Point Cloud Registration Result Evaluation Method and System Based on Distribution Characteristics

Abstract

The present invention provides a point cloud registration result evaluation method and system based on distribution characteristics, including: obtaining a source point cloud and a target point cloud, registering the source point cloud to obtain a registration point cloud; placing the registration point cloud and the target point cloud in the same three-dimensional space, dividing the current three-dimensional space into multiple grids of the same size, the registration point cloud and the target point cloud in the grid are the registration point cloud in the grid and the target point cloud in the grid respectively; for each associated grid, performing feature decomposition on the target point cloud in the grid to obtain a eigenvector and eigenvalue, then determining the distribution characteristics of the target point cloud in the grid according to the size of the eigenvalue and the covariance matrix determinant, and classifying the associated grids according to the distribution characteristics; determining the distribution characteristics of the target point cloud, and classifying the associated grids according to the distribution characteristics; calculating the point cloud registration success rate according to the category; multiplying the registration success rate of each category of point cloud by a weight and summing up the total registration success rate; and finally evaluating the point cloud registration result.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for evaluating point cloud registration results based on distribution characteristics. Background Art

[0002] High-precision mapping and localization technology is a key area of expertise in autonomous driving. Point cloud matching is widely used as a key component in mapping and localization algorithms for accurate pose estimation. The accuracy of point cloud matching directly impacts the accuracy of mapping and localization. Therefore, after using point cloud registration algorithms, it is often necessary to evaluate the registration results and verify their effectiveness.

[0003] Currently, the most commonly used method for evaluating point cloud registration accuracy is to use the average distance between pairs of neighboring points. After registration is complete, the nearest neighbor of each point in the registered point cloud is calculated in the target point cloud. The matching accuracy of the point cloud is then evaluated based on the average matching distance of these point pairs.

[0004] Simply calculating the average matching distance of the point cloud is too brute force. Not only does it have the defect that the score is not stable enough in sparse point clouds, but it is also difficult to detect the registration deviation of a small part of the feature point cloud in a degraded environment.

[0005] However, a more accurate evaluation algorithm means a greater amount of computation. How to complete a more robust point cloud registration accuracy evaluation based on real-time operation has become an urgent problem to be solved.

[0006] Patent document CN113436238A (application number: 202110991625.9) discloses a method, device and electronic device for evaluating the accuracy of point cloud registration. The method includes: after obtaining the source point cloud and the target point cloud, using the initial rigid body transformation matrix to transform the source point cloud to obtain a first point cloud, and aligning the first point cloud and the target point cloud to obtain multiple point pairs; performing plane fitting on the target point cloud to obtain a ground equation, and determining a global motion vector based on the multiple point pairs, the ground equation and the first distance threshold between two points in the point pair; and determining an evaluation index for the registration accuracy of the source point cloud and the target point cloud based on the global motion vector. Summary of the Invention

[0007] In view of the defects in the prior art, the purpose of the present invention is to provide a point cloud registration result evaluation method and system based on distribution characteristics.

[0008] According to the present invention, a point cloud registration result evaluation method based on distribution characteristics is provided, comprising:

[0009] Step S1: obtaining a source point cloud and a target point cloud, and registering the source point cloud to obtain a registered point cloud;

[0010] Step S2: placing the registration point cloud and the target point cloud in the same three-dimensional space, and dividing the three-dimensional space into a plurality of grids of the same size, wherein the registration point cloud and the target point cloud falling within each grid are the in-grid registration point cloud and in-grid target point cloud corresponding to each grid respectively;

[0011] Step S3: All grids are screened based on the condition that the number of points included in the in-grid registration point cloud and the in-grid target point cloud is greater than a first quantity threshold, and N associated grids are obtained after screening;

[0012] Step S4: For each associated grid, perform eigendecomposition on the covariance matrix of the target point cloud within the grid to obtain eigenvectors and eigenvalues, then determine the distribution characteristics of the target point cloud within the grid based on the size of the eigenvalues and the covariance matrix determinant, and classify each associated grid based on the distribution characteristics of the target point cloud within the grid;

[0013] Step S5: Calculate the point cloud registration success rate of each type of associated grid according to the type of associated grid;

[0014] Step S6: multiplying the point cloud registration success rate of each type of associated grid by the corresponding weight parameter and summing them up to obtain the overall point cloud registration success rate;

[0015] Step S7: Evaluate the point cloud registration result according to the overall point cloud registration success rate.

[0016] Preferably, step S4 includes:

[0017] Step S4.1: Number the associated grids from 1 to N, and initialize i to 1;

[0018] Step S4.2: For the i-th associated grid, the covariance matrix C of the target point cloud in the grid t Perform eigendecomposition to obtain eigenvectors and three eigenvalues, and calculate C t The determinant det C t ;

[0019] Step S4.3: Sort the three eigenvalues from largest to smallest to obtain a first eigenvalue λ1, a second eigenvalue λ2, and a third eigenvalue λ3, respectively; divide λ1 by λ2 to obtain a first ratio ζ1; and divide λ2 by λ3 to obtain a second ratio ζ2;

[0020] Step S4.4: According to ζ1, ζ2 and det C t , determine the distribution characteristics of the target point cloud within the grid, if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Less than the first determinant threshold D th, then the distribution characteristics of the target point cloud in the grid belong to concentrated distribution; if ζ1 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to linear distribution; if ζ1 is less than the first ratio threshold ζ th And ζ2 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to the surface distribution; if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Greater than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to discrete distribution;

[0021] Step S4.5: Classify the i-th associated grid according to the distribution characteristics of the point cloud within the grid. If the distribution characteristics of the target point cloud within the grid belong to concentrated distribution, then the i-th associated grid belongs to concentrated distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to linear distribution, then the i-th associated grid belongs to linear distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to surface distribution, then the i-th associated grid belongs to surface distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to discrete distribution, then the i-th associated grid belongs to discrete distribution type associated grid;

[0022] Step S4.6: Add 1 to the value of i, and then determine whether the new value of i is less than or equal to N. If so, jump back to step S4.2 and re-execute the relevant steps in sequence; if not, terminate the execution of all steps.

