CN113192114A - Blade multi-field point cloud registration method based on overlapping features and local distance constraint - Google Patents
Blade multi-field point cloud registration method based on overlapping features and local distance constraint Download PDFInfo
- Publication number
- CN113192114A CN113192114A CN202110739936.6A CN202110739936A CN113192114A CN 113192114 A CN113192114 A CN 113192114A CN 202110739936 A CN202110739936 A CN 202110739936A CN 113192114 A CN113192114 A CN 113192114A
- Authority
- CN
- China
- Prior art keywords
- matrix
- data set
- point
- point cloud
- distance
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Abstract
The invention discloses a blade multi-field point cloud registration method based on overlapping features and local distance constraint, which comprises the following steps of 100: acquiring multi-view field point cloud data of the blade, solving an overlapping area of the point cloud data between two adjacent view fields, and constructing data sets X and Y; step 200: establishing an overlap feature constraint matrix FMN(ii) a Step 300: establishing a local distance constraint matrix DMN(ii) a Step 400: matrix-based Hadamard product, constraining matrix F by overlapping featuresMNAnd a local distance constraint matrix DMNConstructing a membership probability matrix P between data set X and data set YMN,PMN=FMN DMN,Is the Hadamard product; step 500: and calculating a rotation matrix and a translation matrix between the data sets X and Y based on an EM algorithm, and splicing a complete blade section profile through the calculated rotation matrix and translation matrix. The invention provides a blade multi-field-of-view point cloud data registration method independent of the motion precision of a detection system, and aims to improve the blade detection precision and reliability.
Description
Technical Field
The invention belongs to the field of blade detection, and particularly relates to a blade multi-field point cloud registration method based on overlapping features and local distance constraint.
Background
The blade is used as a core component of an aeroengine, and the profile precision of the blade directly influences the energy conversion efficiency and the working reliability of the whole machine. Therefore, accurate blade profile inspection has long been a key step in the blade manufacturing process. In recent years, the blade detection method based on optical measurement has attracted much attention due to the characteristics of good efficiency, flexibility and the like.
At present, the registration accuracy of multi-field point clouds mainly depends on the geometric accuracy and the operational reliability of an established detection system. Chinese invention patent No. 202011134900.7 discloses a method for calibrating a rotation center based on the characteristics of a local leading edge curve of a blade, which can effectively complete the registration of multi-field point cloud data by directly utilizing the motion parameters of a detection platform, but the registration result is easily influenced by multiple factors such as the geometric accuracy and the motion reliability of a detection system, and after all, the motion parameters of the detection platform have certain mechanical errors, so that the final blade detection reliability is poor; from fig. 1a and b, two adjacent joint results are obtained from the result of registering the multi-field point cloud data by using the method, and it can be seen that the overlapping areas between the adjacent fields are not completely matched, and the density is not uniform (point cloud data in the dotted line in fig. 1 a).
Disclosure of Invention
The invention aims to provide a blade multi-field point cloud registration method based on overlapping features and local distance constraints, which is used for realizing blade multi-field point cloud data registration without depending on the motion precision of a detection platform so as to achieve the purpose of improving the blade detection precision and reliability.
In order to achieve the purpose, the invention adopts the following technical scheme:
a blade multi-field point cloud registration method based on overlapping features and local distance constraints comprises the following steps:
step 100: acquiring point cloud data of multiple fields of view of the blade, solving an overlapping area of the point cloud data between two adjacent fields of view, and constructing a data set X = (X) of the overlapping area1,…,xm,…,xM) And Y = (Y)1,…,yn,…yN);
Step 200: establishing an overlap feature constraint matrix FMN;
First, with point X in data set XmDrawing J sampling circles as the center of the circle, and centering on a point xmThe radius of the jth sampling circle of (a) is as in formula i,
in formula I, rmjIs the radius of the jth sampling circle, rmJRadius of the maximum sampling circle, rm1Is the radius of the minimum sampling circle, J is the number of sampling circles, Ln (J) is a logarithmic function of a variable J, and Ln (J) is a logarithmic function value of a fixed value J;
then, construct point xmCovariance matrix C ofmAs shown in the formula II,
in formula II, CmIs a point xmOf (2) covariance matrix, SpjA sampling point corresponding to the jth sampling circle;
secondly, solving a covariance matrix Cmλ 1 and λ 2, and λ 1 ≧ λ 2, point xmLocal characteristic factor fx ofm= λ 1/λ 2; sequentially calculating local characteristic factors of M points in the data set XAnd the local characteristic factor dataset of dataset X is FX=(fx1,…,fxm,…,fxM) (ii) a And calculating a local feature factor data set F of the data set YY=(fy1,…,fyn,…,fyN);
Finally, an overlap feature constraint matrix F between the data sets X and Y is constructedMNOverlap feature constraint matrix FMNThe element in the m-th row and the n-th column is fmn=∣fxm-fyn∣,fmnFor a point X in the data set XmWith point Y in data set YnLocal feature similarity, and sequentially calculating an overlapped feature constraint matrix FMN;
Step 300: establishing a local distance constraint matrix DMN;
