CN115797414A - Complex curved surface measurement point cloud data registration method considering measuring head radius - Google Patents
Complex curved surface measurement point cloud data registration method considering measuring head radius Download PDFInfo
- Publication number
- CN115797414A CN115797414A CN202211344520.5A CN202211344520A CN115797414A CN 115797414 A CN115797414 A CN 115797414A CN 202211344520 A CN202211344520 A CN 202211344520A CN 115797414 A CN115797414 A CN 115797414A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- data
- registration
- cloud data
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Length Measuring Devices With Unspecified Measuring Means (AREA)
Abstract
The invention discloses a complex curved surface measurement point cloud data registration method considering measuring head radius, which is used for realizing the high-precision registration of complex curved surface contact measurement and non-contact measurement data, and the point cloud registration process comprises the following steps: the method comprises the following steps: data preprocessing: preprocessing contact measurement data and non-contact measurement data, and removing outlier noise points in two kinds of original point cloud data; the non-contact measurement data is sampled down, the density of points is reduced, and the operation efficiency of the algorithm is improved; step two: data coarse registration: taking the preprocessed contact type measurement point cloud as a target point cloud and the non-contact type measurement point cloud as a source point cloud, transforming the source point cloud data by adopting an SAC-IA registration algorithm to obtain a coarse registration transformation matrix, and transforming the source point cloud data by utilizing the coarse registration transformation matrix to obtain point cloud data after coarse registration; step three: and (3) data fine registration: and performing multiple iterations on the point cloud data after the coarse registration by adopting an iterative closest point algorithm based on corresponding point distance mean square error improvement to obtain a final transformation result. The invention avoids compensation errors caused by the radius compensation of a contact type measuring data measuring head of a complex curved surface, provides an improved iteration closest point optimization algorithm based on the mean square error of the distance of corresponding points for precise registration, and realizes more precise and rapid contact type measurement of the complex curved surface and high-precision registration of point cloud data of non-contact type measurement.
Description
Technical Field
The invention belongs to the technical field of point cloud registration under multi-source data fusion, and relates to a complex curved surface measurement point cloud data registration method considering measuring head radius.
Background
With the rapid development of modern manufacturing industry, the requirements of complex free-form surface part products represented by blades, dies and the like on the aspects of performance, appearance and the like become higher and higher. The reverse engineering technology is particularly critical in the whole process of design development, simulation and modification, rapid prototyping and digital detection of complex curved surface parts. The first step of reverse engineering is to acquire data of a physical model, and the method mainly comprises two main methods, namely contact measurement represented by a three-coordinate measuring machine and non-contact measurement represented by a laser scanner, wherein the two methods have advantages and disadvantages respectively. Under general conditions, the contact type measuring device has high measuring precision, but has small data volume density and low measuring efficiency, and can achieve good measuring effect on a more regular molded surface; and on the contrary, the non-contact measurement is slightly poorer in measurement accuracy, but the data volume density is large, the measurement speed is high, and the complex curved surface features can be well extracted. When the workpiece comprises a regular profile and has a complex profile, a single measurement is difficult to meet the measurement requirement, and at the moment, the contact type measurement method and the non-contact type measurement method are combined to meet the precision requirement and ensure higher efficiency.
In order to realize contact and non-contact measurement data fusion, point cloud registration of multi-source data is one of key technologies, and how to better complete point cloud registration of the multi-source data is a primary problem in a data processing stage. Because the sources of the multi-source reverse data are not the same measuring equipment, the point cloud data have precision difference and density difference and measuring error, so that the point cloud registration has certain difficulty. When a common three-coordinate measuring machine is used for measurement, the obtained data is not the coordinates of a point where the measuring head touches the surface, but the coordinates of the spherical center of the measuring head, and if the coordinates of the surface point are acquired, the radius compensation of the measuring head must be carried out on the original data. At present, a two-dimensional online automatic compensation method is generally adopted by a coordinate measuring machine, but for a complex curved surface such as an engine blade curved surface, the measurement direction may be inconsistent with the normal vector of a measurement point, and the compensation method introduces a compensation error to influence the subsequent registration result.