[0023] Preferably, step S5 includes:

[0024] The calculation method of the point cloud registration success rate of the concentrated distribution class associated grid is as follows: first determine whether each concentrated distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t1 The coordinates of (x gt1 ,y gt1 ,z gt1 ), the center of gravity G of the grid registration point cloud r1 The coordinates of (x gr1 ,y gr1 ,z gr1 ), calculate the spatial distance between the centers of gravity of two point clouds If d g1 Less than the first distance threshold d th1 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of concentrated distribution class associated grids is N1, and the number of concentrated distribution class associated grids that are successfully registered is K1, then the point cloud registration success rate of concentrated distribution class associated grids is η1=K1 / N1;

[0025] The calculation method of the point cloud registration success rate of linear distribution associated grids is as follows: first, all N2 linear distribution associated grids are screened. Specifically, for each linear distribution associated grid, if the fitting line L of the target point cloud in the grid is t2 The angle α between the vertical axis and the vertical axis is less than the first angle threshold α th Then retain, otherwise remove, and get N after screening 2s candidate linear distribution class associated grids; then determine whether each candidate linear distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t2 The coordinates of (x gt2 ,y gt2 ,z gt2 ), the center of gravity G of the grid registration point cloud r2 The coordinates of (x gr2 ,y gr2 ,z gr2 ), calculate the spatial distance between the centers of gravity of two point clouds In calculating the center of gravity G r2 To line L t2 The vertical distance d2=d g2 cosβ, where β is the line connecting the centroids of the two point clouds and the straight line L t2 If d2 is less than the second distance threshold d th2 , then the registration is judged to be successful, otherwise it is judged to be a registration failure; if the number of candidate linear distribution class associated grids that are successfully registered is K2, then the point cloud registration success rate of linear distribution class associated grids is η2=K2 / N 2s ;

[0026] The calculation method of the point cloud registration success rate of the surface distribution class associated grid is as follows: first determine whether each surface distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t3 The coordinates of (x gt3 ,y gt3 ,z gt3 ), the center of gravity G of the grid registration point cloud r3 The coordinates of (x gr3 ,y gr3 ,z gr3 ), calculate the spatial distance between the centers of gravity of two point clouds Then calculate the center of gravity G r3 Fitting plane S to the target point cloud within the grid t3 The vertical distance d3=d g3 cosγ, where γ is the line connecting the centroids of the two point clouds and the plane S t3 If d3 is less than the third distance threshold d th3, then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of surface distribution class associated grids is N3, and the number of successfully registered surface distribution class associated grids is K3, then the point cloud registration success rate of surface distribution class associated grids is η3=K3 / N3;

[0027] The calculation method of the point cloud registration success rate of discrete distribution class associated grids is as follows: first determine whether each discrete distribution class associated grid is successfully registered. Specifically, for each discrete distribution class associated grid, if the covariance matrix of the target point cloud in the grid is C t4 , whose determinant is det C t4 , the covariance matrix of the grid-registered point cloud is C r4 , whose determinant is det C r4 , then the entropy of the target point cloud within the grid and the registration point cloud within the grid are E t4 =1+ln(det C t4 ) and E r4 =1+ln(det C r4 ), the point cloud entropy difference between the two is ΔE4=E t4 -E r4 , if the absolute value of ΔE4 is less than the first entropy difference threshold ΔE th , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of discrete distribution class associated grids is N4, and the number of successfully registered discrete distribution class associated grids is K4, then the point cloud registration success rate of discrete distribution class associated grids is η4=K4 / N4;

[0028] The sum of the associated grids N1 of the concentrated distribution type, N2 of the linear distribution type, N3 of the surface distribution type, and N4 of the discrete distribution type is N, that is:

[0029] N1+N2+N3+N4=N

[0030] Preferably, the calculation formula for the overall point cloud registration success rate is:

[0031] η whole =a1η1+a2η2+a3η3+a4η4

[0032] Among them, a1, a2, a3 and a4 are the weight parameters corresponding to the associated grids of concentrated distribution, linear distribution, surface distribution and discrete distribution respectively.

[0033] Preferably, the target point cloud and the registration point cloud are stored in a GPU, and the point cloud registration result evaluation method based on distribution characteristics is executed in the GPU.

[0034] According to the present invention, a point cloud registration result evaluation system based on distribution characteristics is provided, comprising:

[0035] Data acquisition module, used to obtain source point cloud and target point cloud;

[0036] A point cloud registration module, configured to register the source point cloud to obtain a registered point cloud;

[0037] A GPU calculation module is used to calculate the overall point cloud registration success rate based on the target point cloud and the registration point cloud using the parallel acceleration function of the GPU;

[0038] The result evaluation module is used to evaluate whether the point cloud registration result is accurate according to the overall point cloud registration success rate.

[0039] Preferably, the GPU computing module includes: a storage unit, a grid processing unit, a screening unit, a classification unit, a partial success rate calculation unit and an overall success rate calculation unit;

[0040] The storage unit is used to store the target point cloud and the registration point cloud;

[0041] The gridding processing unit is used to place the registration point cloud and the target point cloud in the same three-dimensional space, and divide the three-dimensional space into a plurality of grids of the same size, wherein the registration point cloud and the target point cloud falling within each grid are respectively the in-grid registration point cloud and the in-grid target point cloud corresponding to each grid;

[0042] The screening unit is used to screen all grids, and screen all grids based on the condition that the number of points included in the registration point cloud and the target point cloud within the grid is greater than a first number threshold, and obtain N associated grids after screening;

[0043] The classification unit is used to classify each associated grid, and for each associated grid, perform eigendecomposition on the covariance matrix of the target point cloud in the grid to obtain eigenvectors and eigenvalues, and then determine the distribution characteristics of the target point cloud in the grid according to the size of the eigenvalues and the covariance matrix determinant, and classify each associated grid according to the distribution characteristics of the target point cloud in the grid;

[0044] The partial success rate calculation unit is used to calculate the point cloud registration success rate of each type of associated grid according to the type of associated grid;

[0045] The overall success rate calculation unit is used to multiply the point cloud registration success rate of each type of associated grid by the corresponding weight parameter and then add them up to obtain the overall point cloud registration success rate.

[0046] Preferably, the classification unit includes:

[0047] Component 1: Number the associated grids from 1 to N, and initialize i to 1;

[0048] Component 2: For the i-th associated grid, the covariance matrix C of the target point cloud within the grid t Perform eigendecomposition to obtain eigenvectors and three eigenvalues, and calculate C t The determinant det C t ;

[0049] Component 3: Sort the three eigenvalues from largest to smallest to obtain a first eigenvalue λ1, a second eigenvalue λ2, and a third eigenvalue λ3, divide λ1 by λ2 to obtain a first ratio ζ1, and divide λ2 by λ3 to obtain a second ratio ζ2;

[0050] Component 4: Based on ζ1, ζ2 and det C t , determine the distribution characteristics of the target point cloud within the grid, if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Less than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to concentrated distribution; if ζ1 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to linear distribution; if ζ1 is less than the first ratio threshold ζ th And ζ2 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to the surface distribution; if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Greater than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to discrete distribution;

[0051] Component 5: Classify the i-th associated grid according to the distribution characteristics of the point cloud within the grid. If the distribution characteristics of the target point cloud within the grid belong to concentrated distribution, then the i-th associated grid belongs to concentrated distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to linear distribution, then the i-th associated grid belongs to linear distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to surface distribution, then the i-th associated grid belongs to surface distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to discrete distribution, then the i-th associated grid belongs to discrete distribution type associated grid;

[0052] Component 6: Increment the value of i by 1, and then determine whether the new value of i is less than or equal to N. If so, re-execute the relevant components starting from component 2; otherwise, terminate the execution of all components.

[0053] Preferably, in the partial success rate calculation unit:

[0054] The calculation method of the point cloud registration success rate of the concentrated distribution class associated grid is as follows: first determine whether each concentrated distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t1 The coordinates of (x gt1 ,y gt1 ,z gt1 ), the center of gravity G of the grid registration point cloud r1 The coordinates of (x gr1 ,y gr1 ,z gr1 ), calculate the spatial distance between the centers of gravity of two point clouds If d g1 Less than the first distance threshold d th1 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of concentrated distribution class associated grids is N1, and the number of concentrated distribution class associated grids that are successfully registered is K1, then the point cloud registration success rate of concentrated distribution class associated grids is η1=K1 / N1;

[0055] The calculation method of the point cloud registration success rate of linear distribution associated grids is as follows: first, all N2 linear distribution associated grids are screened. Specifically, for each linear distribution associated grid, if the fitting line L of the target point cloud in the grid is t2 The angle α between the vertical axis and the vertical axis is less than the first angle threshold α th Then retain, otherwise remove, and get N after screening 2s candidate linear distribution class associated grids; then determine whether each candidate linear distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t2 The coordinates of (x gt2 ,y gt2 ,z gt2 ), the center of gravity G of the grid registration point cloud r2 The coordinates of (x gr2 ,y gr2 ,z gr2 ), calculate the spatial distance between the centers of gravity of two point clouds In calculating the center of gravity G r2 To line L t2 The vertical distance d2=d g2 cosβ, where β is the line connecting the centroids of the two point clouds and the straight line L t2 If d2 is less than the second distance threshold d th2 , then the registration is judged to be successful, otherwise it is judged to be a registration failure; if the number of candidate linear distribution class associated grids that are successfully registered is K2, then the point cloud registration success rate of linear distribution class associated grids is η2=K2 / N 2s ;

[0056] The calculation method of the point cloud registration success rate of the surface distribution class associated grid is as follows: first determine whether each surface distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t3 The coordinates of (x gt3 ,y gt3 ,z gt3 ), the center of gravity G of the grid registration point cloud r3 The coordinates of (x gr3 ,y gr3 ,z gr3 ), calculate the spatial distance between the centers of gravity of two point clouds Then calculate the center of gravity G r3 Fitting plane S to the target point cloud within the grid t3 The vertical distance d3=d g3 cosγ, where γ is the line connecting the centroids of the two point clouds and the plane S t3 If d3 is less than the third distance threshold d th3 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of surface distribution class associated grids is N3, and the number of successfully registered surface distribution class associated grids is K3, then the point cloud registration success rate of surface distribution class associated grids is η3=K3 / N3;

[0057] The calculation method of the point cloud registration success rate of discrete distribution class associated grids is as follows: first determine whether each discrete distribution class associated grid is successfully registered. Specifically, for each discrete distribution class associated grid, if the covariance matrix of the target point cloud in the grid is C t4 , whose determinant is det C t4 , the covariance matrix of the grid-registered point cloud is C r4 , whose determinant is det C r4 , then the entropy of the target point cloud within the grid and the registration point cloud within the grid are E t4 =1+ln(det C t4 ) and E r4 =1+ln(det C r4 ), the point cloud entropy difference between the two is ΔE4=E t4 -E r4 , if the absolute value of ΔE4 is less than the first entropy difference threshold ΔE th , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of discrete distribution class associated grids is N4, and the number of successfully registered discrete distribution class associated grids is K4, then the point cloud registration success rate of discrete distribution class associated grids is η4=K4 / N4;

[0058] The sum of the associated grids N1 of the concentrated distribution type, N2 of the linear distribution type, N3 of the surface distribution type, and N4 of the discrete distribution type is N, that is:

[0059] N1+N2+N3+N4=N

[0060] Preferably, in the overall success rate calculation unit:

[0061] η whole =a1η1+a2η2+a3η3+a4η4

[0062] Among them, a1, a2, a3 and a4 are the weight parameters corresponding to the associated grids of concentrated distribution, linear distribution, surface distribution and discrete distribution respectively.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. The point cloud registration result evaluation method based on distribution characteristics proposed in this invention addresses the shortcomings of the existing point cloud registration result evaluation based on root mean square error, refines the evaluation of point cloud registration results, and can effectively improve the accuracy of point cloud registration evaluation;

[0065] 2. Taking into account that the refined evaluation of point cloud registration results will introduce a larger amount of calculation, the present invention proposes a complete set of point cloud registration result evaluation algorithms that can be run on GPU. The algorithm process is designed with full consideration of the GPU structure, which solves the problem of high latency when running on CPU and improves the working efficiency of GPU. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0067] Figure 1 A schematic diagram of a process of an embodiment of the present invention;

[0068] Figure 2 for Figure 1 Schematic diagram of the process of step S4;

[0069] Figure 3 A schematic diagram of functional modules of an embodiment. DETAILED DESCRIPTION

[0070] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0071] Example

[0072] According to the present invention, a point cloud registration result evaluation method based on distribution characteristics is provided, such as Figure 1 As shown, including:

[0073] Step S1: obtaining a source point cloud and a target point cloud, and registering the source point cloud to obtain a registered point cloud;

[0074] Step S2: placing the registration point cloud and the target point cloud in the same three-dimensional space, and dividing the three-dimensional space into a plurality of grids of the same size, wherein the registration point cloud and the target point cloud falling within each grid are the in-grid registration point cloud and in-grid target point cloud corresponding to each grid respectively;

[0075] Step S3: All grids are screened based on the condition that the number of points contained in the intra-grid registration point cloud and the intra-grid target point cloud of the grid is greater than a first number threshold, and the remaining N grids are obtained after screening. The N grids are associated grids; the intra-grid registration point cloud in the grid and the intra-grid target point cloud in the grid constitute an associated point cloud pair;

[0076] Step S4: For each associated grid, perform eigendecomposition on the covariance matrix of the target point cloud within the grid to obtain eigenvectors and eigenvalues, then determine the distribution characteristics of the target point cloud within the grid based on the size of the eigenvalues and the covariance matrix determinant, and classify each associated grid based on the distribution characteristics of the target point cloud within the grid;

[0077] Specifically, if Figure 2 As shown, the step S4 adopts:

[0078] Step S4.1: Number the associated grids from 1 to N, and initialize i to 1;

[0079] Step S4.2: For the i-th associated grid, the covariance matrix C of the target point cloud in the grid t Perform eigendecomposition to obtain eigenvectors and three eigenvalues, and calculate C t The determinant det C t ;

[0080] Step S4.3: Sort the three eigenvalues from largest to smallest to obtain a first eigenvalue λ1, a second eigenvalue λ2, and a third eigenvalue λ3, respectively; divide λ1 by λ2 to obtain a first ratio ζ1; and divide λ2 by λ3 to obtain a second ratio ζ2;

[0081] Step S4.4: According to ζ1, ζ2 and det C t , determine the distribution characteristics of the target point cloud within the grid, if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Less than the first determinant threshold Dth , then the distribution characteristics of the target point cloud in the grid belong to concentrated distribution; if ζ1 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to linear distribution; if ζ1 is less than the first ratio threshold ζ th And ζ2 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to the surface distribution; if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Greater than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to discrete distribution;

[0082] Step S4.5: Classify the i-th associated grid according to the distribution characteristics of the point cloud within the grid. If the distribution characteristics of the target point cloud within the grid belong to concentrated distribution, then the i-th associated grid belongs to concentrated distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to linear distribution, then the i-th associated grid belongs to linear distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to surface distribution, then the i-th associated grid belongs to surface distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to discrete distribution, then the i-th associated grid belongs to discrete distribution type associated grid;

[0083] Step S4.6: Add 1 to the value of i, and then determine whether the new value of i is less than or equal to N. If so, jump back to step S4.2 and re-execute the relevant steps in sequence; if not, terminate the execution of all steps.

[0084] Step S5: Calculate the point cloud registration success rate of each type of associated grid according to the type of associated grid;

[0085] Specifically, the step S5 adopts:

[0086] The calculation method of the point cloud registration success rate of the concentrated distribution class associated grid is as follows: first determine whether each concentrated distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t1 The coordinates of (x gt1 ,y gt1 ,z gt1 ), the center of gravity G of the grid registration point cloud r1 The coordinates of (x gr1 ,y gr1 ,z gr1 ), calculate the spatial distance between the centers of gravity of two point clouds If d g1 Less than the first distance threshold d th1, then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of concentrated distribution class associated grids is N1, and the number of concentrated distribution class associated grids that are successfully registered is K1, then the point cloud registration success rate of concentrated distribution class associated grids is η1=K1 / N1;

[0087] The calculation method of the point cloud registration success rate of linear distribution associated grids is as follows: first, all N2 linear distribution associated grids are screened. Specifically, for each linear distribution associated grid, if the fitting line L of the target point cloud in the grid is t2 The angle α between the vertical axis and the vertical axis is less than the first angle threshold α th Then retain, otherwise remove, and get N after screening 2s candidate linear distribution class associated grids; then determine whether each candidate linear distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t2 The coordinates of (x gt2 ,y gt2 ,z gt2 ), the center of gravity G of the grid registration point cloud r2 The coordinates of (x gr2 ,y gr2 ,z gr2 ), calculate the spatial distance between the centers of gravity of two point clouds In calculating the center of gravity G r2 To line L t2 The vertical distance d2=d g2 cosβ, where β is the line connecting the centroids of the two point clouds and the straight line L t2 If d2 is less than the second distance threshold d th2 , then the registration is judged to be successful, otherwise it is judged to be a registration failure; if the number of candidate linear distribution class associated grids that are successfully registered is K2, then the point cloud registration success rate of linear distribution class associated grids is η2=K2 / N 2s ;

[0088] The calculation method of the point cloud registration success rate of the surface distribution class associated grid is as follows: first determine whether each surface distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t3 The coordinates of (x gt3 ,y gt3 ,z gt3 ), the center of gravity G of the grid registration point cloud r3 The coordinates of (x gr3 ,y gr3 ,z gr3 ), calculate the spatial distance between the centers of gravity of two point clouds Then calculate the center of gravity G r3 Fitting plane S to the target point cloud within the grid t3 The vertical distance d3=dg3 cosγ, where γ is the line connecting the centroids of the two point clouds and the plane S t3 If d3 is less than the third distance threshold d th3 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of surface distribution class associated grids is N3, and the number of successfully registered surface distribution class associated grids is K3, then the point cloud registration success rate of surface distribution class associated grids is η3=K3 / N3;

[0089] The calculation method of the point cloud registration success rate of discrete distribution class associated grids is as follows: first determine whether each discrete distribution class associated grid is successfully registered. Specifically, for each discrete distribution class associated grid, if the covariance matrix of the target point cloud in the grid is C t4 , whose determinant is det C t4 , the covariance matrix of the grid-registered point cloud is C r4 , whose determinant is det C r4 , then the entropy of the target point cloud within the grid and the registration point cloud within the grid are E t4 =1+ln(det C t4 ) and E r4 =1+ln(det C r4 ), the point cloud entropy difference between the two is ΔE4=E t4 -E r4 , if the absolute value of ΔE4 is less than the first entropy difference threshold ΔE th , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of discrete distribution class associated grids is N4, and the number of successfully registered discrete distribution class associated grids is K4, then the point cloud registration success rate of discrete distribution class associated grids is η4=K4 / N4;

[0090] The sum of the associated grids N1 of the concentrated distribution type, N2 of the linear distribution type, N3 of the surface distribution type, and N4 of the discrete distribution type is N, that is:

[0091] N1+N2+N3+N4=N

[0092] Step S6: multiplying the point cloud registration success rate of each type of associated grid by the corresponding weight parameter and summing them up to obtain the overall point cloud registration success rate;

[0093] Specifically, the step S6 adopts:

[0094] η whole =a1η1+a2η2+a3η3+a4η4

[0095] Among them, a1, a2, a3 and a4 are the weight parameters corresponding to the associated grids of concentrated distribution, linear distribution, surface distribution and discrete distribution respectively.

[0096] Step S7: Evaluate the point cloud registration result according to the overall point cloud registration success rate.

[0097] Specifically, if Figure 3 As shown, the target point cloud and the registration point cloud are stored in the GPU, and the point cloud registration result evaluation method based on distribution characteristics is run in the GPU.

[0098] According to the present invention, a point cloud registration result evaluation system based on distribution characteristics is provided, comprising:

[0099] Data acquisition module, used to obtain source point cloud and target point cloud;

[0100] A point cloud registration module, configured to register the source point cloud to obtain a registered point cloud;

[0101] A GPU calculation module is used to calculate the overall point cloud registration success rate based on the target point cloud and the registration point cloud using the parallel acceleration function of the GPU;

[0102] Specifically, the GPU computing module includes: a storage unit, a grid processing unit, a screening unit, a classification unit, a partial success rate calculation unit and an overall success rate calculation unit;

[0103] The storage unit is used to store the target point cloud and the registration point cloud;

[0104] The gridding processing unit is used to place the registration point cloud and the target point cloud in the same three-dimensional space, and divide the three-dimensional space into a plurality of grids of the same size, wherein the registration point cloud and the target point cloud falling within each grid are respectively the in-grid registration point cloud and the in-grid target point cloud corresponding to each grid;

[0105] The screening unit is used to screen all grids, and screen all grids based on the condition that the number of points included in the registration point cloud and the target point cloud within the grid is greater than a first number threshold, and obtain N associated grids after screening;

[0106] The classification unit is used to classify each associated grid, and for each associated grid, perform eigendecomposition on the covariance matrix of the target point cloud in the grid to obtain eigenvectors and eigenvalues, and then determine the distribution characteristics of the target point cloud in the grid according to the size of the eigenvalues and the covariance matrix determinant, and classify each associated grid according to the distribution characteristics of the target point cloud in the grid;

[0107] Specifically, the classification units include:

[0108] Component 1: Number the associated grids from 1 to N, and initialize i to 1;

[0109] Component 2: For the i-th associated grid, the covariance matrix C of the target point cloud within the grid t Perform eigendecomposition to obtain eigenvectors and three eigenvalues, and calculate C t The determinant det C t ;

[0110] Component 3: Sort the three eigenvalues from largest to smallest to obtain a first eigenvalue λ1, a second eigenvalue λ2, and a third eigenvalue λ3, divide λ1 by λ2 to obtain a first ratio ζ1, and divide λ2 by λ3 to obtain a second ratio ζ2;

[0111] Component 4: Based on ζ1, ζ2 and det C t , determine the distribution characteristics of the target point cloud within the grid, if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Less than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to concentrated distribution; if ζ1 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to linear distribution; if ζ1 is less than the first ratio threshold ζ th And ζ2 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to the surface distribution; if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Greater than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to discrete distribution;

[0112] Component 5: Classify the i-th associated grid according to the distribution characteristics of the point cloud within the grid. If the distribution characteristics of the target point cloud within the grid belong to concentrated distribution, then the i-th associated grid belongs to concentrated distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to linear distribution, then the i-th associated grid belongs to linear distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to surface distribution, then the i-th associated grid belongs to surface distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to discrete distribution, then the i-th associated grid belongs to discrete distribution type associated grid;

[0113] Component 6: Increment the value of i by 1, and then determine whether the new value of i is less than or equal to N. If so, re-execute the relevant components starting from component 2; otherwise, terminate the execution of all components.

[0114] The partial success rate calculation unit is used to calculate the point cloud registration success rate of each type of associated grid according to the type of associated grid;

[0115] Specifically, in the partial success rate calculation unit:

[0116] The calculation method of the point cloud registration success rate of the concentrated distribution class associated grid is as follows: first determine whether each concentrated distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t1 The coordinates of (x gt1 ,y gt1 ,z gt1 ), the center of gravity G of the grid registration point cloud r1 The coordinates of (x gr1 ,y gr1 ,z gr1 ), calculate the spatial distance between the centers of gravity of two point clouds If d g1 Less than the first distance threshold d th1 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of concentrated distribution class associated grids is N1, and the number of concentrated distribution class associated grids that are successfully registered is K1, then the point cloud registration success rate of concentrated distribution class associated grids is η1=K1 / N1;

[0117] The calculation method of the point cloud registration success rate of linear distribution associated grids is as follows: first, all N2 linear distribution associated grids are screened. Specifically, for each linear distribution associated grid, if the fitting line L of the target point cloud in the grid is t2 The angle α between the vertical axis and the vertical axis is less than the first angle threshold α th Then retain, otherwise remove, and get N after screening 2s candidate linear distribution class associated grids; then determine whether each candidate linear distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t2 The coordinates of (x gt2 ,y gt2 ,z gt2 ), the center of gravity G of the grid registration point cloud r2 The coordinates of (x gr2 ,y gr2 ,z gr2 ), calculate the spatial distance between the centers of gravity of two point clouds In calculating the center of gravity G r2 To line L t2 The vertical distance d2=d g2 cosβ, where β is the line connecting the centroids of the two point clouds and the straight line L t2 If d2 is less than the second distance threshold d th2 , then the registration is judged to be successful, otherwise it is judged to be a registration failure; if the number of candidate linear distribution class associated grids that are successfully registered is K2, then the point cloud registration success rate of linear distribution class associated grids is η2=K2 / N 2s ;

[0118] The calculation method of the point cloud registration success rate of the surface distribution class associated grid is as follows: first determine whether each surface distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t3 The coordinates of (x gt3 ,y gt3 ,z gt3 ), the center of gravity G of the grid registration point cloud r3 The coordinates of (x gr3 ,y gr3 ,z gr3 ), calculate the spatial distance between the centers of gravity of two point clouds Then calculate the center of gravity G r3 Fitting plane S to the target point cloud within the grid t3 The vertical distance d3=d g3 cosγ, where γ is the line connecting the centroids of the two point clouds and the plane S t3 If d3 is less than the third distance threshold d th3 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of surface distribution class associated grids is N3, and the number of successfully registered surface distribution class associated grids is K3, then the point cloud registration success rate of surface distribution class associated grids is η3=K3 / N3;

[0119] The calculation method of the point cloud registration success rate of discrete distribution class associated grids is as follows: first determine whether each discrete distribution class associated grid is successfully registered. Specifically, for each discrete distribution class associated grid, if the covariance matrix of the target point cloud in the grid is C t4 , whose determinant is det C t4 , the covariance matrix of the grid-registered point cloud is C r4 , whose determinant is det C r4 , then the entropy of the target point cloud within the grid and the registration point cloud within the grid are E t4 =1+ln(det C t4 ) and E r4 =1+ln(det C r4 ), the point cloud entropy difference between the two is ΔE4=E t4 -E r4 , if the absolute value of ΔE4 is less than the first entropy difference threshold ΔE th , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of discrete distribution class associated grids is N4, and the number of successfully registered discrete distribution class associated grids is K4, then the point cloud registration success rate of discrete distribution class associated grids is η4=K4 / N4;

[0120] The sum of the associated grids N1 of the concentrated distribution type, N2 of the linear distribution type, N3 of the surface distribution type, and N4 of the discrete distribution type is N, that is:

[0121] N1+N2+N3+N4=N

[0122] The overall success rate calculation unit is used to multiply the point cloud registration success rate of each type of associated grid by the corresponding weight parameter and then add them up to obtain the overall point cloud registration success rate.

[0123] In the overall success rate calculation unit:

[0124] η whole =a1η1+a2η2+a3η3+a4η4

[0125] Among them, a1, a2, a3 and a4 are the weight parameters corresponding to the associated grids of concentrated distribution, linear distribution, surface distribution and discrete distribution respectively.

[0126] The result evaluation module is used to evaluate whether the point cloud registration result is accurate according to the overall point cloud registration success rate.

[0127] Those skilled in the art will appreciate that, in addition to implementing the system, device, and various modules provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, and the like by logically programming the method steps. Therefore, the system, device, and various modules provided by the present invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; the modules for implementing various functions can also be considered both software programs for implementing the method and structures within the hardware component.

[0128] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A point cloud registration result evaluation method based on distribution characteristics, characterized in that: include: Step S1: obtaining a source point cloud and a target point cloud, and registering the source point cloud to obtain a registered point cloud; Step S2: placing the registration point cloud and the target point cloud in the same three-dimensional space, and dividing the three-dimensional space into a plurality of grids of the same size, wherein the registration point cloud and the target point cloud falling within each grid are the in-grid registration point cloud and in-grid target point cloud corresponding to each grid respectively; Step S3: All grids are screened based on the condition that the number of points included in the in-grid registration point cloud and the in-grid target point cloud is greater than a first quantity threshold, and N associated grids are obtained after screening; Step S4: For each associated grid, perform eigendecomposition on the covariance matrix of the target point cloud within the grid to obtain eigenvectors and eigenvalues, then determine the distribution characteristics of the target point cloud within the grid based on the size of the eigenvalues and the covariance matrix determinant, and classify each associated grid based on the distribution characteristics of the target point cloud within the grid; Step S5: Calculate the point cloud registration success rate of each type of associated grid according to the type of associated grid; Step S6: multiplying the point cloud registration success rate of each type of associated grid by the corresponding weight parameter and summing them up to obtain the overall point cloud registration success rate; Step S7: Evaluate the point cloud registration result according to the overall point cloud registration success rate.

2. The point cloud registration result evaluation method based on distribution characteristics according to claim 1, characterized in that: The step S4 comprises: Step S4.1: Number the associated grids from 1 to N, and initialize i to 1; Step S4.2: For the i-th associated grid, the covariance matrix C of the target point cloud in the grid t Perform eigendecomposition to obtain eigenvectors and three eigenvalues, and calculate C t The determinant det C t ; Step S4.3: Sort the three eigenvalues from largest to smallest to obtain a first eigenvalue λ1, a second eigenvalue λ2, and a third eigenvalue λ3, respectively; divide λ1 by λ2 to obtain a first ratio ζ1; and divide λ2 by λ3 to obtain a second ratio ζ2; Step S4.4: According to ζ1, ζ2 and det C t , determine the distribution characteristics of the target point cloud within the grid, if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Less than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to concentrated distribution; if ζ1 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to linear distribution; if ζ1 is less than the first ratio threshold ζ th And ζ2 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to the surface distribution; if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Greater than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to discrete distribution; Step S4.5: Classify the i-th associated grid according to the distribution characteristics of the point cloud within the grid. If the distribution characteristics of the target point cloud within the grid belong to concentrated distribution, then the i-th associated grid belongs to concentrated distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to linear distribution, then the i-th associated grid belongs to linear distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to surface distribution, then the i-th associated grid belongs to surface distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to discrete distribution, then the i-th associated grid belongs to discrete distribution type associated grid; Step S4.6: Add 1 to the value of i, and then determine whether the new value of i is less than or equal to N. If so, jump back to step S4.2 and re-execute the relevant steps in sequence; if not, terminate the execution of all steps.