First, let a distance matrix d between data set X and data set YMNDistance matrix dMNM-th row and n-th column element dmnIs a point xmAnd point ynThe euler distance therebetween;
then, a distance matrix d is searchedMNMinimum value in each line, and constitutes a minimal value data set Dis = (Dis) of distances1,…,dism,…,disM) Calculating the average value Dd of the minimum value data set Dis;
finally, the distance matrix dMNSetting the value of the middle distance larger than the constraint radius Crm as 0, setting the value of the middle distance smaller than or equal to the constraint radius Crm as 1, and establishing a local distance constraint matrix DMN(ii) a Wherein the constraint radius Crm = γ R × Dd, γ R is an amplification factor, and Dd is an average value of the minimum data set Dis;
step 400: constraining the feature bundle matrix F by overlapping based on the Hadamard product between the matricesMNAnd a local distance constraint matrix DMNConstructing a membership probability matrix P between data set X and data set YMNI.e. PMN=FMN DMN,Is the product of Hadamard, PMNThe m-th row and the n-th column of the element are pmn=fmn×dmn,fmnAnd dmnAre respectively FMNAnd DMNRow m and column n elements in (1); updating the membership probability matrix PMNThe method comprises the following steps:
first, a membership probability matrix PMNAll non-0 elements in (a) are included under the framework of a gaussian function, as shown in equation iii,
in formula III, pmnIs a member probability matrix PMNM-th row and n-th column element, deltaF 2Is a member probability matrix PMNThe variance of the non-0 term in the middle,d =2 is the dimension of the point set to be registered, and W is the member probability matrix PMNThe number of non-0 elements;
then to the member probability matrix PMNAll the elements in the solution are normalized; let the current membership probability matrix PMNThe sum of the m-th row elements is SmThen member probability matrix PMNAny element in the m-th row is updated to pmn=pmn/SmAnd sequentially updating the member probability matrix PMNAll of the elements in (A);
step 500: and calculating a rotation matrix and a translation matrix between the data sets X and Y based on an EM algorithm, and splicing the section profile of the blade through the calculated rotation matrix and translation matrix.
Further, the solving of the overlapping area of the point cloud data between two adjacent fields of view and the construction of the data set of the overlapping area in step 100 are as follows:
step 101: setting two adjacent point cloud data sets as V1And V2By a point cloud data set V1End point V of1HFor reference points, points V are calculated separately1HAnd pointCloud data set V2The distance between all the points is solved, and the point corresponding to the minimum distance value is solved as V2pPoint cloud data set V2The overlap region data set Y is the point cloud data set V2End point V of2PAnd V2pAll points in between;
step 102: in the same way, with a point cloud data set V2End point V of2PFor reference points, points V are calculated separately2PWith the point cloud data set V1The distance between all the points is solved, and the point corresponding to the minimum distance value is solved as V1hPoint cloud data set V1The overlap region data set X is the point cloud data set V1End point V of1HAnd V1hAll points in between.
Further, the calculation of the rotation matrix and the translation matrix between the data sets X and Y based on the EM algorithm in step 500 is performed according to the following steps:
step 501: initializing parameters, setting a rotation matrix asAnd translation matrixIteration threshold xi and noise ratio alpha;
step 502: calculating a membership probability matrix PMNThe sum of all elements in the composition is SPAnd respectively calculating the weight centers mu of the data set X and the data set YXAnd muYThen according to the weight center muXAnd muYCalculating central vectors of a data set X and a data set Y;
step 503: calculating a covariance matrix Cov between the data set X and the data set Y according to the central vector, then performing singular value decomposition on the covariance matrix Cov, and updating a rotation matrix and a translation matrix according to a singular value decomposition result;
step 504: calculating a convergence coefficient d tau according to the updated rotation matrix and translation matrix, comparing the convergence coefficient d tau with an iteration threshold xi, repeating the steps 501 to 503 to iterate when d tau is larger than or equal to xi, stopping iterating until d tau is smaller than xi, and outputting an optimal rotation matrix and translation matrix between the data sets X and Y.
Compared with the prior art, the invention has the following beneficial effects:
(1) the registration algorithm specially aiming at multi-field-of-view point cloud data in the blade detection process is provided, the problems that the density of the measured data is non-uniform and the characteristics in an overlapping area are not prominent in the conventional algorithm are solved, and particularly, the algorithm shows good adaptability and robustness aiming at a measured object with complex geometric characteristics, such as a blade;
(2) a new characteristic constraint mechanism is provided, the non-uniformity problem among data points is solved by adopting a uniform sampling mode, and the problem that the characteristics in an overlapping area are not outstanding is solved by utilizing a self-adaptive sampling radius; meanwhile, a constraint mechanism based on local distance is also provided, so that the calculation complexity of the algorithm is effectively reduced, the overall calculation efficiency and accuracy are increased, and the problems of mis-registration and the like are avoided;
(3) based on the proposed multi-field-of-view point cloud registration algorithm, a high-precision and high-efficiency blade detection method is further provided, and the method effectively improves the detection precision and the universality on the premise of not improving the equipment cost of a detection system, and has a good application prospect.
Drawings
Fig. 1 is a registration result of an overlapping area between adjacent fields of view using motion parameters of an inspection platform, where a is a registration result of density unevenness of the overlapping area between two fields of view, and b is a registration result of registration inconsistency of the overlapping area between two fields of view.
FIG. 2 is a schematic diagram of the present invention for solving overlapping region data sets.
FIG. 3 shows a point x according to the present inventionmDrawing a schematic diagram of J sampling circles as the center of the circle.
FIG. 4 is a schematic diagram of the relationship of the sampling radius function of the sampling circle drawn by the present invention.
FIG. 5 shows point x of the present inventionmAs a local spatial schematic of a reference point.
Fig. 6 shows the registration result of the overlapped region between adjacent fields of view by using the method of the present invention, wherein a is the registration result of fig. 1a by using the method of the present invention, and b is the registration result of fig. 1b by using the method of the present invention.
Detailed Description
The blade multi-field-of-view point cloud registration method based on the overlapping features and the local distance constraint provided by the embodiment comprises the following steps:
step 100: the method includes the steps of obtaining a point cloud data set of a blade with multiple view fields, in the embodiment, adopting a four-axis measuring device and a line laser sensor to collect the point cloud data set of the blade, obtaining the point cloud data set of the blade with the multiple view fields by rotating or translating the four-axis measuring device, and obtaining the point cloud data set of the blade by adopting the four-axis measuring device and the line laser sensor, which belong to common technical means in the field, so that the embodiment is not repeated.
Extracting the data sets of the overlapping areas of the point cloud data sets of two adjacent field-of-view blades according to the extraction principle shown in FIG. 2, and specifically implementing the following steps:
step 101: setting point cloud data set of two view field blades as V1And V2By a point cloud data set V1End point V of1HFor reference points, points V are calculated separately1HWith the point cloud data set V2The distance between all the points is solved, and the point corresponding to the minimum distance value is solved as V2pPoint cloud data set V2The overlap region data set Y is the point cloud data set V2End point V of2PAnd V2pAll points in between;
step 102: in the same way, with a point cloud data set V2End point V of2PFor reference points, points V are calculated separately2PWith the point cloud data set V1The distance between all the points is solved, and the point corresponding to the minimum distance value is solved as V1hPoint cloud data set V1The overlap region data set X is the point cloud data set V1End point V of1HAnd V1hAll points in between. Data sets X and Y of the overlap region within the dashed box in fig. 2, data set X and data set Y are renumbered and sorted for ease of subsequent understanding and writing, with the renumbered sorted post data set X = (X) being1,…,xm,…,xM) And data setY=(y1,…,yn,…yN)。
Step 200: establishing an overlap feature constraint matrix FMNThe data of the overlapping region are overlapping features, and the overlapping feature constraint mechanism is to disclose the potential matching probability between two data sets from the feature similarity, xmConstructing a point x for a reference pointmLocal characteristic factor fx ofmAnd further establishing an overlapping feature constraint matrix FMNThe establishment process is as follows:
step 201: at point X in data set XmDrawing J sampling circles for the circle center, as shown in FIG. 3, in order to overcome the non-prominent feature of the overlap feature, more sampling circles far away from xmTo enhance the local characteristic factor fxmSensitivity to local features makes it easier to distinguish the best match from data set X and data set Y, and therefore, for point XmThe radius of the jth sampling circle of (a) is as follows:
in formula I, rmjIs the radius of the jth sampling circle, rmJRadius of the maximum sampling circle, rm1Is the radius of the minimum sampling circle, J is the number of sampling circles, Ln (J) is a logarithmic function of a variable J, and Ln (J) is a logarithmic function value of a fixed value J; the sampling radius of the sampling circle is shown in fig. 4;
step 202: as shown in FIG. 3, let the sampling point corresponding to the jth sampling radius be SpjPoint xmAll sampling point sets SP ofm=(Sp1,…,Spj,…,SpJ) Constructed with a point xmCovariance matrix C ofmAs shown in the formula II,
in formula II, CmIs a point xmOf (2) covariance matrix, SpjIs a sampling point corresponding to the jth sampling circle, J is a point xmThe number of all sampling points is the number of sampling circles;
step 203: solving out covariance matrix Cmλ 1 and λ 2, and λ 1 ≧ λ 2, point xmLocal characteristic factor fx ofm=λ1/λ2;
Step 204: calculating the local characteristic factors of M points in the data set X in sequence according to the steps 201 to 203, wherein the data set of the local characteristic factors of the data set X is FX=(fx1,…,fxm,…,fxM) (ii) a And calculating a local feature factor data set F of the data set YY=(fy1,…,fyn,…,fyN);
Step 205: constructing an overlap feature constraint matrix F between datasets X and YMNOverlap feature constraint matrix FMNThe element in the m-th row and the n-th column is fmn=∣fxm-fyn∣,fmnFor a point X in the data set XmWith point Y in data set YnLocal feature similarity, and sequentially calculating an overlapped feature constraint matrix FMN。
Step 300: establishing a local distance constraint matrix DMNAs shown in FIG. 5, at point xmAs a reference point, and in xmAs a circle center and the radius of which is specified to be CrmIs a point xmBy controlling the radius CrmCan flexibly specify the point xmLocal space, local distance constraint matrix DMNThe establishment process is as follows:
step 301: let the distance matrix d between dataset X and dataset YMNDistance matrix dMNM-th row and n-th column element dmnIs a point xmAnd point ynThe euler distance therebetween;
step 302: searching for a distance matrix dMNMinimum value in each row of, e.g., dis as shown in FIG. 4m,dismIs a point xmAnd point ynMinimum value between, and constitutes a minimal value data set Dis = (Dis) of distance1,…,dism,…,disM) Calculating the average value Dd of the minimum value data set Dis,
in the formula IV, Dd is an average value of the minimum value data set Dis, and M is the minimum value number in the minimum value data set Dis;
step 303: will distance matrix dMNMiddle greater than restraint radius CrmIs set to 0 and is less than or equal to the constraint radius CrmIs set to 1, a local distance constraint matrix D is establishedMN(ii) a Wherein the radius of constraint Crm= γ R × Dd, γ R is an amplification factor, the amplification factor value in this embodiment is 10, Dd is an average value of the minimum value data set Dis, and the local distance constraint matrix DMNM-th row and n-th column element DmnComprises the following steps:
in formula V, DmnConstraining the matrix D for local distancesMNThe element of the m-th row and the n-th column; dmnIs a distance matrix dMNElement of m-th row and n-th column, CrmIs a constraint radius; completing the distance matrix d according to a formulaMNThe judgment of all elements can solve the local distance constraint matrix DMN。
Step 400: constraining the feature bundle matrix F by overlapping based on the Hadamard product between the matricesMNAnd a local distance constraint matrix DMNConstructing a membership probability matrix P between data set X and data set YMNI.e. PMN=FMN DMN,Is the product of Hadamard, PMNThe m-th row and the n-th column of the element are pmn=fmn×dmn,fmnAnd dmnAre respectively FMNAnd DMNRow m and column n elements in (1); updating the membership probability matrix PMNThe updating steps are as follows:
step 401: the member probability matrix PMNAll non-0 elements in (a) are included under the framework of a gaussian function, as shown in equation iii,
in formula III, pmnIs a member probability matrix PMNM-th row and n-th column element, deltaF 2Is a member probability matrix PMNThe variance of the non-0 term in the middle,d =2 is the dimension of the point set to be registered, and W is the member probability matrix PMNThe number of non-0 elements;
step 402: for member probability matrix PMNAll the elements in the solution are normalized; let the current membership probability matrix PMNThe sum of the m-th row elements is SmMember probability matrix PMNAny element in the m-th row is updated to pmn=pmn/SmAnd sequentially updating the member probability matrix PMN。
Membership probability matrix PMNRespectively calculating a density function between each point in the data set (X) and all points in the data set (Y) to be registered by taking one data set (X) as a Gaussian kernel, and constructing a fully-connected Gaussian kernel function between all points in the two data sets; a maximum likelihood function is further constructed. In this embodiment, the data set X is used as the Gaussian kernel, and all rotation and translation are applied to the data set Y, so that the point XmDensity function P (x)m) Comprises the following steps:
in formula VI, P (x)m) Is a point xmA is the noise ratio, and M is in the data set XTotal number of data of pmnIs a member probability matrix PMNThe element of the m-th row and the n-th column;for conditional probability, σ is variance, D =2 is dataset dimension, xmFor the mth data in data set X, dRXYTo rotate the matrix, ynFor the nth data in data set Y, dTXYFor the translation matrix, the maximum likelihood function is further constructed as:
in formula VII, P (X) is an exponent of the logarithmic function Ln, P (X) is a density function of the data set X, P (y)n) Is a member probability pmn,P(xm∣yn) Is a conditional probability.
Step 500: computing a rotation matrix dR between data sets X and Y based on an EM algorithmXYAnd a translation matrix dTXYAnd splicing the section profile characteristics of the blade through the calculated rotation matrix and translation matrix.
Computing a rotation matrix dR between data sets X and Y based on an EM algorithmXYAnd a translation matrix dTXYThe method comprises the following specific steps:
step 501: initialization of parameters, initializing a rotation matrixTranslation matrixIteration threshold ξ =10-5The noise ratio alpha =0.01, and the noise ratio is related to the parameters of the line laser sensor;
step 502: the membership probability matrix P calculated according to the steps 100 to 400MNUpdating a density function and a maximum likelihood function; calculating a membership probability matrix PMNSum of all elements in SPAnd respectively calculating the weight centers of the data set X and YAnd,is a member probability matrix PMNThe column vector of all the middle elements is 1, and then the center mu is determined according to the weightXAnd muYCalculating the central vectors of the data set X and the data set YAnd,is a member probability matrix PMNRow vectors with all the middle elements being 1;
step 503: a covariance matrix Cov between the data set X and the data set Y is calculated from the central vector,then, singular value decomposition is carried out on the covariance matrix Cov, and the decomposition result is set asU and V are matrixes decomposed by singular values;
step 504: setting matrixWherein det (UV)T) As matrices U and VTThe product between corresponds to the value of the determinant, the rotation matrix between the data sets X and Y is dRXY=UηVTThe translation matrix is dTXY=μX-dRXYμY(ii) a Updating the rotation matrix to dRt XY=dRt-1 XYdRXYWherein dRt-1 XYFor the rotation matrix obtained from the previous iteration, dR if the current iteration is the first iterationt-1 XYFor initialising the rotation matrix dR0 XYUpdate the translation matrix dT similarlyt XYCalculating Y according to the updated rotation matrix and translation matrixnew=dRt XYX+dTt XY;
Step 505: calculating convergence coefficient as d τ = | dTt XY-dTt-1 XYII, wherein dTt-1 XYThe translation matrix calculated for the previous iteration, if the first iteration, then dTt-1 XYIt is to initialize the translation matrix dT0 XY. If the convergence coefficient is larger than the iteration coefficient, namely d tau is larger than or equal to xi, repeating the steps 501 to 504 to iterate until d tau is smaller than xi, and stopping the iteration.
Optimal rotation matrix dR updated based on multiple iterationsXYAnd YnewSeparately calculate Y at this timenewAnd the center of gravity of the data set Y, and solving a translation vector between the two centers of gravity to obtain a final translation matrix dT between X and YXYAnd finally completing the registration between X and Y. The currently calculated dRXYAnd dTXYThe original data V1 and V2 are fed back to update the detection data V2newComprises the following steps: v2new=dRXYV1+dTXY。
Sequentially extracting the point cloud data sets of the overlapping areas between the adjacent view fields until the original point cloud data sets of all the view fields are updated, namely completing the multi-view-field data registration and accurate reconstruction of the blade profile, as shown in fig. 6, wherein a is a result of the registration of fig. 1a by adopting the method of the embodiment, b is a result of the registration of fig. 1b by adopting the method of the embodiment, and the result of the point cloud data registration of the adjacent view fields is obviously prior to the result of the motion accuracy registration by using the motion platform; and carrying out deviation analysis on the obtained blade profile and the CAD model of the blade, and finally finishing the precision detection and evaluation of the blade.
The above description is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any modification and replacement based on the technical solution and inventive concept provided by the present invention should be covered within the scope of the present invention.
Claims (3)
1. A blade multi-field point cloud registration method based on overlapping features and local distance constraints is characterized by comprising the following steps:
step 100: acquiring point cloud data of multiple fields of view of the blade, solving an overlapping area of the point cloud data between two adjacent fields of view, and constructing a data set X = (X) of the overlapping area1,…,xm,…,xM) And Y = (Y)1,…,yn,…yN);
Step 200: establishing an overlap feature constraint matrix FMN;
First, with point X in data set XmDrawing J sampling circles as the center of the circle, and centering on a point xmThe radius of the jth sampling circle of (a) is as in formula i,
in formula I, rmjIs the radius of the jth sampling circle, rmJRadius of the maximum sampling circle, rm1Is the radius of the minimum sampling circle, J is the number of sampling circles, Ln (J) is a logarithmic function of a variable J, and Ln (J) is a logarithmic function value of a fixed value J;
then, construct point xmCovariance matrix C ofmAs shown in the formula II,
in formula II, CmIs a point xmOf (2) covariance matrix, SpjA sampling point corresponding to the jth sampling circle;
secondly, solving a covariance matrix Cmλ 1 and λ 2, and λ 1 ≧ λ 2, point xmLocal characteristic factor fx ofm= λ 1/λ 2; sequentially calculating local characteristic factors of M points in the data set X, wherein the data set of the local characteristic factors of the data set X is FX=(fx1,…,fxm,…,fxM) (ii) a And calculating a local feature factor data set F of the data set YY=(fy1,…,fyn,…,fyN);
Finally, an overlap feature constraint matrix F between the data sets X and Y is constructedMNOverlap feature constraint matrix FMNThe element in the m-th row and the n-th column is fmn=∣fxm-fyn∣,fmnFor a point X in the data set XmWith point Y in data set YnLocal feature similarity, and sequentially calculating an overlapped feature constraint matrix FMN;
Step 300: establishing a local distance constraint matrix DMN;
First, let a distance matrix d between data set X and data set YMNDistance matrix dMNM-th row and n-th column element dmnIs a point xmAnd point ynThe euler distance therebetween;
then, a distance matrix d is searchedMNMinimum value in each line, and constitutes a minimal value data set Dis = (Dis) of distances1,…,dism,…,disM) Calculating the average value Dd of the minimum value data set Dis;
finally, the distance matrix dMNSetting the value of the middle distance larger than the constraint radius Crm as 0, setting the value of the middle distance smaller than or equal to the constraint radius Crm as 1, and establishing a local distance constraint matrix DMN(ii) a Wherein the constraint radius Crm = γ R × Dd, γ R is an amplification factor, and Dd is an average value of the minimum data set Dis;
step 400: constraining the feature bundle matrix F by overlapping based on the Hadamard product between the matricesMNAnd a local distance constraint matrix DMNConstructing a membership probability matrix P between data set X and data set YMNI.e. PMN=FMN DMN,Is the product of Hadamard, PMNThe m-th row and the n-th column of the element are pmn=fmn×dmn,fmnAnd dmnAre respectively FMNAnd DMNRow m and column n elements in (1); updating the membership probability matrix PMNThe method comprises the following steps:
first, a membership probability matrix PMNAll non-0 elements in (a) are included under the framework of a gaussian function, as shown in equation iii,
in formula III, pmnIs a member probability matrix PMNM-th row and n-th column element, deltaF 2Is a member probability matrix PMNThe variance of the non-0 term in the middle,d =2 is the dimension of the point set to be registered, and W is the member probability matrix PMNThe number of non-0 elements;
then to the member probability matrix PMNAll the elements in the solution are normalized; let the current membership probability matrix PMNThe sum of the m-th row elements is SmThen member probability matrix PMNAny element in the m-th row is updated to pmn=pmn/SmAnd sequentially updating the member probability matrix PMNAll of the elements in (A);
step 500: and calculating a rotation matrix and a translation matrix between the data sets X and Y based on an EM algorithm, and splicing the section profile of the blade through the calculated rotation matrix and translation matrix.
2. The blade multi-field point cloud registration method based on the overlapped features and the local distance constraint according to claim 1, wherein: in step 100, solving an overlapping area of point cloud data between two adjacent fields of view and constructing a data set of the overlapping area comprises the following steps:
step 101: setting two adjacent point cloud data sets as V1And V2By a point cloud data set V1End point V of1HFor reference points, points V are calculated separately1HWith the point cloud data set V2The distance between all the points is solved, and the point corresponding to the minimum distance value is solved as V2pPoint cloud data set V2The overlap region data set Y is the point cloud data set V2End point V of2PAnd V2pAll points in between;
step 102: in the same way, with a point cloud data set V2End point V of2PFor reference points, points V are calculated separately2PWith the point cloud data set V1The distance between all the points is solved, and the point corresponding to the minimum distance value is solved as V1hPoint cloud data set V1The overlap region data set X is the point cloud data set V1End point V of1HAnd V1hAll points in between.
3. The blade multi-field point cloud registration method based on the overlapped features and the local distance constraint according to claim 1, wherein: in step 500, the rotation matrix and the translation matrix between the data sets X and Y are calculated based on the EM algorithm according to the following steps:
step 501: initializing parameters, setting a rotation matrix asAnd translation matrixIteration threshold xi and noise ratio alpha;
step 502: calculating a membership probability matrix PMNThe sum of all elements in the composition is SPAnd respectively calculating the weight centers mu of the data set X and the data set YXAnd muYThen according to the weight center muXAnd muYCalculate the center of data set X and data set YVector quantity;
step 503: calculating a covariance matrix Cov between the data set X and the data set Y according to the central vector, then performing singular value decomposition on the covariance matrix Cov, and updating a rotation matrix and a translation matrix according to a singular value decomposition result;
step 504: calculating a convergence coefficient d tau according to the updated rotation matrix and translation matrix, comparing the convergence coefficient d tau with an iteration threshold xi, repeating the steps 501 to 503 to iterate when d tau is larger than or equal to xi, stopping iterating until d tau is smaller than xi, and outputting an optimal rotation matrix and translation matrix between the data sets X and Y.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110739936.6A CN113192114B (en) | 2021-07-01 | 2021-07-01 | Blade multi-field point cloud registration method based on overlapping features and local distance constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110739936.6A CN113192114B (en) | 2021-07-01 | 2021-07-01 | Blade multi-field point cloud registration method based on overlapping features and local distance constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113192114A true CN113192114A (en) | 2021-07-30 |
CN113192114B CN113192114B (en) | 2021-09-03 |
Family
ID=76976714
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110739936.6A Active CN113192114B (en) | 2021-07-01 | 2021-07-01 | Blade multi-field point cloud registration method based on overlapping features and local distance constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113192114B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116580069A (en) * | 2023-07-12 | 2023-08-11 | 四川大学 | Multi-view-field data registration method for blade profile from coarse to fine |
Citations (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102779345A (en) * | 2012-07-03 | 2012-11-14 | 河海大学 | Point cloud precise registering method based on gravity center Euclidean distance |
CN103236064A (en) * | 2013-05-06 | 2013-08-07 | 东南大学 | Point cloud automatic registration method based on normal vector |
CN103247041A (en) * | 2013-05-16 | 2013-08-14 | 北京建筑工程学院 | Local sampling-based multi-geometrical characteristic point cloud data splitting method |
US20170161945A1 (en) * | 2015-12-04 | 2017-06-08 | Autodesk, Inc. | Keypoint-based point-pair-feature for scalable automatic global registration of large rgb-d scans |
CN107861920A (en) * | 2017-11-27 | 2018-03-30 | 西安电子科技大学 | cloud data registration method |
CN108053432A (en) * | 2017-11-14 | 2018-05-18 | 华南理工大学 | The method for registering of indoor sparse cloud scene based on local I CP |
CN109493372A (en) * | 2018-10-24 | 2019-03-19 | 华侨大学 | The product point cloud data Fast global optimization method for registering of big data quantity, few feature |
CN109697728A (en) * | 2017-10-20 | 2019-04-30 | 阿里巴巴集团控股有限公司 | Data processing method, device, system and storage medium |
CN109872354A (en) * | 2019-01-28 | 2019-06-11 | 深圳市易尚展示股份有限公司 | Multi-angle of view point cloud registration method and system based on nonlinear optimization |
DE102019108733A1 (en) * | 2018-04-11 | 2019-10-17 | Nvidia Corporation | Fast multi-scale point cloud registration with a hierarchical Gaussian mix |
CN110930495A (en) * | 2019-11-22 | 2020-03-27 | 哈尔滨工业大学(深圳) | Multi-unmanned aerial vehicle cooperation-based ICP point cloud map fusion method, system, device and storage medium |
CN111798570A (en) * | 2020-07-16 | 2020-10-20 | 浙江大学 | Rapid generation method for curved surface gridding of impeller mechanical blade |
CN112017220A (en) * | 2020-08-27 | 2020-12-01 | 南京工业大学 | Point cloud accurate registration method based on robust constraint least square algorithm |
CN112907642A (en) * | 2021-03-01 | 2021-06-04 | 沈阳蓝软智能医疗科技有限公司 | Method, system, storage medium and equipment for accurate registration and coincidence of preoperative CT or nuclear magnetic image and intraoperative corresponding focus |
CN113052881A (en) * | 2021-03-24 | 2021-06-29 | 王程 | Automatic registration method for extracting pole point in indoor three-dimensional point cloud |
-
2021
- 2021-07-01 CN CN202110739936.6A patent/CN113192114B/en active Active
Patent Citations (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102779345A (en) * | 2012-07-03 | 2012-11-14 | 河海大学 | Point cloud precise registering method based on gravity center Euclidean distance |
CN103236064A (en) * | 2013-05-06 | 2013-08-07 | 东南大学 | Point cloud automatic registration method based on normal vector |
CN103247041A (en) * | 2013-05-16 | 2013-08-14 | 北京建筑工程学院 | Local sampling-based multi-geometrical characteristic point cloud data splitting method |
US20170161945A1 (en) * | 2015-12-04 | 2017-06-08 | Autodesk, Inc. | Keypoint-based point-pair-feature for scalable automatic global registration of large rgb-d scans |
CN109697728A (en) * | 2017-10-20 | 2019-04-30 | 阿里巴巴集团控股有限公司 | Data processing method, device, system and storage medium |
CN108053432A (en) * | 2017-11-14 | 2018-05-18 | 华南理工大学 | The method for registering of indoor sparse cloud scene based on local I CP |
CN107861920A (en) * | 2017-11-27 | 2018-03-30 | 西安电子科技大学 | cloud data registration method |
DE102019108733A1 (en) * | 2018-04-11 | 2019-10-17 | Nvidia Corporation | Fast multi-scale point cloud registration with a hierarchical Gaussian mix |
CN109493372A (en) * | 2018-10-24 | 2019-03-19 | 华侨大学 | The product point cloud data Fast global optimization method for registering of big data quantity, few feature |
CN109872354A (en) * | 2019-01-28 | 2019-06-11 | 深圳市易尚展示股份有限公司 | Multi-angle of view point cloud registration method and system based on nonlinear optimization |
CN110930495A (en) * | 2019-11-22 | 2020-03-27 | 哈尔滨工业大学(深圳) | Multi-unmanned aerial vehicle cooperation-based ICP point cloud map fusion method, system, device and storage medium |
CN111798570A (en) * | 2020-07-16 | 2020-10-20 | 浙江大学 | Rapid generation method for curved surface gridding of impeller mechanical blade |
CN112017220A (en) * | 2020-08-27 | 2020-12-01 | 南京工业大学 | Point cloud accurate registration method based on robust constraint least square algorithm |
CN112907642A (en) * | 2021-03-01 | 2021-06-04 | 沈阳蓝软智能医疗科技有限公司 | Method, system, storage medium and equipment for accurate registration and coincidence of preoperative CT or nuclear magnetic image and intraoperative corresponding focus |
CN113052881A (en) * | 2021-03-24 | 2021-06-29 | 王程 | Automatic registration method for extracting pole point in indoor three-dimensional point cloud |
Non-Patent Citations (9)
Title |
---|
CAO S等: "Research on fault diagnosis technology of centrifugal pump blade crack based on PCA and GMM", 《MEASUREMENT》 * |
GHORBANI H等: "Accurate registration of point clouds of damaged aeroengine blades", 《JOURNAL OF MANUFACTURING SCIENCE AND ENGINEERING》 * |
Y. CHOI等: "Measurement of the rotational speed of industrial wind turbine blades using a low cost 24GHz Doppler radar", 《2021 15TH EUROPEAN CONFERENCE ON ANTENNAS AND PROPAGATION (EUCAP)》 * |
Y. GE等: "A point cloud registration method combining enhanced particle swarm optimization and iterative closest point method", 《2016 CHINESE CONTROL AND DECISION CONFERENCE (CCDC)》 * |
YU Y等: "Investigation of microcombing parameters in enhancing the properties of carbon nanotube yarns", 《MATERIALS & DESIGN》 * |
刘晶: "叶片数字化检测中的模型配准技术及应用研究", 《中国博士学位论文全文数据库·工程科技Ⅱ辑》 * |
刘浩浩等: "基于线结构光的叶片型面特征检测方法研究", 《中国测试》 * |
孟禹: "基于采样球和ICP算法的点云配准方法研究", 《中国优秀硕士学位论文全文数据库·信息科技辑》 * |
杜辉等: "一种动态特征匹配的部分重叠点云配准方法", 《电信科学》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116580069A (en) * | 2023-07-12 | 2023-08-11 | 四川大学 | Multi-view-field data registration method for blade profile from coarse to fine |
CN116580069B (en) * | 2023-07-12 | 2023-09-19 | 四川大学 | Multi-view-field data registration method for blade profile from coarse to fine |
Also Published As
Publication number | Publication date |
---|---|
CN113192114B (en) | 2021-09-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105740894B (en) | Semantic annotation method for hyperspectral remote sensing image | |
Robles-Kelly et al. | Graph edit distance from spectral seriation | |
CN107833208B (en) | Hyperspectral anomaly detection method based on dynamic weight depth self-encoding | |
CN107871327A (en) | The monocular camera pose estimation of feature based dotted line and optimization method and system | |
CN103440659B (en) | Based on starry sky image Distortion Detect and the method for estimation of star pattern matching | |
CN111079856B (en) | Multi-period intermittent process soft measurement modeling method based on CSJITL-RVM | |
CN105718999B (en) | A kind of construction method and system of heuristic metabolism coexpression network | |
CN113192114B (en) | Blade multi-field point cloud registration method based on overlapping features and local distance constraint | |
Jung et al. | A line-based progressive refinement of 3D rooftop models using airborne LiDAR data with single view imagery | |
CN116643246A (en) | Deep clustering radar pulse signal sorting method based on inner product distance measurement | |
CN112651317A (en) | Hyperspectral image classification method and system for sample relation learning | |
CN114964313A (en) | RVM-based fiber optic gyroscope temperature compensation method | |
CN112967210B (en) | Unmanned aerial vehicle image denoising method based on full convolution twin network | |
Sima et al. | Sparse representation classification based on flexible patches sampling of superpixels for hyperspectral images | |
CN111553954A (en) | Direct method monocular SLAM-based online luminosity calibration method | |
CN112381070B (en) | Fast robust face recognition method | |
CN112991187B (en) | Convolution twin-point network blade profile splicing system based on multiple spatial similarities | |
CN114627496A (en) | Robust pedestrian re-identification method based on depolarization batch normalization of Gaussian process | |
CN113537108A (en) | Crop chlorophyll content inversion method based on unmanned aerial vehicle hyperspectral remote sensing | |
CN113705731A (en) | End-to-end image template matching method based on twin network | |
CN112417188B (en) | Hyperspectral image classification method based on graph model | |
CN112712050B (en) | Polarized SAR image semantic change detection method based on DS evidence fusion | |
Lachaud et al. | Lightweight Curvature Estimation on Point Clouds with Randomized Corrected Curvature Measures | |
CN117313564B (en) | Method, device and storage medium for inverting ocean-atmosphere optical parameters | |
CN113743363B (en) | Shielded target identification method based on small sample of unmanned aerial vehicle system |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |