Registration method, system and device of point cloud, medium and program product

Abstract

The invention provides a point cloud registration method, system and device, a medium and a program product. The registration method comprises the steps of obtaining a to-be-registered point cloud; performing fitting processing on the to-be-registered point cloud by adopting a preset geometric figure to obtain geometric parameter values of points in the to-be-registered point cloud; selecting a target registration point from the to-be-registered point cloud based on the geometric parameter value; obtaining a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point; and registering the initial point cloud based on the target registration coefficient. The preset geometric figure is adopted to perform fitting processing on the to-be-registered point cloud to obtain the geometric parameter values of the points in the to-be-registered point cloud, and then the target registration points are selected from the to-be-registered point cloud according to the geometric parameter values, so that the number of the to-be-matched target registration points is reduced, and the matching efficiency is improved. The sampling times and the operation time required by a consistency registration algorithm are reduced, and the registration efficiency and the registration accuracy are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer vision technology, and in particular to a point cloud registration method, system, device, medium, and program product. Background Art

[0002] Point cloud registration is a common spatial point cloud data processing technique. It is generally used to convert 3D point cloud data from multiple different coordinate systems into a single coordinate system and calculate the rigid transformation matrix between the coordinate systems. Point cloud registration is of great significance in the field of computer vision. The calculated rigid transformation matrix can provide critical pose information for robot navigation and positioning.

[0003] ICP (Iterative Closest Point) is a conventional point cloud registration algorithm. As an iterative optimization algorithm, ICP is susceptible to the influence of the initial point cloud pose, and can fall into local optimal solutions during calculation. It is often combined with a suitable coarse registration algorithm to obtain the correct initial point cloud pose and ensure that the algorithm converges to the correct result. The accuracy of the coarse registration algorithm directly determines whether ICP can ultimately converge to the global optimal solution.

[0004] Currently, the mainstream coarse registration method utilizes 3D (three-dimensional) point feature descriptors based on local point cloud distribution, such as FPFH (Fast Point Feature Histograms), combined with the RANSAC (Random Sample Consensus) point cloud registration framework to quickly obtain a rigid transformation matrix with high consistency between point clouds. However, this algorithm relies on the specificity of the local distribution of point clouds, requiring a certain degree of variability in the distribution of points in the point clouds to better screen possible candidate matching points, thereby achieving good registration speed and accuracy. However, real-world applications often encounter various regular geometric surfaces with essentially uniform point cloud distributions and low local specificity. This can easily cause feature description methods like FPFH to fail, increase computational complexity, and reduce registration efficiency. Data interference can even lead to incorrectly removing correct matching points, resulting in registration errors and low registration accuracy. Summary of the Invention

[0005] The technical problem to be solved by the present disclosure is to overcome the defects of the prior art in which feature description methods such as FPFH are used as coarse registration methods on regular geometric surfaces, such as large computational complexity, low registration efficiency, and low registration accuracy, and to provide a point cloud registration method, system, equipment, medium and program product.

[0006] The present disclosure solves the above technical problems through the following technical solutions:

[0007] The present disclosure provides a point cloud registration method, the registration method comprising:

[0008] Get the point cloud to be registered;

[0009] Wherein, the point cloud to be registered includes an initial point cloud and a target point cloud;

[0010] Fitting the point cloud to be registered using a preset geometric figure to obtain geometric parameter values of the midpoints of the point cloud to be registered;

[0011] Wherein, the geometric parameter value corresponds to the preset geometric figure;

[0012] Selecting a target registration point from the to-be-registered point cloud based on the geometric parameter value;

[0013] Wherein, the target registration point includes the midpoint of the initial point cloud and the midpoint of the target point cloud;

[0014] Based on the target registration points, obtaining a target registration coefficient between the initial point cloud and the target point cloud;

[0015] The initial point cloud is registered based on the target registration coefficient.

[0016] Optionally, the preset geometric figure includes at least one of a plane, a spherical surface and a cylindrical surface;

[0017] and / or,

[0018] The step of obtaining a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point comprises:

[0019] Based on the target registration point, obtaining an initial registration coefficient between the initial point cloud and the target point cloud;

[0020] Based on the initial registration coefficients, transforming the initial point cloud to obtain a transformed point cloud;

[0021] Obtaining the actual distance between the midpoint of the target point cloud and the transformed point cloud;

[0022] Obtaining the actual number of points in the target point cloud whose actual distance is less than the preset distance;

[0023] In response to the actual number being greater than a preset number, the initial registration coefficient is used as the target registration coefficient.

[0024] Optionally, the step of fitting the point cloud to be registered using a preset geometric figure to obtain geometric parameter values of points in the point cloud to be registered includes:

[0025] Each time, one of the preset geometric figures is used to perform fitting processing on the point cloud to be registered, to obtain a fitting figure corresponding to the preset geometric figure;

[0026] wherein, in each fitting process, the point cloud to be registered is screened based on the fitting graph;

[0027] Repeat the fitting process until the fitting process of each point in the point cloud to be registered is completed;

[0028] Based on the fitting graph, the geometric parameter value of the point in the point cloud to be registered is obtained.

[0029] Optionally, in response to the preset geometric figure being a plane or a cylindrical surface, the step of fitting the point cloud to be registered using one of the preset geometric figures at a time to obtain a fitting figure corresponding to the preset geometric figure includes:

[0030] Obtaining position information and normal information of the midpoint of the point cloud to be registered;

[0031] Based on the position information and the normal information, obtaining the fitting figure corresponding to the preset geometric figure;

[0032] or,

[0033] In response to the preset geometric figure being a sphere, the step of fitting the point cloud to be registered using one of the preset geometric figures at a time to obtain a fitting figure corresponding to the preset geometric figure includes:

[0034] Obtaining the position information of the midpoint of the point cloud to be registered;

[0035] The fitting figure corresponding to the preset geometric figure is obtained based on the position information by adopting the least square method.

[0036] Optionally, the step of selecting a target registration point from the to-be-registered point cloud based on the geometric parameter value includes:

[0037] Selecting a number of initial points from the initial point cloud;

[0038] Selecting a target point from the target point cloud whose geometric parameter value matches the geometric parameter value of the initial point;

[0039] The target registration point is obtained based on the initial point and the corresponding target point.

