Point cloud registration method and device and electronic equipment

Through the point cloud registration method combined with KD-tree algorithm and Huber penalty function, the ICP algorithm has solved the problems of high initial position, long time and low accuracy in ancient building registration, achieving more efficient and higher accuracy Point cloud registration.

CN119027468APending Publication Date: 2024-11-26WSGRI SMART CITY(WUHAN) ENGINEERING TECHNOLOGY CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411115168.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The ICP algorithm in the prior art has limitations due to its high initial position requirements, long registration time and low registration accuracy for ancient buildings.

Method used

The KD-tree algorithm is used to resample the point cloud, filter the feature points and perform coarse registration. The registration quality is evaluated based on the Huber penalty function, and the rotation matrix and the translation matrix are used for precise registration. The global optimal transformation matrix is ​​obtained repeatedly iteratively.

Benefits of technology

Improve the efficiency and accuracy of point cloud registration, provide better initial position and feature point constraints, significantly improve the accuracy of corresponding point pairs, and solve the low accuracy and time of ICP algorithm in ancient building registration Long-term problem.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027468A_ABST
    Figure CN119027468A_ABST
Patent Text Reader

Abstract

The invention relates to a point cloud registration method and device and electronic equipment, and belongs to the technical field of computer vision and point cloud registration, and the point cloud registration method comprises the steps: dividing a point cloud obtained by scanning a building into a source point set # imgabs0 # and a target point set # imgabs1 #; the target point set # imgabs2 is resampled based on a KD-tree algorithm; screening from the source point set # imgabs3 # to obtain a plurality of feature points; mapping points of the feature points are found in the target point set # imgabs4 # after resampling is completed; performing coarse registration on the source point set # imgabs5 # based on the mapping points, the feature points and a Huber penalty function; searching a corresponding point set # imgabs8 #, which is closest to the source point set # imgabs7 # after the coarse registration, in the target point set # imgabs6 #; performing fine registration based on the corresponding point set # imgabs9 # to obtain a rotation matrix and a translation matrix; and completing point cloud registration based on the rotation matrix and the translation matrix. According to the method, the technical problem that the ICP algorithm in the prior art has limitation due to high initial pose requirement, long registration time and low registration precision of the ancient building is effectively solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and point cloud registration, and in particular to a point cloud registration method, device and electronic equipment. Background Art

[0003] Traditional surveying and mapping methods are of low accuracy and are prone to "secondary damage" to ancient buildings. They mainly use total stations, GPS and other instruments to obtain targets. When applied to ancient buildings, traditional surveying and mapping methods can only obtain two-dimensional results of the target, while ground-based three-dimensional laser scanning technology can obtain three-dimensional information of the measured target with detailed dimensions.

[0004] The ground 3D laser scanning technology was proposed at the end of the last century and is known as "another technological revolution in the field of surveying and mapping after GPS technology". The emergence of this technology has transformed the traditional single-point measurement into surface measurement, which has epoch-making significance for the field of surveying and mapping. Nowadays, this technology has been widely used in the restoration of ancient buildings, planning, civil engineering, digital cities, military and other fields, providing many solutions for future construction and having very important significance, but there are still many fields that are trying it out. Summary of the invention

[0005] In view of this, it is necessary to provide a point cloud registration method, device and electronic device to solve the technical problems that the ICP algorithm in the prior art has limitations due to high initial pose requirements, long registration time and low registration accuracy for ancient buildings.

[0006] In order to solve the above problems, the present invention provides a point cloud registration method, comprising: Divide the point cloud obtained by scanning the building into source point sets And the target point set ; Based on the KD-tree algorithm, the target point set Perform resampling; From the source point set A number of feature points are obtained by screening; After completing the resampling of the target point set Find the mapping point of the feature point in ; Based on mapping points, feature points and Huber penalty function, the source point set Perform coarse registration; In the target point set Search and complete the source point set after rough registration The closest corresponding point set ; Based on the corresponding point set Perform precise registration to obtain the rotation matrix and translation matrix; Point cloud registration is completed based on the rotation matrix and translation matrix.

[0007] In a possible implementation, the target point set is Perform resampling, including: Create a target point set KD-tree tree structure; Determining the center of gravity of the tree structure; Searching for the closest point to the center of gravity in each representative area in the tree structure; And replace all points in the representative area with the nearest point.

[0008] In a possible implementation, the source point set Several feature points are screened out, including: In the source point set Set a query point in Taking the query point as the center of the circle, scanning according to a preset radius to obtain the plurality of feature points; Wherein, the intervals between the plurality of feature points are greater than a preset distance threshold.

[0009] In a possible implementation, the mapping points include: a target point set points having a similar fast point feature histogram to the plurality of feature points; The calculation formula of the fast point feature histogram includes:

[0010] In the formula, represents the query point, For query point and i The Euclidean distance between the points in the domain, k represents the number of feature points, For point The neighboring points of SPFP () represents a simplified point feature histogram.

[0011] In a possible implementation, the source point set is subjected to the mapping point, the feature point and the Huber penalty function. Perform a coarse registration, including: Calculate the rigid body transformation matrix based on the singular value decomposition algorithm ; The quality of the current rigid body transformation matrix is ​​evaluated based on the Huber penalty function, and the rigid body transformation matrix with the best global quality is taken as the optimal transformation matrix.

[0012] In a possible implementation, the method of evaluating the quality of the current rigid body transformation matrix based on the Huber penalty function and taking the rigid body transformation matrix with the best global quality as the optimal transformation matrix includes: Based on the rigid body transformation matrix, the source point set Make a transformation; Calculate the transformed source point set The Huber penalty function of the corresponding point pair includes: the query point and the i Field points; The source point set with the minimum result calculated by Huber penalty function The corresponding rigid body transformation matrix is ​​used as the optimal transformation matrix; Wherein, the Huber penalty function includes:

[0013] In the formula, represents the Huber penalty function, Indicates i The Euclidean distance of the corresponding point pairs after transformation; Indicates the preset distance threshold.

[0014] In a possible implementation, the target point set Search and complete the source point set after rough registration The closest corresponding point set ,include: Based on the first formula, find the source point set The corresponding points closest to each other are obtained ; Wherein, the first formula includes:

[0015] represents the distance between each point in the point set P and Q, p is a point in the point set P, q is a point in the point set Q.

[0016] In a possible implementation, the corresponding point set Perform fine registration, including: Based on the corresponding point set Solve the rotation matrix and translation matrix; Update the source point set based on the rotation matrix and translation matrix ; Calculate the loss function; Determine whether an iteration stop condition is met, wherein the iteration stop condition includes: the current iteration number reaches a preset maximum iteration number, or the loss function is less than a preset loss threshold; If not, based on the updated source point set Continue to solve the rotation matrix and the translation matrix until the iteration stop condition is met; If yes, then output the current rotation matrix and translation matrix as the final precise registration result; Wherein, the loss function includes:

[0017] In the formula, and For iteration The rotation matrix and translation matrix of the second Represents the source point set The number of midpoints, Indicates the source point concentration i The initial coordinates of the points, Represents the corresponding point set Middle i The coordinates of the points, represents the loss function, Indicates k The loss function of the iteration is, Represents the loss threshold.

[0018] The present invention also provides a point cloud registration device, comprising: Point cloud processing module, used to divide the point cloud obtained by scanning the building into source point sets And the target point set , and, based on the KD-tree algorithm, the target point set Perform resampling; The coarse registration module is used to A number of feature points whose distance between two points is greater than the preset distance threshold are screened out, and after completing the resampling of the target point set Find the mapping points of the plurality of feature points in the source point set based on the mapping points and the feature points Perform rough registration and evaluate the registration quality based on Huber penalty function; The precise registration module is used to Search and complete the source point set after rough registration The closest corresponding point set , and, based on the corresponding point set Perform precise alignment to obtain the rotation matrix and translation matrix.

[0019] The present invention also provides an electronic device, comprising: Memory, used to store programs; A processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the point cloud registration method described in any one of the above method items.

[0020] The beneficial effects of the present invention are as follows: the present invention provides a point cloud registration method, firstly, preprocessing the point set based on the KD tree to reduce the density of the point set, facilitate registration, and improve the efficiency of point cloud registration, then, extracting feature points from the source point set and performing coarse registration, and evaluating the registration quality based on the Huber penalty function to obtain a globally optimal coarse registration transformation matrix, thereby improving the registration accuracy, and finally, performing fine registration on the basis of the coarse registration, repeatedly iterating to obtain the globally optimal transformation matrix, because the coarse registration provides a better initial posture for the fine registration, and adds feature point constraints, the correct rate of corresponding point pairs is improved, and the registration accuracy is further improved, thereby effectively solving the technical problems that the ICP algorithm in the prior art has limitations due to high initial posture requirements, long registration time, and low registration accuracy for ancient buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A method flow chart of an embodiment of a point cloud registration method provided by the present invention; Figure 2 for Figure 1 A method flow chart of an embodiment of step S102; Figure 3 for Figure 1 A method flow chart of an embodiment of step S103; Figure 4 for Figure 1 A method flow chart of an embodiment of step S105; Figure 5 for Figure 4 A method flow chart of an embodiment of step S402; Figure 6 for Figure 1 A method flow chart of an embodiment of step S107; Figure 7 A method flow chart of an embodiment of a point cloud registration device provided by the present invention; Figure 8 A schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0023] Before showing the embodiments, the following terms are explained: In the description of the embodiments of the present invention, unless otherwise specified, "multiple" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone.

[0024] The descriptions of "first", "second", etc. involved in the embodiments of the present invention are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the technical features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0025] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0026] In order to solve the above problems, Figure 1 As shown, the present invention provides a point cloud registration method, comprising: S101, dividing the point cloud obtained by scanning the building into a source point set And the target point set ; S102, based on KD-tree algorithm, the target point set Perform resampling; S103, from the source point set A number of feature points are obtained by screening; S104, after completing the resampling of the target point set Find the mapping points of the feature points; In a possible implementation, the mapping points include: a target point set Points in the dataset that have similar fast point feature histograms to several feature points; Furthermore, the calculation formula of the fast point feature histogram includes: (1) In formula (1), represents the query point, For query point and i The Euclidean distance between the points in the domain, k represents the number of feature points, For point The neighboring points of SPFP () represents a simplified point feature histogram.

[0027] S105, source point set based on mapping points, feature points and Huber penalty function Perform coarse registration; S106, in the target point set Search and complete the source point set after rough registration The closest corresponding point set ; In a possible implementation, step S106 includes: finding the source point set based on the first formula The corresponding points closest to each other are obtained ; Furthermore, the first formula includes: (2) In formula (2), represents the distance between each point in the point set P and Q, p is a point in the point set P, q is a point in the point set Q.

[0028] S107, based on corresponding point set Perform precise registration to obtain the rotation matrix and translation matrix; S108. Complete point cloud registration based on the rotation matrix and the translation matrix.

[0029] Compared with the prior art, the present invention provides a point cloud registration method. First, the point set is preprocessed based on the KD tree to reduce the density of the point set, facilitate registration, and improve the efficiency of point cloud registration. Then, feature points are extracted from the source point set and coarse registration is performed, and the registration quality is evaluated based on the Huber penalty function to obtain a globally optimal coarse registration transformation matrix to improve the registration accuracy. Finally, fine registration is performed on the basis of coarse registration, and the global optimal transformation matrix is ​​obtained through repeated iterations. Since the coarse registration provides a better initial pose for the fine registration and adds feature point constraints, the correct rate of corresponding point pairs is improved, and the registration accuracy is further improved, thereby effectively solving the technical problems of the ICP algorithm in the prior art due to high initial pose requirements, long registration time and low registration accuracy for ancient buildings, and providing new ideas for the registration of ancient buildings.

[0030] like Figure 2 In a possible implementation, step S102 includes: S201. Establishing a target point set KD-tree tree structure; S202, determining the center of gravity of the tree structure; S203, searching for the closest point to the center of gravity in each representative area in the tree structure; S204: Replace all points in the representative area with the nearest point.

[0031] like Figure 3 In a possible implementation, step S103: S301, in the source point set Set a query point in S302, taking the query point as the center of the circle, scanning according to a preset radius to obtain a number of feature points; The interval between the feature points is greater than a preset distance threshold.

[0032] like Figure 4 In a possible implementation, step S105 includes: S401. Calculate the rigid body transformation matrix based on the singular value decomposition algorithm ; S402, evaluating the quality of the current rigid body transformation matrix based on the Huber penalty function, and taking the rigid body transformation matrix with the best global quality as the optimal transformation matrix.

[0033] like Figure 5 In a possible implementation, step S402 includes: S501, source point set based on rigid body transformation matrix Make a transformation; S502, calculate the transformed source point set The Huber penalty function of the corresponding point pairs; Among them, the corresponding point pairs include: the query point and the i Field points; S503, the source point set with the minimum result calculated by Huber penalty function The corresponding rigid body transformation matrix is ​​used as the optimal transformation matrix; Wherein, the Huber penalty function includes:

[0034] In the formula, represents the Huber penalty function, Indicates i The Euclidean distance of the corresponding point pairs after transformation; Indicates the preset distance threshold.

[0035] like Figure 6 In a possible implementation, step S107 includes: S601, based on corresponding point set Solve the rotation matrix and translation matrix; S602: Update the source point set based on the rotation matrix and the translation matrix ; S603, calculating a loss function; S604, determining whether an iteration stop condition is met; Specifically, the iteration stop conditions include: the current iteration number reaches the preset maximum iteration number, or the loss function is less than the preset loss threshold; S605, if not, then based on the updated source point set Continue to solve the rotation matrix and translation matrix until the iteration stop condition is met; S606: If yes, output the current rotation matrix and translation matrix as the final precise registration result; Among them, the loss function includes: (3) In formula (3), and For iteration The rotation matrix and translation matrix of the second Represents the source point set The number of midpoints, Indicates the source point concentration i The initial coordinates of the points, Represents the corresponding point set Middle i The coordinates of the points, represents the loss function, Indicates k The loss function of the iteration is, Represents the loss threshold.

[0036] As a further technical solution, the precise registration of the present invention includes: solving the coordinates of the target point set under the reference point set , the solution formula is as follows: (4) In formula (4), is the coordinate of the point set after transformation, represents the initial coordinates of the point set, and For iteration The rotation matrix and translation matrix.

[0037] When the number of iterations reaches the maximum or When , the iteration is completed to obtain the rotation matrix and translation matrix at this time, otherwise continue to iterate. The embodiments of the present invention have achieved some positive effects during the development or use process, and indeed have great advantages over the prior art. The following content is described in conjunction with data, charts, etc. of the test process.

[0038] The door handle, rearview mirror and Chengyang Wind and Rain Bridge data were used as experimental objects, and the registration efficiency and accuracy of the algorithm of the present invention and the traditional ICP algorithm were compared horizontally and vertically. Both algorithms completed the point cloud registration, but it can be seen from the local magnified image that the registration effect of relying solely on the traditional ICP algorithm is not good. Combined with the specific data analysis in Table 1, it can be seen that compared with the ICP algorithm, the algorithm disclosed in the present invention has achieved higher registration accuracy and efficiency. In terms of registration accuracy, the registration efficiency and registration accuracy of the algorithm disclosed in the present invention are improved. The reason is that the coarse registration provides a better initial pose for the fine registration, and the feature point constraints are added to improve the accuracy of the corresponding point pairs, so that the registration accuracy is further improved.

[0039]

[0040] The present invention is based on point cloud data and utilizes computer vision, point cloud registration and other technologies to establish a point cloud registration method for complex ancient buildings based on improved ICP, which effectively solves the limitations of the ICP algorithm, such as high initial pose requirements, long registration time and low registration accuracy of ancient buildings, improves accuracy and processing efficiency, and provides new ideas for the registration of ancient buildings.

[0041] like Figure 7 The present invention further provides a point cloud registration device 70, comprising: Point cloud processing module 810 is used to divide the point cloud obtained by scanning the building into source point sets And the target point set , and, based on the KD-tree algorithm, the target point set Perform resampling; The coarse registration module 820 is used to obtain the A number of feature points whose distance between two points is greater than the preset distance threshold are screened out, and after completing the resampling of the target point set Find the mapping points of several feature points, and based on the mapping points and feature points, Perform rough registration and evaluate the registration quality based on Huber penalty function; The fine registration module 730 is used to Search and complete the source point set after rough registration The closest corresponding point set , and, based on the corresponding point set Perform precise alignment to obtain the rotation matrix and translation matrix.

[0042] like Figure 8 The present invention further provides an electronic device 80, comprising: A memory 810, used for storing programs; The processor 820 is coupled to the memory 810 and is used to execute the program stored in the memory 810 to implement the steps in the point cloud registration method described in any one of the above method items.

[0043] Figure 8 Only some of the components of the electronic device 80 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0044] In some embodiments, the processor 820 can be a central processing unit (CPU), a microprocessor or other data processing chip, which is used to run the program code or process data stored in the memory 810, such as the point cloud registration method in the present invention, or the method for generating the control strategy of the unmanned ship.

[0045] In some embodiments, the processor 820 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 820 may be local or remote. In some embodiments, the processor 820 may be implemented in a cloud platform. In one embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination thereof.

[0046] In some embodiments, the memory 810 may be an internal storage unit of the electronic device 80, such as a hard disk or memory of the electronic device 80. In other embodiments, the memory 810 may also be an external storage device of the electronic device 80, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 80.

[0047] Furthermore, the memory 810 may include both an internal storage unit of the electronic device 80 and an external storage device. The memory 810 is used to store application software installed in the electronic device 80 and various data.

[0048] In one embodiment, when the processor 820 executes the point cloud registration program in the memory 810, the following steps may be implemented: Divide the point cloud obtained by scanning the building into source point sets And the target point set ; Based on the KD-tree algorithm, the target point set Perform resampling; From the source point set A number of feature points are obtained by screening; After completing the resampling of the target point set Find the mapping points of the feature points; Based on mapping points, feature points and Huber penalty function, the source point set Perform coarse registration; In the target point set Search and complete the source point set after rough registration The closest corresponding point set ; Based on the corresponding point set Perform precise registration to obtain the rotation matrix and translation matrix; Point cloud registration is completed based on the rotation matrix and translation matrix.

[0049] Furthermore, the embodiment of the present invention does not specifically limit the type of the electronic device 80 mentioned, and the electronic device 80 may be a portable device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable devices include but are not limited to portable devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable devices may also be other portable devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 80 may not be a portable device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0050] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a computer storage medium, wherein the computer storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0051] The point cloud registration method, device and electronic device provided by the present invention are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be modifications in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A point cloud registration method, applied to the surveying and mapping of complex ancient buildings, characterized in that: include: Divide the point cloud obtained by scanning the building into source point sets And the target point set ; Based on the KD-tree algorithm, the target point set Perform resampling; From the source point set A number of feature points are obtained by screening; After completing the resampling of the target point set Find the mapping points of the feature points; Based on mapping points, feature points and Huber penalty function, the source point set Perform coarse registration; At the target point set Search and source point set after coarse registration The closest corresponding point set ; Based on the corresponding point set Perform precise registration to obtain the rotation matrix and translation matrix; Point cloud registration is completed based on the rotation matrix and translation matrix.

2. The point cloud registration method according to claim 1, characterized in that: The target point set is based on the KD-tree algorithm Perform resampling, including: Create a target point set KD-tree tree structure; Determining the center of gravity of the tree structure; Searching for the closest point in each representative area in the tree structure that is closest to the center of gravity; And replace all points in the representative area with the nearest point.

3. The point cloud registration method according to claim 1, characterized in that: The source point set Several feature points are screened out, including: In the source point set Set a query point in Taking the query point as the center of the circle, scanning according to a preset radius to obtain the plurality of feature points; Wherein, the intervals between the plurality of feature points are greater than a preset distance threshold.

4. The point cloud registration method according to claim 3, characterized in that: The mapping points include: target point set points having a similar fast point feature histogram to the plurality of feature points; The calculation formula of the fast point feature histogram includes: In the formula, represents the query point, k represents the number of feature points, The query point and i The Euclidean distance between the points in the domain, For point The neighboring points of SPFP () represents a simplified point feature histogram.

5. The point cloud registration method according to claim 4, characterized in that: The source point set is based on the mapping point, the feature point and the Huber penalty function. Perform a coarse registration, including: Calculate the rigid body transformation matrix based on the singular value decomposition algorithm ; The quality of the current rigid body transformation matrix is ​​evaluated based on the Huber penalty function, and the rigid body transformation matrix with the best global quality is taken as the optimal transformation matrix.

6. The point cloud registration method according to claim 5, characterized in that: The method of evaluating the quality of the current rigid body transformation matrix based on the Huber penalty function and taking the rigid body transformation matrix with the best global quality as the optimal transformation matrix includes: Based on the rigid body transformation matrix, the source point set Make a transformation; Calculate the transformed source point set The Huber penalty function of the corresponding point pair includes: the query point and the i Field points; The source point set with the minimum result calculated by Huber penalty function The corresponding rigid body transformation matrix is ​​used as the optimal transformation matrix; Wherein, the Huber penalty function includes: In the formula, represents the Huber penalty function, Indicates i The Euclidean distance of the corresponding point pairs after transformation; Indicates the preset distance threshold.

7. The point cloud registration method according to claim 1, characterized in that: The target point set Search and source point set after coarse registration The closest corresponding point set ,include: Based on the first formula, find the source point set The corresponding points closest to each other are obtained ; Wherein, the first formula includes: represents the distance between each point in the point set P and Q, p is a point in the point set P, q is a point in the point set Q.

8. The point cloud registration method according to claim 1, characterized in that: Based on the corresponding point set Perform fine registration, including: Based on the corresponding point set Solve the rotation matrix and translation matrix; Update the source point set based on the rotation matrix and translation matrix ; Calculate the loss function; Determine whether an iteration stop condition is met, wherein the iteration stop condition includes: the current iteration number reaches a preset maximum iteration number, or the loss function is less than a preset loss threshold; If not, based on the updated source point set Continue to solve the rotation matrix and the translation matrix until the iteration stop condition is met; If yes, then output the current rotation matrix and translation matrix as the final precise registration result; Wherein, the loss function includes: In the formula, and For iteration The rotation matrix and translation matrix of the second Represents the source point set The number of midpoints, Indicates the source point concentration i The initial coordinates of the points, Represents the corresponding point set Middle i The coordinates of the points, represents the loss function, Indicates k The loss function of the iteration is, Represents the loss threshold.

9. A point cloud registration device, characterized in that: include: Point cloud processing module, used to divide the point cloud obtained by scanning the building into source point sets And the target point set , and, based on the KD-tree algorithm, the target point set Perform resampling; The coarse registration module is used to Several feature points are screened out, and the target point set is completed after the resampling Find the mapping points of the plurality of feature points in the source point set based on the mapping points and the feature points Perform rough registration and evaluate the registration quality based on Huber penalty function; The precise registration module is used to Search and source point set after coarse registration The closest corresponding point set , and, based on the corresponding point set Perform precise alignment to obtain the rotation matrix and translation matrix.

10. An electronic device, characterized in that: include: Memory, used to store programs; A processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the point cloud registration method described in any one of claims 1 to 7.

Citation Information

Cited By

  • Robust measurable modeling method integrating three-dimensional reconstruction and geographic registration

    CN121883726A

  • A robust measurable modeling method fusing three-dimensional reconstruction and geographic registration

    CN121883726B