Disclosure of Invention
Aiming at the problem that compensation errors are introduced into contact measurement data after measuring head radius compensation, and the point cloud registration accuracy of multi-source data is affected, the invention provides a complex curved surface measurement point cloud data registration method considering the measuring head radius.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
the method comprises the following steps: data preprocessing: preprocessing contact measurement data and non-contact measurement data, and removing outlier noise points in two kinds of original point cloud data; the non-contact measurement data is subjected to down-sampling, the density of points is reduced, and the operation efficiency of the algorithm is improved;
step two: data coarse registration: taking the preprocessed contact type measurement point cloud as a target point cloud and the non-contact type measurement point cloud as a source point cloud, transforming the source point cloud data by adopting an SAC-IA registration algorithm to obtain a coarse registration transformation matrix, and transforming the source point cloud data by utilizing the coarse registration transformation matrix to obtain point cloud data after coarse registration;
step three: and (3) data fine registration: and performing multiple iterations on the point cloud data after the coarse registration by adopting a closest point iteration algorithm based on corresponding point distance mean square error improvement to obtain a final transformation result.
The first step comprises the following steps:
step 1-1: carrying out statistical filtering on the two kinds of original point cloud data to remove outlier noise points in the point cloud data;
step 1-2: and uniformly sampling the denoised non-contact measurement point cloud data to reduce the scale of the non-contact measurement point cloud data and obtain source point cloud data required by registration.
The second step comprises the following steps:
step 2-1: at source point cloud P s Sampling s points at random to ensure that the minimum distance between the points is larger than a given threshold value d min ；
Step 2-2: for each sampling point, the point cloud P of the target point t Searching a group of points with similar FPFH characteristics through nearest neighbor search of a K-D tree, and selecting one point from the similar points as a corresponding point;
step 2-3: calculating rigid transformation matrix between n point pairs by SVD singular value decomposition, measuring transformation quality by error between each group of corresponding points after transformation, and returning to minimum error D (e) i ) And transforming the source point cloud data by using the obtained transformation matrix T to obtain point cloud data after coarse registration.
In said step 2-3, an error function D (e) is calculated i ) The method comprises the following steps:
wherein m is e Is a predetermined value, e i Is the distance difference after the ith group corresponding point transformation.
In said step 2-3, the transformation matrix T (R, T) is
Wherein, R is a rotation matrix, and t is a translation vector.
The third step comprises the following steps:
step 3-1: in the above-mentionedSelecting a point set Q = (Q) from the source point cloud after coarse registration 1 ,q 2 ,···,q N ) And selecting a corresponding point set P = (P) from the target point cloud data 1 ,p 2 ,···,p N ) So that each point pair (q) i ,p i ) Satisfy min | | q i -p i || 2 ；
Step 3-2: calculating an initial transformation matrix T through SVD singular value decomposition according to the selected point set Q and the point set P 0 ；
Step 3-3: for Q = (Q) 1 ,q 2 ,···,q N ) Applying an initial transformation T to each point in 0 To yield Q' = (Q) 1 ',q 2 ',···,q N ')；
Step 3-4: calculating the distance d between the point pairs i ＝||p i -q i '|| 2 Averaging their distances
Step 3-5: solving the optimal transformation delta T, and performing iterative computation based on least square method to solve the Euclidean transformation delta T (R, T) so as to make deviation
A minimum value is reached.
Step 3-6: and judging whether convergence occurs according to the iteration errors of the previous iteration and the next iteration, the iteration times and other conditions. If convergence, outputting a final result: t is f ＝ΔT*T 0 Otherwise T 0 ＝ΔT*T 0 And repeating the step 3-1.
The invention has the beneficial effects that: the registration method of the complex curved surface measurement point cloud data considering the measuring head radius provided by the invention avoids the compensation error caused by the complex curved surface contact type measurement data measuring head radius compensation, provides an improved iteration closest point optimization algorithm based on the corresponding point distance mean square error for precise registration, and realizes more accurate and rapid complex curved surface contact type measurement and non-contact type measurement point cloud data high-precision registration.
Drawings
FIG. 1 is a flow chart of the algorithm of the present invention.
FIG. 2 is a schematic diagram of improved iterative closest point algorithm fine registration based on corresponding point distance mean square error.
FIG. 3 is an effect diagram of the contact type and non-contact type point cloud data experimental registration of the blade model by the algorithm.
Detailed Description
The invention is described in detail below with reference to the following figures and embodiments:
the surface of the complex curved surface part is measured by a contact type three-coordinate measuring machine, point cloud data representing the coordinate value of the sphere center of the measuring head can be obtained, radius compensation is not carried out on the contact type measuring data, and the point cloud data is directly used as target point cloud in the registration process; and reconstructing the image through a laser scanner or a CT (computed tomography) slice as a data source of the non-contact measurement point cloud.
As shown in fig. 1, a method for registering complex curved surface measurement point cloud data considering a gauge head radius includes three steps of preprocessing, coarse registration and fine registration. The respective steps will be described in detail below.
