Point cloud registration method and device, storage medium, computer equipment and program product

By combining the flood filling algorithm and the edge-weighted FPFH feature set with the SAC_IA and ICP algorithms, the problems of large computational volume and local optimal solutions of point cloud data are solved, and fast and accurate point cloud registration in industrial scenarios is achieved.

CN118967764APending Publication Date: 2024-11-15GREE ELECTRIC APPLIANCE INC OF ZHUHAI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411175624.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In industrial applications, the introduction of depth information into point cloud data leads to a large amount of data computation, which cannot meet the production cycle requirements. Furthermore, the workpiece has few texture features and high symmetry similarity, making it easy for traditional methods to get stuck in local optima and difficult to locate quickly and accurately.

Method used

A flood filling algorithm is used for salient object segmentation, an edge-weighted FPFH feature set is constructed, and point cloud registration is performed by combining SAC_IA and ICP algorithms. The point cloud registration process is optimized by adaptively identifying salient objects and edge features.

Benefits of technology

It reduces the amount of point cloud computing, improves the accuracy of key point cloud feature extraction, and achieves fast and accurate point cloud registration, adapting to the complex environment and high-speed requirements of industrial scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118967764A_ABST
    Figure CN118967764A_ABST
Patent Text Reader

Abstract

The invention provides a point cloud registration method and device, a storage medium, computer equipment and a program product, and the method comprises the steps: carrying out the significant object segmentation of a 2D depth image of a to-be-registered object through employing a flood filling algorithm, and generating a to-be-registered point cloud; for a preset template point cloud and the to-be-registered point cloud, respectively constructing an FPFH feature set with edge weighting; and point cloud registration is carried out based on the constructed template point cloud and the FPFH feature point set with edge weighting of the to-be-registered point cloud. According to the scheme provided by the invention, the extraction precision of the point cloud key features can be improved while the point cloud calculation amount is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of control, and in particular to a point cloud registration method, device, storage medium, computer equipment and program product. Background Art

[0002] With the improvement of point cloud acquisition hardware equipment, the acquisition of point cloud data has become simpler and more convenient. Point cloud registration refers to the process of integrating data measured at different angles and orientations into a specified unified coordinate system through rigid transformation operations such as point cloud rotation and translation. Point cloud registration is widely used in industrial manufacturing, image synthesis in medical testing, digitization of cultural relics, and environmental perception in intelligent driving. Registration algorithms with high registration accuracy and fast registration speed can effectively improve the performance of these applications.

[0003] Point cloud registration technology works better for objects with rich texture features. However, in actual industrial applications, most workpieces have fewer texture features, and the incoming materials on the assembly line are complex and changeable. The workpieces have high symmetry and similarity, and the production rhythm requirements are high. It is crucial to quickly and accurately locate the workpieces. Point cloud data brings depth information, but also brings a lot of data calculation time. Matching within the entire shooting scene is time-consuming, and the uncertainty of the incoming material position cannot effectively reduce the detection area. Therefore, the strong dependence on the initial pose of the point cloud, the low rate, and the problem of being prone to falling into the local optimal solution are the issues faced by the present invention. Summary of the invention

[0004] The main purpose of the present invention is to overcome the defects of the above-mentioned related technologies and to provide a point cloud registration method, device, storage medium, computer equipment and program product to solve the problem in the related technologies that the point cloud data introduces depth information while bringing about a large amount of data calculation that cannot meet the production rhythm requirements.

[0005] On one hand, the present invention provides a point cloud registration method, comprising: using a flood filling algorithm to perform salient object segmentation on a 2D depth image of an object to be registered, and generating a point cloud to be registered; constructing an edge-weighted FPFH feature set for a preset template point cloud and the point cloud to be registered, respectively; and performing point cloud registration based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered.

[0006] Optionally, a flood filling algorithm is used to segment the salient objects in the 2D depth image of the object to be registered to generate a point cloud to be registered, including: setting the pixel point (0, 0) in the 2D depth image as the initial seed point, and setting the maximum height difference threshold of the flood filling and the minimum area threshold of the segmentation mask to be retained; starting from the initial seed point according to the maximum height difference threshold, flood filling is performed, and neighboring points whose height difference with the current seed point is less than the maximum height difference threshold are retained, and the current seed point is marked as traversed; when the flood filling is completed, it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold, if If it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold, the current segmentation mask is retained; traversing from the current seed point in row and column order, finding the next unmarked pixel point in the 2D depth image as the next seed point, and filling it with water until the end of the 2D depth image is traversed; calculating the minimum bounding rectangle and the average depth for each retained segmentation mask, determining the segmentation masks with similar average depth values ​​and intersections of the minimum bounding rectangles as the same object, and merging them to obtain a salient object mask; generating a salient object point cloud according to the obtained salient object mask as the point cloud to be registered.

[0007] Optionally, for the preset template point cloud and the point cloud to be registered, respectively constructing an FPFH feature set with edge weighting, including: for each point in the preset template point cloud and the point cloud to be registered, calculating the FPFH feature with edge weighting according to the following formula, to obtain the FPFH feature set with edge weighting of the preset template point cloud and the point cloud to be registered:

[0008]

[0009] Among them, P i is any point in the preset template point cloud and the point cloud to be registered, w k For point P i Its nearest neighbor point P k The Euclidean distance between i ) and SPFH(P k ) are points P i and its neighbor point P k The simplified point feature histogram of For point P i The neighbor point P k The edge constraint, θ th To set the edge constraint threshold.

