved;

[0080] Step two two, the normal vector n P' , n Q' of each point in the reconstruction point cloud P' and the docking ring template point cloud Q' is calculated respectively;

[0081] Step two three, the FPFH feature of each point in the docking ring template point cloud Q' is calculated, and a spatial index structure based on KDTree is constructed thereon;

[0082] Step two four, n sampling points are selected from the reconstruction point cloud P' to obtain P''={p" i ,||p" i -p" j ||<d,i≠j,i<n}, the FPFH feature of the sampling points is calculated; according to the feature similarity principle, the K nearest neighbor search is used in the point set Q' to find the points closest to the feature distance of p" i , select the points with the smallest difference from the normal vector of p" i as the corresponding points, and obtain the corresponding point set {P'', Q''};

[0083] Step 25: Calculate the rigid body transformation matrix between corresponding points. Then, determine the performance of the current registration transformation by solving the distance error sum function after the transformation of the corresponding points. The distance error sum function is often represented by the Huber penalty function, denoted as... in:

[0084]

[0085] Where: m l For a pre-defined value, such as 0.1; i Let be the distance difference between the corresponding points in the i-th group after transformation;

[0086] Step 26: Repeat steps 24 to 25 until a set of pose transformations is found that meets the error threshold requirement or reaches the required number of iterations. The transformation at this point is the final registration transformation matrix {R0, T0}.

[0087] Step 27: Transform the point cloud P through {R0, T0} to obtain point cloud P0 = R0P + T0;

[0088] Step 28: Use the BIRCH clustering algorithm to segment the point cloud, reconstructing the point cloud P0 into multiple parts {P 0i};

[0089] Step 29: Calculate Q' and the set of points P for each part. 0i The three-dimensional point coordinate overlap rate is used to select the point set with the largest overlap rate, that is, to retain the point cloud of the docking ring component.

[0090] In step three, the ICP algorithm is used to perform fine matching between the regional point cloud containing the docking ring component and the standard point cloud of the docking ring, and the three-dimensional point coordinates of the docking ring component are extracted using nearest neighbor search. According to an embodiment of the present invention, using the rigid body transformation matrix obtained from coarse registration as the initial value, the ICP algorithm is used to perform fine registration between the regional point cloud containing the docking ring component and the standard docking ring point cloud. The projection transformation relationship between the two point sets is iteratively solved to accurately identify the docking ring in the reconstructed point cloud. Nearest neighbor search based on Euclidean distance is used to find the nearest reconstructed point corresponding to the standard point cloud, and the three-dimensional point coordinates of the docking ring component are extracted. Specific steps include:

[0091] Step 31: Using the region point cloud P containing the docking ring component... part The pose transformation matrix {R0, T0} is used as the initial value;

[0092] Step 3.2: Calculate the point cloud P part For each point in the vector, find the nearest point in the point cloud Q to obtain the corresponding set of points.

[0093] Step 3: Obtain the average distance between the corresponding points mentioned above through SVD singular value decomposition. The minimum rigid body transformation {R', T'}, wherein n is the number of points, R' is a rotation matrix, and T' is a translation vector, is further calculated to obtain a new transformed point set P' part = R'P part + T'.

[0094] Step three four, the transformed point set P' part is calculated, and the average distance between the transformed point set P' and the template point cloud Q is calculated When the average distance is less than a given threshold value, the calculation is stopped, otherwise, the current {R', T'} is taken as an initial value, and the above steps are repeated to calculate a new pose transformation matrix until When the threshold value is met or the number of iterations reaches a given value, the iterative calculation is stopped, and at this time, a final pose transformation matrix {R, T} and a transformed point cloud P are obtained

[0095] Step three five, based on a KDTree spatial index structure in the docking ring template point cloud Q, the nearest neighbor search is used in the point cloud to find the points closest to the points in Q, and the coordinates P of these points are extracted, that is, the docking ring component of the point cloud is accurately identified. roll

[0096] Through the above steps, the docking ring component can be accurately identified from the reconstructed point cloud, and the spatial position coordinates of the docking ring component point cloud are obtained.