[0040] Optionally, the step of obtaining the target registration point based on the initial point and the corresponding target point includes:

[0041] Obtaining an initial distance between the initial points and a target distance between the target points;

[0042] Obtaining an actual difference value between the initial distance and the target distance;

[0043] In response to the actual gap value being less than or equal to a preset gap value, taking the initial point and the corresponding target point as the target registration point;

[0044] In response to the actual gap value being greater than the preset gap value, returning to the step of selecting a plurality of initial points from the initial point cloud.

[0045] The present disclosure also provides a point cloud registration system, the registration system comprising:

[0046] Point cloud acquisition module, used to obtain the point cloud to be registered;

[0047] Wherein, the point cloud to be registered includes an initial point cloud and a target point cloud;

[0048] A fitting module, configured to perform fitting processing on the point cloud to be registered using a preset geometric figure to obtain geometric parameter values of the midpoints of the point cloud to be registered;

[0049] Wherein, the geometric parameter value corresponds to the preset geometric figure;

[0050] A registration point acquisition module, configured to select a target registration point from the to-be-registered point cloud based on the geometric parameter value;

[0051] Wherein, the target registration point includes the midpoint of the initial point cloud and the midpoint of the target point cloud;

[0052] A registration coefficient acquisition module, configured to obtain a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point;

[0053] A registration module is used to register the initial point cloud based on the target registration coefficient.

[0054] Optionally, the preset geometric figure includes at least one of a plane, a spherical surface and a cylindrical surface;

[0055] and / or,

[0056] The registration coefficient acquisition module includes:

[0057] an initial coefficient acquisition unit, configured to obtain initial registration coefficients between the initial point cloud and the target point cloud based on the target registration point;

[0058] a registration point cloud acquisition unit, configured to transform the initial point cloud based on the initial registration coefficients to obtain a transformed point cloud;

[0059] an actual distance acquisition unit, configured to acquire an actual distance between a midpoint of the target point cloud and the transformed point cloud;

[0060] an actual number obtaining unit, configured to obtain the actual number of points in the target point cloud whose actual distance is less than the preset distance;

[0061] The target coefficient acquisition unit is configured to use the initial registration coefficient as the target registration coefficient in response to the actual number being greater than a preset number.

[0062] Optionally, the fitting module includes:

[0063] a fitting processing unit, configured to perform fitting processing on the point cloud to be registered using one of the preset geometric figures at a time to obtain a fitting figure corresponding to the preset geometric figure;

[0064] wherein, in each fitting process, the point cloud to be registered is screened based on the fitting graph;

[0065] A fitting completion unit, configured to repeat the fitting process until the fitting process for each point in the point cloud to be registered is completed;

[0066] A parameter value acquisition unit is used to obtain the geometric parameter value of the point in the point cloud to be registered based on the fitting graph.

[0067] Optionally, in response to the preset geometric figure being a plane or a cylindrical surface, the fitting processing unit includes:

[0068] A first information acquisition subunit is used to obtain the position information and normal information of the midpoint of the point cloud to be registered;

[0069] A first fitting subunit, configured to obtain the fitting figure corresponding to the preset geometric figure based on the position information and the normal information;

[0070] or,

[0071] In response to the preset geometric figure being a sphere, the fitting processing unit includes:

[0072] A second information acquisition subunit is used to obtain the position information of the point in the point cloud to be registered;

[0073] The second fitting subunit is configured to obtain the fitting figure corresponding to the preset geometric figure based on the position information by adopting a least square method.

[0074] Optionally, the registration point acquisition module includes:

[0075] An initial point acquisition unit, configured to select a number of initial points from the initial point cloud;

[0076] a target point acquisition unit, configured to select a target point whose geometric parameter value matches the geometric parameter value of the initial point from the target point cloud;

[0077] A registration point acquisition unit is configured to obtain the target registration point based on the initial point and the corresponding target point.

[0078] Optionally, the registration point acquisition unit includes:

[0079] a distance acquisition subunit, configured to acquire an initial distance between the initial points and a target distance between the target points;

[0080] a gap value obtaining subunit, configured to obtain an actual gap value between the initial distance and the target distance;

[0081] a first response subunit, configured to use the initial point and the corresponding target point as the target registration point in response to the actual gap value being less than or equal to a preset gap value;

[0082] The second response subunit is configured to call the initial point acquisition unit in response to the actual gap value being greater than the preset gap value.

[0083] The present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and used to run on the processor, and when the processor executes the computer program, the above-mentioned point cloud registration method is implemented.

[0084] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the point cloud registration method described above is implemented.

[0085] The present disclosure also provides a computer program product, including a computer program, which implements the point cloud registration method as described above when the computer program is executed by a processor.

[0086] On the basis of conforming to the common sense in this field, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0087] The positive progress of this disclosure is:

[0088] The present invention adopts preset geometric figures to fit the point cloud to be registered, obtains the geometric parameter values of the points in the point cloud to be registered, and then selects the target registration points from the point cloud to be registered according to the geometric parameter values, and replaces the local 3D features with the overall fitting geometric features, thereby overcoming the problems of large computational complexity, low registration efficiency and low registration accuracy of local 3D features when registering on the surface of regular geometric objects, reducing the number of target registration points to be matched, reducing the number of sampling times and running time required for the consistency registration algorithm, and improving the registration efficiency and registration accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 This is a first flow chart of the point cloud registration method of Example 1 of the present disclosure;

[0090] Figure 2 This is a second flow chart of the point cloud registration method of Example 1 of the present disclosure;

[0091] Figure 3 This is a first flow chart of step S121 in the point cloud registration method of embodiment 1 of the present disclosure;

[0092] Figure 4 This is a second flow chart of step S121 in the point cloud registration method of embodiment 1 of the present disclosure;

[0093] Figure 5 This is a flowchart of step S13 in the point cloud registration method of embodiment 1 of the present disclosure;

[0094] Figure 6 This is a specific example diagram of the point cloud registration method of Example 1 of the present disclosure;

[0095] Figure 7 This is a schematic diagram of the first module of the point cloud registration system of Example 2 of the present disclosure;

[0096] Figure 8 This is a schematic diagram of the second module of the point cloud registration system according to Embodiment 2 of the present disclosure;

[0097] Figure 9 This is a structural diagram of an electronic device according to embodiment 3 of the present disclosure. DETAILED DESCRIPTION

[0098] The present disclosure is further illustrated below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0099] In the embodiments of the present disclosure, prefixes such as "first" and "second" are used only to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. In the embodiments of the present disclosure, the use of prefixes such as ordinal numbers to distinguish description objects does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and no unnecessary limitations should be constituted due to the use of such prefixes. In addition, in the description of this embodiment, unless otherwise specified, the meaning of "plurality" is two or more.

[0100] Example 1

[0101] This embodiment provides a point cloud registration method, such as Figure 1 As shown, the registration method includes:

[0102] S11, obtaining the point cloud to be registered;

[0103] Among them, the point cloud to be registered includes the initial point cloud and the target point cloud;

[0104] S12, fitting the point cloud to be registered using a preset geometric figure to obtain geometric parameter values of the midpoints of the point cloud to be registered;

[0105] Among them, the geometric parameter values correspond to the preset geometric figures;

[0106] S13. Selecting a target registration point from the point cloud to be registered based on the geometric parameter value;

[0107] Among them, the target registration point includes the midpoint of the initial point cloud and the midpoint of the target point cloud;

[0108] S14. Obtaining a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point;

[0109] S15. Register the initial point cloud based on the target registration coefficient.

[0110] Specifically, a point cloud to be registered is obtained, which includes an initial point cloud and a target point cloud. The point cloud to be registered can be a point cloud of a regular geometric object. Position information of the points in the point cloud to be registered is available. The point cloud to be registered can be preprocessed, and then a preset geometric figure is used to fit the preprocessed point cloud to obtain the geometric parameter values of the points in the point cloud to be registered.

[0111] Preprocessing includes at least one of filtering, downsampling, clustering, and normal estimation. Filtering uses statistical filtering and radius filtering to remove interference points. Normal estimation is performed by local plane fitting, using principal component analysis (PCA) to obtain the plane normal, which is then inverted to face the viewpoint. Downsampling uses voxel downsampling. Clustering uses Euclidean clustering, ultimately dividing the target point cloud into multiple clusters, retaining the largest cluster for subsequent fitting.

[0112] The initial point cloud and the target point cloud are fitted using the preset geometric shapes to obtain the geometric parameter values corresponding to the midpoints of the initial point cloud and the target point cloud. The fitting process of the initial point cloud and the target point cloud is similar.

[0113] According to the geometric parameter values, some points are selected from the initial point cloud and the target point cloud as target registration points.

[0114] The target registration coefficients between the initial point cloud and the target point cloud are obtained based on the target registration points. The target registration coefficients can be in the form of a matrix, that is, the transformation matrix from the initial point cloud to the target point cloud is obtained based on the position information of the target registration points.

[0115] The initial point cloud is registered using the target registration coefficients, that is, the initial point cloud is transformed using the transformation matrix.

[0116] Through clustering and fitting, the overall features of the point cloud to be registered are extracted, and the points of the point cloud to be registered are expressed in the form of geometric parameter values. Then, the target registration points are selected from the point cloud to be registered according to the geometric parameter values, which reduces the number of target registration points to be matched and improves the efficiency of registration.

[0117] In this scheme, the preset geometric figures are used to fit the point cloud to be registered, and the geometric parameter values of the points in the point cloud to be registered are obtained. Then, the target registration points are selected from the point cloud to be registered according to the geometric parameter values, and the local 3D features are replaced by the overall fitting geometric features. This overcomes the problems of large computational complexity, low registration efficiency, and low registration accuracy of local 3D features when registering on the surface of regular geometric objects, reduces the number of target registration points to be matched, reduces the number of sampling times and running time required for the consistency registration algorithm, and improves the registration efficiency and accuracy.

[0118] In one feasible solution, the preset geometric figure includes at least one of a plane, a spherical surface, and a cylindrical surface.

[0119] In this scheme, the point cloud to be registered is fitted by preset geometric figures such as planes, spheres and cylinders to obtain the geometric parameter values of the points in the point cloud to be registered. The points of the point cloud to be registered are expressed in the form of geometric parameter values, and then the target registration points are selected from the point cloud to be registered according to the geometric parameter values, which reduces the number of target registration points to be matched and improves the efficiency of registration.

[0120] In one feasible solution, Figure 2 As shown, step S12 includes:

[0121] S121, each time using a preset geometric figure to perform fitting processing on the point cloud to be registered, to obtain a fitting figure corresponding to the preset geometric figure;

[0122] In each fitting process, the point cloud to be registered is screened based on the fitting graph;

[0123] S122, repeat the fitting process until the fitting process of each point in the point cloud to be registered is completed;

[0124] S123. Based on the fitted graph, obtain the geometric parameter values of the points in the point cloud to be registered.

[0125] Specifically, the fitting method used is RANSAC, and each pre-set geometric shape is fitted independently. Each feature is fitted multiple times, and each fitting removes inliers and then fits the remaining points until the number of remaining points is less than a certain number, or the fitting result has too few inliers.

[0126] The fitting process will generate multiple fitting graphs, and the geometric parameter values of the points in the class corresponding to the fitting graphs are obtained based on the fitting graphs. The geometric parameter values include the spatial position information, normal information and a floating-point descriptor of the points in the class.

[0127] For example, the preset geometric figures include planes, spheres, and cylinders. First, fit a plane, then perform screening processing, that is, eliminate in-class points, and then fit the next plane. If there are not many points left or the fitting cannot be done, use spherical fitting again, and so on, until the points in the point cloud to be aligned are fitted. In-class points are points whose distance from a plane is within a first preset distance, whose distance from a sphere is within a second preset distance, or whose distance from a cylinder is within a third preset distance. The first preset distance, the second preset distance, and the third preset distance can be set or adjusted according to actual conditions. Eliminate in-class points, that is, eliminate points near the plane, sphere, or cylinder.

[0128] In this scheme, the point cloud to be registered is fitted with a preset geometric figure to obtain a fitting figure, and then the collective parameter values of the points in the point cloud to be registered are obtained. The points of the point cloud to be registered are expressed in the form of geometric parameter values, and then the target registration points are selected from the point cloud to be registered according to the geometric parameter values, which reduces the number of target registration points to be matched and improves the efficiency of registration.

[0129] In one feasible solution, in response to the preset geometric figure being a plane or a cylinder, as Figure 3 As shown, step S121 includes:

[0130] S1211, obtaining the position information and normal information of the midpoint of the point cloud to be registered;

[0131] S1212: Obtain a fitting figure corresponding to a preset geometric figure based on the position information and the normal information.

[0132] Specifically, the position information is expressed as coordinates, and the normal information is expressed as normals.

[0133] The plane is fitted using single-point fitting, extracting a three-dimensional point X with a normal each time d (x d ,y d , z d ) to obtain the three-dimensional plane parameters. The expression formula of the three-dimensional plane parameters is:

[0134] ax+by+cz+d=0;

[0135] Among them, (x d ,y d , z d ) represents the sampling point X d The coordinates of the three-dimensional point on the plane are (x, y, z), and (a, b, c) represent the corresponding sampling point X in the normal fitting process of the plane. d The normal of is a unit vector, d represents the position parameter of the plane, and the calculation formula of d is:

[0136] d=-(ax d +by d +cy d ).

[0137] Normal information N of the plane's internal points l is the unit vector of the plane normal, the descriptor of the plane's in-class points is the number of in-class points in the plane during the fitting process, and the spatial position information of the plane's in-class points is O l C is the geometric mean center point of all the points within the class during the fitting process l The projection formula on the plane is as follows:

[0138]

[0139] in, represents the unit normal of the plane, dist represents C l The distance from the point to the plane.

[0140] The formula corresponding to dist is as follows:

[0141]

[0142] in, Indicates that the origin points to C l vector.

[0143] Cylinder fitting uses two-point fitting based on the normal. The two points with normals are regarded as two straight lines. The common perpendicular line of the two straight lines is the axis of the cylinder. The closest point on the two straight lines is the point on the axis, thereby determining the position of the axis. The cylinder radius is determined by the distance from one of the two points to the axis. The distance from the point to the axis is used as the cylinder radius, and the distance from the other point to the cylinder axis is calculated at the same time. If the difference with the cylinder radius is too large, the two points are resampled for calculation.

[0144] The formula for the distance from a point to an axis is as follows:

[0145]

[0146] Where, distance represents the distance from the point to the axis, A represents the point to be calculated, and P represents a point on the axis of the cylinder. represents the vector from A to P, The unit vector representing the axis of the cylinder.

[0147] Normal information N of the inner points of the cylindrical surface c In order to fit the unit vector of the cylinder axis, the descriptor of the in-class point of the cylinder surface is the cylinder radius, and the spatial position information of the in-class point of the cylinder surface is O c is the center of the point within the class C c The projection point dis on the axis, the projection point formula to the straight line is:

[0148]

[0149] in, Indicates that P points to C c vector.

[0150] Due to the ambiguity in direction, for the geometric parameter values of the in-class points of the plane and cylindrical surfaces, a redundant copy needs to be added, that is, the redundant geometric parameter value. The spatial position information and descriptor of the copy are the same as the spatial position information and descriptor of the geometric parameter value, but the direction of the normal information is opposite to the direction of the normal information of the geometric parameter value.

[0151] In this scheme, the position information and normal information of the midpoint of the point cloud to be registered are used to obtain a fitting figure corresponding to the plane or cylindrical surface, which ensures the accuracy and reliability of the fitting processing of the point cloud to be registered; the plane fitting is performed through the normal, which reduces the number of samples of the RANSAC algorithm to a minimum and reduces the time complexity.

[0152] In one feasible solution, in response to the preset geometric figure being a sphere, such as Figure 4 As shown, step S121 includes:

[0153] S1213, obtaining the position information of the midpoint of the point cloud to be registered;

[0154] S1214: Using the least squares method, based on the position information, obtain a fitting graph corresponding to the preset geometric graph.

[0155] Specifically, the sphere is fitted by extracting four sampling points and using the least square method, that is, a sphere with the minimum distance from the four points is obtained.

[0156] The normal information of the inner point of the sphere is zero vector, and the spatial position information of the inner point of the sphere is O s The center of the fitted sphere corresponds to the point in the class of the sphere, and the descriptor of the inlier point is the radius of the sphere.

[0157] In this solution, the fitting graphics corresponding to the sphere are obtained through the position information and normal information of the midpoint of the point cloud to be registered, which ensures the accuracy and reliability of the fitting processing of the point cloud to be registered.

[0158] In one feasible solution, Figure 5 As shown, step S13 includes:

[0159] S131, selecting a number of initial points from the initial point cloud;

[0160] S132, selecting a target point from the target point cloud whose geometric parameter value matches the geometric parameter value of the initial point;

[0161] S133: Obtain a target registration point based on the initial point and the corresponding target point.

[0162] Specifically, a set number of initial points are extracted from the initial point cloud. Based on the type and descriptor of each initial point, the points in the target point cloud are screened. Points of the same type and with descriptor differences within a threshold are selected as candidate points. Finally, a point is randomly selected from the candidate points as the target point corresponding to the initial point. The setting of a number greater than or equal to 3 is merely an example. The type of initial point, i.e., the type of fitted shape to which the initial point belongs, can be a plane, a sphere, or a cylinder.

[0163] In this scheme, an initial point is selected from the initial point cloud according to the geometric parameter value, and a target point corresponding to the initial point is selected from the target point cloud. The target registration point is obtained according to the initial point and the corresponding target point, which reduces the number of target registration points to be matched and improves the efficiency of registration.

[0164] In one feasible solution, Figure 2 As shown, step S133 includes:

[0165] S1331, obtaining the initial distance between the initial points and the target distance between the target points;

[0166] S1332, obtaining an actual difference value between the initial distance and the target distance;

[0167] In response to the actual gap value being less than or equal to the preset gap value, step S1333 is executed; in response to the actual gap value being greater than the preset gap value, the process returns to step S131.

[0168] S1333. Use the initial point and the corresponding target point as target registration points.

[0169] Specifically, the initial distances between a set number of initial points extracted from the initial point cloud and the target distances between the corresponding matched target points are calculated. The actual difference between the initial distances and the target distances is obtained. The actual difference can be a difference or a ratio. The difference is the difference between the initial distance and the target distance, and the ratio is the quotient between the difference and the initial distance or the target distance.

[0170] If the actual gap value is not greater than the preset gap value, the initial point and the corresponding target point are used as target registration points; if the actual gap value is greater than the preset gap value, the sampled initial point and target point are rejected, and the process returns to step S131 to reselect several initial points from the initial point cloud.

[0171] The initial distance can be the sum of multiple first distances, and the target distance can be the sum of multiple second distances. The first distance is the distance between the two initial points, and the second distance is the distance between the two target points. For example, if there are three initial points, there will be three first distances and three second distances. Specifically, the distance between any two of the initial points is calculated, and the sum of these distances is used as the initial distance. The distance between any two of the target points is calculated, and the sum of these distances is used as the target distance. The distance between the two selected points is calculated only once and not repeatedly.

[0172] The initial distance may also be an average value of multiple first distances, and the target distance may be an average value of multiple second distances.

[0173] In this scheme, by obtaining the actual gap values of the initial distance between the initial points and the target distance between the target points, different operations are adopted according to the size relationship between the actual gap value and the preset gap value. The rejection mechanism based on the distance angle is used to screen out the time waste caused by some invalid sampling, thereby improving efficiency and ensuring the accuracy and reliability of the target registration points.

[0174] In one feasible solution, Figure 2 As shown, step S14 includes:

[0175] S141. Obtaining an initial registration coefficient between the initial point cloud and the target point cloud based on the target registration point;

[0176] S142. Transforming the initial point cloud based on the initial registration coefficients to obtain a transformed point cloud;

[0177] S143, obtaining the actual distance between the midpoint of the target point cloud and the transformed point cloud;

[0178] S144, obtaining the actual number of points in the target point cloud whose actual distance is less than the preset distance;

[0179] S145 . In response to the actual number being greater than the preset number, use the initial registration coefficient as the target registration coefficient.

[0180] Specifically, the initial center point mean of the initial point cloud is calculated by averaging source And the target center point mean of the target point cloud target Randomly extract two points from the initial points and extract two corresponding points from the target points. Connect the two randomly extracted points from the initial points to the initial center point to obtain two initial vectors. Normalize these two initial vectors to obtain two normalized initial direction vectors s1 and s2. Similarly, obtain the two target direction vectors t1 and t2 corresponding to the target point. s1, s2, t1, and t2 are all 3×1 vectors.

[0181] Check the angle to ensure that s1 and s2 are not collinear, and the angle between s1 and s2 and the angle between t1 and t2 are not too different. Use the dot product of the vectors to express the size of the angle. The corresponding formula is as follows:

[0182] cosα=s1·s2;

[0183] cosβ=t1·t2;

[0184] Wherein, α represents the angle between s1 and s2, and β represents the angle between t1 and t2.

[0185] If the angle satisfies the first condition, which is that cosα and cosβ are greater than the first set threshold or the difference between them is greater than the second set threshold, and the difference between them is the difference between cosα and cosβ, then return to step S131 and reselect several initial points from the initial point cloud.

[0186] If the angle does not meet the first condition, calculate the initial registration coefficient. To calculate the initial registration coefficient, first construct a local coordinate system using the two extracted points:

[0187] S rot =[s1,s2,s1×s2];

[0188] T rot =[t1,t2,t1×t2];

[0189] Among them, S rot represents the local coordinate system constructed by two points extracted from the initial point, T rot Represents a local coordinate system constructed from two points extracted from the target point.

[0190] The initial registration coefficients include rotation coefficients and translation coefficients. The initial registration coefficients are specifically a transformation matrix, the rotation coefficients are specifically a rotation matrix, and the translation coefficients are specifically a translation matrix.

[0191] The rotation matrix R is the local coordinate system T constructed by two points extracted from the target point rot It is obtained by multiplying the transpose of the local coordinate system Srot constructed by the two points extracted from the initial point. The corresponding formula is as follows:

[0192]

[0193] The translation matrix t passes through the target center point mean of the target point cloud target And the initial center point mean of the rotated initial point cloud source The corresponding formula is as follows:

[0194] t=mean target -R mean source ;

[0195] The transformation matrix M is a 4×4 matrix, and the corresponding expression is as follows:

[0196]

[0197] The initial registration coefficient, that is, the transformation matrix M, is used to transform the coordinates of the initial point cloud to obtain the transformed point cloud.

[0198] The initial registration coefficients meet the requirements based on the actual number of points in the overlap between the transformed point cloud and the target point cloud. The overlapping points are the points in the target point cloud whose actual distance from the transformed point cloud is less than the preset distance.

[0199] If the actual number is greater than the preset number, the initial registration coefficient is retained as the target registration coefficient. The preset number can be the product of the total number of points in the target point cloud and the preset percentage. The actual number is greater than the preset number, that is, the actual percentage of the overlapping points in the entire target point cloud is greater than the preset percentage.

[0200] Determine whether the number of target registration coefficients currently selected has reached the preset number of iterations. If not, return to step S131 and reselect several initial points from the initial point cloud to select target registration coefficients again until the preset number of iterations is reached. Select the target registration coefficient with the largest actual number from the multiple selected target registration coefficients as the final target registration coefficient, and register the initial point cloud using the final target registration coefficient.

[0201] In this scheme, the initial point cloud is transformed by the initial registration coefficient to obtain the transformed point cloud, and then the target registration coefficient is determined according to the actual number of points in the target point cloud whose actual distance to the transformed point cloud is less than the preset distance. This realizes the evaluation of the initial registration coefficient and improves the accuracy and reliability of the target registration coefficient.

[0202] The working principle of the point cloud registration method of this embodiment is explained below with reference to specific examples. Figure 6 As shown:

[0203] During industrial automation production, it's necessary to locate a regular part for subsequent operations. The sensor used for this positioning is a 3D structured light camera, which captures a point cloud of the part, the target, for precise positioning using registration. The camera is approximately 1 meter away from the part, which is approximately 5 cm long. The main body of the part is composed of relatively regular geometric shapes, such as planes and cylinders.

[0204] S21. Obtain the point cloud to be registered.

[0205] S22. Preprocess the point cloud to be registered. Use statistical filtering and radius filtering to reduce interference points. The MeanK parameter of the statistical filter is set to 5, the radius parameter of the radius filter is 5 mm, the minimum number of neighboring points is set to 10, the search radius of the normal estimation is 2.5 mm, the downsampled voxel size is set to 1 mm, and the clustering method uses Euclidean clustering. The clustering interval is twice the downsampled voxel size, that is, 2 mm.

[0206] S23, fitting processing. The pre-processed point cloud to be registered is fitted with a preset geometric figure to obtain the geometric parameter values of the points in the point cloud to be registered. When performing RANSAC geometric fitting segmentation, the intra-class point distance threshold is set to 0.4, that is, the first preset distance, the second preset distance and the third preset distance are set to 0.4. Each preset geometric figure fits and extracts up to 7 intra-class points. The number of intra-class points accounts for a threshold of 0.01 in the overall point cloud. If it is less than 0.01, it is considered that the fitting extraction has failed, and the extraction of this preset geometric figure is terminated and the fitting extraction of the next preset geometric figure is started. The fitting process will obtain multiple fitting figures, and the geometric parameter values of the intra-class points corresponding to the fitting figures are obtained according to the fitting figures. The geometric parameter values include the spatial position information, normal information and a floating-point type descriptor of the intra-class points. Since the plane of the part has a relatively thin plate structure, the same plane with two normals can be observed from both sides. Therefore, in addition to the cylindrical surface, a redundant copy of the plane will be added, whose spatial position information and descriptors are the same as those of the geometric parameter values, and the normal direction, that is, the direction of the normal information is opposite to the direction of the normal information of the geometric parameter values.

[0207] S24. Select initial points and target points. Sampling matching first extracts a set number of initial points from the initial point cloud. Then, based on the type and descriptor of each initial point, the points in the target point cloud are screened. Points with the same type and descriptor difference within a threshold are selected as candidate points. The difference is represented by the ratio between the two descriptors. Points with a ratio between 0.1 and 10 are retained as candidate points. Finally, a point is randomly selected from the candidate points as the target point corresponding to the initial point.

[0208] S25. Calculate the initial distance and target distance. Use the square of the Euclidean distance to calculate the initial distance between the initial points and the target distance between the corresponding matching target points.

[0209] S26. Determine whether to reselect. Compare the actual difference between the initial distance and the target distance. If the actual difference is greater than a preset difference value, such as 0.2, reject the sampling and return to step S24 to reselect the initial and target points. If the actual difference is not greater than the preset difference value, the initial point and the corresponding target point are used as the target registration points.

[0210] S27, calculate the initial registration coefficient. That is, estimate the transformation matrix. The initial center point mean of the initial point cloud is calculated by averaging. source And the target center point mean of the target point cloud target Randomly extract two points from the initial points and extract two corresponding points from the target points. Connect the two randomly extracted points from the initial points to the initial center point to obtain two initial vectors. Normalize the two initial vectors to obtain two normalized initial direction vectors s1 and s2. Similarly, obtain the two target direction vectors t1 and t2 corresponding to the target point.

[0211] Check the angle to ensure that s1 and s2 are not collinear, and the angle between s1 and s2 and the angle between t1 and t2 are not too different. Use the dot product of the vectors to express the size of the angle. The corresponding formula is as follows:

[0212] cosα=s1·s2;

[0213] cosβ=t1·t2;

[0214] Wherein, α represents the angle between s1 and s2, and β represents the angle between t1 and t2.

[0215] If the angle satisfies the first condition, which is that cosα and cosβ are greater than a first set threshold or the difference between them is greater than a second set threshold, where the first set threshold is set to 0.9 and the second set threshold is set to 0.3, then the process returns to step S24 and reselects the initial point and target point.

[0216] If the angle does not meet the first condition, calculate the initial registration coefficient. To calculate the initial registration coefficient, first construct a local coordinate system using the two extracted points:

[0217] S rot =[s1,s2,s1×s2];

[0218] T rot =[t1,t2,t1×t2];

[0219] The initial registration coefficients include rotation coefficients and translation coefficients. The initial registration coefficients are specifically a transformation matrix, the rotation coefficients are specifically a rotation matrix, and the translation coefficients are specifically a translation matrix.

[0220] The rotation matrix R is the local coordinate system T constructed by two points extracted from the target point rot It is obtained by multiplying the transpose of the local coordinate system Srot constructed by the two points extracted from the initial point. The corresponding formula is as follows:

[0221]

[0222] The translation matrix t passes through the target center point mean of the target point cloud target And the initial center point mean of the rotated initial point cloud source The corresponding formula is as follows:

[0223] t=mean target -R mean source ;

[0224] The transformation matrix M is a 4×4 matrix, and the corresponding expression is as follows:

[0225]

[0226] S28. Evaluation of the initial registration coefficients. Use the initial registration coefficients, i.e., the transformation matrix M, to perform coordinate transformation on the initial point cloud to obtain a transformed point cloud. Determine whether the initial registration coefficients meet the requirements based on the actual number of points in the overlap between the transformed point cloud and the target point cloud. Overlapping points, i.e., points in the target point cloud whose actual distance from the transformed point cloud is less than a preset distance. If the actual percentage of the actual number of points in the target point cloud to the total number of points is greater than a preset percentage, retain the initial registration coefficients as the target registration coefficients. The preset percentage is set to 0.4.

[0227] S29. Determine whether to stop iteration. Determine whether the number of times the target registration coefficient has been selected reaches the preset number of iterations. If not, return to step S24 and reselect the initial point and target point to select the target registration coefficient again. If not, stop iteration.

[0228] S30, point cloud registration: The initial point cloud is registered based on the target registration coefficient.

[0229] In the registration scenario of this regular part, the traditional FPFH and the consistency sampling registration algorithm based on pre-rejection were used, and 3 million samples were sampled, which took about 6 minutes and 20 seconds, but still could not achieve a good coarse registration effect. The point cloud registration method of this embodiment only needed 2,000 samples to achieve an ideal coarse registration effect. The overall time from selecting the target registration point to coarse registration plus ICP precise registration was about 16 seconds. The coarse registration method of this embodiment greatly improved the registration efficiency.

