A three-dimensional data registration method and system based on curvature feature geometry sampling

Abstract

The application discloses a three-dimensional data registration method and system based on curvature feature geometry sampling, which comprises the following steps: extracting feature point sets according to the curvature features of a test point set and a template point set respectively, so as to obtain a test feature point set and a template feature point set; performing coarse matching by using an SVD singular value decomposition algorithm, so as to correspond the test feature point set and the template feature point set one by one, and calculating the scores after the correspondence, so as to obtain an initial transformation matrix set sorted according to the scores; performing geometry sampling on the test point set and the template point set respectively according to the curvature features of the test point set and the template point set, so as to obtain a test sampling point set and a template sampling point set; and applying the initial transformation matrix set as the initial position relationship of fine matching to the test sampling point set and the template sampling point set in sequence, and performing fine matching by using an ICP algorithm. The application improves the accuracy, robustness and efficiency of matching, is suitable for a wide range of scenes, and has high market promotion value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a three-dimensional data registration method and system based on curvature feature geometric sampling. BACKGROUND

[0002] With the rapid development of three-dimensional vision technology and three-dimensional acquisition equipment, the application of depth images or three-dimensional point cloud data is becoming more and more widespread in industrial measurement, detection and matching positioning scenes.

[0003] Since the registration accuracy of three-dimensional point cloud directly affects the reliability of the results of subsequent processes such as detection and matching, point cloud registration is a key step in the fields of computer vision and computer-aided geometry such as stereo matching, target recognition and image positioning.

[0004] However, the existing coarse matching process only estimates a single feature of the point cloud data, which cannot fully describe the performance and state of the point cloud data, and when the size of the point cloud data is large, it takes a long time to estimate the features of all the point cloud data and the accuracy is low, which makes the point cloud registration method have poor robustness, low computational efficiency and cannot provide a good initial value for the fine registration process, also making the accuracy and efficiency of the fine registration process low.

[0005] Therefore, it is necessary to improve the prior art.

[0006] The above information is given as background information only to assist with an understanding of the present disclosure, and should not be taken as an acknowledgement or admission that any of the above-mentioned details form part of the prior art with respect to the present disclosure. SUMMARY

[0007] The present application provides a three-dimensional data registration method and system based on curvature feature geometric sampling to solve the problems of the prior art.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0009] In a second aspect, the present application provides a three-dimensional data registration method based on curvature feature geometric sampling, which comprises:

[0010] According to the curvature features corresponding to the test point set and the template point set, the feature point sets of the test point set and the template point set are extracted respectively to obtain test feature point sets and template feature point sets;

[0011] The SVD singular value decomposition algorithm is used for coarse matching to one-to-one correspond the test feature point sets and the template feature point sets, and the scores after correspondence are calculated to obtain an initial transformation matrix set sorted by scores;

[0012] According to the curvature characteristics corresponding to the test point set and the template point set, the test point set and the template point set are respectively geometrically sampled to obtain a test sampling point set and a template sampling point set;

[0013] The initial transformation matrix set is taken as an initial position relationship of fine matching, and is sequentially applied to the test sampling point set and the template sampling point set, and a symmetric error improved ICP algorithm is used for fine matching.

[0014] Further, in the three-dimensional data registration method based on curvature characteristic geometric sampling, before the step of extracting the test feature point set and the template feature point set from the test point set and the template point set according to the curvature characteristics corresponding to the test point set and the template point set, the method further comprises:

[0015] A test point set of a test image and a template point set of a template image are obtained, and curvature characteristics of the test point set and the template point set are respectively calculated.

[0016] Further, in the three-dimensional data registration method based on curvature characteristic geometric sampling, the step of using the SVD singular value decomposition algorithm for coarse matching to one-to-one correspond the test feature point set and the template feature point set, and calculating a score after correspondence to obtain an initial transformation matrix set sorted according to the score, comprises:

[0017] The SVD singular value decomposition algorithm is used for coarse matching to one-to-one correspond the test feature point set and the template feature point set;

[0018] A point in the template feature point set is taken as a starting point, the number of points in the test feature point set within a search radius range is calculated, and the percentage of the corresponding test feature point set is calculated, so as to obtain an overlap score after correspondence;

[0019] The distance between the corresponding points in the test feature point set and the template feature point set is calculated, and according to the set initial search radius of 10%, 25%, 50%, 75% and 100% respectively corresponding to 16 points, 8 points, 4 points, 2 points and 1 point, the quality score after correspondence is accumulated to obtain;

[0020] The score is calculated according to the following formula to obtain an initial transformation matrix set sorted according to the score:

[0021] Score=(overlap score*0.5+quality score*0.5) / 16.

[0022] Further, in the three-dimensional data registration method based on curvature feature geometry sampling, the step of sampling the test point set and the template point set according to the curvature features corresponding to the test point set and the template point set respectively to obtain a test sampling point set and a template sampling point set comprises:

[0023] setting a curvature threshold value;

[0024] dividing the test point set into a test feature region and a non-test feature region according to the curvature features of the test point set and taking the curvature threshold value as a boundary;

[0025] sampling the test feature region and the non-test feature region, and the sampling rate of the test feature region is higher than that of the non-test feature region to obtain a test sampling point set;

[0026] dividing the template point set into a template feature region and a non-template feature region according to the curvature features of the template point set and taking the curvature threshold value as a boundary;

[0027] sampling the template feature region and the non-template feature region, and the sampling rate of the template feature region is higher than that of the non-template feature region to obtain a template sampling point set.

[0028] Further, in the three-dimensional data registration method based on curvature feature geometry sampling, the step of taking the initial transformation matrix set as an initial position relationship for fine matching, applying the initial transformation matrix set to the test sampling point set and the template sampling point set in turn, and performing fine matching by using the ICP algorithm improved in terms of symmetry error comprises:

[0029] taking the initial transformation matrix set as an initial position relationship for fine matching, and applying the initial transformation matrix set to the test sampling point set and the template sampling point set in turn;

[0030] performing nearest neighbor search on the test sampling point set and the template sampling point set respectively, obtaining the correspondence between points and points, and iteratively calculating an error until the error is less than a set convergence value, and fine matching is completed.

[0031] In a second aspect, an embodiment of the present application provides a three-dimensional data registration system based on curvature feature geometry sampling, and the system comprises:

[0032] a feature point extraction module configured to extract feature point sets of the test point set and the template point set according to the curvature features corresponding to the test point set and the template point set respectively to obtain a test feature point set and a template feature point set;

[0033] a coarse matching module configured to perform coarse matching by using an SVD singular value decomposition algorithm to correspond the test feature point set and the template feature point set one by one, and to calculate a score after correspondence to obtain an initial transformation matrix set sorted according to the score;

[0034] a point set sampling module configured to perform geometric sampling on the test point set and the template point set respectively according to the curvature features corresponding to the test point set and the template point set to obtain a test sampling point set and a template sampling point set;

[0035] a fine matching module configured to apply the initial transformation matrix set as an initial positional relationship of fine matching to the test sampling point set and the template sampling point set in turn, and to perform fine matching by using a symmetric error improved ICP algorithm.

[0036] Further, in the three-dimensional data registration system based on curvature feature geometric sampling, the system further comprises:

[0037] a point set acquisition module configured to acquire a test point set of a test image and a template point set of a template image, and to calculate curvature features of the test point set and the template point set respectively before the step of extracting a test feature point set and a template feature point set from the test point set and the template point set respectively according to the curvature features corresponding to the test point set and the template point set.

[0038] Further, in the three-dimensional data registration system based on curvature feature geometric sampling, the coarse matching module is specifically configured to:

[0039] perform coarse matching by using an SVD singular value decomposition algorithm to correspond the test feature point set and the template feature point set one by one;

[0040] calculate how many points in the test feature point set are within a search radius range from a starting point in the template feature point set, and calculate a percentage of the corresponding test feature point set, thereby obtaining an overlap score after correspondence;

[0041] calculate distances between corresponding points in the test feature point set and the template feature point set, and respectively assign 16 points, 8 points, 4 points, 2 points and 1 point according to 10%, 25%, 50%, 75% and 100% of the set initial search radius, thereby obtaining a quality score after correspondence by accumulation;

[0042] calculate a score according to the following formula to obtain an initial transformation matrix set sorted according to the score:

[0043] Score = (Overlap Score * 0.5 + Quality Score * 0.5) / 16.

[0044] Further, in the three-dimensional data registration system based on curvature feature geometry sampling, the point set sampling module is specifically used for:

[0045] Setting a curvature threshold value;

[0046] According to the curvature feature of the test point set, the test point set is divided into a test feature region and a non-test feature region by taking the curvature threshold value as a boundary;

[0047] Geometric sampling is performed on the test feature region and the non-test feature region, and the sampling rate of the test feature region is higher than that of the non-test feature region, to obtain a test sampling point set;

[0048] According to the curvature feature of the template point set, the template point set is divided into a template feature region and a non-template feature region by taking the curvature threshold value as a boundary;

[0049] Geometric sampling is performed on the template feature region and the non-template feature region, and the sampling rate of the template feature region is higher than that of the non-template feature region, to obtain a template sampling point set.

[0050] Further, in the three-dimensional data registration system based on curvature feature geometry sampling, the precision matching module is specifically used for:

[0051] The initial transformation matrix set is taken as an initial position relationship of precision matching, and is applied to the test sampling point set and the template sampling point set in sequence;

[0052] The test sampling point set and the template sampling point set are respectively subjected to nearest neighbor search, point-to-point correspondence is obtained, and error is iteratively calculated until the error is less than a set convergence value, and precision matching is completed.

[0053] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0054] The three-dimensional data registration method and system based on curvature feature geometry sampling provided by the embodiments of the present application can realize matching of three-dimensional data by sequentially performing feature point extraction, coarse matching, point set sampling and precision matching on the test point set of a test image and the template point set of a template image, not only improve the accuracy and robustness of matching, meet the precision and efficiency requirements in industrial detection matching, but also are very sensitive to shape deformation, can be applied to some matching objects with insufficient surface features, thus have more applicable scenarios and have high market promotion value. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to make the technical solutions of the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only need to be some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0056] Figure 1 is a flowchart of a three-dimensional data registration method based on curvature feature geometric sampling provided by the first embodiment of the present application;

[0057] Figure 2 is a feature point diagram of test point set extraction in the nut matching example in the first embodiment of the present application;

[0058] Figure 3 is a feature point diagram of template point set extraction in the nut matching example in the first embodiment of the present application;

[0059] Figure 4 is a diagram of geometric sampling results in the first embodiment of the present application;

[0060] Figure 5 is a diagram of the intercepted part of the workpiece with unclear surface features in the first embodiment of the present application;

[0061] Figure 6 is a diagram of the corresponding results after matching in the first embodiment of the present application;

[0062] Figure 7 is a diagram of a single target in multiple identical target images in the first embodiment of the present application;

[0063] Figure 8 is a diagram of the corresponding results after matching in the first embodiment of the present application;

[0064] Figure 9 is a functional module diagram of a three-dimensional data registration system based on curvature feature geometric sampling provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0065] In order to make the technical solutions of the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only need to be some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.

[0066] In the description of the present application, it needs to be understood that when one component is considered to be "connected" to another component, it can be directly connected to the other component or there can be a component arranged in the middle. When one component is considered to be "arranged on" another component, it can be directly arranged on the other component or there can be a component arranged in the middle.

[0067] In addition, the terms "long", "short", "inner", "outer", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, which is only for the convenience of describing the present application, and does not indicate or imply that the device or element referred to must have this particular orientation, be constructed in a particular orientation, and cannot be understood as a limitation of the present application.

[0068] The technical solutions of the present application will be further illustrated below in conjunction with the drawings and through specific embodiments.

[0069] Embodiment one

[0070] In view of the defects of the existing three-dimensional data registration technology, the applicant, based on many years of rich practical experience and professional knowledge in designing and manufacturing such products, and in cooperation with the application of theory, actively researches and innovates, in order to create a technology that can solve the defects in the prior art, so that the three-dimensional data registration technology is more practical. After continuous research, design, and repeated trial of samples and improvement, the present application is finally created.

[0071] Please refer to Figure 1 The embodiment of the present application provides a three-dimensional data registration method based on curvature feature geometric sampling, which comprises:

[0072] S101, respectively according to the curvature features corresponding to the test point set and the template point set, extracting the feature point set of the test point set and the template point set, obtaining the test feature point set and the template feature point set.

[0073] It should be noted that the present embodiment is applicable to the three-dimensional matching requirements of two different scenes, that is, the local data is used as the matching data, the global template data is matched, and then the position and attitude of the current local data are estimated; and the scene data is used as the matching data to match the template sample data, so that the workpiece in the scene and the corresponding pose can be matched and extracted, which is used for object positioning and mechanical arm application in a cluttered stacking scene.

[0074] Taking nut matching as an example, as Figure 2 The feature points extracted for the test point set, as Figure 3 The feature points extracted for the template point set.

[0075] In the present embodiment, before the step S101, the method further comprises:

[0076] Obtaining a test point set of a test image and a template point set of a template image, and calculating curvature features of the test point set and the template point set respectively.

[0077] It should be noted that the IVD curvature feature is used in the embodiment, which is sensitive to shape deformation and can be applied to some matching objects with insufficient surface features, so that the matching method has more applicable scenarios.

[0078] S102, coarse matching is performed by using an SVD singular value decomposition algorithm to correspond the test feature point set and the template feature point set one by one, and a score after correspondence is calculated to obtain an initial transformation matrix set sorted according to the score.

[0079] In the embodiment, the step S102 can further include the following steps:

[0080] Coarse matching is performed by using an SVD singular value decomposition algorithm to correspond the test feature point set and the template feature point set one by one;

[0081] Taking a point in the template feature point set as a starting point, the number of points in the test feature point set within a search radius range is calculated, and the percentage of the corresponding test feature point set is calculated, so as to obtain an overlap score after correspondence;

[0082] The distance between the corresponding points in the test feature point set and the template feature point set is calculated, and according to the set initial search radius of 10%, 25%, 50%, 75% and 100%, 16 points, 8 points, 4 points, 2 points and 1 point are respectively assigned, so as to obtain a quality score after correspondence by accumulation;

[0083] The score is calculated according to the following formula to obtain an initial transformation matrix set sorted according to the score:

[0084] Score = (overlap score * 0.5 + quality score * 0.5) / 16.

[0085] It should be noted that before coarse matching, point pair information needs to be formed according to the feature points obtained in step S101, including distance between two points, angle, normal vector and other features.

[0086] The embodiment innovatively designs a 5-order scoring system for calculating the score after correspondence, which is divided into an overlap score and a quality score. Among them, the overlap score is that taking a point in the template feature point set as a starting point, the number of points in the test feature point set within a search radius range is calculated, and the percentage of the corresponding test feature point set is the overlap score;

[0087] The quality score is calculated by the distance between the corresponding points in the test feature point set and the template feature point set. According to the initial search radius of 10%, 25%, 50%, 75% and 100%, 16 points, 8 points, 4 points, 2 points and 1 point are respectively assigned, and the corresponding quality score is accumulated.

[0088] S103, according to the curvature feature corresponding to the test point set and the template point set, respectively, the test point set and the template point set are geometrically sampled to obtain the test sampling point set and the template sampling point set.

[0089] It should be noted that the geometric sampling according to the curvature feature can retain more feature points, more geometric features of the restored data, and the sampling reduces the amount of point set data, so that the iteration efficiency of the fine matching is faster and the time consumption cost is reduced.

[0090] In the embodiment, the step S103 can further include the following steps:

[0091] Setting a curvature threshold;

[0092] According to the curvature feature of the test point set, the test point set is divided into a test feature region and a non-test feature region with the curvature threshold as the limit;

[0093] Geometric sampling is performed on the test feature region and the non-test feature region, and the sampling rate of the test feature region is higher than that of the non-test feature region to obtain a test sampling point set;

[0094] According to the curvature feature of the template point set, the template point set is divided into a template feature region and a non-template feature region with the curvature threshold as the limit;

[0095] Geometric sampling is performed on the template feature region and the non-template feature region, and the sampling rate of the template feature region is higher than that of the non-template feature region to obtain a template sampling point set.

[0096] It should be noted that this step is geometric sampling according to the curvature feature in step S101. The sampling object is determined according to the actual scale of the point set data, and the smaller scale object is sampled so that enough corresponding points can be found in the fine matching. The sampling process is as follows: setting a curvature threshold, dividing into a feature region and a non-feature region with the threshold as the limit, and the sampling rate of the feature region is higher than that of the non-feature region to retain more features.

[0097] Different curvatures have different types of calculation methods, and their common feature is a kind of expression of spatial geometric shape feature. Referring to the related literature, when representing the curvature feature, the volume integral of the ball convolution symbol is constructed to estimate the curvature, Where the integral kernel Br (p) is defined as a sphere with center point p and radius r. S is the surface of the object. The volume is the volume of the sphere and the object overlap. The volume integral describes the curvature shape in the following way:

[0098]

[0099] where p represents an arbitrary point in space, C r (p) represents the volume integral under the current sphere radius r. The first term is the volume of the sphere. H represents the average curvature of the point. From the above formula, it can be seen that the volume integral and the average curvature have a linear relationship, that is, the larger the average curvature, the smaller the volume integral value. On the depth map, the discrete pixel sphere convolution operator shown in the following formula can be used to conveniently calculate:

[0100] V(p) = (G B *G o )(p)

[0101] Please refer to Figure 4 , Figure 4 for a schematic diagram of the geometric sampling result.

[0102] S104, apply the initial transformation matrix set as the initial position relationship of fine matching to the test sampling point set and the template sampling point set in turn, and perform fine matching using the ICP algorithm improved by symmetric error.

[0103] In the embodiment, the step S104 can further include the following steps:

[0104] applying the initial transformation matrix set as the initial position relationship of fine matching to the test sampling point set and the template sampling point set;

[0105] performing nearest neighbor search on the test sampling point set and the template sampling point set respectively, obtaining the correspondence between points and points, and iteratively calculating the error until the error is less than a set convergence value, and the fine matching is completed.

[0106] It should be noted that the set convergence value is set by the technical personnel through experience, which is based on specific experimental results and can be any numerical value.

[0107] By taking the initial transformation matrix set obtained by coarse matching as the initial position relationship of fine matching, and then performing fine matching on the sampled point set, the calculation amount of fine matching can be effectively reduced, and the matching task can be completed within milliseconds.

[0108] The fine matching adopts an ICP algorithm, performs nearest neighbor search on the sampled point set, finds the corresponding matching points, and then performs error calculation, and there are three error calculation options, point-to-point, point-to-plane, and symmetric point-plane error. Figure 5 For the intercepted part of the workpiece with unclear surface features, such as Figure 6 For the corresponding result after matching, such as Figure 7 For a single target in multiple identical target images, such as Figure 8 For the corresponding result after matching.

[0109] The three-dimensional data registration method based on curvature feature geometric sampling provided by the embodiment of the application can realize matching of three-dimensional data by sequentially performing feature point extraction, coarse matching, point set sampling and fine matching on the test point set of the test image and the template point set of the template image, not only improves the accuracy and robustness of matching, meets the precision and efficiency requirements in industrial detection matching, is very sensitive to shape deformation, can be applied to some matching objects with insufficient surface features, has more applicable scenarios, and has high market promotion value.

[0110] Embodiment two

[0111] Please refer to the accompanying Figure 9 , Figure 9 The function module schematic diagram of the three-dimensional data registration system based on curvature feature geometric sampling provided by the second embodiment of the application is suitable for executing the three-dimensional data registration method based on curvature feature geometric sampling provided by the embodiment of the application. The system specifically comprises the following modules:

[0112] The feature point extraction module 201 is configured to extract feature point sets of the test point set and the template point set according to corresponding curvature features of the test point set and the template point set, respectively, to obtain a test feature point set and a template feature point set.

[0113] The coarse matching module 202 is configured to perform coarse matching by adopting an SVD singular value decomposition algorithm, so as to correspond the test feature point set and the template feature point set one by one, calculate scores after correspondence, and obtain an initial transformation matrix set sorted according to the scores.

[0114] The point set sampling module 203 is configured to perform geometric sampling on the test point set and the template point set according to corresponding curvature features of the test point set and the template point set, respectively, to obtain a test sampling point set and a template sampling point set.

[0115] The fine matching module 204 is configured to apply the initial transformation matrix set as the fine matching initial position relationship to the test sample point set and the template sample point set in sequence, and perform fine matching by using the ICP algorithm improved based on symmetric error.

[0116] Preferably, the system further comprises:

[0117] The point set acquisition module is configured to acquire the test point set of the test image and the template point set of the template image, and calculate the curvature features of the test point set and the template point set respectively before the step of extracting the feature point sets of the test point set and the template point set according to the curvature features of the test point set and the template point set respectively to obtain the test feature point set and the template feature point set.

[0118] Preferably, the coarse matching module 202 is specifically configured to:

[0119] The SVD singular value decomposition algorithm is used for coarse matching to one-to-one correspond the test feature point set and the template feature point set;

[0120] A point in the template feature point set is taken as a starting point, the number of points in the test feature point set within a search radius range is calculated, and the corresponding percentage of the test feature point set is calculated to obtain the overlap score after correspondence;

[0121] The distance between the corresponding points in the test feature point set and the template feature point set is calculated, and the quality score after correspondence is accumulated according to that 10%, 25%, 50%, 75% and 100% of the initial search radius are respectively assigned 16 points, 8 points, 4 points, 2 points and 1 point;

[0122] The score is calculated according to the following formula to obtain the initial transformation matrix set ranked according to the score:

[0123] Score=(overlap score*0.5+quality score*0.5) / 16.

[0124] Preferably, the point set sampling module 203 is specifically configured to:

[0125] A curvature threshold is set;

[0126] According to the curvature features of the test point set, the test point set is divided into a test feature region and a non-test feature region by taking the curvature threshold as a limit;

[0127] The test feature region and the non-test feature region are geometrically sampled, and the sampling rate of the test feature region is higher than that of the non-test feature region to obtain a test sample point set;

[0128] According to the curvature characteristics of the template point set, the template point set is divided into a template feature region and a non-template feature region, with the curvature threshold as a boundary;

[0129] The template feature region and the non-template feature region are geometrically sampled, and a sampling rate of the template feature region is higher than that of the non-template feature region, to obtain a template sampling point set.

[0130] Preferably, the fine matching module 204 is specifically used for:

[0131] The initial transformation matrix set is applied to the test sampling point set and the template sampling point set as an initial positional relationship of fine matching in sequence;

[0132] The test sampling point set and the template sampling point set are respectively subjected to nearest neighbor search, point-to-point correspondence, and iterative error calculation until the error is less than a set convergence value, and fine matching is completed.

[0133] The three-dimensional data registration system based on curvature feature geometric sampling provided by the embodiment of the application can realize matching of three-dimensional data by sequentially performing feature point extraction, coarse matching, point set sampling, and fine matching on the test point set of the test image and the template point set of the template image, improve the accuracy and robustness of matching, meet the precision and efficiency requirements in industrial detection matching, is sensitive to shape deformation, can be applied to some matching objects with insufficient surface features, has more applicable scenarios, and has high market promotion value.

[0134] The above description of the embodiments is provided for the purpose of illustration and description. It is not intended to be exhaustive or to limit the disclosure. The individual elements or features of a particular embodiment are generally not limited to that particular embodiment, but are interchangeable with each other, and with other selected embodiments, if applicable, even if not specifically shown or described. In many aspects, the same elements or features can change. Such changes are not considered to be a departure from the disclosure, and all such modifications are intended to be included within the scope of the disclosure.

[0135] Example embodiments are provided so that this disclosure will be thorough, and will fully convey the scope to those skilled in the art. Numerous specific details are set forth such as examples of specific parts, devices, and methods, to provide a thorough understanding of embodiments of the present disclosure. It will be apparent to those skilled in the art that specific details need not be employed, that example embodiments can be embodied in many different forms and that neither should be construed to limit the scope of the disclosure. In some example embodiments, well-known procedures, well-known device structures, and well-known technologies are not described in detail.

[0136] Professional terms are used herein only for the purpose of reciting particular example embodiments and are not meant to be limiting. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises" and "comprising" are inclusive and therefore specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The method steps, processes, and operations described herein are not to be construed as necessarily requiring their performance in the particular order in which they are discussed and illustrated unless explicitly stated otherwise. It is also to be understood that additional or alternative steps can be employed.

[0137] When an element or layer is referred to as being "on," "engaged," "connected," or "coupled" to another element or layer, it can be directly on, engaged, connected, or coupled to the other element or layer or intervening elements or layers can be present. In contrast, when an element or layer is referred to as being "directly on," "directly engaged," "directly connected," or "directly coupled" to another element or layer, there are no intervening elements or layers present. Other words used to describe the relationship between elements should be interpreted in a like fashion (e.g., "between" versus "directly between," "adjacent" versus "directly adjacent," etc.). The term "and / or" as used herein refers to and encompasses any and all combinations of one or more of the associated listed items. Although the terms first, second, third, etc. can be used herein to describe various elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms can be only used to distinguish one element, component, region, layer or section from another element, component, region, layer or section. Terms such as "first," "second," and other numerical terms when used herein do not permit a sequence or order to be implied unless explicitly stated by the context. Thus, a first element, component, region, layer or section discussed below could be termed a second element, component, region, layer or section without departing from the teachings of the example embodiments.

[0138] Spatially relative terms, such as "inner," "outer," "beneath," "below," "lower," "above," "upper," and the like, can be used herein for ease of description to describe one element or feature's relationship to another element(s) or feature(s) as illustrated in the figures. Spatially relative terms can be intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if the device in the figures is turned over, elements described as "below" or "beneath" other elements or features would then be oriented "above" the other elements or features. Thus, the example term "below" can encompass both an orientation of above and below. The device can be otherwise oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly.

Claims

1. A method for 3D data registration based on curvature feature geometry sampling, characterized in that, The method comprises: According to the curvature characteristics corresponding to the test point set and the template point set, the feature point sets of the test point set and the template point set are extracted respectively to obtain the test feature point set and the template feature point set; SVD singular value decomposition algorithm is used for coarse matching to one-to-one correspond the test feature point set and the template feature point set, and the score after correspondence is calculated to obtain an initial transformation matrix set sorted according to the score; According to the curvature characteristics corresponding to the test point set and the template point set, the test point set and the template point set are geometrically sampled respectively to obtain the test sampling point set and the template sampling point set; The initial transformation matrix set is applied to the test sampling point set and the template sampling point set in turn as the initial position relationship of fine matching, and the ICP algorithm improved by symmetric error is used for fine matching; The step of using SVD singular value decomposition algorithm for coarse matching to one-to-one correspond the test feature point set and the template feature point set, and calculating the score after correspondence to obtain an initial transformation matrix set sorted according to the score comprises: SVD singular value decomposition algorithm is used for coarse matching to one-to-one correspond the test feature point set and the template feature point set; Taking a point in the template feature point set as a starting point, the number of points in the test feature point set within the search radius range is calculated, and the corresponding percentage of the test feature point set is calculated to obtain the overlap score after correspondence; The distance between the corresponding points in the test feature point set and the template feature point set is calculated, and according to the set initial search radius of 10%, 25%, 50%, 75% and 100%, 16 points, 8 points, 4 points, 2 points and 1 point are respectively assigned, so as to accumulate to obtain the quality score after correspondence; The score is calculated according to the following formula to obtain an initial transformation matrix set sorted according to the score: Score=(overlap score*0.5+quality score*0.5) / 16.

2. The method of claim 1, wherein the curvature feature geometry-based sampling of 3D data registration is characterized by, Before the step of extracting the feature point sets of the test point set and the template point set according to the curvature characteristics corresponding to the test point set and the template point set respectively to obtain the test feature point set and the template feature point set, the method further comprises: Obtaining the test point set of the test image and the template point set of the template image, and calculating the curvature characteristics of the test point set and the template point set respectively.

3. The method of claim 1, wherein the curvature feature geometry-based sampling of 3D data registration is characterized by, The step of geometrically sampling the test point set and the template point set according to the curvature characteristics corresponding to the test point set and the template point set to obtain the test sampling point set and the template sampling point set comprises: Setting a curvature threshold; According to the curvature characteristics of the test point set, the test point set is divided into a test feature region and a non-test feature region with the curvature threshold as the boundary; Geometrically sampling the test feature region and the non-test feature region, the sampling rate of the test feature region is higher than that of the non-test feature region to obtain the test sampling point set; According to the curvature characteristics of the template point set, the template point set is divided into a template feature region and a non-template feature region with the curvature threshold as the boundary; Geometrically sampling the template feature region and the non-template feature region, a sampling rate of the template feature region being higher than that of the non-template feature region, to obtain a template sample point set.

4. The method of claim 1, wherein, The step of applying the initial transformation matrix set as an initial position relation for fine matching to the test sample point set and the template sample point set in sequence and performing fine matching by using the ICP algorithm improved in terms of symmetric error comprises: applying the initial transformation matrix set as an initial position relation for fine matching to the test sample point set and the template sample point set in sequence; performing nearest neighbor search on the test sample point set and the template sample point set respectively, obtaining point-to-point correspondence, and iteratively calculating error until the error is less than a set convergence value, and fine matching is completed.

5. A three-dimensional data registration system based on curvature feature geometry sampling, characterized by, The system comprises: a feature point extraction module configured to extract feature point sets of the test point set and the template point set according to corresponding curvature features of the test point set and the template point set respectively, to obtain a test feature point set and a template feature point set; a coarse matching module configured to perform coarse matching by using the SVD singular value decomposition algorithm to correspond the test feature point set and the template feature point set one by one, and to calculate a score after correspondence to obtain an initial transformation matrix set ranked according to the score; a point set sampling module configured to perform geometric sampling on the test point set and the template point set respectively according to corresponding curvature features of the test point set and the template point set, to obtain a test sample point set and a template sample point set; a fine matching module configured to apply the initial transformation matrix set as an initial position relation for fine matching to the test sample point set and the template sample point set in sequence, and to perform fine matching by using the ICP algorithm improved in terms of symmetric error; the coarse matching module is specifically configured to: perform coarse matching by using the SVD singular value decomposition algorithm to correspond the test feature point set and the template feature point set one by one; calculate how many points in the test feature point set are within a search radius range with a point in the template feature point set as a starting point, and calculate a percentage of the test feature point set corresponding to the points, to obtain an overlap score after correspondence; calculate distances between corresponding points in the test feature point set and the template feature point set, and assign 16 points, 8 points, 4 points, 2 points and 1 point respectively corresponding to 10%, 25%, 50%, 75% and 100% of a set initial search radius, to obtain a quality score after correspondence by accumulation; calculate a score according to the following formula to obtain an initial transformation matrix set ranked according to the score: score=(overlap score*0.5+quality score*0.5) / 16.

6. The three-dimensional data registration system based on curvature feature geometry sampling of claim 5, wherein, The system further comprises: a point set acquisition module configured to acquire a test point set of a test image and a template point set of a template image, and calculate curvature features of the test point set and the template point set respectively, before the step of extracting feature point sets of the test point set and the template point set according to corresponding curvature features of the test point set and the template point set respectively to obtain a test feature point set and a template feature point set.

7. The three-dimensional data registration system based on curvature feature geometry sampling of claim 5, wherein, The point set sampling module is specifically configured to: set a curvature threshold; According to the curvature characteristics of the test point set, the test point set is divided into a test feature region and a non-test feature region with the curvature threshold as a boundary; Geometric sampling is performed on the test feature region and the non-test feature region, and a sampling rate of the test feature region is higher than that of the non-test feature region, to obtain a test sampling point set; According to the curvature characteristics of the template point set, the template point set is divided into a template feature region and a non-template feature region with the curvature threshold as a boundary; Geometric sampling is performed on the template feature region and the non-template feature region, and a sampling rate of the template feature region is higher than that of the non-template feature region, to obtain a template sampling point set.

8. The three-dimensional data registration system based on curvature feature geometry sampling of claim 5, wherein, The fine matching module is specifically used for: applying the initial transformation matrix set as an initial position relationship of fine matching to the test sampling point set and the template sampling point set in sequence; performing nearest neighbor search on the test sampling point set and the template sampling point set respectively, to obtain a point-to-point corresponding relationship, and iteratively calculating an error until the error is less than a set convergence value, and fine matching is completed.

Images