[0010] Optionally, the edge constraints of any point in the preset template point cloud and the point cloud to be registered are determined in the following manner: for the neighboring points of each point in the template point cloud and the point cloud to be registered, project to the tangent plane of the point; the tangent plane of the point refers to the plane perpendicular to the local surface normal vector of the point; taking the current point as the corner point, connect the projected neighboring points with the current point in pairs in a counterclockwise order to form n angles, where n is equal to the number of neighboring points of the current point minus 1; the largest angle among the n angles formed is determined as the edge constraint of the current point.

[0011] Optionally, point cloud registration is performed based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered, including: based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered, using the SAC_IA algorithm to establish an initial matching relationship between the template point cloud and the point cloud to be registered; based on the established initial matching relationship between the template point cloud and the point cloud to be registered, using the ICP algorithm to perform precise registration of the template point cloud and the point cloud to be registered.

[0012] On the other hand, the present invention provides a point cloud registration device, including: a segmentation unit, used to use a flood filling algorithm to perform salient object segmentation on a 2D depth image of an object to be registered, and generate a point cloud to be registered; a construction unit, used to construct an FPFH feature set with edge weighting for a preset template point cloud and the point cloud to be registered, respectively; and a registration unit, used to perform point cloud registration based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered.

[0013] Optionally, the segmentation unit uses a flood filling algorithm to perform salient object segmentation on the 2D depth image of the object to be registered to generate a point cloud to be registered, including: setting the pixel point (0, 0) in the 2D depth image as the initial seed point, and setting a maximum height difference threshold for flood filling and a minimum area threshold for segmentation mask retention; starting flood filling from the initial seed point according to the maximum height difference threshold, retaining neighbor points whose height difference with the current seed point is less than the maximum height difference threshold, and marking the current seed point as traversed; when the flood filling is completed, judging whether the area of ​​the current segmentation mask is greater than the minimum area threshold value, if it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold, the current segmentation mask is retained; traversing from the current seed point in row and column order, finding the next unmarked pixel point in the 2D depth image as the next seed point, and filling it with water until the end of the 2D depth image is traversed; calculating the minimum bounding rectangle and the average depth for each retained segmentation mask, determining the segmentation masks with similar average depth values ​​and intersections of the minimum bounding rectangles as the same object, and merging them to obtain a salient object mask; generating a salient object point cloud according to the obtained salient object mask as the point cloud to be registered.

[0014] Optionally, the construction unit constructs edge-weighted FPFH feature sets for the preset template point cloud and the point cloud to be registered, respectively, including: calculating the edge-weighted FPFH feature according to the following formula for each point in the preset template point cloud and the point cloud to be registered, to obtain the edge-weighted FPFH feature sets of the preset template point cloud and the point cloud to be registered:

[0015]

[0016] Among them, P i is any point in the preset template point cloud and the point cloud to be registered, w k For point P i Its nearest neighbor point P k The Euclidean distance between i ) and SPFH(P k ) are points P i and its neighbor point P k The simplified point feature histogram of For point P i The neighbor point P k The edge constraint, θ th To set the edge constraint threshold.

[0017] Optionally, the edge constraints of any point in the preset template point cloud and the point cloud to be registered are determined in the following manner: for the neighboring points of each point in the template point cloud and the point cloud to be registered, project to the tangent plane of the point; the tangent plane of the point refers to the plane perpendicular to the local surface normal vector of the point; taking the current point as the corner point, connect the projected neighboring points with the current point in pairs in a counterclockwise order to form n angles, where n is equal to the number of neighboring points of the current point minus 1; the largest angle among the n angles formed is determined as the edge constraint of the current point.

[0018] Optionally, the registration unit performs point cloud registration based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered, including: based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered, using the SAC_IA algorithm to establish an initial matching relationship between the template point cloud and the point cloud to be registered; based on the established initial matching relationship between the template point cloud and the point cloud to be registered, using the ICP algorithm to perform precise registration of the template point cloud and the point cloud to be registered.

[0019] Another aspect of the present invention provides a storage medium having a computer program stored thereon, wherein the program implements the steps of any of the aforementioned methods when executed by a processor.

[0020] In another aspect, the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the aforementioned methods when executing the program.

[0021] In another aspect, the present invention provides a computer device, comprising any of the aforementioned point cloud registration devices.

[0022] According to the technical solution of the present invention, in view of the complex environment, many types of objects, large posture changes, high positioning difficulty and high beat requirements in industrial scenarios, the depth image collected by the 3D sensor is adaptively flood-filled and segmented, and significant objects are identified and merged. The objects to be matched and the initial matching center are quickly confirmed, thereby reducing the computing volume of the point cloud and improving the extraction accuracy of the key features of the point cloud.

[0023] According to the technical solution of the present invention, in order to further locate the precise posture of the target object, the edge-weighted FPFH feature descriptor integrating the AC algorithm idea is used to construct the corresponding feature point set, and then the initial posture of the two groups of point clouds are matched based on the random sampling consistency algorithm, and finally the ICP algorithm is used for iterative approximation to obtain the precise matching relationship between the two point clouds. The edge-weighted FPFH feature can effectively focus on the edge features of the object, and can effectively avoid the local optimal matching results caused by the large initial posture difference of the ICP precise registration algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0025] Figure 1 is a method schematic diagram of an embodiment of the point cloud registration method provided by the present invention;

[0026] Figure 2 A flowchart of a specific implementation of the steps of using a flood filling algorithm to segment a depth image of an object to be registered and generate a point cloud to be registered according to the present invention is shown;

[0027] Figure 3a All segmentation masks are shown;

[0028] Figure 3b The final merged salient object mask is shown;

[0029] Figure 4 A flowchart of a specific implementation of the step of constructing an FPFH feature set with edge weighting for a preset template point cloud and the point cloud to be registered according to the present invention is shown;

[0030] Figure 5 A schematic diagram of determining edge constraints is shown;

[0031] Figure 6 A flow chart of a specific implementation of the step of performing point cloud registration based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered according to the present invention is shown;

[0032] Figure 7 It is a flowchart of a specific embodiment of the point cloud registration method provided by the present invention;

[0033] Figure 8 A flowchart showing a specific implementation of depth image salient object segmentation according to the present invention is shown;

[0034] Fig. 9 A flow chart for performing point cloud registration is shown;

[0035] Fig.10 It is a structural block diagram of an embodiment of the point cloud registration device provided by the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] The invention provides a point cloud registration method.

[0039] Figure 1 It is a method schematic diagram of an embodiment of the point cloud registration method provided by the present invention.

[0040] like Figure 1 As shown, according to one embodiment of the present invention, the point cloud registration method at least includes step S110, step S120 and step S130.

[0041] Step S110 , using a flood filling algorithm to segment the 2D depth image of the object to be registered into salient objects, and generating a point cloud to be registered.

[0042] Figure 2 The flowchart of a specific implementation of the step of using the flood filling algorithm to segment the depth image of the object to be registered and generate the point cloud to be registered is shown in the present invention. Figure 2 As shown, in a specific implementation, step S110 includes steps S111 to S114.

[0043] Step S111 : setting the pixel point (0, 0) in the 2D depth image as the initial seed point, and setting the maximum height difference threshold for flood filling and the minimum area threshold for segmentation mask retention.

[0044] The maximum height threshold h is used in the flood filling algorithm to determine which neighboring points should be included; the segmentation mask retains the minimum area threshold a to filter out too small segmentation areas, which may be noise or irrelevant objects.

[0045] Step S112, starting from the initial seed point to perform flood filling according to the maximum height difference threshold, retaining neighboring points whose height difference with the current seed point is less than the maximum height difference threshold, and marking the current seed point as traversed.

[0046] Step S113: when the flood filling is completed, it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold; if it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold, the current segmentation mask is retained.

[0047] Step S114, traverse from the current seed point in row and column order, find the next unmarked pixel point in the 2D depth image as the next seed point, and fill it with water until the end of the 2D depth image is traversed.

[0048] Specifically, flood filling is performed starting from the initial seed point, and eight connected directions (i.e., eight directions of up, down, left, right, upper left, upper right, lower left, and lower right) are used to diffuse outward during filling, and neighbor points whose height difference with the current seed point is less than the maximum height difference threshold h are retained, and the current seed point is marked as traversed. When the diffusion is completed and the flood filling is completed, the area of ​​the current segmentation mask is calculated, and it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold a. If it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold a, the current segmentation mask is retained. From the current seed point, the traversal is continued in row and column order to find the next unmarked point as the next seed point for flood filling. When the filling is completed, the area of ​​the current segmentation mask is calculated, and it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold a. If it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold a, the current segmentation mask is retained, and so on, until the end of the depth image is traversed, and the automatic selection of seed points can be realized.

[0049] Step S115 , calculating the minimum bounding rectangle and average depth of each retained segmentation mask, determining segmentation masks with similar average depth values ​​and intersections of minimum bounding rectangles as the same object, and merging them to obtain a salient object mask.

[0050] Specifically, after the entire depth image is traversed, the minimum bounding rectangle and average depth of all retained segmentation masks are calculated, and segmentation masks with similar average depths and intersections of minimum bounding rectangles are determined to be the same object, and are merged to obtain a salient object mask. Figure 3a and Figure 3bAs shown, Figure 3a All segmentation masks are shown. Figure 3b An example of the final merged salient object mask is shown.

[0051] Step S116: generating a salient object point cloud according to the obtained salient object mask as a point cloud to be registered.

[0052] Specifically, according to the obtained salient object mask, the corresponding pixel points in the depth image are extracted and converted to generate a three-dimensional point cloud, which is the salient object point cloud to be registered.

[0053] Step S120: constructing an FPFH feature set with edge weighting for the preset template point cloud and the point cloud to be registered.

[0054] Specifically, for each point in the preset template point cloud and the point cloud to be registered, the FPFH feature with edge weighting is calculated according to the following formula to obtain the FPFH feature set of the preset template point cloud and the point cloud to be registered with edge weighting;

[0055]

[0056] Among them, P i is any point in the preset template point cloud and the point cloud to be registered, w k For point P i Its nearest neighbor point P k The Euclidean distance between i ) and SPFH(P k ) are points P i and its neighbor point P k The simplified point feature histogram of For point P i The neighbor point P k The edge constraint, θ th To set the edge constraint threshold.

[0057] Before constructing the FPFH feature set with edge weighting, the template point cloud and the point cloud to be registered are first subjected to the same preprocessing, including at least one of downsampling, outlier removal, and normal vector calculation.

[0058] Figure 4 The flowchart of a specific implementation of the step of constructing an FPFH feature set with edge weighting for a preset template point cloud and the point cloud to be registered according to the present invention is shown. Figure 4 As shown, in a specific implementation, step S120 includes steps S121 to S125.

[0059] Specifically, for the template point cloud and the point cloud to be registered, respectively, an FPFH feature set with edge weighting is constructed through the following steps.

[0060] Step S121, calculating the FPFH feature of the neighborhood distance constraint for each point in the template point cloud and the point cloud to be registered;

[0061] Specifically, for each point P i , the FPFH (Fast Point Feature Histogram) feature of its neighborhood distance constraint is calculated using the following formula:

[0062]

[0063] In the formula, Represents the weight coefficient, which is used to compensate for the unevenness of sampling points. k Indicates the center point P i and its neighbor point P k The Euclidean distance between the two represents the center point P i The kth neighboring point P k The scope of influence, SPFH(P i ) and SPFH(P k ) are P i and its neighboring point P k Simplified point feature histogram of .

[0064] After obtaining the FPFH feature of the neighborhood distance constraint of each point, the edge constraint of each point is determined, and the edge constraint of the neighboring points of each point in the point cloud is added to its FPFH feature to obtain the edge-weighted FPFH feature of each point in the template point cloud and the point cloud to be registered.

[0065] The edge constraints of any point in the preset template point cloud and the point cloud to be registered are determined through the following steps S122 to S124.

[0066] Step S122 , projecting the neighboring points of each point in the template point cloud and the point cloud to be registered onto the tangent plane of the point.

[0067] The tangent plane of a point is a plane perpendicular to the local surface normal vector of the point. Specifically, the tangent plane of any point in the point cloud is defined as: a plane perpendicular to the local surface normal vector of the point. The normal vector can be estimated by mathematical methods based on the geometric positions of the point's neighboring points, reflecting the local surface characteristics of the point cloud at the point position.

[0068] The steps for projecting the neighboring points of any point p onto the tangent plane of that point are:

[0069] Determine the normal vector: Use the least squares method to estimate the normal vector of the point. Approximate the K nearest neighbor points of the point on a local plane, and then fit the plane equation using the least squares method to find the normal vector.

[0070] Construct a local coordinate system: Select the normal vector of the point as the z-axis of the local coordinate system. Then define the x-axis and y-axis so that the x, y, and z axes form a right-handed coordinate system. This can be done by cross-multiplying the normal vector and any non-parallel vector.

[0071] Projection calculation: For each point in the neighborhood (neighboring points), calculate their vectors to the reference point (i.e. the selected center point). Then, use the rotation matrix to transform these vectors from the global coordinate system to the local coordinate system (x, y, z). The rotation matrix is ​​usually composed of the three axes of the local coordinate system.

[0072] Get the projection point on the tangent plane: In the local coordinate system, the z-axis component represents the perpendicular distance between the vector and the tangent plane. Ignore the z-coordinate value of the projection point and only keep the x and y coordinate values ​​to get the projection of the neighborhood point on the tangent plane.

[0073] The neighboring points of any point are projected onto the tangent plane of the point. This process actually converts the neighboring point information in three-dimensional space into a two-dimensional plane, making it easier to perform subsequent processing and analysis.

[0074] Step S123, taking the current point as a corner point, connect the projected neighboring points with the current point in pairs in a counterclockwise order to form n angles, where n is equal to the number of neighboring points of the current point minus 1.

[0075] Step S124: determining the largest angle among the series of angles formed as the edge constraint of the current point.

[0076] Specifically, for each point P i Neighborhood (Neighborhood Point P i The neighboring points of the projected point are projected onto the tangent plane. i As corner points, in counterclockwise order, two by two with P i Connect to form a series of angles θ = {θ1, θ2, …, θ n},in, That is, the number of angles formed is equal to the number of neighboring points minus 1. The current point P i The neighboring point is point P i As the center of the circle, a sphere is constructed according to the given radius r (r is related to the downsampling in the preprocessing), and the other points in the sphere are P i The neighboring point of point P i The edge constraints are It can be expressed as max(Θ), The larger the value, the point P i It is more likely to be an edge point.

[0077] Figure 5 A schematic diagram of determining edge constraints is shown. Figure 5 As shown, the edge constraints include both edge points and non-edge points. Figure 5 As shown, the neighboring points after projection are expressed as P i As corner points, in counterclockwise order, two by two with P i Connect and take the largest angle θ as point P i In point cloud matching, edge points are more sensitive and can effectively represent the contour features of objects, which helps to improve the accuracy of coarse matching and avoid the subsequent ICP from falling into the local optimum.

[0078] Step S125 , for each point, adding the edge constraint of each of its neighboring points to its FPFH feature, so as to obtain the edge-weighted FPFH feature of each point in the template point cloud and the point cloud to be registered.

[0079] After obtaining the edge constraints of each point in the template point cloud and the point cloud to be registered, set the edge constraint threshold θ th Based on the edge constraint θ between neighboring points, the FPFH feature operator is improved to make its edge features more sensitive. The final edge-weighted FPFH feature description is:

[0080]

[0081] in, For point P i The neighbor point P k edge constraints.

[0082] Step S130 , performing point cloud registration based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered.

[0083] Figure 6 The flowchart of a specific implementation of the step of performing point cloud registration based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered according to the present invention is shown. Figure 6 As shown, in a specific implementation, step S130 includes steps S131 to S132.

[0084] Step S131, based on the constructed FPFH feature point set with edge weighting of the template point cloud and the point cloud to be registered, the SAC_IA algorithm (sampling consistency initial registration algorithm) is used to establish an initial matching relationship between the template point cloud and the point cloud to be registered.

[0085] Specifically, n (n>=3) pairs of matching features (having similar FPFH features) are randomly selected from the edge-weighted FPFH feature point sets of the template point cloud and the point cloud to be registered, and the rotation and displacement of the point cloud to be registered relative to the template point cloud in this matching situation are solved by SVD (singular value decomposition), and the error between the corresponding edge-weighted FPFH features is calculated at this time. The iteration is repeated until the preset error threshold is met, and the obtained rotation and displacement of the point cloud to be registered relative to the template point cloud is used as the initial matching relationship.

[0086] Step S132 , based on the established initial matching relationship between the template point cloud and the point cloud to be registered, an ICP (Iterative Closest Point) algorithm is used to perform precise registration of the template point cloud and the point cloud to be registered.

[0087] Specifically, according to the initial matching relationship between the point cloud to be registered and the template point cloud, a source point cloud P is obtained; the nearest neighbor point (p) between the source point cloud P and the point cloud to be registered Q is determined. i ,q i ), for each point pi in the source point cloud, find the closest corresponding point qi in the point cloud to be registered to form an initial corresponding point pair (p i ,q i ), that is, the nearest neighbor point; calculate the optimal matching parameters R and t to minimize the error function. The error function E(R,t) is:

[0088]

[0089] Among them, n is the number of nearest neighbor pairs, p i is a point in the source point cloud P, q i is the point cloud Q to be registered and p i The corresponding nearest point, R is the rotation matrix, and t is the translation vector.

[0090] In order to clearly illustrate the technical solution of the present invention, the execution process of the point cloud registration method provided by the present invention is described below with reference to several specific embodiments.

[0091] Figure 7 FIG. 1 is a flow chart of a specific embodiment of the point cloud registration method provided by the present invention. Figure 7As shown, a 2D depth image is input (the picture is for illustration only), salient objects are segmented, and the entire image is traversed and flood filled to obtain a segmentation mask. The segmentation mask that meets the requirements is selected, and the same object judgment is performed and merged into a salient object mask to generate a point cloud to be registered. The template point cloud and the salient object point cloud to be registered are quickly registered, and the edge constraints of each point in the template point cloud and the point cloud to be registered are determined. Then, the edge constraints of the neighboring points of each point are added to its FPFH feature, thereby obtaining the edge-weighted FPFH feature set of the template point cloud and the point cloud to be registered. Point cloud registration is performed based on the edge-weighted FPFH feature point set of the template point cloud and the point cloud to be registered. The SAC_IA algorithm is used to establish a preliminary correspondence between point clouds, and the ICP fine registration method is used to achieve the final precise positioning.

[0092] Figure 8 A flowchart of a specific implementation of depth image salient object segmentation according to the present invention is shown.

[0093] Step S1, input a 2D depth image, set pixel (0, 0) as the initial seed point, set the maximum height difference threshold h for flood filling, and set the minimum area threshold a for segmentation mask retention.

[0094] Step S2: Use the current seed point to perform flood filling on the depth image. When filling, diffuse outward in the eight-connected direction, retain neighbor points with a height difference less than h, and mark them as traversed. When the diffusion is completed, judge the current segmentation mask and retain it if the area is greater than a.

[0095] Step S3, continue traversing from the current seed point in row and column order, looking for the next unmarked point, and perform flood filling. If no new seed point is found after traversing to the end of the image, proceed to step S4, otherwise, update the new seed point and return to step S2;

[0096] Step S4: After the entire image is traversed, the minimum bounding rectangle and average depth are calculated for all retained segmentation masks, and segmentation masks with similar average depths and intersections of minimum bounding rectangles are determined to be the same object, and are merged to obtain a salient object mask.

[0097] Step S5, generating a point cloud according to the mask, and outputting the point cloud to be registered.

[0098] Fig. 9 The flowchart of point cloud registration is shown in FIG. Fig. 9 As shown in FIG. 1 , the template point cloud is registered with the salient object point cloud obtained in the previous step to achieve fast and precise positioning of the object.

[0099] The template point cloud and the point cloud to be registered are subjected to the same preprocessing, including downsampling, outlier removal, and normal vector calculation. The edge-weighted point cloud FPFH feature set of the template point cloud and the point cloud to be registered is constructed based on the AC algorithm. The SAC_IA (sampling consistency initial registration algorithm) algorithm is used to establish the preliminary correspondence between the template point cloud and the point cloud to be registered. The ICP (iterative closest point algorithm) precise registration method is used to achieve the final precise positioning.

[0100] The invention provides a point cloud registration method.

[0101] Fig.10 1 is a structural block diagram of an embodiment of the point cloud registration device provided by the present invention. Fig.10 As shown, the point cloud registration device 100 includes: a segmentation unit 110 , a construction unit 120 and a registration unit 130 .

[0102] The segmentation unit 110 is used to perform salient object segmentation on the 2D depth image of the object to be registered by using a flood filling algorithm to generate a point cloud to be registered.

[0103] In a specific implementation, the segmentation unit 110 uses a flood filling algorithm to perform significant object segmentation on the 2D depth image of the object to be registered, and generates a point cloud to be registered, including: setting the pixel point (0, 0) in the 2D depth image as the initial seed point, and setting the maximum height difference threshold of the flood filling and the minimum area threshold retained by the segmentation mask; starting from the initial seed point according to the maximum height difference threshold, performing flood filling, retaining neighbor points whose height difference with the current seed point is less than the maximum height difference threshold, and marking the current seed point as traversed; when the flood filling is completed, judging whether the area of ​​the current segmentation mask is greater than the Minimum area threshold, if it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold, the current segmentation mask is retained; traversing from the current seed point in row and column order, looking for the next unmarked pixel point in the 2D depth image as the next seed point, and filling it with water until the end of the 2D depth image is traversed; calculating the minimum bounding rectangle and the average depth for each retained segmentation mask, determining the segmentation masks with similar average depth values ​​and intersections of the minimum bounding rectangles as the same object, and merging them to obtain a salient object mask; generating a salient object point cloud based on the obtained salient object mask as the point cloud to be registered.

[0104] The maximum height threshold h is used to determine which neighboring points should be included in the flood filling algorithm; the segmentation mask retains the minimum area threshold a to filter out too small segmented areas, which may be noise or irrelevant objects. Specifically, flood filling is performed starting from the initial seed point, and eight connected directions (i.e., eight directions of up, down, left, right, upper left, upper right, lower left, and lower right) are used to diffuse outward during filling, and neighboring points whose height difference with the current seed point is less than the maximum height difference threshold h are retained, and the current seed point is marked as traversed. When the diffusion is completed and the flood filling is completed, the area of ​​the current segmentation mask is calculated, and it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold a. If it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold a, the current segmentation mask is retained. Continue traversing from the current seed point in row and column order, find the next unmarked point as the next seed point, and perform flood filling. When the filling is completed, calculate the area of ​​the current segmentation mask, and determine whether the area of ​​the current segmentation mask is greater than the minimum area threshold a. If the area of ​​the current segmentation mask is greater than the minimum area threshold a, retain the current segmentation mask, and so on, until the end of the depth image is traversed.

[0105] After the entire depth image is traversed, the minimum bounding rectangle and average depth of all retained segmentation masks are calculated, and segmentation masks with similar average depths and intersections of minimum bounding rectangles are determined to be the same object, and merged to obtain a salient object mask. Figure 3a and Figure 3b As shown, Figure 3a All segmentation masks are shown. Figure 3b An example of the final merged salient object mask is shown. The pixels in the obtained salient object mask are converted from two dimensions to three dimensions, thereby generating a three-dimensional point cloud, which is the salient object point cloud to be registered.

[0106] The construction unit 120 is used to construct FPFH feature sets with edge weighting for the preset template point cloud and the point cloud to be registered.

[0107] Specifically, for each point in the preset template point cloud and the point cloud to be registered, the FPFH feature with edge weighting is calculated according to the following formula to obtain the FPFH feature set of the preset template point cloud and the point cloud to be registered with edge weighting;

[0108]

[0109] Among them, P i is any point in the preset template point cloud and the point cloud to be registered, w k For point P i Its nearest neighbor point P kThe Euclidean distance between i ) and SPFH(P k ) are points P i and its neighbor point P k The simplified point feature histogram of For point P i The neighbor point P k The edge constraint, θ th To set the edge constraint threshold.

[0110] Before constructing the FPFH feature set with edge weighting, the template point cloud and the point cloud to be registered are first subjected to the same preprocessing, including at least one of downsampling, outlier removal, and normal vector calculation.

[0111] Specifically, the construction unit 120 constructs an FPFH feature set with edge weighting for the template point cloud and the point cloud to be registered through the following steps.

[0112] Step S121, calculating the FPFH feature of the neighborhood distance constraint for each point in the template point cloud and the point cloud to be registered.

[0113] Specifically, for each point P i , the FPFH (Fast Point Feature Histogram) feature of its neighborhood distance constraint is calculated using the following formula:

[0114]

[0115] In the formula, Represents the weight coefficient, which is used to compensate for the unevenness of sampling points. k Indicates the center point P i and its neighbor point P k The Euclidean distance between the two represents the center point P i The kth neighboring point P k The scope of influence, SPFH(P i ) and SPFH(P k ) are P i and its neighboring point P k Simplified point feature histogram of .

[0116] After obtaining the FPFH feature of the neighborhood distance constraint of each point, the edge constraint of each point is determined, and the edge constraint of the neighboring points of each point in the point cloud is added to its FPFH feature to obtain the edge-weighted FPFH feature of each point in the template point cloud and the point cloud to be registered.

[0117] The edge constraints of any point in the preset template point cloud and the point cloud to be registered are determined through the following steps S122 to S124.

[0118] Step S122 , projecting the neighboring points of each point in the template point cloud and the point cloud to be registered onto the tangent plane of the point.

[0119] The tangent plane of a point is a plane perpendicular to the local surface normal vector of the point. Specifically, the tangent plane of any point in the point cloud is defined as: a plane perpendicular to the local surface normal vector of the point. The normal vector can be estimated by mathematical methods based on the geometric positions of the point's neighboring points, reflecting the local surface characteristics of the point cloud at the point position.

[0120] The steps for projecting the neighboring points of any point p onto the tangent plane of that point are:

[0121] Determine the normal vector: Use the least squares method to estimate the normal vector of the point. Approximate the K nearest neighbor points of the point on a local plane, and then fit the plane equation using the least squares method to find the normal vector.

[0122] Construct a local coordinate system: Select the normal vector of the point as the z-axis of the local coordinate system. Then define the x-axis and y-axis so that the x, y, and z axes form a right-handed coordinate system. This can be done by cross-multiplying the normal vector and any non-parallel vector.

[0123] Projection calculation: For each point in the neighborhood (neighboring points), calculate their vectors to the reference point (i.e. the selected center point). Then, use the rotation matrix to transform these vectors from the global coordinate system to the local coordinate system (x, y, z). The rotation matrix is ​​usually composed of the three axes of the local coordinate system.

[0124] Get the projection point on the tangent plane: In the local coordinate system, the z-axis component represents the perpendicular distance between the vector and the tangent plane. Ignore the z-coordinate value of the projection point and only keep the x and y coordinate values ​​to get the projection of the neighborhood point on the tangent plane.

[0125] The neighboring points of any point are projected onto the tangent plane of the point. This process actually converts the neighboring point information in three-dimensional space into a two-dimensional plane, making it easier to perform subsequent processing and analysis.

[0126] Step S123, taking the current point as a corner point, connect the projected neighboring points with the current point in pairs in a counterclockwise order to form n angles, where n is equal to the number of neighboring points of the current point minus 1.

[0127] Step S124: determining the largest angle among the series of angles formed as the edge constraint of the current point.

[0128] Specifically, for each point P i Neighborhood (Neighborhood Point P iThe neighboring points of the projected point are projected onto the tangent plane. i As corner points, in counterclockwise order, two by two with P i Connect to form a series of angles θ = {θ1, θ2, …, θ n},in, That is, the number of angles formed is equal to the number of neighboring points minus 1. The current point P i The neighboring point is point P i As the center of the circle, a sphere is constructed according to the given radius r (r is related to the downsampling in the preprocessing), and the other points in the sphere are P i The neighboring point of point P i The edge constraint is θ Pi , it can be expressed as max(θ),θ Pi The larger the value, the point P i It is more likely to be an edge point.

[0129] Figure 5 A schematic diagram of determining edge constraints is shown. Figure 5 As shown, the edge constraints include both edge points and non-edge points. Figure 5 As shown, the neighboring points after projection are expressed as P i As corner points, in counterclockwise order, two by two with P i Connect and take the largest angle θ as point P i edge constraints.

[0130] Step S125 , for each point, adding the edge constraint of each of its neighboring points to its FPFH feature, so as to obtain the edge-weighted FPFH feature of each point in the template point cloud and the point cloud to be registered.

[0131] After obtaining the edge constraints of each point in the template point cloud and the point cloud to be registered, set the edge constraint threshold θ th Based on the edge constraint θ between neighboring points, the FPFH feature operator is improved to make its edge features more sensitive. The final edge-weighted FPFH feature description is:

[0132]

[0133] in, For point P i The neighbor point P k edge constraints.

[0134] The registration unit 130 is used to perform point cloud registration based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered.

[0135] In a specific embodiment, the registration unit 130 performs point cloud registration based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered, including: based on the constructed template point cloud and the edge-weighted FPFH feature point set of the point cloud to be registered, using the SAC_IA algorithm to establish an initial matching relationship between the template point cloud and the point cloud to be registered; based on the established initial matching relationship between the template point cloud and the point cloud to be registered, using the ICP algorithm to perform precise registration of the template point cloud and the point cloud to be registered.

[0136] Specifically, n (n>=3) pairs of matching features (having similar FPFH features) are randomly selected from the edge-weighted FPFH feature point sets of the template point cloud and the point cloud to be registered, and the rotation and displacement of the point cloud to be registered relative to the template point cloud in this matching situation are solved by SVD, and the error between the corresponding edge-weighted FPFH features is calculated at this time. The iteration is repeated until the preset error threshold is met, and the obtained rotation and displacement of the point cloud to be registered relative to the template point cloud is used as the initial matching relationship.

[0137] According to the initial matching relationship between the point cloud to be registered and the template point cloud, a source point cloud P is obtained; the nearest neighbor point (p i ,q i ), for each point pi in the source point cloud, find the closest corresponding point qi in the point cloud to be registered to form an initial corresponding point pair (p i ,q i ), that is, the nearest neighbor point; calculate the optimal matching parameters R and t to minimize the error function. The error function E(R,t) is:

[0138]

[0139] Among them, n is the number of nearest neighbor pairs, p i is a point in the source point cloud P, q i is the point cloud Q to be registered and p i The corresponding nearest point, R is the rotation matrix, and t is the translation vector.

[0140] The present invention also provides a storage medium corresponding to the point cloud registration method, on which a computer program is stored, and when the program is executed by a processor, the steps of any of the aforementioned methods are implemented.

[0141] The present invention also provides a computer device corresponding to the point cloud registration method, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the aforementioned methods when executing the program.

[0142] The present invention also provides a computer device corresponding to the point cloud registration device, comprising any of the aforementioned point cloud registration devices.

[0143] The background of industrial scenes is complex, the incoming materials of the assembly line are complex and changeable, and there are many interference factors. It is impossible to effectively identify them with only 2D vision. The introduction of depth information in 3D point clouds also brings a large amount of data calculation, which cannot meet the production cycle requirements. Therefore, the present invention first performs significant object segmentation on the 2D depth image, adaptively identifies and segments the objects to be matched, greatly reduces the amount of calculation for point cloud processing, and improves matching efficiency.

[0144] Point cloud registration technology has a better registration effect for objects with rich texture features. However, in actual industrial applications, most workpiece objects have fewer texture features and have a high symmetric similarity problem. The traditional FPFH feature operator is prone to reverse problems during initial positioning, which leads to local optimality during ICP registration. Therefore, the present invention proposes an optimized FPFH feature descriptor that is sensitive to edge features based on the AC (Angle Criterion) algorithm, which effectively improves the accuracy of the initial pose of the point cloud, thereby achieving precise positioning of objects in industrial scenarios.

[0145] Based on this, the solution provided by the present invention, in view of the complex environment, many types of objects, large posture changes, high positioning difficulty, and high rhythm requirements in industrial scenarios, performs adaptive flood filling segmentation on the depth images collected by the 3D sensor, identifies and merges significant objects, and quickly confirms the objects to be matched and the initial matching center, thereby reducing the computing volume of the point cloud and improving the extraction accuracy of the key features of the point cloud.

[0146] The solution provided by the present invention adopts the edge-weighted FPFH feature descriptor that integrates the AC algorithm concept to construct the corresponding feature point set, and then matches the initial poses of the two groups of point clouds based on the random sampling consistency algorithm, and finally uses the ICP algorithm for iterative approximation to obtain the precise matching relationship between the two point clouds. The edge-weighted FPFH feature can effectively focus on the edge features of the object, and can effectively avoid the local optimal matching results caused by the large initial pose differences of the ICP precise registration algorithm.

[0147] The solution provided by the present invention quickly segments the masks of connected areas with obvious height differences based on the flood filling idea, and then screens and merges the masks according to the depth information and the minimum circumscribed rectangle collision constraint, so as to adaptively identify significant objects in industrial scenes, reduce the cloud computing volume of point clouds, and improve the extraction accuracy of key features of point clouds, thereby assisting in quickly locating the initial posture of the target object.

[0148] The solution provided by the present invention, based on the traditional FPFH feature operator, adds rotation change constraints between neighboring points, constructs an FPFH optimization operator that is sensitive to edge features, accurately locates the initial pose of point cloud matching, can solve the problem of easily falling into local optimality during ICP precise alignment, and realizes fast and accurate point cloud alignment.

[0149] The functions described herein may be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions may be stored as one or more instructions or codes on a computer-readable medium or transmitted via a computer-readable medium. Other examples and implementations are within the scope and spirit of the present invention and the appended claims. For example, due to the nature of software, the functions described above may be implemented using software executed by a processor, hardware, firmware, hard wiring, or a combination of any of these. In addition, each functional unit may be integrated into a processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit.

[0150] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0151] The units described as separate components may or may not be physically separated, and the components of the control device may or may not be physical units, that is, they may be located in one place or distributed in multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0152] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the relevant technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0153] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A point cloud registration method, characterized in that: include: The 2D depth image of the object to be registered is segmented using the flood filling algorithm to generate a point cloud to be registered. For the preset template point cloud and the point cloud to be registered, constructing FPFH feature sets with edge weighting respectively; Point cloud registration is performed based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered.

2. The method according to claim 1, characterized in that The 2D depth image of the object to be registered is segmented using the flood filling algorithm to generate a point cloud to be registered, including: The pixel point (0, 0) in the 2D depth image is set as the initial seed point, and the maximum height difference threshold of the flood filling and the minimum area threshold of the segmentation mask are set; Perform flood filling from the initial seed point according to the maximum height difference threshold, retain neighboring points whose height difference with the current seed point is less than the maximum height difference threshold, and mark the current seed point as traversed; When the flood filling is completed, it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold, and if it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold, the current segmentation mask is retained; Traverse from the current seed point in row and column order, find the next unmarked pixel point in the 2D depth image as the next seed point, and fill it with water until the end of the 2D depth image is traversed; The minimum bounding rectangle and average depth are calculated for each retained segmentation mask, and the segmentation masks with similar average depth values ​​and intersections of the minimum bounding rectangles are determined to be the same object, and then merged to obtain a salient object mask; A salient object point cloud is generated according to the obtained salient object mask as a point cloud to be registered.

3. The method according to claim 1, characterized in that: For the preset template point cloud and the point cloud to be registered, FPFH feature sets with edge weighting are constructed respectively, including: For each point in the preset template point cloud and the point cloud to be registered, the FPFH feature with edge weighting is calculated according to the following formula to obtain the FPFH feature set with edge weighting of the preset template point cloud and the point cloud to be registered: Among them, P i is any point in the preset template point cloud and the point cloud to be registered, w k For point P i Its nearest neighbor point P k The Euclidean distance between i ) and SPFH(P k ) are points P i and its neighbor point P k The simplified point feature histogram of For point P i The neighbor point P k The edge constraint, θ th To set the edge constraint threshold.

4. The method according to claim 3, characterized in that The edge constraint of any point in the preset template point cloud and the point cloud to be registered is determined by: For each neighboring point of the template point cloud and the point cloud to be registered, project the point onto a tangent plane; the tangent plane of the point refers to a plane perpendicular to the local surface normal vector of the point; Take the current point as the corner point, and connect the projected neighboring points with the current point in pairs in a counterclockwise order to form n angles, where n is equal to the number of neighboring points of the current point minus 1; The largest angle among the n formed angles is determined as the edge constraint of the current point.

5. The method according to any one of claims 1 to 4, characterized in that: Point cloud registration is performed based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered, including: Based on the constructed FPFH feature point set with edge weights of the template point cloud and the point cloud to be registered, an initial matching relationship between the template point cloud and the point cloud to be registered is established using a SAC_IA algorithm; According to the established initial matching relationship between the template point cloud and the point cloud to be registered, an ICP algorithm is used to perform precise registration of the template point cloud and the point cloud to be registered.

6. A point cloud registration device, characterized in that: include: A segmentation unit, used to segment the 2D depth image of the object to be registered using a flood filling algorithm to perform salient object segmentation and generate a point cloud to be registered; A construction unit, used for constructing FPFH feature sets with edge weighting for a preset template point cloud and the point cloud to be registered; A registration unit is used to perform point cloud registration based on the constructed template point cloud and the FPFH feature point set with edge weighting of the point cloud to be registered.

7. The device according to claim 1, characterized in that The segmentation unit uses a flood filling algorithm to perform salient object segmentation on the 2D depth image of the object to be registered, and generates a point cloud to be registered, including: The pixel point (0, 0) in the 2D depth image is set as the initial seed point, and the maximum height difference threshold of the flood filling and the minimum area threshold of the segmentation mask are set; Perform flood filling from the initial seed point according to the maximum height difference threshold, retain neighboring points whose height difference with the current seed point is less than the maximum height difference threshold, and mark the current seed point as traversed; When the flood filling is completed, it is determined whether the area of ​​the current segmentation mask is greater than the minimum area threshold, and if it is determined that the area of ​​the current segmentation mask is greater than the minimum area threshold, the current segmentation mask is retained; Traverse from the current seed point in row and column order, find the next unmarked pixel point in the 2D depth image as the next seed point, and fill it with water until the end of the 2D depth image is traversed; The minimum bounding rectangle and average depth are calculated for each retained segmentation mask, and the segmentation masks with similar average depth values ​​and intersections of the minimum bounding rectangles are determined to be the same object, and then merged to obtain a salient object mask; A salient object point cloud is generated according to the obtained salient object mask as a point cloud to be registered.

8. A storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps of any method described in claims 1-5 are implemented.

9. A computer device, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any method described in claims 1-5 when executing the program, or comprises a point cloud registration device described in any one of claims 6-7.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the steps of any one of the methods of claims 1 to 5 when executed by a processor.