A non-spherical 3D point cloud data registration method
Through data preprocessing and aspherical shape analysis technology, combined with least squares method and ICP algorithm, the problem of low data registration accuracy and efficiency of large-diameter aspherical point clouds is solved, and more accurate and efficient point cloud registration is achieved.
Patent Information
- Application Number
- CN202311435886.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-31
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2043-10-31
AI Technical Summary
When processing large-diameter aspherical point cloud data, the registration accuracy and efficiency between point clouds are low.
Through data preprocessing, denoising and filtering, surface features and edge features are extracted, curvature and normal vectors are calculated in combination with the aspherical shape analysis module, the least squares iterative optimization algorithm is used for registration, and the initial transformation matrix is generated through the ICP algorithm, and iterative optimization is performed. Until the predetermined accuracy is reached.
It improves the accuracy and efficiency of point cloud registration, can better process large-diameter aspherical point cloud data, and provides a more accurate and reliable point cloud registration method.
Smart Images
Figure CN117495926B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a three-dimensional point cloud data registration method, and belongs to the technical field of three-dimensional point cloud data processing. Background Art
[0002] In the field of 3D point cloud data processing, point cloud data is a collection of a large number of 3D coordinate points, usually generated by devices such as laser scanners or cameras; in addition to geometric positions, some point cloud data also have color information; color information is usually obtained by acquiring color images through a camera, and then the color information of the pixels at the corresponding positions is assigned to the corresponding points in the point cloud; intensity information is obtained by the echo intensity collected by the laser scanner receiving device, and this intensity information is related to the surface material, roughness, incident angle direction of the target, as well as the reflection energy of the instrument and the laser wavelength; in recent years, point cloud data processing technology has been widely used in many fields, but when processing large-aperture aspheric point cloud data, there are often problems with low registration accuracy and efficiency between point clouds. Summary of the invention
[0003] The present invention aims to solve the problem of low registration accuracy and efficiency between point clouds when processing large-caliber aspheric point cloud data, and further proposes a method for registering aspheric three-dimensional point cloud data.
[0004] The technical solution adopted by the present invention to solve the above-mentioned problem is: the steps of the present invention include:
[0005] Step 1: De-noising, filtering and data cleaning of point cloud data are achieved through data preprocessing;
[0006] Step 2, extracting surface features and edge features from the point cloud data processed in step 1;
[0007] Step 3: Use surface features and edge features to perform preliminary registration and generate an initial transformation matrix;
[0008] Step 4: Calculate the curvature, normal vector and curvature radius of the point cloud data through the aspherical shape analysis module;
[0009] Step 5: By combining the aspheric shape analysis results with the preliminary registration results, a more accurate point cloud registration result is obtained;
[0010] Step 6: Use the least squares iterative optimization algorithm to improve the registration result and minimize the residual of the point cloud data; iterate multiple times until the predetermined registration accuracy is achieved;
[0011] Step 7: Verify the final registration result and use the root mean square error to evaluate the accuracy of the registration.
[0012] Furthermore, data preprocessing is to denoise the collected point cloud data, use Gaussian filtering to reduce noise, apply outlier removal of point cloud data, and remove abnormal points that are obviously deviated.
[0013] Furthermore, the method for extracting surface features and edge features in step 2 is to calculate the normal vector of each feature point through SIFT and Harris corner detection to determine the directionality of the point cloud data.
[0014] Furthermore, the aspherical shape analysis in step 4 uses the PointNet deep learning network architecture to process the point cloud and extract the curvature and normal vector information in the point cloud data.
[0015] Furthermore, in step 5, the preliminary registration uses the traditional registration algorithm ICP to preliminarily register the feature points from different data sets; an initial transformation matrix is obtained to align one point cloud data set to other data sets.
[0016] The beneficial effects of the present invention are:
[0017] 1. The present invention provides a more accurate and reliable point cloud registration method, which improves the accuracy and efficiency of registration by introducing aspheric shape analysis technology;
[0018] 2. The present invention conducts in-depth analysis on the properties of aspherical point cloud data, such as curvature, normal vector and curvature radius. The algorithm can better process large-aperture aspherical point cloud data. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a flowchart of the algorithm of the present invention. DETAILED DESCRIPTION
[0020] Specific implementation method 1: Combination Figure 1 To illustrate this embodiment, the steps of a three-dimensional point cloud data registration method for an aspherical surface described in this embodiment include:
[0021] Step 1: De-noising, filtering and data cleaning of point cloud data are achieved through data preprocessing;
[0022] Step 2, extracting surface features and edge features from the point cloud data processed in step 1;
[0023] Step 3: Use surface features and edge features to perform preliminary registration and generate an initial transformation matrix;
[0024] Step 4: Calculate the curvature, normal vector and curvature radius of the point cloud data through the aspherical shape analysis module;
[0025] Step 5: By combining the aspheric shape analysis results with the preliminary registration results, a more accurate point cloud registration result is obtained;
[0026] Step 6: Use the least squares iterative optimization algorithm to improve the registration result and minimize the residual of the point cloud data; iterate multiple times until the predetermined registration accuracy is achieved;
[0027] Step 7: Verify the final registration result and use the root mean square error to evaluate the accuracy of the registration.
[0028] Specific implementation method 2: Combination Figure 1 To illustrate this embodiment, the data preprocessing of a non-spherical three-dimensional point cloud data registration method described in this embodiment is to denoise the collected point cloud data, use Gaussian filtering to reduce noise, apply wild value removal of the point cloud data, and remove abnormal points that are obviously deviated.
[0029] Specific implementation method three: Combination Figure 1 To illustrate this embodiment, the method for extracting surface features and edge features in step 2 of a non-spherical three-dimensional point cloud data registration method described in this embodiment is to calculate the normal vector of each feature point through SIFT and Harris corner point detection to determine the directionality of the point cloud data.
[0030] Specific implementation method four: Combination Figure 1 To illustrate this embodiment, in step 4 of a method for aligning aspherical three-dimensional point cloud data, the aspherical shape analysis described in this embodiment uses the PointNet deep learning network architecture to process the point cloud and extract the curvature and normal vector information in the point cloud data.
[0031] Specific implementation method five: Combination Figure 1 To illustrate this embodiment, in step 5 of a non-spherical three-dimensional point cloud data registration method described in this embodiment, the preliminary registration uses a traditional registration algorithm ICP to preliminarily align feature points from different data sets; an initial transformation matrix is obtained to align one point cloud data set to other data sets.
[0032] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.
Claims
1. A method for registering three-dimensional point cloud data of an aspherical surface, characterized in that: The steps of the aspherical three-dimensional point cloud data registration method include: Step 1: De-noising, filtering and data cleaning of point cloud data are achieved through data preprocessing; Step 2, extracting surface features and edge features from the point cloud data processed in step 1; Step 3: Use surface features and edge features to perform preliminary registration and generate an initial transformation matrix; Step 4: Calculate the curvature, normal vector and curvature radius of the point cloud data through the aspherical shape analysis module; Step 5: By combining the aspheric shape analysis results with the preliminary registration results, a more accurate point cloud registration result is obtained; Step 6: Use the least squares iterative optimization algorithm to improve the registration result and minimize the residual of the point cloud data; iterate multiple times until the predetermined registration accuracy is achieved; Step 7: Verify the final registration result and use the root mean square error to evaluate the accuracy of the registration.
2. The method for aligning aspherical three-dimensional point cloud data according to claim 1, characterized in that: Data preprocessing is to denoise the collected point cloud data, use Gaussian filtering to reduce noise, apply outlier removal of point cloud data, and remove abnormal points with obvious deviations.
3. The method for aligning aspherical three-dimensional point cloud data according to claim 1, characterized in that: The method for extracting surface features and edge features in step 2 is to calculate the normal vector of each feature point through SIFT and Harris corner detection to determine the directionality of the point cloud data.
4. The method for aligning aspherical three-dimensional point cloud data according to claim 1, characterized in that: In step 4, the aspherical shape analysis uses the PointNet deep learning network architecture to process the point cloud and extract the curvature and normal vector information in the point cloud data.
5. The method for registering three-dimensional point cloud data of an aspherical surface according to claim 1, characterized in that: In step 5, the initial registration uses the traditional registration algorithm ICP to preliminarily register feature points from different data sets; the initial transformation matrix is obtained to align one point cloud data set to other data sets.