The method comprises the following steps: data preprocessing: preprocessing contact measurement data and non-contact measurement data, and removing outlier noise points in two kinds of original point cloud data; the non-contact measurement data is subjected to down-sampling, the density of points is reduced, and the operation efficiency of the algorithm is improved;
the first step comprises the following steps:
step 1-1: carrying out statistical filtering on the two kinds of original point cloud data to remove outlier noise points in the point cloud data;
step 1-2: and uniformly sampling the non-contact type measuring point cloud data after denoising processing to reduce the scale of the non-contact type measuring point cloud data and obtain source point cloud data required by registration.
Step two: data coarse registration: taking the preprocessed contact type measurement point cloud as a target point cloud and the non-contact type measurement point cloud as a source point cloud, transforming the source point cloud data by adopting an SAC-IA registration algorithm to obtain a coarse registration transformation matrix, and transforming the source point cloud data by utilizing the coarse registration transformation matrix to obtain point cloud data after coarse registration;
the purpose of coarse registration is to provide a better initial condition for subsequent fine registration, namely to enable two registered point clouds to be closer, the SAC-IA registration algorithm selected herein adopts FPFH characteristic registration, compared with coarse matching algorithms such as NDT and 4PCS, the SAC-IA registration algorithm can achieve a good registration effect, registration errors between the point clouds are calculated iteratively by matching similar points of an FPFH characteristic diagram, an optimal transformation matrix is found, and the coarse registration of data is completed.
The second step comprises the following steps:
step 2-1: at source point cloud P s S points are randomly sampled to ensure that the minimum distance between the points is larger than a given threshold value d min ；
Step 2-2: for each sampling point, the point cloud P of the target point t Searching a group of points with similar FPFH characteristics through nearest neighbor search of a K-D tree, and selecting one point from the similar points as a corresponding point;
step 2-3: calculating rigid transformation matrix between n point pairs by SVD singular value decomposition, measuring transformation quality by error between each group of corresponding points after transformation, and returning to minimum error D (e) i ) And transforming the source point cloud data by using the obtained transformation matrix T to obtain point cloud data after coarse registration.
In this step 2-3, an error function D (e) is calculated i ) The method comprises the following steps:
wherein m is e Is a predetermined value, e i Is the distance difference after the ith group corresponding point transformation.
In this step 2-3, the transformation matrix T (R, T) is
Wherein, R is a rotation matrix, and t is a translation vector.
Step three: and (3) data fine registration: and performing multiple iterations on the point cloud data after the coarse registration by adopting a closest point iteration algorithm improved based on the corresponding point distance mean square error to obtain a final transformation result.
The ICP closest point iterative algorithm is used as the most common precise registration mode at present, and has the advantages of accurate registration effect, no need of segmentation and feature extraction on the processed point cloud, good algorithm convergence under a better initial value condition and the like. Generally, measurement head radius compensation needs to be performed on contact point clouds before contact and non-contact point clouds on complex curved surfaces are registered, a two-dimensional online automatic compensation method is usually adopted at present, but for complex curved surfaces such as engine blade curved surfaces, the measurement direction may be inconsistent with the normal vector of a measurement point, and compensation errors are introduced by using the compensation method to influence subsequent registration results. The method provides an improved iteration closest point optimization algorithm based on the corresponding point distance mean square error for precise registration, and avoids the influence of measuring head radius compensation on registration precision.
As shown in fig. 2, for the contact-type measurement point cloud data without radius compensation, the distance between the spherical center of the stylus and the contact point is the radius r of the stylus, which means that for the same measurement target, the difference of the distance between the point cloud of non-contact measurement and the contact-type point cloud without radius compensation is the radius r of the stylus, and according to the target function of ICP, the difference is found
Ideally, when two groups of point clouds are completely overlapped, i.e. | | q i -p i || 2 In the case of =0, the two groups of point cloud registration obtain the optimal solution, and as known above, the two groups of point cloud registration are different, and it is impossible to achieve registration. Therefore, the objective function of the iterative closest point algorithm is modified, and the distance difference between the two is utilized to improve the objective function into
At the moment, for the contact point cloud without radius compensation, the condition of the optimal solution is changed into
Therefore, the accurate registration of the contact measurement and non-contact measurement data of the complex curved surface is realized through an improved closest point iterative algorithm.
The third step comprises the following steps:
step 3-1: selecting a point set Q = (Q) in the source point cloud after coarse registration 1 ,q 2 ,···,q N ) And selecting a corresponding point set P = (P) from the target point cloud data 1 ,p 2 ,···,p N ) So that each point pair (q) i ,p i ) Satisfy min | | q i -p i || 2 ；
Step 3-2: calculating an initial transformation matrix T through SVD singular value decomposition according to the selected point set Q and the point set P 0 ；
Step 3-3: for Q = (Q) 1 ,q 2 ,···,q N ) Applying an initial transformation T to each point in 0 To yield Q' = (Q) 1 ',q 2 ',···,q N ')；
Step 3-4: calculating the distance d between the point pairs i ＝||p i -q i '|| 2 Averaging their distances
Step 3-5: solving the optimal transformation delta T, and performing iterative computation based on least square method to obtain Euclidean transformation delta T (R, T) to obtain
A minimum value is reached.
Step 3-6: and judging whether convergence occurs according to the iteration errors of the previous iteration and the next iteration, the iteration times and other conditions. If convergence, outputting a final result: t is f ＝ΔT*T 0 Otherwise T 0 ＝ΔT*T 0 And repeating the step 3-1.
Fig. 3 shows the effect of the registration algorithm, in which the point cloud data obtained by contact-type measurement of the blade model surface is used as the point cloud data with a smaller density, the point cloud data obtained by non-contact measurement is used as the point cloud data with a larger density, fig. 3 (a) shows the distribution of the two point clouds before registration, and fig. 3 (b) shows the effect after registration.
Claims (6)
1. A registration method of complex curved surface measurement point cloud data considering measuring head radius is characterized by comprising the following steps:
the method comprises the following steps: data preprocessing: preprocessing contact measurement data and non-contact measurement data, and removing outlier noise points in two kinds of original point cloud data; the non-contact measurement data is subjected to down-sampling, the density of points is reduced, and the operation efficiency of the algorithm is improved;
step two: data coarse registration: taking the preprocessed contact type measurement point cloud as a target point cloud and the non-contact type measurement point cloud as a source point cloud, and transforming the source point cloud data by adopting an SAC-IA (sample consensus-IA) registration algorithm to obtain a rough registration transformation matrix, so that the source point cloud data is transformed by utilizing the rough registration transformation matrix to obtain point cloud data after rough registration;
step three: and (3) data fine registration: and performing multiple iterations on the point cloud data after the coarse registration by adopting an iteration nearest point algorithm improved based on the mean square error of the distances of the corresponding points to obtain a final transformation result.
2. The method for registering the complex curved surface measurement point cloud data by considering the radius of the measuring head according to claim 1, wherein in the first step, the specific data preprocessing step comprises the following steps:
step 1-1: carrying out statistical filtering on the two kinds of original point cloud data to remove outlier noise points in the point cloud data;
step 1-2: and uniformly sampling the denoised non-contact measurement point cloud data to reduce the scale of the non-contact measurement point cloud data and obtain source point cloud data required by registration.
3. The method for registering the complex curved surface measurement point cloud data considering the measuring head radius according to claim 1, wherein in the second step, the data rough registration comprises the following specific steps:
step 2-1: at the sourcePoint cloud P s S points are randomly sampled to ensure that the minimum distance between the points is larger than a given threshold value d min ；
Step 2-2: for each sampling point, the point cloud P is located in the target point t Searching a group of points with similar FPFH characteristics through nearest neighbor search of a K-D tree, and selecting one point from the similar points as a corresponding point;
step 2-3: calculating rigid transformation matrix between n point pairs by SVD singular value decomposition, measuring transformation quality by error between each group of corresponding points after transformation, and returning to minimum error D (e) i ) And transforming the source point cloud data by using the obtained transformation matrix T to obtain point cloud data after coarse registration.
4. The method for registering the complex curved surface measurement point cloud data considering the measuring head radius according to claim 3, wherein in the step 2-3, an error function D (e) is calculated i ) The method comprises the following steps:
wherein m is e Is a predetermined value, e i Is the distance difference after the ith group of corresponding points is transformed.
5. The method for registering the complex curved surface measurement point cloud data considering the measuring head radius according to claim 3, wherein in the step 2-3, the transformation matrix T (R, T) is
Wherein, R is a rotation matrix, and t is a translation vector.
