A Bidirectional Nearest Neighbor Similarity Ratio Algorithm Excluding Neighborhood

Abstract

The present invention discloses a bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood, which is used to construct the correspondence between the model point cloud and the scene point cloud, and includes the following steps: S1, input key points and descriptors; S2, preliminarily match the correspondence; S3, apply bidirectional neighborhood constraints; S4, bidirectionally implement the Nannsr algorithm; S5, output corresponding point pairs. The present invention constructs an initial corresponding point pair set by similarity matching, and then preliminarily excludes the wrong corresponding relationship in the point pair set by applying bidirectional neighborhood constraints. Finally, the Nannsr algorithm is implemented on the scene and model key points respectively, and the point pairs with uniqueness in both the scene and the model are extracted as the final output corresponding point pairs, thereby constructing a robust corresponding point pair construction method. The performance of the algorithm is significantly better than the existing corresponding point pair construction algorithm, avoiding the problem that the uniqueness of the point pair cannot be correctly judged, and has better performance even on point cloud data interfered by noise, occlusion and data anomalies.

Description

Technical Field

[0001] The invention relates to the technical field of image data processing, in particular to a bidirectional nearest neighbor similarity ratio algorithm excluding neighborhood. Background Art

[0002] In the field of 3D computer vision, building correspondences in point clouds is widely used in 3D reconstruction, 3D object recognition, point cloud registration, shape retrieval and positioning. Building a good correspondence is crucial for the subsequent successful estimation of the transformation relationship between two or more point clouds.

[0003] Existing methods for constructing corresponding point pairs mainly include similarity score (SS), double similarity score (BSS), and nearest neighbors similarity ratio (NNSR) algorithms. SS calculates the similarity between feature descriptors on two point clouds and selects corresponding point pairs with high similarity to establish corresponding relationships. The similarity is determined by the feature distance between descriptors. However, SS is easily affected by interference such as noise, occlusion, and overlapping areas, thereby outputting incorrect corresponding relationships. BSS is similar to SS, except that the BSS algorithm imposes a bidirectional constraint on the initial corresponding relationship. This constraint retains the best matching correspondence between the model and the scene feature descriptor, and then selects corresponding point pairs with high similarity from the retained correspondence to establish the corresponding relationship. However, the BSS algorithm is too strict and can easily exclude correct corresponding relationships. In order to obtain highly unique corresponding relationships, NNSR penalizes the corresponding relationship by the ratio of the nearest and second closest distances in the feature space, so that the selected point pairs have more unique features and are not easily affected by interference. However, NNSR cannot judge the correctness of the correspondence between the first and second feature neighbors. Due to the limitations of existing methods, constructing an accurate correspondence set is still a challenging task.

[0004] Existing point pair construction algorithms either easily exclude correct correspondences or are not robust enough to deal with noise, clutter, occlusion and other interferences, and are difficult to fully adapt to different types of point cloud data. These limitations have affected the adaptability and stability of the algorithm in complex environments to a certain extent. Summary of the invention

[0005] The present invention proposes an improved nearest neighbor similarity ratio method based on NNSR, which is called the nearest neighbor similarity ratio on non-adjacent region (Nannsr) algorithm. By implementing the Nannsr algorithm on the scene and model key points respectively, point pairs that are unique on both the scene and the model are extracted as the final output corresponding point pairs. By implementing bidirectional neighborhood constraints and bidirectional Nannsr algorithms, a robust corresponding point pair construction method is formed, which is called the double nearest neighbors similarity ratio on non-adjacent region (NaDnnsr) algorithm, which avoids the problem that the uniqueness of point pairs cannot be correctly judged, so as to solve the problem that the existing algorithms are either prone to mistakenly exclude the correct corresponding relationship, or are not robust enough when dealing with interference such as noise, clutter and occlusion, and are difficult to fully adapt to different types of point cloud data.

[0006] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0007] A bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood is used to construct the correspondence between the model point cloud and the scene point cloud, including the following steps:

[0008] S1. Input key points and descriptors: create model point cloud object P and scene point cloud object Q. and are the key point sets on the model point cloud object P and the scene object Q respectively. and They are respectively the model key point sets and scene key point set The set of local feature descriptors corresponding to the key points in ;

[0009] S2. Preliminary matching correspondence: Generate the initial correspondence by similarity matching the feature descriptors of the model point cloud and the scene point cloud, and build an initial corresponding point pair set between the model and the scene. ;

[0010] S3, impose a two-way neighborhood constraint: impose a relaxed neighborhood constraint on the initial correspondence relationship, and Perform preliminary verification to exclude obviously wrong corresponding point pairs, and obtain a set of corresponding relationships that preliminarily exclude wrong corresponding point pairs. ; This neighborhood constraint method does not require that the corresponding points must strictly correspond to the original points when searching in reverse, but guarantees that they are within a certain neighborhood of the original points. Compared with the bidirectional constraint in the BSS algorithm, this neighborhood constraint method reduces the false exclusion of correct corresponding point pairs.

[0011] S4, bidirectional implementation of Nannsr algorithm: the corresponding relationship set that has been initially excluded As input, the Nannsr algorithm is implemented on the scene points and model points in the set, the uniqueness score of each point pair is calculated, and the corresponding point pairs with relatively good scores are output to obtain the final corresponding point pair set The Nannsr algorithm uses feature neighbor points outside a certain neighborhood range to replace the second feature neighbor points in the NNSR algorithm, avoiding the problem in the NNSR algorithm that the correctness of the corresponding relationship cannot be determined due to the close spatial distance between the first and second feature neighbor points.

[0012] S5. Output corresponding point pairs: The final corresponding point pair set extracted is unique in both directions. For output, the correspondence between the model point cloud and the scene point cloud is reconstructed. The output corresponding point pairs are highly unique and not easily affected by interference, which improves the accuracy of subsequent estimation of the conversion relationship between two or more point clouds.

[0013] Furthermore, in step S2, an initial corresponding point pair set is constructed The specific method is: for each model key point , from the scene key point set Search for features in descending order of similarity k Key points of the scene , the corresponding descriptor is ,in k is a positive integer, the key point of the model and scene key points Forming initial corresponding point pairs , the initial corresponding point pairs constructed on all model key points That is the initial corresponding point pair set .

[0014] Furthermore, the initial correspondence between the model key points and the scene key points is determined based on the similarity of the corresponding local feature descriptors on the key points. and , similarity It is expressed as:

[0015] (1)

[0016] in, represents the L2 norm.

[0017] Further, in step S3, a corresponding relationship set is obtained that preliminarily excludes erroneous corresponding point pairs. The specific steps are:

[0018] S3.1. For the initial corresponding point pair , defined by model key points is the two-way neighborhood constraint space region with the center of the sphere and radius r1;

[0019] S3.2. Find the key points of the scene through reverse search Corresponding points on the model point cloud ;

[0020] S3.3 If and The spatial distance is no greater than r1, and the initial corresponding point pair is retained. , otherwise it will be excluded.

[0021] Furthermore, in step S4, the final corresponding point pair set is obtained The specific steps are:

[0022] S4.1. Key points of the model To scene key point and scene key points To model key points The Nannsr algorithm is implemented once in each direction to calculate their nearest neighbor similarity scores, which are recorded as and ;

[0023] S4.2, respectively and Assign corresponding weights and sum them up to get the final score S :

[0024] (3)

[0025] in, and Respectively represent the smaller and larger values, and Represent the weights of smaller and larger nearest neighbor similarity ratio scores, respectively;

[0026] S4.3. Scoring of corresponding points based on the above S , the score of each point pair S Sort from small to large, from the set Select the first one after sorting K Score S The corresponding point pairs are taken as the final corresponding point pair set.

[0027] Furthermore, the specific steps of the Nannsr algorithm are:

[0028] S10, at each key point of the model The corresponding k Key points of the scene In the definition, the first characteristic neighbor point is the separated space region with the center of the sphere and radius r2;

[0029] S20, scene key points As the first feature neighbor , excluding scene key points Except for the first characteristic neighbor point All feature neighbor points located in the separated space area except for , is the new second feature neighbor point;

[0030] S30, calculate the nearest neighbor similarity score :

[0031] (2)

[0032] in , and They are , and The corresponding feature descriptor.

[0033] Furthermore, if k If all scene key points are located in the separated space area, then select the distance The farthest feature neighbor point is .

[0034] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the program in the memory to execute the algorithm as described above.

[0035] The present invention further provides a storage medium, wherein processor-executable instructions are stored, and the processor-executable instructions are used to execute the algorithm as described above when executed by the processor.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] 1. The present invention constructs an initial corresponding point pair set through similarity matching, and then preliminarily eliminates erroneous correspondences in the point pair set by applying bidirectional neighborhood constraints. Finally, the Nannsr algorithm is implemented on the scene and model key points respectively to extract point pairs that are unique in both the scene and the model as the final output corresponding point pairs, thereby constructing a robust corresponding point pair construction method, namely, the bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood (NaDnnsr). The performance of this algorithm is significantly better than the existing corresponding point pair construction algorithm, avoiding the problem that the uniqueness of the point pair cannot be correctly judged, and it has good performance even on point cloud data interfered by noise, occlusion, data anomalies, etc.

[0038] 2. The present invention proposes a bidirectional neighborhood constraint method. Compared with the bidirectional constraint in the BSS algorithm, this relaxed neighborhood constraint is more flexible and can further improve the accuracy of corresponding point pairs while retaining more correct corresponding relationships, thereby improving the accuracy of subsequent registration algorithms.

[0039] 3. The Nannsr algorithm proposed in the present invention selects feature neighbor points outside a certain neighborhood range to replace the second feature neighbor points in the traditional NNSR, thereby preventing the first and second feature neighbor points from being too close and the correctness of the corresponding relationship from being calculated correctly. By implementing the Nannsr algorithm in both directions, it is possible to ensure the output of highly robust corresponding point pairs. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flowchart of the improved nearest neighbor similarity ratio algorithm of the present invention;

[0041] Figure 2 A schematic diagram of a bidirectional constraint process in an embodiment;

[0042] Figure 3 Schematic diagram of a two-way neighborhood constraint process in an embodiment;

[0043] Figure 4 Schematic diagram of the process of obtaining a new feature neighbor point sequence to obtain the nearest neighbor similarity ratio in the embodiment. DETAILED DESCRIPTION

[0044] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0045] It should be noted that when a component is referred to as being "mounted on" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component at the same time. When a component is considered to be "fixed to" another component, it may be directly fixed on the other component or there may be a central component at the same time.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0047] See also Figure 1 The present invention provides a bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood, which is used to construct the correspondence between the model point cloud and the scene point cloud, including the following steps:

[0048] S1. Input key points and descriptors: create model point cloud object P and scene point cloud object Q. and are the key point sets on the model point cloud object P and the scene object Q respectively. and They are respectively the model key point sets and scene key point set The algorithm goal of this application is to use P F and Q F The local feature descriptors in the point cloud are used to construct a more accurate point-to-point correspondence set, thereby improving the success rate of subsequent point cloud registration.

[0049] S2. Preliminary matching correspondence: Generate the initial correspondence by similarity matching the feature descriptors of the model point cloud and the scene point cloud, and build an initial corresponding point pair set between the model and the scene. .

[0050] The initial correspondence between the model key points and the scene key points is determined based on the similarity of the local feature descriptors corresponding to the key points. The principle is that the similarity of the feature descriptors reflects the similarity of the corresponding local surfaces. Specifically: for two local feature descriptors and , similarity It is expressed as:

[0051] (1)

[0052] in, represents the L2 norm.

[0053] The following takes 6 key points as an example to explain in detail the specific use of the algorithm.

[0054] For each model key point , from the scene key point set Search for 6 scene key points in descending order of feature similarity , the corresponding descriptor is Since the first feature neighbor point (scene key point ) and model key points The similarity is higher than that of other feature neighboring points, so the key point of the model and scene key points Forming initial corresponding point pairs , the initial corresponding point pairs constructed on all model key points That is the initial corresponding point pair set The purpose of searching for multiple scene key points is to provide more neighboring relationships for the Nannsr algorithm, thereby constructing a new nearest neighbor similarity ratio.

[0055] S3, impose a two-way neighborhood constraint: impose a relaxed neighborhood constraint on the initial correspondence relationship, and Perform preliminary verification to exclude obviously wrong corresponding point pairs, and obtain a set of corresponding relationships that preliminarily exclude wrong corresponding point pairs. .

[0056] The specific steps are:

[0057] S3.1. For the initial corresponding point pair , defined by model key points is a two-way neighborhood constraint space region with a sphere center and a radius of r1. In this embodiment, the region radius r1 is set to 6mr, where mr represents the average grid resolution on the scene point cloud.

[0058] S3.2. Find the key points of the scene through reverse search Corresponding points on the model point cloud ;

[0059] S3.3 If and The spatial distance is no greater than r1, and the initial corresponding point pair is retained. , otherwise it will be excluded.

[0060] like Figure 2 As shown (the forward solid arrow in the figure indicates the key point of the model Matching scene key points The reverse dashed arrows represent the key points of the scene. Reverse search to obtain corresponding points on the model point cloud The process is the same in the following figures). Among the existing bidirectional constraints, only when and Completely overlap, the feature descriptors corresponding to the two points are closest to each other, indicating that the initial corresponding point pair If the bidirectional constraints are satisfied, the point pair is retained, such as Figure 2 The forward solid arrow and the reverse dotted arrow shown in the lower middle part coincide with each other; if and Do not overlap, the initial corresponding point pair If the bidirectional constraint is not satisfied, the point pair is excluded, such as Figure 2 The situation where the forward solid arrow and the reverse dashed arrow shown in the lower middle part do not overlap.

[0061] Compared with the bidirectional constraint method used in the BSS algorithm, the bidirectional neighborhood constraint method is more flexible and can retain more correct correspondences. Bidirectional constraints usually require that two points completely coincide with each other, but this strict constraint can easily lead to the exclusion of some correct correspondences. The bidirectional neighborhood constraint solves this problem by introducing spatial distance tolerance. Specifically, unlike the existing bidirectional constraints that strictly require and Completely overlapped, the two-way neighborhood constraint in this application only requires and The spatial distance is kept within a certain area (i.e. Located in the two-way neighborhood constraint space area), indicating that the initial corresponding point pair Satisfying the bidirectional domain constraints, the point pair is retained ,like Figure 3 The situation shown in the upper middle part; if and The spatial distance of is greater than the above r1, indicating that the initial corresponding point pair If the bidirectional domain constraint is not satisfied, the point pair is excluded. and Close enough (i.e. and The spatial distance is less than r1), using or Substituting into the subsequent algorithm for calculation, the final output results are almost the same. Therefore, these approximately closest point pairs are as likely to be correct corresponding point pairs as the strictly closest point pairs. The bidirectional neighborhood constraint retains such point pairs, which can initially exclude outliers while retaining more inliers to improve the accuracy of the subsequent registration algorithm.

[0062] By initial corresponding point pair set By implementing the bidirectional neighborhood constraint, we can obtain a set of correspondence relationships that preliminarily excludes erroneous corresponding point pairs. Compared with the bidirectional constraint in the BSS algorithm, this relaxed neighborhood constraint can further improve the accuracy of corresponding point pairs while retaining more internal points.

[0063] S4, bidirectional implementation of Nannsr algorithm: the corresponding relationship set that has been initially excluded As input, the Nannsr algorithm is implemented on the scene points and model points in the set, the uniqueness score of each point pair is calculated, and the corresponding point pairs with relatively good scores (the smaller the score, the better) are output to obtain the final corresponding point pair set .

[0064] The Nannsr algorithm proposed in this application is optimized on the basis of the traditional NNSR algorithm and improves the existing problems. The Nannsr algorithm calculates the nearest neighbor similarity ratio by selecting key points that are far away from the first feature neighbor point as the new second feature neighbor point, thereby avoiding the problem in the traditional NNSR algorithm that the first and second feature neighbor points are too close, resulting in the uniqueness cannot be calculated correctly, thereby mistakenly excluding the correct corresponding point pairs.

[0065] The specific steps of the Nannsr algorithm are:

[0066] S10, each model key point obtained in step S2 The corresponding 6 scene key points In the definition, the first characteristic neighbor point The center of the sphere is the separated space region with a radius of r2. In this embodiment, r2 is the same as the aforementioned r1, that is, also 6mr.

[0067] S20, scene key points As the first feature neighbor , excluding 6 scene key points Except for the first characteristic neighbor point All feature neighbor points located in the separated space area except for , is the new second feature neighbor point.

[0068] S30, calculate the nearest neighbor similarity score :

[0069] (2)

[0070] in , and They are , and The smaller the nearest neighbor similarity score, the higher the uniqueness of the key point and the stronger its resistance to interference.

[0071] like Figure 4 As shown, for the initial corresponding point pair , scene key points and The first feature neighbor The spatial distance between them is no greater than r2, that is, the two points are in the exclusion area. At this time, if we use or Calculating the nearest neighbor similarity ratio cannot reflect the uniqueness of the neighborhood. That is , the new second feature neighbor point is , and These points are selected according to their nearest neighbor points to the first feature. The spatial distances between them correspond to the new feature neighbor point sequence in the order from near to far. , and ,Right now It is the closest to the first feature neighbor point among the three new feature neighbor points point.

[0072] For the new feature neighbor point sequence, Nannsr selects To replace the original second feature neighbor point The spatial distance between the new second feature neighbor point and the first feature neighbor point is far, which effectively overcomes the problem that the uniqueness judgment of the corresponding relationship is easily disturbed by the neighboring points. If the six scene key points are all located in the separated space area, that is, The spatial distance of is no greater than r2, then choose the distance The farthest feature neighbor point is .

[0073] Executing the Nannsr algorithm once can only determine the uniqueness of one key point (such as the model key point in the previous process ), which is important for the entire corresponding point (such as the initial corresponding point pair mentioned above ) For the pair, the uniqueness judgment is not perfect. To ensure that both points in the extracted point pair are unique, it is necessary to perform uniqueness judgment on the scene key points and the model key points respectively.

[0074] Therefore, the final corresponding point pair set is obtained The specific steps are:

[0075] S4.1. Key points of the model To scene key point and scene key points To model key points The Nannsr algorithm is implemented once in each direction to calculate their nearest neighbor similarity scores, which are recorded as and ;

[0076] S4.2, respectively and Assign corresponding weights and sum them up to get the final score S :

[0077] (3)

[0078] in, and Respectively represent the smaller and larger values, and Respectively represent the weights of the smaller and larger nearest neighbor similarity scores. In this embodiment, set , .

[0079] S4.3. Scoring of corresponding points based on the above S , the score of each point pair S Sort from small to large, from the set Select the first one after sorting K Score S The corresponding point pairs are taken as the final corresponding point pair set.

[0080] By verifying the dual uniqueness of model key points and scene key points, it can be ensured that the selected correspondences have obvious characteristics and are not easily affected by interference.

[0081] S5. Output corresponding point pairs: The final corresponding point pair set extracted is unique in both directions. For output, the correspondence between the model point cloud and the scene point cloud is reconstructed.

[0082] By implementing the Nannsr algorithm bidirectionally, we can ensure the output of corresponding point pairs with strong robustness, and perform well even on point cloud data under interference conditions such as noise, occlusion, and data anomalies.

[0083] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the program in the memory to execute the algorithm as described above.

[0084] An embodiment of the present invention further provides a storage medium, in which processor-executable instructions are stored. When the processor-executable instructions are executed by the processor, they are used to execute the algorithm as described above.

[0085] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0086] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0089] Finally, it should be noted that the various technical features of the above-described embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood, used to construct the correspondence between the model point cloud and the scene point cloud, characterized in that: The following steps are involved: S1. Input key points and descriptors: create model point cloud object P and scene point cloud object Q. and are the key point sets on the model point cloud object P and the scene object Q respectively. and They are respectively the model key point sets and scene key point set The set of local feature descriptors corresponding to the key points in ; S2. Preliminary matching correspondence: Generate the initial correspondence by similarity matching the feature descriptors of the model point cloud and the scene point cloud, and build an initial corresponding point pair set between the model and the scene. ; S3, impose a two-way neighborhood constraint: impose a relaxed neighborhood constraint on the initial correspondence relationship, and Perform preliminary verification to exclude obviously wrong corresponding point pairs, and obtain a set of corresponding relationships that preliminarily exclude wrong corresponding point pairs. ; S4, bidirectional implementation of Nannsr algorithm: the corresponding relationship set that has been initially excluded As input, the Nannsr algorithm is implemented on the scene points and model points in the set, the uniqueness score of each point pair is calculated, and the corresponding point pairs with relatively good scores are output to obtain the final corresponding point pair set ; S5. Output corresponding point pairs: The final corresponding point pair set extracted is unique in both directions. For output, the correspondence between the model point cloud and the scene point cloud is reconstructed.

2. The bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood according to claim 1, characterized in that: In step S2, construct an initial corresponding point pair set The specific method is: for each model key point , from the scene key point set Search for features in descending order of similarity k Key points of the scene , the corresponding descriptor is ,in k is a positive integer, the key point of the model and scene key points Forming initial corresponding point pairs , the initial corresponding point pairs constructed on all model key points That is the initial corresponding point pair set .

3. The bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood according to claim 2, characterized in that: The initial correspondence between the model key points and the scene key points is determined by the similarity of the corresponding local feature descriptors on the key points. and , similarity It is expressed as: （1） in, represents the L2 norm.

4. The bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood according to claim 2, characterized in that: In step S3, a corresponding relationship set is obtained that preliminarily excludes erroneous corresponding point pairs. The specific steps are: S3.

1. For the initial corresponding point pair , defined by model key points is the two-way neighborhood constraint space region with the center of the sphere and radius r1; S3.

2. Find the key points of the scene through reverse search Corresponding points on the model point cloud ; S3.3 If and The spatial distance is no greater than r1, and the initial corresponding point pair is retained. , otherwise it will be excluded.

5. A bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood according to any one of claims 2 to 4, characterized in that: In step S4, the final corresponding point pair set is obtained The specific steps are: S4.

1. Key points of the model To scene key point and scene key points To model key points The Nannsr algorithm is implemented once in each direction to calculate their nearest neighbor similarity scores, which are recorded as and ; S4.2, respectively and Assign corresponding weights and sum them up to get the final score S : （3） in, and Respectively represent the smaller and larger values, and Represent the weights of smaller and larger nearest neighbor similarity ratio scores, respectively; S4.

3. Scoring of corresponding points based on the above S , the score of each point pair S Sort from small to large, from the set Select the first one after sorting K Score S The corresponding point pairs are taken as the final corresponding point pair set.

6. The bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood according to claim 5, characterized in that: The specific steps of the Nannsr algorithm are: S10, at each key point of the model The corresponding k Key points of the scene In the definition, the first characteristic neighbor point is the separated space region with the center of the sphere and radius r2; S20, scene key points As the first feature neighbor , excluding scene key points Except for the first characteristic neighbor point All feature neighbor points located in the separated space area except for , is the new second feature neighbor point; S30, calculate the nearest neighbor similarity score : （2） in , and They are , and The corresponding feature descriptor.

7. The bidirectional nearest neighbor similarity ratio algorithm excluding the neighborhood according to claim 6, characterized in that: like k If all scene key points are located in the separated space area, then select the distance The farthest feature neighbor point is .

8. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store at least one program, and the processor is used to load the program in the memory to execute the algorithm as claimed in any one of claims 1 to 7.

9. A storage medium storing instructions executable by a processor, characterized in that: The processor-executable instructions are used to perform the algorithm according to any one of claims 1 to 7 when executed by the processor.

Images