3. The point cloud registration result evaluation method based on distribution characteristics according to claim 1, characterized in that: The step S5 comprises: The calculation method of the point cloud registration success rate of the concentrated distribution class associated grid is as follows: first determine whether each concentrated distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t1 The coordinates of (x gt1 ,y gt1 ,z gt1 ), the center of gravity G of the grid registration point cloud r1 The coordinates of (x gr1 ,y gr1 ,z gr1 ), calculate the spatial distance between the centers of gravity of two point clouds If d g1 Less than the first distance threshold d th1 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of concentrated distribution class associated grids is N1, and the number of concentrated distribution class associated grids that are successfully registered is K1, then the point cloud registration success rate of concentrated distribution class associated grids is η1=K1 / N1; The calculation method of the point cloud registration success rate of linear distribution associated grids is as follows: first, all N2 linear distribution associated grids are screened. Specifically, for each linear distribution associated grid, if the fitting line L of the target point cloud in the grid is t2 The angle α between the vertical axis and the vertical axis is less than the first angle threshold α th Then retain, otherwise remove, and get N after screening 2s candidate linear distribution class associated grids; then determine whether each candidate linear distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t2 The coordinates of (x gt2 ,y gt2 ,z gt2 ), the center of gravity G of the grid registration point cloud r2 The coordinates of (x gr2 ,y gr2 ,z gr2 ), calculate the spatial distance between the centers of gravity of two point clouds In calculating the center of gravity G r2 To line L t2 The vertical distance d2=d g2 cosβ, where β is the line connecting the centroids of the two point clouds and the straight line L t2 If d2 is less than the second distance threshold d th2 , then the registration is judged to be successful, otherwise it is judged to be a registration failure; if the number of candidate linear distribution class associated grids that are successfully registered is K2, then the point cloud registration success rate of linear distribution class associated grids is η2=K2 / N 2s ; The calculation method of the point cloud registration success rate of the surface distribution class associated grid is as follows: first determine whether each surface distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t3 The coordinates of (x gt3 ,y gt3 ,z gt3 ), the center of gravity G of the grid registration point cloud r3 The coordinates of (x gr3 ,y gr3 ,z gr3 ), calculate the spatial distance between the centers of gravity of two point clouds Then calculate the center of gravity G r3 Fitting plane S to the target point cloud within the grid t3 The vertical distance d3=d g3 cosγ, where γ is the line connecting the centroids of the two point clouds and the plane S t3 If d3 is less than the third distance threshold d th3 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of surface distribution class associated grids is N3, and the number of successfully registered surface distribution class associated grids is K3, then the point cloud registration success rate of surface distribution class associated grids is η3=K3 / N3; The calculation method of the point cloud registration success rate of discrete distribution class associated grids is as follows: first determine whether each discrete distribution class associated grid is successfully registered. Specifically, for each discrete distribution class associated grid, if the covariance matrix of the target point cloud in the grid is C t4 , whose determinant is det C t4 , the covariance matrix of the grid-registered point cloud is C r4 , whose determinant is det C r4 , then the entropy of the target point cloud within the grid and the registration point cloud within the grid are E t4 =1+ln(det C t4 ) and E r4 =1+ln(det C r4 ), the point cloud entropy difference between the two is ΔE4=E t4 -E r4 , if the absolute value of ΔE4 is less than the first entropy difference threshold ΔE th , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of discrete distribution class associated grids is N4, and the number of successfully registered discrete distribution class associated grids is K4, then the point cloud registration success rate of discrete distribution class associated grids is η4=K4 / N4; The sum of the associated grids N1 of the concentrated distribution type, N2 of the linear distribution type, N3 of the surface distribution type, and N4 of the discrete distribution type is N, that is: N1+N2+N3+N4=N.

4. The point cloud registration result evaluation method based on distribution characteristics according to claim 3 is characterized in that: The calculation formula for the overall point cloud registration success rate is: or whole =a1η1+a2η2+a3η3+a4η4 Among them, a1, a2, a3 and a4 are the weight parameters corresponding to the associated grids of concentrated distribution, linear distribution, surface distribution and discrete distribution respectively.

5. The point cloud registration result evaluation method based on distribution characteristics according to claim 1, characterized in that: The target point cloud and the registration point cloud are stored in a GPU, and the point cloud registration result evaluation method based on distribution characteristics is executed in the GPU.

6. A point cloud registration result evaluation system based on distribution characteristics, the system being used in the point cloud registration result evaluation method based on distribution characteristics according to any one of claims 1 to 5, characterized in that: include: Data acquisition module, used to obtain source point cloud and target point cloud; A point cloud registration module, configured to register the source point cloud to obtain a registered point cloud; A GPU calculation module is used to calculate the overall point cloud registration success rate based on the target point cloud and the registration point cloud using the parallel acceleration function of the GPU; The result evaluation module is used to evaluate whether the point cloud registration result is accurate according to the overall point cloud registration success rate.

7. The point cloud registration result evaluation system based on distribution characteristics according to claim 6, characterized in that: The GPU computing module includes: a storage unit, a grid processing unit, a screening unit, a classification unit, a partial success rate calculation unit and an overall success rate calculation unit; The storage unit is used to store the target point cloud and the registration point cloud; The gridding processing unit is used to place the registration point cloud and the target point cloud in the same three-dimensional space, and divide the three-dimensional space into a plurality of grids of the same size, wherein the registration point cloud and the target point cloud falling within each grid are respectively the in-grid registration point cloud and the in-grid target point cloud corresponding to each grid; The screening unit is used to screen all grids, and screen all grids based on the condition that the number of points included in the registration point cloud and the target point cloud within the grid is greater than a first number threshold, and obtain N associated grids after screening; The classification unit is used to classify each associated grid, and for each associated grid, perform eigendecomposition on the covariance matrix of the target point cloud in the grid to obtain eigenvectors and eigenvalues, and then determine the distribution characteristics of the target point cloud in the grid according to the size of the eigenvalues and the covariance matrix determinant, and classify each associated grid according to the distribution characteristics of the target point cloud in the grid; The partial success rate calculation unit is used to calculate the point cloud registration success rate of each type of associated grid according to the type of associated grid; The overall success rate calculation unit is used to multiply the point cloud registration success rate of each type of associated grid by the corresponding weight parameter and then add them up to obtain the overall point cloud registration success rate.

8. The point cloud registration result evaluation system based on distribution characteristics according to claim 7, characterized in that: Included in the taxonomic units are: Component 1: Number the associated grids from 1 to N, and initialize i to 1; Component 2: For the i-th associated grid, the covariance matrix C of the target point cloud within the grid t Perform eigendecomposition to obtain eigenvectors and three eigenvalues, and calculate C t The determinant det C t ; Component 3: Sort the three eigenvalues from largest to smallest to obtain a first eigenvalue λ1, a second eigenvalue λ2, and a third eigenvalue λ3, divide λ1 by λ2 to obtain a first ratio ζ1, and divide λ2 by λ3 to obtain a second ratio ζ2; Component 4: Based on ζ1, ζ2 and det C t , determine the distribution characteristics of the target point cloud within the grid, if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Less than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to concentrated distribution; if ζ1 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to linear distribution; if ζ1 is less than the first ratio threshold ζ th And ζ2 is greater than the first ratio threshold ζ th , then the distribution characteristics of the target point cloud in the grid belong to the surface distribution; if ζ1 and ζ2 are both less than the first ratio threshold ζ th , and det C t Greater than the first determinant threshold D th , then the distribution characteristics of the target point cloud in the grid belong to discrete distribution; Component 5: Classify the i-th associated grid according to the distribution characteristics of the point cloud within the grid. If the distribution characteristics of the target point cloud within the grid belong to concentrated distribution, then the i-th associated grid belongs to concentrated distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to linear distribution, then the i-th associated grid belongs to linear distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to surface distribution, then the i-th associated grid belongs to surface distribution type associated grid; if the distribution characteristics of the target point cloud within the grid belong to discrete distribution, then the i-th associated grid belongs to discrete distribution type associated grid; Component 6: Increment the value of i by 1, and then determine whether the new value of i is less than or equal to N. If so, re-execute the relevant components starting from component 2; otherwise, terminate the execution of all components.

9. The point cloud registration result evaluation system based on distribution characteristics according to claim 7, characterized in that: In the partial success rate calculation unit: The calculation method of the point cloud registration success rate of the concentrated distribution class associated grid is as follows: first determine whether each concentrated distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t1 The coordinates of (x gt1 ,y gt1 ,z gt1 ), the center of gravity G of the grid registration point cloud r1 The coordinates of (x gr1 ,y gr1 ,z gr1 ), calculate the spatial distance between the centers of gravity of two point clouds If d g1 Less than the first distance threshold d th1 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of concentrated distribution class associated grids is N1, and the number of concentrated distribution class associated grids that are successfully registered is K1, then the point cloud registration success rate of concentrated distribution class associated grids is η1=K1 / N1; The calculation method of the point cloud registration success rate of linear distribution associated grids is as follows: first, all N2 linear distribution associated grids are screened. Specifically, for each linear distribution associated grid, if the fitting line L of the target point cloud in the grid is t2 The angle α between the vertical axis and the vertical axis is less than the first angle threshold α th Then retain, otherwise remove, and get N after screening 2s candidate linear distribution class associated grids; then determine whether each candidate linear distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t2 The coordinates of (x gt2 ,y gt2 ,z gt2 ), the center of gravity G of the grid registration point cloud r2 The coordinates of (x gr2 ,y gr2 ,z gr2 ), calculate the spatial distance between the centers of gravity of two point clouds In calculating the center of gravity G r2 To line L t2 The vertical distance d2=d g2 cosβ, where β is the line connecting the centroids of the two point clouds and the straight line L t2 If d2 is less than the second distance threshold d th2 , then the registration is judged to be successful, otherwise it is judged to be a registration failure; if the number of candidate linear distribution class associated grids that are successfully registered is K2, then the point cloud registration success rate of linear distribution class associated grids is η2=K2 / N 2s ; The calculation method of the point cloud registration success rate of the surface distribution class associated grid is as follows: first determine whether each surface distribution class associated grid is successfully registered. Specifically, if the centroid G of the target point cloud in the grid t3 The coordinates of (x gt3 ,y gt3 ,z gt3 ), the center of gravity G of the grid registration point cloud r3 The coordinates of (x gr3 ,y gr3 ,z gr3 ), calculate the spatial distance between the centers of gravity of two point clouds Then calculate the center of gravity G r3 Fitting plane S to the target point cloud within the grid t3 The vertical distance d3=d g3 cosγ, where γ is the line connecting the centroids of the two point clouds and the plane S t3 If d3 is less than the third distance threshold d th3 , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of surface distribution class associated grids is N3, and the number of successfully registered surface distribution class associated grids is K3, then the point cloud registration success rate of surface distribution class associated grids is η3=K3 / N3; The calculation method of the point cloud registration success rate of discrete distribution class associated grids is as follows: first determine whether each discrete distribution class associated grid is successfully registered. Specifically, for each discrete distribution class associated grid, if the covariance matrix of the target point cloud in the grid is C t4 , whose determinant is det C t4 , the covariance matrix of the grid-registered point cloud is C r4 , whose determinant is det C r4 , then the entropy of the target point cloud within the grid and the registration point cloud within the grid are E t4 =1+ln(det C t4 ) and E r4 =1+ln(det C r4 ), the point cloud entropy difference between the two is ΔE4=E t4 -E r4 , if the absolute value of ΔE4 is less than the first entropy difference threshold ΔE th , then it is judged that the registration is successful, otherwise it is judged that the registration fails; if the total number of discrete distribution class associated grids is N4, and the number of successfully registered discrete distribution class associated grids is K4, then the point cloud registration success rate of discrete distribution class associated grids is η4=K4 / N4; The sum of the associated grids N1 of the concentrated distribution type, N2 of the linear distribution type, N3 of the surface distribution type, and N4 of the discrete distribution type is N, that is: N1+N2+N3+N4=N.

10. The point cloud registration result evaluation system based on distribution characteristics according to claim 7, characterized in that: In the overall success rate calculation unit: or whole =a1η1+a2η2+a3η3+a4η4 Among them, a1, a2, a3 and a4 are the weight parameters corresponding to the associated grids of concentrated distribution, linear distribution, surface distribution and discrete distribution respectively.

Images