Point cloud registration method and system based on ISS feature points
By employing the point cloud registration method based on ISS feature points, and utilizing random downsampling, RANSAC plane fitting, and FPFH descriptors, the problem of slow registration speed in large-scale point cloud registration is solved, achieving efficient and accurate point cloud registration.
Patent Information
- Application Number
- CN202411169993.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-26
AI Technical Summary
Existing point cloud registration methods are computationally complex, especially for large-scale point cloud data, resulting in slow algorithm speed and low matching accuracy under noise and occlusion conditions.
A point cloud registration method based on ISS feature points is adopted. Through random downsampling, RANSAC plane fitting, ISS feature point detection and FPFH descriptor calculation, combined with least squares method to estimate rigid body transformation matrix, the accurate registration of point clouds is achieved.
It improves the accuracy and speed of point cloud registration, can effectively handle large-scale, high-noise point cloud data, and has high robustness.
Smart Images

Figure CN118674760B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a point cloud registration technology, and in particular to a point cloud registration method and system based on ISS feature points. Background Art
[0002] 3D point cloud registration is to align point cloud data taken from different perspectives or at different times so that they represent the same scene or object in the same coordinate system. This is one of the key tasks in fields such as computer vision, robotics, and map building. The main goal of registration is to find the best transformation between two or more point clouds to minimize the difference between them. This process is usually divided into two main stages: coarse registration and fine registration.
[0003] Coarse registration refers to a relatively rough registration when the transformation between two point clouds is completely unknown. The main purpose is to provide a good initial value for the transformation for the fine registration. Coarse registration provides a good starting point for the precise registration so that the subsequent iterative precise matching is more effective. Since most of the point cloud data actually acquired contains noise, noise and outliers may affect the feature extraction and matching of the point cloud, thereby having a negative impact on the registration. The feature-based point cloud registration algorithm is sensitive to noise. In the presence of noise or partial occlusion of the point cloud, the algorithm may produce incorrect matches.
[0004] The existing registration method based on points in point clouds has the problem of complex calculation process, especially for large-scale point clouds, which makes the algorithm slow in practical applications. Summary of the invention
[0005] The embodiments of the present invention provide a point cloud registration method and system based on ISS feature points, which can solve the problems in the prior art.
[0006] According to a first aspect of the embodiments of the present invention,
[0007] Provides point cloud registration methods based on ISS feature points, including:
[0008] Reading original point cloud data from a storage device, and performing random downsampling processing on the original point cloud data to obtain downsampled point cloud data; converting the Euler angle into a rotation matrix according to a preset rotation angle and a translation vector, and performing a rigid transformation on the downsampled point cloud data using the rotation matrix to obtain rigidly transformed point cloud data;
[0009] Applying a random sample consensus algorithm to perform plane fitting on the original point cloud data and the point cloud data after the rigid transformation, respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation; performing intrinsic shape feature ISS feature point detection on the first fitting plane and the second fitting plane, respectively, and calculating the fast point feature histogram FPFH descriptor of the ISS feature point, to obtain a first FPFH descriptor of the original point cloud data and a second FPFH descriptor of the point cloud data after the rigid transformation; calculating the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor, to obtain a Euclidean distance matrix;
[0010] The best matching point pair between the original point cloud data and the point cloud data after the rigid transformation is determined according to the Euclidean distance matrix; the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation is estimated by the least squares method using the best matching point pair; the optimal rigid body transformation matrix is applied to the original point cloud data to achieve alignment between the original point cloud data and the point cloud data after the rigid transformation.
[0011] In an optional embodiment,
[0012] Applying a random sample consensus algorithm to the original point cloud data and the point cloud data after the rigid transformation to perform plane fitting respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation comprises:
[0013] The RANSAC algorithm is used to iteratively select three points to calculate the plane model to perform plane fitting on the original point cloud data and the point cloud data after rigid transformation, where the plane fitting formula is:
[0014] ;
[0015] Among them, (x, y, z) represents the x, y, z axes, (a, b, c) represents the plane normal vector, d Indicates the offset from the origin of the plane.
[0016] In an optional embodiment,
[0017] The intrinsic shape feature ISS feature point detection is performed on the first fitting plane and the second fitting plane respectively, and the fast point feature histogram FPFH descriptor of the ISS feature point is calculated, which includes:
[0018] For each point p on the first fitting plane and the second fitting plane, determine the r-neighborhood formed by all points whose distances are less than a preset radius r, calculate the covariance matrix of all points in the r-neighborhood, perform eigenvalue decomposition on the covariance matrix, and select points whose non-zero eigenvalues are greater than a preset screening threshold as ISS feature points;
[0019] The calculation of the FPFH descriptor includes: for each of the ISS feature points and the normal vector of each point in its neighborhood, calculating the angle between the ISS feature point and each point in the neighborhood, constructing a simplified point feature histogram according to the angle, and performing weighted averaging of the simplified point feature histogram of the ISS feature point and the simplified point feature histogram of each point in the neighborhood to obtain the FPFH descriptor of the ISS feature point.
[0020] In an optional embodiment,
[0021] Calculating the covariance matrix of all points in the r-neighborhood includes:
[0022] ;
[0023] in, C represents the covariance matrix, N r (P) represents the set of all points in the r-neighborhood, P represents the original point cloud data, q represents any point in the r-neighborhood, μ represents the average position of all points in the r-neighborhood.
[0024] In an optional embodiment,
[0025] Calculating the angle between the ISS feature point and each point in the neighborhood, and constructing a simplified point feature histogram according to the angle includes:
[0026] ;
[0027] ;
[0028] ;
[0029] in, α Indicates at point l The angle from the u direction to the w direction on the normal plane; ∅ represents the angle from point l The normal direction to the vector ls The angle of direction, v Indicate point l The vector to point s and the point l The dot product of the normal of θRepresents a vector ls The projection on the normal plane is u The angle between the axes, used to determine the exact quadrant of the vector within the plane; w Indicate point l To point s The vector projection of l of y Components on the normal plane; u Indicate point l To point s The vector projection at point l of x Component on the normal plane.
[0030] In an optional embodiment,
[0031] Estimating the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation by the least square method includes:
[0032] ;
[0033] in, Indicates the point x i Applying a rotation matrix R As a result, t represents the optimal translation vector, y i Indicates the target point cloud and point x i The corresponding matching points, R, t An expression representing the least squares method.
[0034] In the second aspect of the embodiments of the present application,
[0035] Provides a point cloud registration system based on ISS feature points, including:
[0036] The first unit is used to read the original point cloud data from the storage device, and perform random downsampling processing on the original point cloud data to obtain downsampled point cloud data; according to a preset rotation angle and translation vector, convert the Euler angle into a rotation matrix, and use the rotation matrix to perform a rigid transformation on the downsampled point cloud data to obtain rigidly transformed point cloud data;
[0037] The second unit is used to apply a random sample consensus algorithm to perform plane fitting on the original point cloud data and the point cloud data after the rigid transformation, respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation; perform intrinsic shape feature ISS feature point detection on the first fitting plane and the second fitting plane, respectively, and calculate the fast point feature histogram FPFH descriptor of the ISS feature point to obtain a first FPFH descriptor of the original point cloud data and a second FPFH descriptor of the point cloud data after the rigid transformation; calculate the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor to obtain a Euclidean distance matrix;
[0038] The third unit is used to determine the best matching point pair between the original point cloud data and the point cloud data after the rigid transformation according to the Euclidean distance matrix; use the best matching point pair to estimate the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation through the least squares method; apply the optimal rigid body transformation matrix to the original point cloud data to achieve alignment between the original point cloud data and the point cloud data after the rigid transformation.
[0039] According to a third aspect of the embodiments of the present invention,
[0040] An electronic device is provided, comprising:
[0041] processor;
[0042] a memory for storing processor-executable instructions;
[0043] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0044] A fourth aspect of the embodiments of the present invention is:
[0045] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0046] This application reduces the amount of data and improves the efficiency of subsequent processing by randomly downsampling the original point cloud data; uses a random sample consistency algorithm to perform plane fitting on the original point cloud data and the point cloud data after rigid transformation, effectively extracting the main plane features in the point cloud; accurately describes the local geometric features of the point cloud through ISS feature point detection and FPFH descriptor calculation; uses the Euclidean distance matrix to determine the best matching point pair, and estimates the optimal rigid body transformation matrix through the least squares method, achieving accurate registration of the original point cloud and the transformed point cloud. The present invention improves the accuracy and speed of point cloud registration, has high robustness, and can effectively process large-scale, high-noise point cloud data. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 Schematic diagram of the process of the point cloud registration method based on ISS feature points according to an embodiment of the present invention;
[0048] Figure 2 Another schematic diagram of the process of the point cloud registration method based on ISS feature points according to an embodiment of the present invention;
[0049] Figure 3 Schematic diagram of the structure of a point cloud registration system based on ISS feature points according to an embodiment of the present invention. DETAILED DESCRIPTION
[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0051] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0052] Figure 1 FIG. 1 is a flow chart of a point cloud registration method based on ISS feature points according to an embodiment of the present invention. Figure 1 As shown, the method includes:
[0053] S101. Reading original point cloud data from a storage device, and performing random downsampling processing on the original point cloud data to obtain downsampled point cloud data; converting the Euler angle into a rotation matrix according to a preset rotation angle and a translation vector, and performing a rigid transformation on the downsampled point cloud data using the rotation matrix to obtain rigidly transformed point cloud data;
[0054] Exemplarily, point cloud data preprocessing is an important step in point cloud registration, which mainly includes two parts: reading and downsampling of point cloud data. This application uses PCL (Point Cloud Library) to read and process point cloud data. PCL is an open source point cloud processing library that provides read and write interfaces for a variety of point cloud data formats. This article uses PCD (Point Cloud Data) format to store point cloud data. PCD is a point cloud data format customized by PCL, which can store information such as coordinates, colors, and normal vectors of point clouds. By calling relevant functions in the PCL library, point cloud data can be easily read from the PCD file and stored in the point cloud object. The original point cloud data usually contains a large number of data points, and direct processing will affect the efficiency of the algorithm. In order to reduce the amount of point cloud data and improve the registration speed, the original point cloud needs to be downsampled. This application adopts a random downsampling method to approximate the original point cloud by randomly selecting some points in the point cloud. According to the preset rotation angle and translation vector, the Euler angle is converted into a rotation matrix, and the downsampled point cloud data is rigidly transformed using the rotation matrix to obtain the rigidly transformed point cloud data.
[0055] S102. Apply the random sample consensus algorithm to perform plane fitting on the original point cloud data and the point cloud data after the rigid transformation, respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation; perform intrinsic shape feature ISS feature point detection on the first fitting plane and the second fitting plane, respectively, and calculate the fast point feature histogram FPFH descriptor of the ISS feature point to obtain a first FPFH descriptor of the original point cloud data and a second FPFH descriptor of the point cloud data after the rigid transformation; calculate the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor to obtain a Euclidean distance matrix;
[0056] Exemplarily, in order to extract the main geometric features in the point cloud, it is necessary to perform plane fitting on the point cloud. This application uses the random sample consensus (RANSAC) algorithm for plane fitting. RANSAC is a robust model fitting algorithm that can estimate model parameters in data containing noise and outliers. It iteratively solves the optimal model parameters by randomly selecting a data subset, fitting a model, and evaluating the model.
[0057] Specifically, for the input point cloud data, the main steps of RANSAC plane fitting are as follows:
[0058] (1) Randomly select three points in the point cloud and calculate the parameters of the plane model determined by these three points;
[0059] (2) Calculate the distances of all points to the plane model and count the number of inliers whose distances are less than a preset threshold;
[0060] (3) Repeat steps (1) and (2) several times and select the plane model with the largest number of internal points as the optimal model;
[0061] (4) Using all interior points, the parameters of the optimal plane model are refined by the least squares method.
[0062] Through the RANSAC algorithm, the main plane features can be stably extracted from the original point cloud data containing noise and outliers. This application performs RANSAC plane fitting on the original point cloud data and the point cloud data after rigid transformation, respectively, to obtain two fitting planes, which are respectively recorded as the first fitting plane and the second fitting plane. These two fitting planes reflect the main geometric features of the original point cloud and the transformed point cloud, providing a basis for subsequent feature extraction and matching.
[0063] In order to extract stable and repeatable feature points on the fitting plane, this paper uses the intrinsic shape signature (ISS) algorithm for feature point detection. ISS feature points are selected by analyzing the eigenvalue distribution of the local surface of the point cloud. They are scale and rotation invariant and can accurately describe the local geometric features of the point cloud.
[0064] In order to accurately characterize the local geometric features of ISS feature points, this paper uses the fast point feature histogram (FPFH) algorithm for feature description. FPFH is a local feature descriptor based on point normal vectors. It generates a compact and efficient feature vector by calculating the angular relationship between the feature point and its neighborhood points. Through the FPFH algorithm, a compact and efficient feature descriptor can be generated for each ISS feature point, which characterizes the geometric features of the local neighborhood of the feature point. This paper performs FPFH feature description on the ISS feature points of the first fitting plane and the second fitting plane, respectively, and obtains two groups of FPFH feature descriptors, which are recorded as the first FPFH descriptor and the second FPFH descriptor.
[0065] S103. Determine the best matching point pair between the original point cloud data and the point cloud data after the rigid transformation according to the Euclidean distance matrix; use the best matching point pair to estimate the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation by the least squares method; apply the optimal rigid body transformation matrix to the original point cloud data to achieve alignment between the original point cloud data and the point cloud data after the rigid transformation.
[0066] Exemplarily, by calculating the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor, a distance matrix is obtained, which reflects the similarity between the two sets of feature points. In order to establish the correspondence between the original point cloud and the transformed point cloud, it is necessary to determine the best matching point pair from the distance matrix.
[0067] The present application adopts the nearest neighbor distance ratio (NNDR) criterion to screen reliable matching point pairs. For each descriptor in the first FPFH descriptor set, two descriptors with the closest Euclidean distance to it are found in the second FPFH descriptor set, and their distance ratio is calculated.
[0068] In order to further improve the accuracy of matching, this paper uses a cross-check to verify the reliability of candidate matching point pairs. Specifically, for each candidate matching point pair, find the descriptor closest to any point in the second FPFH descriptor set. If the feature points corresponding to the two are the same, then the matching point pair is retained, otherwise it is removed.
[0069] Through the NNDR criterion and cross consistency test, a set of reliable best matching point pairs can be screened out from the candidate matching point pairs, and an accurate correspondence between the original point cloud and the transformed point cloud can be established.
[0070] The point cloud registration method of the present application realizes the accurate registration of the original point cloud and the transformed point cloud through a series of steps such as point cloud preprocessing, rigid transformation, plane fitting, feature point detection, feature description, feature matching and rigid body transformation estimation. The method uses the RANSAC algorithm to extract the main plane features of the point cloud, characterizes the local geometric features of the point cloud through ISS feature point detection and FPFH feature description, uses the NNDR criterion and cross consistency test to screen reliable matching point pairs, and uses the least squares method to estimate the optimal rigid body transformation matrix, finally realizing robust and efficient point cloud registration.
[0071] In an optional embodiment,
[0072] Applying a random sample consensus algorithm to the original point cloud data and the point cloud data after the rigid transformation to perform plane fitting respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation comprises:
[0073] The RANSAC algorithm is used to iteratively select three points to calculate the plane model to perform plane fitting on the original point cloud data and the point cloud data after rigid transformation, where the plane fitting formula is:
[0074] ;
[0075] Among them, (x, y, z) represents the x, y, z axes, (a, b, c) represents the plane normal vector, d Indicates the offset from the origin of the plane.
[0076] In an optional embodiment,
[0077] The intrinsic shape feature ISS feature point detection is performed on the first fitting plane and the second fitting plane respectively, and the fast point feature histogram FPFH descriptor of the ISS feature point is calculated, which includes:
[0078] For each point p on the first fitting plane and the second fitting plane, determine the r-neighborhood formed by all points whose distances are less than a preset radius r, calculate the covariance matrix of all points in the r-neighborhood, perform eigenvalue decomposition on the covariance matrix, and select points whose non-zero eigenvalues are greater than a preset screening threshold as ISS feature points;
[0079] The calculation of the FPFH descriptor includes: for each of the ISS feature points and the normal vector of each point in its neighborhood, calculating the angle between the ISS feature point and each point in the neighborhood, constructing a simplified point feature histogram according to the angle, and performing weighted averaging of the simplified point feature histogram of the ISS feature point and the simplified point feature histogram of each point in the neighborhood to obtain the FPFH descriptor of the ISS feature point.
[0080] In an optional embodiment,
[0081] Calculating the covariance matrix of all points in the r-neighborhood includes:
[0082] ;
[0083] in, C represents the covariance matrix, N r (P) represents the set of all points in the r-neighborhood, P represents the original point cloud data, q represents any point in the r-neighborhood, μ represents the average position of all points in the r-neighborhood.
[0084] In an optional embodiment,
[0085] Calculating the angle between the ISS feature point and each point in the neighborhood, and constructing a simplified point feature histogram according to the angle includes:
[0086] ;
[0087] ;
[0088] ;
[0089] in, α Indicates at point l The angle from the u direction to the w direction on the normal plane; ∅ represents the angle from point lThe normal direction to the vector ls The angle of direction, v Indicate point l The vector to point s and the point l The dot product of the normal of θ Represents a vector ls The projection on the normal plane is u The angle between the axes, used to determine the exact quadrant of the vector within the plane; w Indicate point l To point s The vector projection at point l of y Components on the normal plane; u Indicate point l To point s The vector projection at point l of x Component on the normal plane.
[0090] In an optional embodiment,
[0091] Estimating the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation by the least square method includes:
[0092] ;
[0093] in, Indicates the point x i Applying a rotation matrix R As a result, t represents the optimal translation vector, y i Indicates the target point cloud and point x i The corresponding matching points, R, t An expression representing the least squares method.
[0094] Another embodiment of the present application is as follows:
[0095] The original point cloud data is read from the storage device and random downsampling is performed to reduce the amount of data and improve the efficiency of subsequent processing steps. After random downsampling, predefined rotation and translation transformations are applied to the point cloud data. This process involves converting the Euler angles into corresponding rotation matrices and combining them with the translation vectors to form a rigid transformation matrix. After applying the rigid transformation matrix, the original point cloud and the transformed point cloud are fitted with planes using the random sample consensus (RANSAC) algorithm. Through this method, the main planes in each point cloud can be effectively identified and extracted, and then key point cloud data can be extracted from each plane. Subsequently, the point clouds in these planes are enhanced by feature analysis, by detecting Intrinsic Shape Signatures (ISS) feature points and calculating the Fast Point Feature Histograms (FPFH) descriptors for each feature point. These descriptors are used to describe the local geometric structure in the point cloud, which is convenient for subsequent feature matching. The best matching relationship between the feature points in the original and transformed point clouds is determined by calculating the Euclidean distance matrix between the two sets of FPFH descriptors. The rigid transformation relationship is estimated using matching point pairs, and the transformation matrix obtained through this relationship is used to adjust the transformed point cloud data to achieve accurate alignment with the original point cloud.
[0096] Figure 2 FIG. 4 is another flow chart of the point cloud registration method based on ISS feature points according to an embodiment of the present invention. Figure 2 As shown, the specific implementation steps are:
[0097] S1: Read the original point cloud data (point cloud 1) from the storage device and perform random downsampling on the original point cloud data (point cloud 1).
[0098] S2: The Euler angles are converted into rotation matrices through the rotation angle and translation vector of the randomly downsampled point cloud (the rotation angle and translation vector parameters can be selected arbitrarily). The rotation matrix is used to perform a rigid transformation on the randomly downsampled point cloud to obtain the rigidly transformed point cloud data (point cloud 2).
[0099] S3: Apply the random sample consensus algorithm (RANSAC algorithm) to the original point cloud data (point cloud 1) and the point cloud data after rigid transformation (point cloud 2) to perform plane fitting to obtain the first fitting plane and the second fitting plane. The first fitting plane is the fitting plane of the original point cloud data (point cloud 1), and the second fitting plane is the fitting plane of the point cloud data (point cloud 2) after rigid transformation.
[0100] S4: Perform ISS (Intrinsic Shape Signatures) feature point detection on the first fitting plane and the second fitting plane respectively and calculate the FPFH (Fast Point Feature Histograms) descriptors of the ISS feature points to obtain two sets of FPFH descriptors, namely the FPFH descriptors of the original point cloud data (point cloud 1) and the FPFH descriptors of the point cloud data after rigid transformation (point cloud 2).
[0101] S5: Calculate the Euclidean distance between the two groups of FPFH descriptors to obtain a Euclidean distance matrix between the two groups of FPFH descriptor features.
[0102] S6: Determine the best matching relationship between the original point cloud data (point cloud 1) and the point cloud data after rigid transformation (point cloud 2) through the Euclidean distance matrix.
[0103] S7: Obtain a set of matching point pairs through the Euclidean distance matrix and the best matching relationship.
[0104] S8: According to the least squares method, the rotation and translation parameters that minimize the distance between the matching point pairs are found, and the optimal rigid body transformation matrix between the original point cloud (point cloud 1) and the point cloud after rigid transformation (point cloud 2) is obtained.
[0105] S9: Apply the optimal rigid body transformation matrix to perform rotation and translation transformation on the original point cloud (point cloud 1) to align it with the point cloud after rigid transformation (point cloud 2).
[0106] Figure 3 FIG. 1 is a schematic diagram of the structure of a point cloud registration system based on ISS feature points according to an embodiment of the present invention. Figure 3 As shown, the system comprises:
[0107] The first unit is used to read the original point cloud data from the storage device, and perform random downsampling processing on the original point cloud data to obtain downsampled point cloud data; according to a preset rotation angle and translation vector, convert the Euler angle into a rotation matrix, and use the rotation matrix to perform a rigid transformation on the downsampled point cloud data to obtain rigidly transformed point cloud data;
[0108] The second unit is used to apply a random sample consensus algorithm to perform plane fitting on the original point cloud data and the point cloud data after the rigid transformation, respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation; perform intrinsic shape feature ISS feature point detection on the first fitting plane and the second fitting plane, respectively, and calculate the fast point feature histogram FPFH descriptor of the ISS feature point to obtain a first FPFH descriptor of the original point cloud data and a second FPFH descriptor of the point cloud data after the rigid transformation; calculate the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor to obtain a Euclidean distance matrix;
[0109] The third unit is used to determine the best matching point pair between the original point cloud data and the point cloud data after the rigid transformation according to the Euclidean distance matrix; use the best matching point pair to estimate the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation through the least squares method; apply the optimal rigid body transformation matrix to the original point cloud data to achieve alignment between the original point cloud data and the point cloud data after the rigid transformation.
[0110] According to a third aspect of the embodiments of the present invention,
[0111] An electronic device is provided, comprising:
[0112] processor;
[0113] a memory for storing processor-executable instructions;
[0114] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0115] A fourth aspect of the embodiments of the present invention is:
[0116] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0117] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.
[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. The point cloud registration method based on ISS feature points is characterized by: include: Reading original point cloud data from a storage device, and performing random downsampling processing on the original point cloud data to obtain downsampled point cloud data; According to a preset rotation angle and translation vector, the Euler angle is converted into a rotation matrix, and the down-sampled point cloud data is rigidly transformed using the rotation matrix to obtain rigidly transformed point cloud data; Applying a random sample consensus algorithm to perform plane fitting on the original point cloud data and the point cloud data after the rigid transformation, respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation; performing intrinsic shape feature ISS feature point detection on the first fitting plane and the second fitting plane, respectively, and calculating the fast point feature histogram FPFH descriptor of the ISS feature point, to obtain a first FPFH descriptor of the original point cloud data and a second FPFH descriptor of the point cloud data after the rigid transformation; calculating the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor, to obtain a Euclidean distance matrix; Determine the best matching point pair between the original point cloud data and the point cloud data after the rigid transformation according to the Euclidean distance matrix; use the best matching point pair to estimate the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation by the least square method; apply the optimal rigid body transformation matrix to the original point cloud data to achieve registration of the original point cloud data and the point cloud data after the rigid transformation; The intrinsic shape feature ISS feature point detection is performed on the first fitting plane and the second fitting plane respectively, and the fast point feature histogram FPFH descriptor of the ISS feature point is calculated, which includes: For each point p on the first fitting plane and the second fitting plane, determine the r-neighborhood formed by all points whose distances are less than a preset radius r, calculate the covariance matrix of all points in the r-neighborhood, perform eigenvalue decomposition on the covariance matrix, and select points whose non-zero eigenvalues are greater than a preset screening threshold as ISS feature points; The calculation of the FPFH descriptor includes: for each of the ISS feature points and the normal vector of each point in its neighborhood, calculating the angle between the ISS feature point and each point in the neighborhood, constructing a simplified point feature histogram according to the angle, and performing weighted averaging on the simplified point feature histogram of the ISS feature point and the simplified point feature histogram of each point in the neighborhood to obtain the FPFH descriptor of the ISS feature point; The nearest neighbor distance ratio (NNDR) criterion is used to screen matching point pairs. For each descriptor in the first FPFH descriptor set, the two descriptors with the closest Euclidean distance to it are found in the second FPFH descriptor set as candidate matching point pairs, and their distance ratio is calculated. A cross consistency check is used to verify the reliability of candidate matching point pairs. For each candidate matching point pair, the descriptor closest to any point is found in the second FPFH descriptor set. If the feature points corresponding to the two are the same, the matching point pair is retained, otherwise it is removed. Through the NNDR criterion and cross consistency test, a set of best matching point pairs can be screened out from the candidate matching point pairs, and the precise correspondence between the original point cloud and the transformed point cloud can be established; Estimating the optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation by the least square method includes: ; in, Indicates the point x i Applying a rotation matrix R As a result, t represents the optimal translation vector, y i Indicates the target point cloud and point x i The corresponding matching points, R, t An expression representing the least squares method.
2. The method according to claim 1, characterized in that Applying a random sample consensus algorithm to the original point cloud data and the point cloud data after the rigid transformation to perform plane fitting respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation comprises: The RANSAC algorithm is used to iteratively select three points to calculate the plane model to perform plane fitting on the original point cloud data and the point cloud data after rigid transformation, where the plane fitting formula is: ; Among them, (x, y, z) represents the x, y, z axes, (a, b, c) represents the plane normal vector, d Indicates the offset from the origin of the plane.
3. The method according to claim 1, characterized in that Calculating the covariance matrix of all points in the r-neighborhood includes: ; in, C represents the covariance matrix, N r (P) represents the set of all points in the r-neighborhood, P represents the original point cloud data, q represents any point in the r-neighborhood, μ represents the average position of all points in the r-neighborhood.
4. The method according to claim 1, characterized in that: Calculating the angle between the ISS feature point and each point in the neighborhood, and constructing a simplified point feature histogram according to the angle includes: ; ; ; in, α Indicates at point l The angle from the u direction to the w direction on the normal plane; ∅ represents the angle from point l The normal direction to the vector ls The angle of direction, v Indicate point l The vector to point s and the point l The dot product of the normal of θ Represents a vector ls The projection on the normal plane is u The angle between the axes, used to determine the exact quadrant of the vector within the plane; w Indicate point l To point s The vector projection at point l of y Components on the normal plane; u Indicate point l To point s The vector projection of l of x Component on the normal plane.
5. A point cloud registration system based on ISS feature points, used to implement the method described in any one of claims 1 to 4, characterized in that: include: The first unit is used to read the original point cloud data from the storage device, and perform random downsampling processing on the original point cloud data to obtain downsampled point cloud data; According to a preset rotation angle and translation vector, the Euler angle is converted into a rotation matrix, and the down-sampled point cloud data is rigidly transformed using the rotation matrix to obtain rigidly transformed point cloud data; The second unit is used to apply a random sample consensus algorithm to perform plane fitting on the original point cloud data and the point cloud data after the rigid transformation, respectively, to obtain a first fitting plane of the original point cloud data and a second fitting plane of the point cloud data after the rigid transformation; perform intrinsic shape feature ISS feature point detection on the first fitting plane and the second fitting plane, respectively, and calculate the fast point feature histogram FPFH descriptor of the ISS feature point to obtain a first FPFH descriptor of the original point cloud data and a second FPFH descriptor of the point cloud data after the rigid transformation; calculate the Euclidean distance between the first FPFH descriptor and the second FPFH descriptor to obtain a Euclidean distance matrix; A third unit is used to determine the best matching point pair between the original point cloud data and the point cloud data after the rigid transformation according to the Euclidean distance matrix; The optimal rigid body transformation matrix between the original point cloud data and the point cloud data after the rigid transformation is estimated by the least squares method using the best matching point pair; the optimal rigid body transformation matrix is applied to the original point cloud data to achieve the registration of the original point cloud data and the point cloud data after the rigid transformation.
6. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 4.
7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.