[0230] In this embodiment, the point cloud to be registered is fitted with preset geometric figures to obtain the geometric parameter values of the points in the point cloud to be registered, and then the target registration points are selected from the point cloud to be registered according to the geometric parameter values, and the local 3D features are replaced by the overall fitting geometric features. This overcomes the problems of large computational complexity, low registration efficiency, and low registration accuracy of local 3D features when registering on the surface of regular geometric objects, reduces the number of target registration points to be matched, reduces the number of sampling times and running time required for the consistency registration algorithm, and improves the registration efficiency and registration accuracy.

[0231] Example 2

[0232] Corresponding to the aforementioned point cloud registration method embodiment, the present disclosure also provides an embodiment of a point cloud registration system.

[0233] like Figure 7 As shown, the registration system includes:

[0234] Point cloud acquisition module 1, used to obtain the point cloud to be registered;

[0235] Among them, the point cloud to be registered includes the initial point cloud and the target point cloud;

[0236] Fitting module 2, used for fitting the point cloud to be registered using a preset geometric figure to obtain the geometric parameter value of the midpoint of the point cloud to be registered;

[0237] Among them, the geometric parameter values correspond to the preset geometric figures;

[0238] The registration point acquisition module 3 is used to select the target registration point from the point cloud to be registered based on the geometric parameter value;

[0239] Among them, the target registration point includes the midpoint of the initial point cloud and the midpoint of the target point cloud;

[0240] A registration coefficient acquisition module 4 is used to obtain a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point;

[0241] The registration module 5 is used to register the initial point cloud based on the target registration coefficient.

[0242] In this scheme, the preset geometric figures are used to fit the point cloud to be registered, and the geometric parameter values of the points in the point cloud to be registered are obtained. Then, the target registration points are selected from the point cloud to be registered according to the geometric parameter values, and the local 3D features are replaced by the overall fitting geometric features. This overcomes the problems of large computational complexity, low registration efficiency, and low registration accuracy of local 3D features when registering on the surface of regular geometric objects, reduces the number of target registration points to be matched, reduces the number of sampling times and running time required for the consistency registration algorithm, and improves the registration efficiency and accuracy.

[0243] In one feasible solution, the preset geometric figure includes at least one of a plane, a spherical surface, and a cylindrical surface.

[0244] In one feasible solution, Figure 8 As shown, the registration coefficient acquisition module 4 includes:

[0245] An initial coefficient acquisition unit 41 is used to obtain initial registration coefficients between the initial point cloud and the target point cloud based on the target registration point;

[0246] A registration point cloud acquisition unit 42 is configured to transform the initial point cloud based on the initial registration coefficients to obtain a transformed point cloud;

[0247] The actual distance obtaining unit 43 is used to obtain the actual distance between the midpoint of the target point cloud and the transformed point cloud;

[0248] An actual number obtaining unit 44 is used to obtain the actual number of points in the target point cloud whose actual distance is less than the preset distance;

[0249] The target coefficient acquiring unit 45 is configured to use the initial registration coefficient as the target registration coefficient in response to the actual number being greater than the preset number.

[0250] In one feasible solution, the fitting module 2 includes:

[0251] A fitting processing unit 21 is configured to perform fitting processing on the point cloud to be registered using a preset geometric figure each time to obtain a fitting figure corresponding to the preset geometric figure;

[0252] In each fitting process, the point cloud to be registered is screened based on the fitting graph;

[0253] A fitting completion unit 22 is used to repeat the fitting process until the fitting process of each point in the point cloud to be registered is completed;

[0254] The parameter value obtaining unit 23 is used to obtain the geometric parameter values of the points in the point cloud to be registered based on the fitted graph.

[0255] In one feasible solution, in response to the preset geometric figure being a plane or a cylinder, the fitting processing unit 21 includes:

[0256] The first information acquisition subunit 211 is used to obtain the position information and normal information of the point in the point cloud to be registered;

[0257] The first fitting subunit 212 is configured to obtain a fitting figure corresponding to a preset geometric figure based on the position information and the normal information.

[0258] In one feasible solution, in response to the preset geometric figure being a sphere, the fitting processing unit 21 includes:

[0259] The second information acquisition subunit 213 is used to obtain the position information of the point in the point cloud to be registered;

[0260] The second fitting subunit 214 is configured to obtain a fitting figure corresponding to a preset geometric figure based on the position information by adopting a least square method.

[0261] In one feasible solution, the registration point acquisition module 3 includes:

[0262] An initial point acquisition unit 31 is used to select a number of initial points from the initial point cloud;

[0263] A target point acquisition unit 32 is used to select a target point whose geometric parameter value matches the geometric parameter value of the initial point from the target point cloud;

[0264] The registration point acquisition unit 33 is configured to obtain a target registration point based on the initial point and the corresponding target point.

[0265] In one feasible solution, the registration point acquisition unit 33 includes:

[0266] The distance acquisition subunit 331 is used to acquire the initial distance between the initial points and the target distance between the target points;

[0267] The gap value obtaining subunit 332 is used to obtain the actual gap value between the initial distance and the target distance;

[0268] a first response subunit 333 for using the initial point and the corresponding target point as target registration points in response to the actual gap value being less than or equal to the preset gap value;

[0269] The second response subunit 334 is configured to call the initial point acquisition unit in response to the actual gap value being greater than the preset gap value.

[0270] In this embodiment, the point cloud to be registered is fitted with preset geometric figures to obtain the geometric parameter values of the points in the point cloud to be registered, and then the target registration points are selected from the point cloud to be registered according to the geometric parameter values, and the local 3D features are replaced by the overall fitting geometric features. This overcomes the problems of large computational complexity, low registration efficiency, and low registration accuracy of local 3D features when registering on the surface of regular geometric objects, reduces the number of target registration points to be matched, reduces the number of sampling times and running time required for the consistency registration algorithm, and improves the registration efficiency and registration accuracy.

[0271] Since the system embodiments generally correspond to the method embodiments, reference will be made to the description of the method embodiments for relevant details. The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components of the units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the disclosed solution.

[0272] Example 3

[0273] Figure 9 This is a structural diagram of an electronic device showing an example embodiment of the present disclosure, wherein the electronic device includes a memory, a processor, and a computer program stored in the memory and for running on the processor, and when the processor executes the computer program, the point cloud registration method described in any of the above embodiments is implemented. Figure 9 The electronic device 90 shown is only an example and should not limit the functionality and scope of use of the embodiments of the present disclosure.

[0274] like Figure 9 As shown, the electronic device 90 may be a general-purpose computing device, such as a server device. Components of the electronic device 90 may include, but are not limited to, the at least one processor 91, the at least one memory 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0275] The bus 93 includes a data bus, an address bus, and a control bus.

[0276] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922 , and may further include a read-only memory (ROM) 923 .

[0277] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) of program modules 924, such program modules 924 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include an implementation of a network environment.

[0278] The processor 91 executes various functional applications and data processing by running the computer program stored in the memory 92, such as the point cloud registration method provided in any of the above embodiments.

[0279] The electronic device 90 can also communicate with one or more external devices 94 (e.g., a keyboard, pointing device, etc.). Such communication can occur via an input / output (I / O) interface 95. Furthermore, the electronic device 90 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 96. As shown, the network adapter 96 communicates with other modules of the electronic device 90 via a bus 93. It should be understood that, although not shown, other hardware and / or software modules can be used in conjunction with the electronic device 90, including but not limited to microcode, device drivers, redundant processors, external disk drive arrays, RAID (RAID) systems, tape drives, and data backup storage systems.

[0280] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0281] Example 4

[0282] An embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the point cloud registration method provided in any of the above embodiments.

[0283] The readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0284] Example 5

[0285] An embodiment of the present disclosure further provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described point cloud registration methods.

[0286] The program code for executing the computer program product of the present disclosure may be written in any combination of one or more programming languages, and the program code may be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on the remote device.

[0287] While specific embodiments of the present disclosure have been described above, those skilled in the art will appreciate that these are merely illustrative and that the scope of protection of the present disclosure is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present disclosure, and such changes and modifications are intended to fall within the scope of protection of the present disclosure.

Claims

1. A point cloud registration method, characterized in that: The registration method comprises: Get the point cloud to be registered; Wherein, the point cloud to be registered includes an initial point cloud and a target point cloud; Fitting the point cloud to be registered using a preset geometric figure to obtain geometric parameter values of the midpoints of the point cloud to be registered; Wherein, the geometric parameter value corresponds to the preset geometric figure; Selecting a target registration point from the to-be-registered point cloud based on the geometric parameter value; Wherein, the target registration point includes the midpoint of the initial point cloud and the midpoint of the target point cloud; Based on the target registration points, obtaining a target registration coefficient between the initial point cloud and the target point cloud; The initial point cloud is registered based on the target registration coefficient.

2. The point cloud registration method according to claim 1, wherein: The preset geometric figure includes at least one of a plane, a spherical surface and a cylindrical surface; and / or, The step of obtaining a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point comprises: Based on the target registration point, obtaining an initial registration coefficient between the initial point cloud and the target point cloud; Based on the initial registration coefficients, transforming the initial point cloud to obtain a transformed point cloud; Obtaining the actual distance between the midpoint of the target point cloud and the transformed point cloud; Obtaining the actual number of points in the target point cloud whose actual distance is less than the preset distance; In response to the actual number being greater than a preset number, the initial registration coefficient is used as the target registration coefficient.

3. The point cloud registration method according to claim 1, wherein: The step of fitting the point cloud to be registered using a preset geometric figure to obtain the geometric parameter value of the midpoint of the point cloud to be registered comprises: Each time, one of the preset geometric figures is used to perform fitting processing on the point cloud to be registered, to obtain a fitting figure corresponding to the preset geometric figure; wherein, in each fitting process, the point cloud to be registered is screened based on the fitting graph; Repeat the fitting process until the fitting process of each point in the point cloud to be registered is completed; Based on the fitting graph, the geometric parameter value of the point in the point cloud to be registered is obtained.

4. The point cloud registration method according to claim 3, wherein: In response to the preset geometric figure being a plane or a cylindrical surface, the step of fitting the point cloud to be registered using one of the preset geometric figures at a time to obtain a fitting figure corresponding to the preset geometric figure includes: Obtaining position information and normal information of the midpoint of the point cloud to be registered; Based on the position information and the normal information, obtaining the fitting figure corresponding to the preset geometric figure; or, In response to the preset geometric figure being a sphere, the step of fitting the point cloud to be registered using one of the preset geometric figures at a time to obtain a fitting figure corresponding to the preset geometric figure includes: Obtaining the position information of the midpoint of the point cloud to be registered; The fitting figure corresponding to the preset geometric figure is obtained based on the position information by adopting the least square method.

5. The point cloud registration method according to any one of claims 1 to 4, characterized in that: The step of selecting a target registration point from the to-be-registered point cloud based on the geometric parameter value comprises: Selecting a number of initial points from the initial point cloud; Selecting a target point from the target point cloud whose geometric parameter value matches the geometric parameter value of the initial point; The target registration point is obtained based on the initial point and the corresponding target point.

6. The point cloud registration method according to claim 5, wherein: The step of obtaining the target registration point based on the initial point and the corresponding target point includes: Obtaining an initial distance between the initial points and a target distance between the target points; Obtaining an actual difference value between the initial distance and the target distance; In response to the actual gap value being less than or equal to a preset gap value, taking the initial point and the corresponding target point as the target registration point; In response to the actual gap value being greater than the preset gap value, returning to the step of selecting a plurality of initial points from the initial point cloud.

7. A point cloud registration system, characterized in that: The registration system comprises: Point cloud acquisition module, used to obtain the point cloud to be registered; Wherein, the point cloud to be registered includes an initial point cloud and a target point cloud; A fitting module, configured to perform fitting processing on the point cloud to be registered using a preset geometric figure to obtain geometric parameter values of the midpoints of the point cloud to be registered; Wherein, the geometric parameter value corresponds to the preset geometric figure; A registration point acquisition module, configured to select a target registration point from the to-be-registered point cloud based on the geometric parameter value; Wherein, the target registration point includes the midpoint of the initial point cloud and the midpoint of the target point cloud; A registration coefficient acquisition module, configured to obtain a target registration coefficient between the initial point cloud and the target point cloud based on the target registration point; A registration module is used to register the initial point cloud based on the target registration coefficient.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and configured to run on the processor, wherein: When the processor executes the computer program, the point cloud registration method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the point cloud registration method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the point cloud registration method according to any one of claims 1 to 6 is implemented.

Images