6. The method for registering the complex curved surface measurement point cloud data considering the measuring head radius according to claim 1, wherein in the third step, the specific data fine registration step comprises the following steps:
step 3-1: sources after the coarse registrationSelecting a point set Q = (Q) from the point cloud 1 ,q 2 ,…,q N ) And selecting a corresponding point set P = (P) from the target point cloud data 1 ,p 2 ,…,p N ) Such that each point pair (q) i ,p i ) Satisfy min | | q i -p i || 2 ；
Step 3-2: calculating an initial transformation matrix T through SVD singular value decomposition according to the selected point set Q and the point set P 0 ；
Step 3-3: for Q = (Q) 1 ,q 2 ,…,q N ) Applying an initial transformation T to each point in 0 To yield Q' = (Q) 1 ',q 2 ',…,q N ')；
Step 3-4: calculating the distance d between the point pairs i ＝||p i -q i '|| 2 Averaging their distances
Step 3-5: solving the optimal transformation delta T, and performing iterative computation based on least square method to solve Euclidean transformation delta T (R, T) to make deviation
A minimum value is reached.
Step 3-6: and judging whether convergence occurs according to the iteration errors of the previous iteration and the next iteration, the iteration times and other conditions. If convergence, outputting a final result: t is f ＝ΔT*T 0 Otherwise T 0 ＝ΔT*T 0 And repeating the step 3-1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211344520.5A CN115797414A (en) | 2022-10-31 | 2022-10-31 | Complex curved surface measurement point cloud data registration method considering measuring head radius |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211344520.5A CN115797414A (en) | 2022-10-31 | 2022-10-31 | Complex curved surface measurement point cloud data registration method considering measuring head radius |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115797414A true CN115797414A (en) | 2023-03-14 |
Family
ID=85434501
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211344520.5A Pending CN115797414A (en) | 2022-10-31 | 2022-10-31 | Complex curved surface measurement point cloud data registration method considering measuring head radius |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115797414A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116777961A (en) * | 2023-06-26 | 2023-09-19 | 安徽开源路桥有限责任公司 | Parallelization point cloud registration method using KD tree search |
| CN117495926A (en) * | 2023-10-31 | 2024-02-02 | 哈尔滨工业大学 | Three-dimensional point cloud data registration method for large-caliber aspheric surface |
-
2022
- 2022-10-31 CN CN202211344520.5A patent/CN115797414A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116777961A (en) * | 2023-06-26 | 2023-09-19 | 安徽开源路桥有限责任公司 | Parallelization point cloud registration method using KD tree search |
| CN117495926A (en) * | 2023-10-31 | 2024-02-02 | 哈尔滨工业大学 | Three-dimensional point cloud data registration method for large-caliber aspheric surface |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114170279B (en) | Point cloud registration method based on laser scanning | |
| CN115797414A (en) | Complex curved surface measurement point cloud data registration method considering measuring head radius | |
| CN109118574A (en) | A kind of fast reverse modeling method extracted based on three-dimensional feature | |
| CN105654483B (en) | The full-automatic method for registering of three-dimensional point cloud | |
| CN113628263A (en) | Point cloud registration method based on local curvature and neighbor characteristics thereof | |
| CN109101741B (en) | Complex surface detection self-adaptive sampling method based on triangular mesh simplification | |
| CN111179321B (en) | Point cloud registration method based on template matching | |
| CN111539070B (en) | Wing body butt joint gap distribution control method based on measured data | |
| CN116402866A (en) | Point cloud-based part digital twin geometric modeling and error assessment method and system | |
| CN111340862B (en) | Point cloud registration method and device based on multi-feature fusion and storage medium | |
| CN109483887B (en) | Online detection method for contour accuracy of forming layer in selective laser melting process | |
| CN113516695B (en) | Point cloud registration strategy in laser profiler flatness measurement | |
| WO2021082380A1 (en) | Laser radar-based pallet recognition method and system, and electronic device | |
| CN116401794B (en) | Blade three-dimensional accurate reconstruction method based on attention-guided depth point cloud registration | |
| CN114240871B (en) | Point cloud data processing method for contour detection in workpiece forming process | |
| CN113192116A (en) | Aviation blade thickness parameter measuring method based on structured light camera | |
| CN115100277A (en) | Method for determining position and pose of complex curved surface structure part | |
| CN116310355A (en) | Laser point cloud denoising and defect detection method for complex structural member | |
| CN116309026A (en) | Point cloud registration method and system based on statistical local feature description and matching | |
| CN116642904A (en) | Aircraft skin defect detection and measurement method based on three-dimensional point cloud | |
| CN115147433A (en) | Point cloud registration method | |
| CN112509018B (en) | Quaternion space optimized three-dimensional image registration method | |
| CN112884057B (en) | Point cloud data-based three-dimensional curved surface quality classification method and system and storage medium | |
| CN115147471A (en) | Laser point cloud automatic registration method based on curvature density characteristics | |
| CN114898069A (en) | Three-dimensional data splicing method for curved surface skin |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |