Point cloud registration method and system based on multi-scale curvature point pair features

Abstract

The invention provides a point cloud registration method and system based on multi-scale curvature point pair features, and relates to the technical field of three-dimensional point cloud processing, and the method comprises the steps: obtaining a preoperative target point cloud and an intra-operative source point cloud; extracting initial key points in the target point cloud and the source point cloud in combination with adaptive voxel down-sampling and weighted ISS sampling; determining an intersection point set of a neighborhood to which the initial key point belongs and the multi-scale sphere; calculating the self-similarity of the intersection point set, and determining the geometric feature score of the neighborhood; repositioning each initial key point in the target domain according to the geometric feature score, and respectively obtaining key point sets of the target point cloud and the source point cloud; calculating an adaptive main neighborhood radius of the key point set; calculating weighted curvature values of the key points according to the self-adaptive main neighborhood radius; according to the weighted curvature value, determining a multi-scale weighted curvature point pair characteristic value; and performing registration between the target point cloud and the source point cloud according to the multi-scale curvature point pair feature values.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional point cloud processing, and in particular to a point cloud registration method and system based on multi-scale curvature point pair features. Background Art

[0002] Multi-scale curvature point pair features refer to a technology that simultaneously utilizes the local geometric information of points (curvature) and the relative relationship between points (point pair features) to align point clouds obtained from different perspectives or data sources into a unified coordinate system. The point cloud registration method based on multi-scale curvature point pair features extracts the curvature information of key points and the relationship between point pairs through multi-scale geometric analysis, and constructs an effective point cloud feature descriptor to achieve high-precision matching and registration of two sets of point cloud data.

[0003] Point cloud registration technology has important applications in medical surgical navigation, reverse engineering, robotic vision and other fields. Through point cloud registration, it can achieve the conversion from scattered data to a unified spatial model, which is of great significance to reducing noise interference, improving the efficiency and accuracy of registration, promoting technological progress, and intelligent and precise technology applications.

[0004] However, existing point cloud registration technologies face multiple challenges in practical applications, such as the low overlap problem. The global point cloud acquired before surgery (such as CT scans) covers the complete anatomical structure, while the local point cloud acquired during surgery (such as structured light scans) only contains partial anatomical features. Due to the characteristics of low-overlap point clouds, traditional global matching algorithms are prone to fall into local optimality. Secondly, intraoperative point clouds are limited by sensor accuracy, surgical environment, and instrument occlusion, and are often accompanied by high noise and external point interference, resulting in poor adaptability to noise and sparse point clouds, making it difficult to capture global geometric structure information. The matching accuracy is limited in complex geometric scenes, which reduces the robustness and accuracy of the registration. Summary of the Invention

[0005] In order to solve the technical problems that the existing point cloud registration technology in the prior art faces multiple challenges in practical applications, such as poor adaptability to noisy and sparse point clouds, difficulty in capturing global geometric structure information, limited matching accuracy in complex geometric scenes, and reduced robustness and accuracy of registration, the present invention provides a point cloud registration method and system based on multi-scale curvature point pair features.

[0006] The technical solutions provided by the embodiments of the present invention are as follows:

[0007] First aspect

[0008] An embodiment of the present invention provides a point cloud registration method based on multi-scale curvature point pair features, comprising:

[0009] S1: Obtain the patient's preoperative target point cloud and intraoperative source point cloud;

[0010] S2: Combine adaptive voxel down sampling and weighted ISS sampling to extract the initial key points in the target point cloud and the source point cloud;

[0011] S3: Determine the set of intersections between the neighborhood of each initial key point and the multi-scale spheres with different radii;

[0012] S4: Calculate the self-similarity of each intersection set, and determine the geometric feature score of the neighborhood based on the self-similarity;

[0013] S5: Based on the geometric feature scores, each initial key point is relocated in the target area to obtain the key point sets of the target point cloud and the source point cloud respectively. The target neighborhood is specifically the neighborhood whose geometric feature score is less than the preset geometric feature score;

[0014] S6: Calculate the adaptive main neighborhood radius of each key point in the key point set;

[0015] S7: Calculate the weighted curvature value of the key point at different scaling scales based on the adaptive main neighborhood radius;

[0016] S8: determining the eigenvalues of the multi-scale weighted curvature point pairs according to the weighted curvature values;

[0017] S9: Combined with the voting mechanism, the key point sets corresponding to the target point cloud and the source point cloud are aligned based on the eigenvalues of the multi-scale curvature points.

[0018] Second aspect

[0019] An embodiment of the present invention provides a point cloud registration system based on multi-scale curvature point pair features, comprising:

[0020] processor;

[0021] A memory having computer-readable instructions stored thereon, which, when executed by a processor, implements the point cloud registration method based on multi-scale curvature point pair features as in the first aspect.

[0022] The third aspect

[0023] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the point cloud registration method based on multi-scale curvature point pair features according to the first aspect is implemented.

[0024] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0025] In the present invention, by combining adaptive voxel downsampling and weighted ISS sampling, initial key point extraction is performed on the acquired target point cloud and source point cloud. The intersection set between the neighborhood to which each initial key point belongs and multi-scale spheres of different radii is determined, and the self-similarity of each intersection set and the geometric feature score of the neighborhood are calculated. Based on the geometric feature score, each initial key point is relocated to a neighborhood with a geometric feature score less than a preset geometric feature score, thereby obtaining the key point sets of the target point cloud and the source point cloud, respectively, to ensure that the key points are located in areas with more obvious geometric features and less noise. The adaptive main neighborhood radius of each key point in the key point set is calculated. Based on the adaptive main neighborhood radius, the weighted curvature value of the key point at different scaling scales is calculated. Then, based on the weighted curvature value, a multi-scale weighted curvature point pair feature is constructed. Finally, combined with a voting mechanism, the initial registration between the corresponding key point sets of the target point cloud and the source point cloud is performed based on the multi-scale curvature point pair feature value, effectively improving the robustness and accuracy of the registration. The present invention proposes a point cloud registration method based on multi-scale complex analysis and global optimization, which combines multi-scale key point extraction and weighted curvature descriptors to achieve efficient and accurate point cloud registration in complex geometric and noisy scenes, effectively reduce the interference of high noise and outliers, improve the ability and robustness of the registration method to capture the global structure of the scene, improve matching efficiency and accuracy, enhance the convergence of fine registration, and effectively solve the interference of low overlap problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A schematic diagram of a process flow of a point cloud registration method based on multi-scale curvature point pair features provided by an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of an overall registration framework of a point cloud registration method based on multi-scale curvature point pair features provided by an embodiment of the present invention;

[0029] Figure 3 A schematic diagram of a process for determining multi-scale weighted curvature point pair features of a given point pair provided by an embodiment of the present invention;

[0030] Figure 4 A schematic structural diagram of a point cloud registration system based on multi-scale curvature point pair features provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0032] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0033] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0034] Reference Manual Figure 1 , shows a flow chart of a point cloud registration method based on multi-scale curvature point pair features provided by an embodiment of the present invention.

[0035] An embodiment of the present invention provides a point cloud registration method based on multi-scale curvature point pair features. This method can be implemented by a point cloud registration device based on multi-scale curvature point pair features, which can be a terminal or a server. The processing flow of the point cloud registration method based on multi-scale curvature point pair features can include the following steps:

[0036] S1: Obtain the patient's preoperative target point cloud and intraoperative source point cloud.

[0037] Among them, the target point cloud refers to the three-dimensional scanning data of the patient's body parts collected before the operation, which is usually used as the reference data for alignment. The source point cloud refers to the three-dimensional scanning data of the patient's body parts collected in real time during the operation, which needs to be aligned with the target point cloud.

[0038] It is important to note that obtaining the preoperative target point cloud and the intraoperative source point cloud lays the foundation for subsequent point cloud registration. This step, combining preoperative and intraoperative data, effectively addresses point cloud mismatches caused by factors such as tissue deformation and displacement during surgery, improving the accuracy and robustness of registration and providing reliable data support for real-time navigation and precise operation during surgery.

[0039] S2: Combine adaptive voxel down-sampling and weighted ISS sampling to extract initial keypoints in the target point cloud and source point cloud.

[0040] Among them, adaptive voxel downsampling is a sampling method based on the spatial distribution of point clouds. By adjusting the voxel size, it adaptively and evenly samples the point cloud data, reducing redundant points while retaining key geometric information. Weighted ISS (Intrinsic Shape Signature) sampling is a key point detection method based on the intrinsic shape characteristics of the points. It uses a weighted strategy to prioritize points with significant geometric features as key points. The initial key points are a set of points with significant geometric features extracted from the point cloud data. These points exhibit strong spatial recognition capabilities and are usually located at the edges, corners, or other areas with unique curvature characteristics of the point cloud.

[0041] It should be noted that the combination of adaptive voxel downsampling and weighted ISS sampling can reduce the size of point cloud data while retaining key geometric features and improving the efficiency and accuracy of subsequent calculations.

[0042] In a possible implementation, S2 specifically includes:

[0043] S201: Downsample the source point cloud and the target point cloud through a voxel filter to obtain a candidate point set.

[0044] S202: Pre-locate key points in the candidate point set using an intrinsic shape feature (ISS) detector to generate initial key points of the target point cloud and the source point cloud.

[0045] It should be noted that adaptive voxel sampling ensures the uniform distribution of point cloud data, and weighted ISS sampling further selects key points with high discriminability, making the registration more robust, especially in complex or noisy point clouds.

[0046] Reference Manual Figure 2 , showing a schematic diagram of the overall registration framework of a point cloud registration method based on multi-scale curvature point pair features provided by an embodiment of the present invention.

[0047] Figure 2In this paper, the source point cloud represents the current point cloud (intraoperative point cloud), and the target point cloud represents the reference point cloud (preoperative point cloud). The source point cloud and the target point cloud are input, and the spatial position of the source point cloud relative to the target point cloud is determined through point cloud registration. Secondly, based on the multi-scale key point repositioning algorithm, the key points of the point cloud are optimized to obtain the source key point set and the target key point set, making them more robust and able to effectively reflect the geometric characteristics of the point cloud. The optimized key points are described to generate a multi-scale geometric feature descriptor for subsequent point cloud feature matching. Based on the global model description, all point pair features of the source point cloud and the target point cloud are stored to construct a feature matching model. The weighted curvature voting is similar to the Hough voting mechanism to find the best pose transformation (rotation matrix and translation vector) to complete the initial registration. The initial registration result is further optimized through global optimization to ensure the registration accuracy, and the final point cloud registration is completed to obtain a high-precision alignment result of the source point cloud and the target point cloud, that is, the registration result.

[0048] S3: Determine the set of intersections between the neighborhood of each initial key point and the multi-scale spheres with different radii.

[0049] Among them, the neighborhood refers to the local range of a key point in the point cloud, which is usually used to analyze the geometric characteristics around the key point. The multi-scale sphere refers to the spherical area constructed around the initial key point at different radius scales, which is used to explore the changes of local geometric features at different scales. The intersection set refers to the point set composed of the intersection of the multi-scale sphere and the point cloud data in the neighborhood of the key point, which reflects the local geometric distribution at different scales.

[0050] It is important to note that by analyzing the intersection of the neighborhood of the initial keypoints and the multi-scale sphere, we can effectively capture the diversity of local geometric features in the point cloud as they vary with scale. This multi-scale analysis enables the method to adapt to point cloud data of varying resolutions while significantly enhancing the geometric description and global recognition of keypoints.

[0051] S4: Calculate the self-similarity of each intersection set, and determine the geometric feature score of the neighborhood based on the self-similarity.

[0052] Among them, self-similarity is a measure of the geometric consistency of the intersection point set in the point cloud, reflecting the distribution characteristics of the local structure at different scales and the consistency of the normal vector direction. The geometric feature score is an indicator that quantifies the strength of the geometric features of the neighborhood of key points in the point cloud, and is used to evaluate the geometric characteristics of the neighborhood point set.

[0053] It should be noted that a multi-scale analysis method combining self-similarity and geometric feature scores with Gaussian weighting factors can accurately quantify the geometric properties of keypoint neighborhoods at different scales. Self-similarity makes this method more robust to local geometric features, while the use of geometric feature scores ensures the scientific nature of feature selection.

[0054] In a possible implementation, S4 specifically includes:

[0055] S401: Calculate the self-similarity of each intersection set:

[0056]

[0057] Among them, SS i represents the self-similarity at the i-th scale, c represents the total number of points in the intersection set, j represents the index number of the points in the intersection set, n0 represents the normal vector, n(q j ) represents the jth point q in the intersection set j The normal vector, q j Represents the jth point in the intersection set.

[0058] S402: Calculate the geometric feature score of the neighborhood based on the self-similarity:

[0059]

[0060] Among them, g represents the geometric feature score, λ i represents the Gaussian weight factor at the i-th scale, exp represents the natural exponential function, R i represents the neighborhood radius of the i-th scale, and σ represents the scale parameter that controls the Gaussian function.

[0061] It should be noted that the application of Gaussian weighting factors balances the contribution of multi-scale features and avoids the dominance of a certain scale in the registration results. This method not only improves the accuracy of key point positioning, but also enhances the applicability of point cloud registration in complex geometric structures, laying a high-quality foundation for subsequent feature relocation and key point extraction.

[0062] S5: Based on the geometric feature scores, each initial key point is relocated in the target area to obtain the key point sets of the target point cloud and the source point cloud respectively. The target neighborhood is specifically the neighborhood whose geometric feature score is less than the preset geometric feature score.

[0063] Among them, the preset geometric feature score refers to a threshold value used to filter out key point neighborhoods with low geometric feature scores. These areas usually contain information with more significant geometric features. The key point set is a group of specific points extracted from the point cloud. These points are significant in geometric structure and can represent the core features of the point cloud, reducing the amount of calculation while retaining geometric information.

[0064] Among them, those skilled in the art can set the size of the preset geometric feature score according to actual conditions, and the present invention does not limit it.

[0065] It should be noted that through the evaluation and relocalization mechanism of geometric feature scores, the initial keypoints are moved to areas with more significant geometric features, making the final keypoint set more stable and representative. This method can effectively filter out noise and redundant points, while enhancing the keypoints' ability to describe the overall structure of the point cloud. This lays a more solid foundation for subsequent feature extraction and registration steps, and improves the accuracy and robustness of point cloud registration.

[0066] S6: Calculate the adaptive main neighborhood radius of each key point in the key point set.

[0067] The adaptive primary neighborhood radius is a radius dynamically calculated based on the distribution of the nearest neighbors around each keypoint. It reflects the local geometric characteristics of the keypoint and adapts to the needs of areas with varying densities. By calculating the adaptive primary neighborhood radius, the neighborhood range can be dynamically adjusted based on changes in the point cloud density around the keypoint, allowing for detailed local characterization in dense areas while maintaining global integrity in sparse areas.

[0068] In a possible implementation, S6 specifically includes:

[0069]

[0070] Among them, r i represents the adaptive primary neighborhood radius of the i-th key point, k represents the total number of nearest neighbors, d represents the Euclidean distance symbol, and p i represents the i-th key point, p j represents the jth neighbor point, d(p i ,p j ) represents the key point p i The corresponding neighbor point p j The Euclidean distance between .

[0071] It should be noted that adaptability significantly improves the accuracy and robustness of key point description, helps to more accurately capture the geometric characteristics of the point cloud, and provides a reliable foundation for feature extraction and curvature calculation in subsequent steps.

[0072] S7: Calculate the weighted curvature value of the key point at different scaling scales based on the adaptive main neighborhood radius.

[0073] The weighted curvature value is a local curvature value calculated by weighting neighborhood points at different scales. It reflects the curvature of the point cloud surface at that location. By calculating the weighted curvature value based on an adaptive primary neighborhood radius and different scaling scales, it can capture the multi-layered geometric features of the point cloud in detail. Multi-scale analysis enables the algorithm to flexibly adapt to point cloud data of varying resolutions and accurately extract both local and global curvature information.

[0074] In a possible implementation, S7 specifically includes:

[0075] S701: Calculate the neighborhood radius of the key point at different zoom scales based on the adaptive main neighborhood radius:

[0076] r i,j =s j ×r i

[0077] Among them, r i,j represents the neighborhood radius of the i-th key point at the j-th scale, and sj represents the j-th scale factor.

[0078] S702: Determine a multi-scale neighborhood point set of the key point according to the neighborhood radius.

[0079] S703: Calculate the geometric center of the neighborhood point set at the current scale:

[0080]

[0081] Among them, c i Indicates the current scale s j The geometric center of the neighborhood point set, m represents the total number of neighborhood points, q k represents the kth neighborhood point, N i,j Represents the key point p at the jth scale i The neighborhood point set of .

[0082] S704: Determine the offset vector of each neighborhood point based on the geometric center:

[0083] q′ k =q k -c i

[0084] Among them, q′ k Represents the offset vector of the kth neighborhood point about the geometric center.

[0085] S705: Construct the covariance matrix of the offset vector:

[0086]

[0087] Among them, C i,j Indicates that the i-th key point is at scale s j The covariance matrix within the neighborhood, T represents the matrix transpose.

[0088] S706: Perform eigenvalue decomposition on the covariance matrix to determine the local curvature value at each scale:

[0089]

[0090] Among them, curvature i,j Represents the i-th key point p i At the jth scale s j The local curvature values under λ1, λ2 and λ3 all represent the eigenvalues of the covariance matrix greater than 0, λ1≤λ2≤λ3.

[0091] S707: Calculate the weighted curvature value based on the local curvature values at each scale:

[0092]

[0093] Among them, w j represents the Gaussian weight of the jth scale, s j represents the jth scale, mc i Represents the key point p i The multi-scale weighted curvature value of .

[0094] It should be noted that the geometric features of the point cloud are effectively extracted through multi-scale neighborhood analysis, adaptive curvature calculation, and multi-scale curvature fusion. Specifically, the adaptive main neighborhood radius dynamically adjusts the neighborhood range to capture geometric information in sparse and dense areas. The calculation of the geometric center and the introduction of the covariance matrix improve the stability of local feature modeling. Finally, through the Gaussian weighted integration of multi-scale curvature values, local features are smoothed and the incompleteness of single-scale descriptions is avoided. This holistic approach can capture more comprehensive and robust geometric features in complex scenes, providing high-precision support for point cloud registration and subsequent processing.

[0095] Reference Manual Figure 3 , shows a schematic flow chart of a multi-scale weighted curvature point pair feature of a given point pair provided by an embodiment of the present invention.

[0096] Figure 3 In, The multiscale neighborhood of p i Represents point p i The multiscale neighborhood of p j Represents point p j The normal vectors of p i Represents point p i The normal vectors of p j Represents point p j The normal vector, p i and p j Represents the two key points in the point pair, p i is the key point in the source point cloud, p jis the key point in the target point cloud, n i Represents point p i Normal vector, n j Represents point p j The normal vector of point pair (p i , p j ), r i,1 、r i,2 、r i,3 、r j,1 、r j,2 and r j,3 They represent the neighborhood range at different scales, i.e., the radius, and {pi, εi} represents point p i A scale neighborhood point set, ∠(n i ,d) represents the normal vector n i The angle between the point vector d, ∠(n j ,d) represents the normal vector n j The angle between the point vector d, ∠(n i ,n j ) represents the normal vector n i With the normal vector n j The angle between them.

[0097] The multi-scale neighborhood analysis and point pair feature description shown in the figure provide a robust basis for point cloud registration, including multi-scale neighborhood, multi-scale geometric feature extraction, point pair feature description, etc., highlighting the robustness and adaptability of the method, and providing solid theoretical support for point cloud matching and registration.

[0098] S8: Determine the eigenvalues of the multi-scale weighted curvature point pairs according to the weighted curvature values.

[0099] Among them, the multi-scale weighted curvature point pair eigenvalue (FMC-PPF) is a point cloud feature descriptor based on multi-scale weighted curvature. It comprehensively considers the geometric structure, normal vector, curvature value and relative position relationship of the point pair, and is used to describe the geometric characteristics of two points in the point cloud.

[0100] It is important to note that by constructing a multi-scale weighted curvature point pair feature (FMC-PPF), we fully utilize the local geometric information and global topological relationships of the point pairs. This feature combines information such as curvature value, normal vector direction, and inter-point distance to describe the characteristics of the point cloud in a highly robust and discriminative form, thereby improving the accuracy and stability of the registration process.

[0101] In a possible implementation, the formula for constructing the eigenvalue of the multi-scale weighted curvature point pair is specifically:

[0102] F MC-PPF =MC-PPF(p i ,pj ,n i ,n j ,mc i ,mc j )=(||d||2,∠(n i ,d),∠(n j ,d),∠(n i ,n j ),mc i ,mc j ) T

[0103] Among them, F MC-PPF Represents the multi-scale weighted curvature point pair eigenvalue, p j represents the jth key point, n i Represents the i-th key point p i The normal vector, n j Represents the jth key point p j The normal vector, mc i Represents the key point p i The multi-scale weighted curvature value, mc j Represents the key point p j The multi-scale weighted curvature value of ||d||2 represents the two-norm of vector d, that is, the key point p i and key point p j The Euclidean distance between i ,d) represents the normal vector n i The angle between the point vector d, ∠(n j ,d) represents the normal vector n j The angle between the point vector d, ∠(n i ,n j ) represents the normal vector n i With the normal vector n j The angle between them.

[0104] Specifically, the eigenvalues of the multi-scale curvature point pairs are stored in a hash table, and the eigenvalues of the multi-scale curvature point pairs are used as keys. The correspondence between the source key point set and the target key point set is found in the hash table through feature matching.

[0105] It should be noted that the multi-scale feature construction method can adapt to point clouds of different resolutions, enhance the model's generalization ability for diverse scenes, and provide a solid foundation for subsequent fine alignment and application.

[0106] S9: Combined with the voting mechanism, the key point sets corresponding to the target point cloud and the source point cloud are aligned based on the eigenvalues of the multi-scale curvature points.

[0107] Among them, the voting mechanism is a statistical-based method used to select the optimal registration result among multiple possible matching point pairs. The multi-scale curvature point pair eigenvalue is a descriptor defined between two points to express the geometric relationship between them. Registration refers to aligning multiple point cloud data into the same coordinate system for unified analysis or application.

[0108] It should be noted that by combining feature matching with geometric consistency constraints for coarse registration, the geometric consistency between feature points is exploited to improve matching accuracy while effectively removing the interference of noise point pairs. This method can quickly find the approximate alignment position of the point cloud in complex scenes, providing a reliable initial estimate for subsequent fine registration.

[0109] In a possible implementation, S9 specifically includes:

[0110] S901: Calculate the difference between the weighted curvature values of each scale.

[0111] S902: Based on the difference, the voting value of the point pair feature in the weighted voting is calculated using the Hough voting mechanism:

[0112]

[0113] Among them, V α represents the voting value, λ represents the adjustment factor, log represents the logarithmic function, Represents the key point m in the target point cloud r The multi-scale weighted curvature descriptor of Represents the key point m in the source point cloud i Multi-scale weighted curvature descriptor of , +1 represents the translation operation.

[0114] It should be noted that the Hough voting mechanism is a voting-based pattern recognition method commonly used to detect geometric shapes (such as lines, circles, and ellipses) and for feature matching. It maps data from the input space to the parameter space and counts the number of "votes" for parameter combinations that meet certain conditions, thereby identifying shapes or features that meet the conditions. Using a voting mechanism similar to Hough, when a correspondence is obtained, votes are accumulated in a two-dimensional accumulator. The more reliable the vote, the higher its weight in the voting process. Finally, based on the voting results, the pose transformation with the highest number of votes is selected to obtain the rotation matrix and translation vector between the two point clouds, completing the initial registration of the two point clouds.

[0115] S903: According to the voting value, the pose transformation with the highest number of votes is selected to complete the initial registration of the target point cloud and the source point cloud.

[0116] Among them, pose transformation is a mathematical representation method that describes the changes in the position (Position) and posture (Orientation) of an object in space. It is widely used in robotics, computer vision, point cloud registration and other fields.

[0117] It should be noted that by combining the global representativeness of multi-scale features with the efficiency of the voting mechanism, a highly robust, efficient, and accurate initial registration of complex point cloud data is achieved at a low computational cost. This lays a solid foundation for subsequent fine-grained registration and point cloud processing in application scenarios.

[0118] Specifically, a precise registration algorithm (such as the ICP algorithm) is used to achieve precise registration of the preoperative point cloud and the intraoperative point cloud, calculate a more accurate rotation matrix and translation vector, and obtain the spatial transformation relationship between the image space and the patient space, thereby optimizing the initial registration.

[0119] In a possible implementation, it is characterized in that the posture transformation specifically includes: a rotation matrix and a translation vector.

[0120] A rotation matrix is a matrix used to describe the rotation of an object in three-dimensional space. It can be used to represent the rotation of a point or vector relative to a fixed coordinate system. A translation vector is a vector that describes the movement of an object from one position to another. It represents the amount of translation of a point or object along certain directions in space.

[0121] It's important to note that rotation matrices and translation vectors are indispensable tools for three-dimensional spatial transformation. Rotation matrices precisely describe the rotation of an object, while translation vectors simply and intuitively represent its translation. Combined, these two tools can efficiently and accurately accomplish complex spatial transformation tasks, finding widespread application in areas such as point cloud registration, image processing, and robot positioning.

[0122] In this paper, a novel registration framework is introduced that leverages point-pair features derived from multi-scale weighted curvature descriptors. This framework combines a multi-scale keypoint relocalization strategy with a weighted voting mechanism to improve robustness and accuracy during the registration process. Compared to local feature-based registration algorithms, the proposed framework combines local weighted curvature features with global geometric information, enhancing feature description capabilities and demonstrating stronger noise immunity compared to existing point-pair feature-based methods. First, a multi-scale keypoint relocalization strategy is employed to extract keypoints from both partial and complete point sets, reducing computational complexity while mitigating keypoint anomalies caused by noise. Second, a geometric feature descriptor based on multi-scale weighted curvature is used to encode each keypoint in the extracted set, ensuring that it is robust to rotation, translation, and scale. Third, a voting-based pairwise registration method integrates local multi-scale curvature features with point-pair features (MC-PPF) for comprehensive global modeling. A weighted voting factor is introduced based on the inter-point curvature difference to prioritize geometrically significant pairs, accelerating coarse registration. A global optimization strategy is then used to refine the registration, which effectively solves the point cloud registration problem with low overlap rate.

[0123] In a specific embodiment, in order to verify the point cloud registration effect under different overlap rates, the point clouds of the actual intraoperative surgical space and the preoperative image space were compared. The results showed that under five groups of different overlap rates, the tooth point cloud and the mandibular point cloud could achieve good registration effects.

[0124] Specifically, to validate the robustness of the proposed registration framework in the presence of noise and anomalies, a series of evaluations were conducted under different noise conditions. Partial tooth point clouds were segmented from the full mandibular point cloud and subjected to rigid random transformations, including five sets of random rotations (Rtrue) and random translations (ttrue). Gaussian noise of varying percentages and distributions was then injected into the selected tooth point clouds to simulate noise interference commonly encountered in clinical scenarios, such as sensor errors or manual measurement errors when acquiring intraoperative tooth exposure point clouds. The number of generated noise points was based on the total number of points in the point cloud and the specified noise percentage, simulating various noise levels. Four different outlier percentages were tested: 10%, 30%, 50%, and 90%. Isotropic and anisotropic noise were applied along the three axes by adjusting the covariance matrix of the Gaussian distribution.

[0125] The robustness of the MC-PPF (Multi-Scale Weighted Curvature Feature)-based method was evaluated by comparing it with four other registration methods: GICP, Fast Global Registration (FGR), PPF, and Cur-PPF. The proposed method demonstrated its resilience to noise and outlier interference. The MC-PPF-based method achieved the lowest rotation and translation error values under all outlier percentages. Under the most challenging noise conditions (90% outliers and anisotropic noise), the MC-PPF method achieved a rotation error of 0.6552° and a translation error of 0.7824mm. This demonstrates superior calibration accuracy compared to the other methods. GICP produced average rotation and translation errors of 156.149° and 400.3561mm, respectively. While GICP minimizes the distance between corresponding points on two surfaces, it suffers from rotational misalignment due to its reliance on point-to-point correspondences. The FGR method reduced the translation error to 146.68mm, but exhibited a larger rotation error of 161.4938°. This method utilizes the Fast Point Feature Histogram (FPFH) feature to enhance point correspondence calculation. However, it remains sensitive to noise and outliers, especially under anisotropic noise conditions. When the outliers reach 90%, the translation error increases from 108.7289mm to 146.68mm and the rotation error increases from 133.1895° to 161.4938°. The PPF method produces an average error of 84.2967mm and 22.4072°. The Cur-PPF method is an enhanced version of the PPF method, with an average error of 31.035mm and 8.959°. The introduction of curvature features helps to improve the model representation and enhance its noise resistance, thereby achieving higher accuracy than the main PPF method. The MC-PPF-based method consistently outperforms all other methods, with an average translation error of 0.7824mm and a rotation error of 0.6552°. This is attributed to the method's ability to fuse multidimensional point pair features with multi-scale weighted curvature, as well as the application of a weighted curvature voting mechanism. These innovations provide a more accurate and stable description of the local geometry, effectively reducing the registration process's sensitivity to noise and making the features more robust to random noise in the point cloud data, thereby improving matching accuracy. Furthermore, the method demonstrates robustness under both isotropic and anisotropic noise conditions. For isotropic noise, the method achieves an average translation error of 1.1456 mm and an average rotation error of 0.5014°. Under anisotropic noise, the average translation error increases slightly to 1.2306 mm, while the rotation error is 0.5076°. A keypoint relocalization strategy further improves these results by re-evaluating and selecting the most representative keypoints during the registration process, reducing the impact of noise and outliers. This improvement enhances the method's robustness in practical applications.

[0126] These advances will enhance the applicability, efficiency, and effectiveness of this framework in various settings, ultimately contributing to improved patient outcomes in surgery.

[0127] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0128] In the present invention, by combining adaptive voxel downsampling and weighted ISS sampling, initial key point extraction is performed on the acquired target point cloud and source point cloud. The intersection set between the neighborhood to which each initial key point belongs and multi-scale spheres of different radii is determined, and the self-similarity of each intersection set and the geometric feature score of the neighborhood are calculated. Based on the geometric feature score, each initial key point is relocated to a neighborhood with a geometric feature score less than a preset geometric feature score, thereby obtaining the key point sets of the target point cloud and the source point cloud, respectively, to ensure that the key points are located in areas with more obvious geometric features and less noise. The adaptive main neighborhood radius of each key point in the key point set is calculated. Based on the adaptive main neighborhood radius, the weighted curvature value of the key point at different scaling scales is calculated. Then, based on the weighted curvature value, a multi-scale weighted curvature point pair feature is constructed. Finally, combined with a voting mechanism, the initial registration between the corresponding key point sets of the target point cloud and the source point cloud is performed based on the multi-scale curvature point pair feature value, effectively improving the robustness and accuracy of the registration. The present invention proposes a point cloud registration method based on multi-scale complex analysis and global optimization, which combines multi-scale key point extraction and weighted curvature descriptors to achieve efficient and accurate point cloud registration in complex geometric and noisy scenes, effectively reduce the interference of high noise and outliers, improve the ability and robustness of the registration method to capture the global structure of the scene, improve matching efficiency and accuracy, enhance the convergence of fine registration, and effectively solve the interference of low overlap problems.

[0129] Reference Manual Figure 4 , which shows a structural schematic diagram of a point cloud registration system based on multi-scale curvature point pair features provided by the present invention.

[0130] The present invention further provides a point cloud registration system 20 based on multi-scale curvature point pair features, which is applied to the above-mentioned point cloud registration method based on multi-scale curvature point pair features, comprising:

[0131] Processor 201.

[0132] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201 , the point cloud registration method based on multi-scale curvature point pair features as in the method embodiment is implemented.

[0133] The point cloud registration system 20 based on multi-scale curvature point pair features provided by the present invention can execute the above-mentioned point cloud registration method based on multi-scale curvature point pair features and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0134] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0135] In the present invention, by combining adaptive voxel downsampling and weighted ISS sampling, initial key point extraction is performed on the acquired target point cloud and source point cloud. The intersection set between the neighborhood to which each initial key point belongs and multi-scale spheres of different radii is determined, and the self-similarity of each intersection set and the geometric feature score of the neighborhood are calculated. Based on the geometric feature score, each initial key point is relocated to a neighborhood with a geometric feature score less than a preset geometric feature score, thereby obtaining the key point sets of the target point cloud and the source point cloud, respectively, to ensure that the key points are located in areas with more obvious geometric features and less noise. The adaptive main neighborhood radius of each key point in the key point set is calculated. Based on the adaptive main neighborhood radius, the weighted curvature value of the key point at different scaling scales is calculated. Then, based on the weighted curvature value, a multi-scale weighted curvature point pair feature is constructed. Finally, combined with a voting mechanism, the initial registration between the corresponding key point sets of the target point cloud and the source point cloud is performed based on the multi-scale curvature point pair feature value, effectively improving the robustness and accuracy of the registration. The present invention proposes a point cloud registration method based on multi-scale complex analysis and global optimization, which combines multi-scale key point extraction and weighted curvature descriptors to achieve efficient and accurate point cloud registration in complex geometric and noisy scenes, effectively reduce the interference of high noise and outliers, improve the ability and robustness of the registration method to capture the global structure of the scene, improve matching efficiency and accuracy, enhance the convergence of fine registration, and effectively solve the interference of low overlap problems.

[0136] It should be understood that the processor in the embodiments of the present invention may be a central processing unit (CPU), but may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0137] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0138] The above embodiments can be implemented in whole or in part through software, hardware (such as circuits), firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function according to the embodiments of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired method (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available media can be magnetic media (such as floppy disks, hard disks, tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0139] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0140] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0141] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0142] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0143] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0144] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms.

[0145] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0146] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0147] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program codes.

[0148] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the point cloud registration method based on multi-scale curvature point pair features as described in the method embodiment is implemented.

[0149] The computer-readable storage medium provided by the present invention can implement the steps and effects of the point cloud registration method based on multi-scale curvature point pair features of the above method embodiment. To avoid repetition, the present invention will not go into details.

[0150] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0151] In the present invention, by combining adaptive voxel downsampling and weighted ISS sampling, initial key point extraction is performed on the acquired target point cloud and source point cloud. The intersection set between the neighborhood to which each initial key point belongs and multi-scale spheres of different radii is determined, and the self-similarity of each intersection set and the geometric feature score of the neighborhood are calculated. Based on the geometric feature score, each initial key point is relocated to a neighborhood with a geometric feature score less than a preset geometric feature score, thereby obtaining the key point sets of the target point cloud and the source point cloud, respectively, to ensure that the key points are located in areas with more obvious geometric features and less noise. The adaptive main neighborhood radius of each key point in the key point set is calculated. Based on the adaptive main neighborhood radius, the weighted curvature value of the key point at different scaling scales is calculated. Then, based on the weighted curvature value, a multi-scale weighted curvature point pair feature is constructed. Finally, combined with a voting mechanism, the initial registration between the corresponding key point sets of the target point cloud and the source point cloud is performed based on the multi-scale curvature point pair feature value, effectively improving the robustness and accuracy of the registration. The present invention proposes a point cloud registration method based on multi-scale complex analysis and global optimization, which combines multi-scale key point extraction and weighted curvature descriptors to achieve efficient and accurate point cloud registration in complex geometric and noisy scenes, effectively reduce the interference of high noise and outliers, improve the ability and robustness of the registration method to capture the global structure of the scene, improve matching efficiency and accuracy, enhance the convergence of fine registration, and effectively solve the interference of low overlap problems.

[0152] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0153] There are a few points to note:

[0154] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0155] (2) For the sake of clarity, the thickness of layers or regions in the drawings used to describe the embodiments of the present invention are exaggerated or reduced, that is, these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region, or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or intervening elements may be present.

[0156] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to form new embodiments.

[0157] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A point cloud registration method based on multi-scale curvature point pair features, characterized in that: include: S1: Obtain the patient's preoperative target point cloud and intraoperative source point cloud; S2: extracting initial key points from the target point cloud and the source point cloud by combining adaptive voxel down sampling and weighted ISS sampling; S3: Determine the set of intersections between the neighborhood of each initial key point and the multi-scale spheres with different radii; S4: Calculating the self-similarity of each intersection set, and determining the geometric feature score of the neighborhood based on the self-similarity; S5: relocating each initial key point in the target area according to the geometric feature score to obtain key point sets of the target point cloud and the source point cloud respectively, wherein the target neighborhood is specifically a neighborhood having a geometric feature score less than a preset geometric feature score; S6: Calculating the adaptive main neighborhood radius of each key point in the key point set; S7: Calculating weighted curvature values of the key point at different scaling scales according to the adaptive main neighborhood radius; S8: Determine a multi-scale weighted curvature point pair eigenvalue according to the weighted curvature value; S9: In combination with a voting mechanism, registration is performed between the target point cloud and the key point set corresponding to the source point cloud according to the multi-scale curvature point pair eigenvalues.

2. The point cloud registration method based on multi-scale curvature point pair features according to claim 1, characterized in that: The S2 specifically includes: S201: downsampling the source point cloud and the target point cloud through a voxel filter to obtain a candidate point set; S202: Pre-locating key points in the candidate point set using an intrinsic shape feature (ISS) detector to generate initial key points of the target point cloud and the source point cloud.

3. The point cloud registration method based on multi-scale curvature point pair features according to claim 1, characterized in that: The S4 specifically includes: S401: Calculate the self-similarity of each intersection set: Among them, SS i represents the self-similarity at the i-th scale, c represents the total number of points in the intersection set, j represents the index number of the points in the intersection set, n0 represents the normal vector, n(q j ) represents the jth point q in the intersection set j The normal vector, q j represents the jth point in the intersection set; S402: Calculate the geometric feature score of the neighborhood based on the self-similarity: Among them, g represents the geometric feature score, λ i represents the Gaussian weight factor at the i-th scale, exp represents the natural exponential function, R i represents the neighborhood radius of the i-th scale, and σ represents the scale parameter that controls the Gaussian function.

4. The point cloud registration method based on multi-scale curvature point pair features according to claim 1, characterized in that: The S6 is specifically: Among them, r i represents the adaptive primary neighborhood radius of the i-th key point, k represents the total number of nearest neighbors, d represents the Euclidean distance symbol, and p i represents the i-th key point, p j represents the jth neighbor point, d(p i ,p j ) represents the key point p i The corresponding neighbor point p j The Euclidean distance between .

5. The point cloud registration method based on multi-scale curvature point pair features according to claim 1, characterized in that: The S7 specifically includes: S701: Calculate the neighborhood radius of the key point at different zoom scales according to the adaptive main neighborhood radius: r i,j ＝s j ×r i Among them, r i,j represents the neighborhood radius of the i-th key point at the j-th scale, and sj represents the j-th scale factor; S702: Determine a multi-scale neighborhood point set of the key point according to the neighborhood radius; S703: Calculate the geometric center of the current scale neighborhood point set: Among them, c i Indicates the current scale s j The geometric center of the neighborhood point set, m represents the total number of neighborhood points, q k represents the kth neighborhood point, N i,j Represents the key point p at the jth scale i The neighborhood point set of ; S704: Determine the offset vector of each neighborhood point according to the geometric center: q′ k =q k -c i Among them, q′ k Represents the offset vector of the kth neighborhood point about the geometric center; S705: Construct the covariance matrix of the offset vector: Among them, C i,j Indicates that the i-th key point is at scale s j The covariance matrix within the neighborhood, T represents the matrix transpose; S706: Perform eigenvalue decomposition on the covariance matrix to determine the local curvature value of each scale: Among them, curvature i,j Represents the i-th key point p i At the jth scale s j The local curvature values under , λ1, λ2 and λ3 all represent the eigenvalues of the covariance matrix greater than 0, λ1≤λ2≤λ3; S707: Calculate the weighted curvature value according to the local curvature values of each scale: Among them, w j represents the Gaussian weight of the jth scale, s j represents the jth scale, mc i Represents the key point p i The multi-scale weighted curvature value of .

6. The point cloud registration method based on multi-scale curvature point pair features according to claim 1, characterized in that: The formula for constructing the eigenvalue of the multi-scale weighted curvature point pair is specifically: F MC-PPF =MC-PPF(p i ,p j ,n i ,n j ,mc i ,mc j )=(||d||2,∠(n i ,d),∠(n j ,d),∠(n i ,n j ),mc i ,mc j ) T Among them, F MC-PPF Represents the multi-scale weighted curvature point pair eigenvalue, p j represents the jth key point, n i Represents the i-th key point p i The normal vector, n j Represents the jth key point p j The normal vector, mc i Represents the key point p i The multi-scale weighted curvature value, mc j Represents the key point p j The multi-scale weighted curvature value of ||d||2 represents the two-norm of vector d, that is, the key point p i and key point p j The Euclidean distance between i ,d) represents the normal vector n i The angle between the point vector d, ∠(n j ,d) represents the normal vector n j The angle between the point vector d, ∠(n i ,n j ) represents the normal vector n i With the normal vector n j The angle between them.

7. The point cloud registration method based on multi-scale curvature point pair features according to claim 1, characterized in that: The S9 specifically includes: S901: Calculating the difference between the weighted curvature values of each scale; S902: Calculate the voting value of the point pair feature in the weighted voting based on the difference using the Hough voting mechanism: Among them, V α represents the voting value, λ represents the adjustment factor, log represents the logarithmic function, Represents the key point m in the target point cloud r The multi-scale weighted curvature descriptor of Represents the key point m in the source point cloud i Multi-scale weighted curvature descriptor, +1 represents the translation operation; S903: Selecting the pose transformation with the highest number of votes according to the voting value to complete the initial registration of the target point cloud and the source point cloud.

8. The point cloud registration method based on multi-scale curvature point pair features according to claim 7, characterized in that: The posture transformation specifically includes: a rotation matrix and a translation vector.

9. A point cloud registration system based on multi-scale curvature point pair features, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the point cloud registration method based on multi-scale curvature point pair features according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the point cloud registration method based on multi-scale curvature point pair features according to any one of claims 1 to 8 is implemented.

Images