[0097] In step four, according to the three-dimensional point coordinates of the docking ring component, the principal component analysis method is used to project and calculate the center coordinates and the diameter of the docking ring component; according to the embodiment of the application, the main direction of the identified reconstructed docking ring point cloud is obtained by using the principal component analysis method, and the docking ring diameter size is taken as the average of the two-dimensional envelope length and width, the two-dimensional position coordinates of the center of the envelope are taken as the envelope center point, and the other dimension position coordinates are taken as the maximum and minimum values of the third dimension coordinates of the point cloud. The maximum and minimum values are selected according to the rules of the body coordinate system. The specific steps include:

[0098] Step four one, the mean value of each dimension of the point cloud coordinates P roll (X roll , Y roll , Z roll ) is calculated, and the zero mean value of the three-dimensional point coordinates is obtained

[0099] Step four two, the covariance matrix of the point cloud coordinate matrix P' roll is solved

[0100] Step four three, the eigenvalue decomposition of the covariance is performed, V -1 ​CV=D, where D is a diagonal matrix of eigenvalues of the covariance matrix 3, V is the eigenvector matrix also called principal component coefficient matrix; take the first two rows of the principal component coefficient matrix to form a matrix E, and utilize P roll = E*P roll Project the point cloud to the XY plane;

[0101] Step four, calculate the diameter of the docking ring Let the spatial position coordinates of the center of the circle be (X0, Y0, Z0), wherein Z0 takes the maximum or minimum value of the third dimension data, and the selection of the maximum or minimum value is determined according to the rules of the body coordinate system, such as Figure 2 (a) If the body coordinate system is established, Z0 takes the minimum value, and if Figure 2 (b) If the body coordinate system is established, the center coordinate Z0 of the docking ring takes the maximum value.

[0102] Through the above steps, the center of the circle three-dimensional coordinates and the diameter of the docking ring can be calculated from the reconstructed docking ring point cloud, thereby providing information for the subsequent approaching, capturing and other tasks of the target spacecraft.

[0103] The technical effects of the present application are further verified through docking ring identification simulation experiments.

[0104] Three-dimensional reconstruction is performed on the target, and the experimental object is a simulated satellite model. The relative distance between the camera and the center of the satellite is 10 m. When shooting, the camera rotates 360 degrees around the center while keeping the distance from the center unchanged, and the optical axis of the camera always points to the center of the satellite. An image is collected every 2 degrees of rotation. A number of sequence pictures are selected for the front of the satellite. The camera parameters are set as follows: focal length 45 mm, field of view angle 14 degrees, and image resolution 5120*5120. Dense point cloud is obtained by using a three-dimensional reconstruction algorithm. According to Figure 2 (b) The body coordinate system is established, the point cloud is converted to the body system, and the scale is recovered by using the ranging information. The visualization result is shown in Figure 3 .

[0105] The BIRCH clustering algorithm is used for segmentation processing on the reconstructed point cloud, and the point cloud is divided into three parts, as shown in Figure 4 . Figure 5 For the docking ring standard point cloud (white) and the complete reconstructed point cloud SAC-IA coarse matching result, the area with the maximum overlap rate with the standard point cloud is reconstructed, and the standard docking ring point cloud is precisely matched. The visualization result is shown in Figure 6 . Figure 7 The reconstructed docking ring point cloud is extracted based on the nearest neighbor search of the Euclidean distance. The comparison results of the center of the docking ring and the radius solved and the standard values are shown in Table 1, and the error is controlled within 10%.

[0106] Table 1

[0107] Parameter Identification value True value Error Diameter of the adapter ring / m 1.16 1.2 3.125％ Coordinate of the center (0.66，0.00422，0.02849) (0.6，0，0) 6.65％

[0108] Another embodiment of the present application provides a space invalid satellite docking ring identification system based on three-dimensional point cloud matching, which comprises:

[0109] a reconstructed point cloud acquisition module configured to acquire a three-dimensional reconstructed point cloud containing a satellite;

[0110] a coarse matching module configured to coarsely match the three-dimensional reconstructed point cloud with a docking ring standard point cloud by using a sample consistency initial registration algorithm, and acquire a regional point cloud containing a docking ring component;

[0111] a fine matching module configured to finely match the regional point cloud containing the docking ring component with the docking ring standard point cloud by using an iterative closest point algorithm, and extract three-dimensional point coordinates of the docking ring component by using nearest neighbor search;

[0112] a docking ring identification module configured to calculate a center coordinate and a diameter of the docking ring component by using principal component analysis projection according to the three-dimensional point coordinates of the docking ring component.

[0113] In the embodiment, preferably, the specific process of acquiring the regional point cloud containing the docking ring component in the coarse matching module comprises:

[0114] Step two one, uniformly down-sample the three-dimensional reconstructed point cloud P and the docking ring standard point cloud Q respectively to obtain P' and Q';

[0115] Step two two, calculate the normal vector n P' , n Q' of each point in P' and Q' respectively;

[0116] Step two three, calculate the point fast feature histogram feature of each point in Q', and construct a KDTree-based spatial index structure for the point fast feature histogram feature;

[0117] Step two four, select n sample points p” i from P', calculate the point fast feature histogram feature of the sample points, and according to the feature similarity principle, use K nearest neighbor search in Q' to find the points closest to p” i in feature distance, select the points with the smallest difference from p” i in normal vector as the corresponding points to obtain a corresponding point set {P”, Q”};

[0118] Step two five, calculate the rigid transformation matrix between the corresponding points, and judge the performance of the current registration transformation by solving the distance error sum of squares function after the transformation of the corresponding points;

[0119] Step two six, iteratively repeat step two four to step two five until the distance error sum function reaches a preset threshold or reaches a maximum number of iterations, at which time the corresponding rigid transformation matrix is the final registration transformation matrix {R0, T0};

[0120] Step two seven, pose-convert the three-dimensional reconstructed point cloud P through the registration transformation matrix {R0, T0} to obtain a point cloud P0 = R0P + T0;

[0121] Step two eight, perform segmentation processing on the point cloud P0 using a BIRCH clustering algorithm to obtain a plurality of partial point sets {P 0i};

[0122] Step two nine, calculate the three-dimensional point coordinate overlap rate of Q' and each partial point set P 0i , and select the point set with the largest overlap rate, which is the region point cloud containing the butt joint ring component.

[0123] In this embodiment, preferably, the distance error sum function in the coarse matching module is expressed using a Huber penalty function, denoted as wherein:

[0124]

[0125] In the formula, m l is a pre-given value, l i is the distance difference after transformation of the i-th group of corresponding points.

[0126] In this embodiment, preferably, the fine matching module uses an iterative closest point algorithm to perform fine matching on the region point cloud containing the butt joint ring component and the butt joint ring standard point cloud, and uses nearest neighbor search to extract the three-dimensional point coordinates of the butt joint ring component. The specific steps include:

[0127] Step three one, taking the region point cloud P part containing the butt joint ring component and the registration transformation matrix {R0, T0} as initial values, calculate the closest point of each point in the point cloud P part in the butt joint ring standard point cloud Q to obtain a corresponding point set (P part , Q j );

[0128] Step three two, obtain a rigid transformation {R', T'} that minimizes the average distance part in the corresponding point set (P j , Q part ) through singular value decomposition; wherein n is the number of points, R' is a rotation matrix, and T' is a translation vector. Then, a new transformed point set P' part = R'P part + T';

[0129] Step three, calculate the transformed point set P' part The average distance between the docking ring standard point cloud Q and the transformed point cloud P When the average distance is less than a given threshold, stop the calculation, otherwise, take the current {R', T'} as the initial value, repeat the above steps to calculate a new registration transformation matrix until The threshold requirement is met or the number of iterations reaches a given value, stop the iterative calculation; at this time, the final registration transformation matrix {R, T} and the transformed point cloud P are obtained

[0130] Step three four, in the docking ring standard point cloud Q, based on the KDTree spatial index structure constructed in step two three, find the nearest point in the transformed point cloud P Using nearest neighbor search, extract the coordinates P of the nearest point roll , that is, accurately identify the point cloud of the docking ring component.

[0131] In this embodiment, preferably, the specific steps of the docking ring identification module for calculating the center coordinates and diameter of the docking ring component according to the three-dimensional point coordinates of the docking ring component include:

[0132] Step four one, calculate the mean value of each dimension of the point cloud coordinates P roll (X roll , Y roll , Z roll ), zero-mean the three-dimensional point coordinates to obtain the point cloud coordinate matrix P' roll ;

[0133] Step four two, solve the covariance matrix of P' roll

[0134] Step four three, perform eigenvalue decomposition V -1 CV = D on the covariance matrix, where D is a diagonal matrix composed of eigenvalues of the covariance matrix, and V is an eigenvector matrix, also known as a principal component coefficient matrix; take the first two rows of the principal component coefficient matrix V to form a matrix E, and project the point cloud to the XY plane using P" roll = E * P roll ;

[0135] Step four four, calculate the diameter of the docking ring where X max , X min , Y max , Y min represent the maximum value of the coordinates, let the spatial position coordinates of the center be (X0, Y0, Z0), then Z0 takes the maximum value of the third dimension data.

[0136] ​The function of the space failure satellite docking ring identification system based on three-dimensional point cloud matching according to the embodiment of the application can be described by the space failure satellite docking ring identification method based on three-dimensional point cloud matching described above, and therefore, the unexplained part of the system embodiment can be referred to the method embodiment described above, and will not be described here again.

[0137] Although the application has been described in terms of limited embodiments, those skilled in the art will appreciate that other embodiments are possible, within the scope of the application described herein, in light of the above teachings. The disclosure of the application is illustrative only and not restrictive of the application, the scope of which is defined by the appended claims.

Claims

1. A method for identifying space-failed satellite docking loops based on three-dimensional point cloud matching, characterized in that, Includes the following steps: Step 1: Obtain a 3D reconstructed point cloud containing satellite data; Step 2: Use the sample consistency initial registration algorithm to coarsely match the 3D reconstructed point cloud with the standard point cloud of the docking ring to obtain the regional point cloud containing the docking ring component; Step 3: Use the iterative nearest neighbor algorithm to perform a fine match between the point cloud of the region containing the docking ring component and the standard point cloud of the docking ring, and use nearest neighbor search to extract the 3D point coordinates of the docking ring component, including: Step 31: Using the point cloud P of the region containing the docking ring component... part Using the registration transformation matrix {R0, T0} as initial values, calculate the point cloud P. part For each point in the loop, the nearest point in the standard point cloud Q of the docking ring is obtained, resulting in the corresponding point set (P). part Q j ); Step 3.2: Obtain the set of corresponding points (P) through singular value decomposition. part Q j Average distance in The minimum rigid body transformation {R', T'}, where n is the number of points, R' is the rotation matrix, and T' is the translation vector, then the new transformed point set P' part =R'P part +T'; Step 3: Calculate the transformed point set P' part Average distance between the docking ring and the standard point cloud Q The calculation stops when the average distance is less than a given threshold; otherwise, the above steps are repeated to calculate a new registration transformation matrix, using the current {R', T'} as the initial value, until... The iterative calculation stops when the threshold requirement is met or the number of iterations reaches a given value; at this point, the final registration transformation matrix {R, T} and the transformed point cloud are obtained. Steps 3 and 4: Based on the spatial index structure of the KDTree constructed in Step 2, transform the point cloud in the standard point cloud Q of the docking loop. The nearest neighbor search is used to find the point closest to a point in Q, and the coordinates P of the nearest point are extracted. roll That is, to accurately identify the point cloud of the docking ring component; Step 4: Based on the three-dimensional point coordinates of the docking ring component, use principal component analysis to project and calculate the center coordinates and diameter of the docking ring component.

2. The method for identifying space-failed satellite docking loops based on three-dimensional point cloud matching according to claim 1, characterized in that, Step two includes the following specific steps: Step 2: Perform uniform downsampling on the 3D reconstructed point cloud P and the standard point cloud Q of the docking ring to obtain P' and Q' respectively; Step 2: Calculate the normal vector n for each point in P' and Q' respectively. P' n Q' ; Steps 2 and 3: Calculate the point fast feature histogram features for each point in Q', and construct a spatial index structure based on KDTree for the point fast feature histogram features; Step 24: Select n sampling points p″ from P' i Calculate the point fast feature histogram features of the sampling points; based on the feature similarity principle, use K-nearest neighbor search and p” in Q’. i Select the point closest to the feature distance, which is closest to p″. i The point whose normal vectors differ the least is taken as the corresponding point, and the corresponding point set {P”, Q”} is obtained; Step 25: Calculate the rigid body transformation matrix between corresponding points, and determine the performance of the current registration transformation by solving the distance error and function of the corresponding points after transformation; Step 26: Iterate and repeat steps 24 to 25 until the distance error and function reach a preset threshold or the maximum number of iterations. At this time, the corresponding rigid body transformation matrix is ​​the final registration transformation matrix {R0, T0}. Step 27: Transform the pose of the 3D reconstructed point cloud P using the registration transformation matrix {R0, T0} to obtain point cloud P0 = R0P + T0; Step 28: Use the BIRCH clustering algorithm to segment the point cloud P0 to obtain multiple partial point sets {P 0i }; Step 29: Calculate Q' and the set of points P for each part. 0i The overlap rate of the three-dimensional point coordinates is used to select the point set with the largest overlap rate, which is the point cloud of the region containing the docking ring component.

3. The method for identifying space-failed satellite docking loops based on three-dimensional point cloud matching according to claim 2, characterized in that, The distance error and function described in step two five are represented using the Huber penalty function, denoted as . in: In the formula, m l For a pre-given value, l i Let be the distance difference between the corresponding points in the i-th group after transformation.

4. The method for identifying space-failed satellite docking loops based on three-dimensional point cloud matching according to claim 3, characterized in that, Step four includes the following specific steps: Step 41: Calculate the point cloud coordinates P roll (X roll ,Y roll Z roll The mean of each dimension is used to zero-mean the 3D point coordinates, resulting in the point cloud coordinate matrix P'. roll ; Step 42: Solve for P' roll covariance matrix Step 4.3: Perform eigenvalue decomposition V on the covariance matrix. -1 CV = D, where D is a diagonal matrix composed of the eigenvalues ​​of the covariance matrix, and V is the eigenvector matrix, also known as the principal component coefficient matrix; the first two rows of the principal component coefficient matrix V are taken to form matrix E, and P” is used. roll =E*P roll Project the point cloud onto the XY plane; Step 4: Calculate the diameter of the mating ring. Where X max X min Y max Y min Let X represent the extreme values ​​of the coordinates. Let the spatial coordinates of the center of the circle be (X0, Y0, Z0). Z0 takes the maximum or minimum value of the third dimension of data.

5. A space-failed satellite docking loop identification system based on three-dimensional point cloud matching, characterized in that, include: The reconstructed point cloud acquisition module is configured to acquire a 3D reconstructed point cloud containing satellites; The coarse matching module is configured to use the sample consistency initial registration algorithm to coarsely match the three-dimensional reconstructed point cloud with the standard point cloud of the docking ring to obtain the regional point cloud containing the docking ring component. The fine matching module is configured to perform fine matching between the point cloud of the region containing the docking ring component and the standard point cloud of the docking ring using an iterative nearest neighbor algorithm, and to extract the 3D point coordinates of the docking ring component using nearest neighbor search, including: Step 31: Using the point cloud P of the region containing the docking ring component... part Using the registration transformation matrix {R0, T0} as initial values, calculate the point cloud P. part For each point in the loop, the nearest point in the standard point cloud Q of the docking ring is obtained, resulting in the corresponding point set (P). part Q j ); Step 3.2: Obtain the set of corresponding points (P) through singular value decomposition. part Q j Average distance in The minimum rigid body transformation {R', T'}, where n is the number of points, R' is the rotation matrix, and T' is the translation vector, then the new transformed point set P' part =R'P part +T'; Step 3: Calculate the transformed point set P' part Average distance between the docking ring and the standard point cloud Q The calculation stops when the average distance is less than a given threshold; otherwise, the above steps are repeated to calculate a new registration transformation matrix, using the current {R', T'} as the initial value, until... The iterative calculation stops when the threshold requirement is met or the number of iterations reaches a given value; at this point, the final registration transformation matrix {R, T} and the transformed point cloud are obtained. Steps three and four: Based on the spatial index structure of KDTree constructed in the coarse matching module, transform the point cloud in the standard point cloud Q of the docking loop. The nearest neighbor search is used to find the point closest to a point in Q, and the coordinates P of the nearest point are extracted. roll That is, to accurately identify the point cloud of the docking ring component; The docking ring identification module is configured to calculate the center coordinates and diameter of the docking ring component by projecting the three-dimensional point coordinates of the docking ring component using principal component analysis.

6. A space-failed satellite docking ring identification system based on three-dimensional point cloud matching according to claim 5, characterized in that, The specific process of obtaining the region point cloud containing the docking ring component in the coarse matching module includes: Step 2: Perform uniform downsampling on the 3D reconstructed point cloud P and the standard point cloud Q of the docking ring to obtain P' and Q' respectively; Step 2: Calculate the normal vector n for each point in P' and Q' respectively. P' n Q' ; Steps 2 and 3: Calculate the point fast feature histogram features for each point in Q', and construct a spatial index structure based on KDTree for the point fast feature histogram features; Step 24: Select n sampling points p″ from P' i Calculate the point fast feature histogram features of the sampling points; based on the feature similarity principle, use K-nearest neighbor search and p” in Q’. i Select the point closest to the feature distance, which is closest to p″. i The point whose normal vectors differ the least is taken as the corresponding point, and the corresponding point set {P”, Q”} is obtained; Step 25: Calculate the rigid body transformation matrix between corresponding points, and determine the performance of the current registration transformation by solving the distance error and function of the corresponding points after transformation; Step 26: Iterate and repeat steps 24 to 25 until the distance error and function reach a preset threshold or the maximum number of iterations. At this time, the corresponding rigid body transformation matrix is ​​the final registration transformation matrix {R0, T0}. Step 27: Transform the pose of the 3D reconstructed point cloud P using the registration transformation matrix {R0, T0} to obtain point cloud P0 = R0P + T0; Step 28: Use the BIRCH clustering algorithm to segment the point cloud P0 to obtain multiple partial point sets {P 0i }; Step 29: Calculate Q' and the set of points P for each part. 0i The overlap rate of the three-dimensional point coordinates is used to select the point set with the largest overlap rate, which is the point cloud of the region containing the docking ring component.

7. A space-failed satellite docking ring identification system based on three-dimensional point cloud matching according to claim 6, characterized in that, The distance error sum function in the coarse matching module is represented using the Huber penalty function, denoted as... in: In the formula, m l For a pre-given value, l i Let be the distance difference between the corresponding points in the i-th group after transformation.

8. A space-failed satellite docking ring identification system based on three-dimensional point cloud matching according to claim 7, characterized in that, The specific steps in the docking ring identification module to calculate the center coordinates and diameter of the docking ring component using principal component analysis based on the three-dimensional point coordinates of the docking ring component include: Step 41: Calculate the point cloud coordinates P roll (X roll ,Y roll Z roll The mean of each dimension is used to zero-mean the 3D point coordinates, resulting in the point cloud coordinate matrix P'. roll ; Step 42: Solve for P' roll covariance matrix Step 4.3: Perform eigenvalue decomposition V on the covariance matrix. -1 CV = D, where D is a diagonal matrix composed of the eigenvalues ​​of the covariance matrix, and V is the eigenvector matrix, also known as the principal component coefficient matrix; the first two rows of the principal component coefficient matrix V are taken to form matrix E, and P” is used. roll =E*P roll Project the point cloud onto the XY plane; Step 4: Calculate the diameter of the mating ring. Where X max X min Y max Y min Let X represent the extreme values ​​of the coordinates. Let the spatial coordinates of the center of the circle be (X0, Y0, Z0). Z0 takes the maximum or minimum value of the third dimension of data.

Images