CN112017220A - Point cloud accurate registration method based on robust constraint least square algorithm - Google Patents
Point cloud accurate registration method based on robust constraint least square algorithm Download PDFInfo
- Publication number
- CN112017220A CN112017220A CN202010875627.7A CN202010875627A CN112017220A CN 112017220 A CN112017220 A CN 112017220A CN 202010875627 A CN202010875627 A CN 202010875627A CN 112017220 A CN112017220 A CN 112017220A
- Authority
- CN
- China
- Prior art keywords
- point
- point cloud
- registered
- cloud
- points
- 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
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 32
- 239000013598 vector Substances 0.000 claims abstract description 39
- 239000011159 matrix material Substances 0.000 claims description 32
- 238000013519 translation Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000000007 visual effect Effects 0.000 abstract 1
- 238000012937 correction Methods 0.000 description 5
- 230000005484 gravity Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000005192 partition Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000036544 posture Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Operations Research (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention provides a point cloud accurate registration method based on robust constraint least square algorithm, which comprises the steps of firstly, acquiring point cloud data with certain overlapping degree observed from different visual angles of a measured object according to a ground three-dimensional laser scanner, and roughly registering the point cloud to be registered and a target point cloud through a PCA algorithm to enable the two point clouds to have a relatively good initial position; secondly, establishing a kd-tree in the target point cloud, searching points corresponding to the point cloud to be registered, and reserving points with larger feature degrees according to the normal vector features of the point cloud; and finally, unifying the point cloud to be registered and the target point cloud to the same coordinate system by using an robust constraint least square algorithm for the obtained corresponding point pairs. The invention effectively solves the problems of large data volume and low precision of the original point cloud and complex solving of the trigonometric function, avoids the trapping of partial optimal solution in the registration process, shortens the registration time and improves the precision of point cloud registration.
Description
Technical Field
The invention belongs to the field of computer vision, and particularly relates to a point cloud accurate registration method based on Robust Constrained Least Square (RCLS) algorithm.
Background
The 3D point cloud registration is one of key research problems in the field of computer vision, and has important application in the aspects of reverse engineering, SLAM, image processing, pattern recognition and the like. The aim of point cloud registration is to solve a transformation matrix of point clouds with different postures under the same coordinate, realize the accurate registration of multi-view scanning point clouds by using the matrix, and finally obtain a complete 3D digital model.
For 3D point cloud registration, a large amount of research is performed by scholars at home and abroad, but most research methods usually focus on the previous corresponding point pair search, the corresponding point pair search efficiency is improved by extracting relevant features of the point cloud such as normal vectors, curvatures and the like, and then the rotation matrix R and the translation vector T are solved by analytical solution methods such as an SVD method or a quaternion method, but these methods do not consider gross errors contained in the point cloud data and the complexity of trigonometric function calculation, and limit the accuracy of point cloud registration to a certain extent.
Disclosure of Invention
The purpose of the invention is as follows: in order to solve the problems, the invention provides a point cloud accurate registration method based on an robust constraint least square algorithm, which can perform point cloud registration at any angle, shorten registration time and improve registration accuracy.
The technical scheme is as follows: the invention discloses a point cloud accurate registration method based on an robust constraint least square algorithm, which comprises the following steps:
(1) acquiring a cloud of points to be registered and a target point cloud which are subjected to denoising and filtering;
(2) roughly registering the point cloud to be registered and the target point cloud by a PCA algorithm;
(3) searching corresponding points of the point cloud to be registered in the target point cloud by using a kd-tree, setting a search distance threshold eta, and deleting wrong point pairs;
(4) extracting points with larger feature degree according to the normal vector features of the point cloud;
(5) and registering the cloud of the point to be registered to the target point cloud by utilizing an anti-difference constraint least square algorithm to the obtained corresponding point pair until the two point clouds meet a distance threshold (1 e-6).
Further, the step (1) is realized as follows:
denoising and filtering the two-point cloud by adopting an SOA outlier noise removing algorithm, and setting the point cloud to be processed as P ═ Pi∈R31,2,3, …, n, where n is the number of sampling points in the point cloud to be processed, and for any sampling point piEstablishing piK neighborhood of (a), calculating a sampling point piAnd (3) average distance to k neighborhood points of the point is obtained, then the average distance is judged, if the distance is within a set range, the point is judged as a main point cloud, otherwise, the point is judged as a noise point, and then the noise point is removed.
Further, the step (2) is realized as follows:
(21) calculating the coordinates of the mass center points of the point cloud P to be registered and the target point cloud Q:
wherein N and M represent the number of point clouds P and Q to be registered, UPAnd UQRepresenting the coordinates of the centroid points of P and Q;
(22) the covariance of the two sets of point clouds was calculated:
covP＝(P-UP)(P-UP)T
covQ＝(Q-UQ)(Q-UQ)T
in the formula of UPAnd UQCentroid point coordinates representing P, Q, cov representing covariance, T representing transpose;
(23) respectively solving the characteristic vectors of the two groups of point clouds through the covariance matrix, arranging the characteristic vectors from large to small according to characteristic values, solving a rotation matrix R and a translation matrix T of the coarse registration, unifying coordinate axes formed by the two groups of point clouds to the same coordinate system, and finishing the coarse registration.
Further, the search distance threshold η in step (3) is 1.5 times of the average distance value of all corresponding point pairs.
Further, the step (4) is realized as follows:
(41) a certain point p in the point cloudiDegree of change of normal vector, i.e. its feature degree fiDefined as the arithmetic mean of the angles between its normal vector and its k neighbors:
wherein, thetaijIs a point piNormal vector of and its neighboring point pjThe included angle of the normal vector of (a);
(42) selecting proper threshold value1Removing the flatter portion of the point cloud and retaining fi>1For any p in the reserved pointsmIf it satisfies:
pmf(pm)＝max[f(pm1),f(pm2),…,f(pmk)]
p is to bemAs a feature point, wherein f (p)m1),f(pm2),…,f(pmk) Is a point pmK is the feature degree of the neighboring point;
(43) setting two point clouds as P and Q, wherein P is the point cloud to be registered, Q is the target point cloud, respectively extracting the characteristic points of the two point clouds to obtain a characteristic point set of P as Pt＝{ptl,pt2,pt3…,ptn′Q, the characteristic point set of Q is Qt＝{qtl,qt2,qt3,…,qtm′N 'and m' are the numbers of characteristic points of P and Q, respectively.
Further, the step (5) is realized as follows:
(51) carrying out barycenter processing on the coordinates of the point cloud to be registered and the target point cloud, and establishing an error equation after barycenter processing;
(52) obtaining a parameter approximate value by using a basic formula of indirect adjustment as an initial value of a parameter to be estimated;
(53) obtain 5 constraint stripsThe initial value of the parameter obtained from (52) is used to obtain WxAnd a C matrix;
(54) using indirect adjustment with a limiting condition to update parameter values to obtain a rotation matrix R;
(55) judging whether the convergence condition is met according to the size of the parameter, and if the two norms of the parameter to be estimated are met
The parameters are considered to satisfy convergence, and the parameter estimation values are updated; if the condition is not met, repeating the iteration process;
(56) and restoring the translation parameters, and calculating the coordinates of the point cloud to be registered in the new coordinate system according to the obtained conversion parameters.
Has the advantages that: compared with the prior art, the invention has the beneficial effects that: the method effectively solves the problems of low accuracy and large data volume of the original point cloud data, greatly shortens the registration time, is easier to obtain the initial value compared with the traditional analytic solution method, does not need to use complex trigonometric function calculation, avoids trapping in the local optimal solution in the registration process, and obtains a more accurate registration result.
Drawings
FIG. 1 is a schematic flow diagram of a construction method according to the present invention;
FIG. 2 is a schematic diagram of establishing a point cloud kd-tree;
FIG. 3 is a diagram of normal vectors of different regions.
Detailed Description
The present invention is described in further detail below with reference to the attached drawing figures.
The invention provides a point cloud accurate registration method based on robust constraint least square algorithm, which comprises the steps of firstly carrying out rough registration on a cloud of a point to be registered and a target point cloud which are subjected to denoising and filtering by using PCA algorithm, and then searching corresponding point pairs in the cloud of the point to be registered and the target point cloud by using kd-tree. Secondly, according to the normal vector characteristics of the point cloud, preserving the point pairs with larger characteristic degree; and finally, solving the rotation matrix R and the translational vector T of the obtained corresponding point pair according to a constrained least square robust method, and registering the cloud of the point to be registered to the target point cloud through rotation and translation until the absolute value of the distance difference between the two points of cloud is minimum, thereby realizing the accurate registration of the point cloud. As shown in fig. 1, the method specifically comprises the following steps:
step one, acquiring a point cloud to be registered and a target point cloud which are subjected to denoising and filtering.
Denoising and filtering two-point cloud by using an Outlier noise Removal algorithm of a Statistical Outlier Removal (SOA), wherein the method has the main principle that the point cloud to be processed is P ═ Pi∈R31,2,3, …, n, where n is the number of sampling points in the point cloud to be processed, and for any sampling point piFirst, p is establishediK neighborhood of (a), then calculate the sampling point piAnd (3) average distance to k neighborhood points of the point, then judging the average distance, if the distance is within a set range (defined variance and global distance average), judging the point cloud as a main body point cloud, otherwise, judging the point cloud as a noise point, and then removing the noise points and diffusing the noise points to the neighborhood of the point. Thereby enabling the curved surface to be smoother.
Two thresholds need to be set during the SOR denoising, one is the number k of neighborhood points, and the other is a multiple n of the standard deviation, as follows:
wherein the content of the first and second substances,
denotes the mean distance, σ denotes the standard deviation, when the distance exceeds dmaxIt is determined that the noise point is deleted.
And step two, roughly registering the point cloud to be registered and the target point cloud through a PCA algorithm.
Firstly, calculating the coordinates of the mass center points of two groups of point clouds P and Q:
wherein N and M represent the number of point clouds P and Q to be registered, UPAnd UQRepresenting the centroid point coordinates of P, Q.
Secondly, calculating the covariance of the two groups of point clouds:
covP＝(P-UP)(P-UP)T
covQ＝(Q-UQ)(Q-UQ)T
in the formula of UPAnd UQRepresenting the centroid point coordinates of P, Q, cov representing the covariance, and T representing the transpose.
And respectively solving the characteristic vectors of the two groups of point clouds through the covariance matrix, arranging the characteristic vectors from large to small according to characteristic values, solving a rotation matrix R and a translation matrix T, unifying coordinate axes formed by the two groups of point clouds to the same coordinate system, and finishing the related process of coarse registration.
And step three, searching corresponding points of the point cloud to be registered in the target point cloud by using the kd-tree, setting a search distance threshold eta (eta is 1.5 times of the distance average value of all corresponding point pairs), and deleting wrong point pairs.
The kd-tree is a K-dimensional tree for short, and is a data structure divided by space. Are often used for searches in high dimensional spaces such as range searches and nearest neighbor searches. A kd-Tree is a special case of a binary space partition tree, and in a three-dimensional point cloud, the dimension of the kd-Tree is 3. Each level of the kd-Tree separates all the child nodes in a given dimension. All children nodes at the root of the tree are separated in the first designated dimension, each level of the tree is separated in the next dimension, and all other dimensions are used up and then returned to the first dimension, so the kd-Tree is highly efficient for interval and neighbor search.
The three-dimensional point cloud data kd-tree generation process firstly selects a YZ plane, an XZ plane or an XY plane as a partition plane according to the variance value of the point cloud in a certain direction and from large to small. And repeating the segmentation process until the segmented region does not contain any point, and finishing the segmentation. The point cloud kd-tree is established as shown in FIG. 2.
And step four, extracting points with larger feature according to the normal vector features of the point cloud.
Defining a point p in a point cloudiDegree of change of normal vector, i.e. its feature degree fiIs the arithmetic mean value of the included angle between the normal vector and the k adjacent points:
in the formula [ theta ]ijIs a point piNormal vector of and its neighboring point pjThe angle of the normal vector of (a).
According to this definition, points with a greater degree of feature represent a greater fluctuation of the area, so the feature point in the point cloud can be extracted using this degree of feature. Selecting proper threshold value1Removing the flatter portion of the point cloud and retaining fi>1For any p in the reserved pointsmIf it satisfies:
pmf(pm)＝max[f(pm1),f(pm2),…,f(pmk)]
p is to bemAs a feature point, wherein f (p)m1),f(pm2),…,f(pmk) Is a point pmK of (2) is the feature of the neighboring point. And setting two point clouds as P and Q, wherein P is the point cloud to be registered, and Q is the target point cloud. Respectively extracting the feature points of the two point clouds to obtain a feature point set P of the Pt＝{ptl,pt2,pt3…,ptn′Q, the characteristic point set of Q is Qt＝{qtl,qt2,qt3,…,qtm′N 'and m' are the numbers of characteristic points of P and Q, respectively.
Because the normal vectors of different areas are different, as shown in fig. 3, the points with larger feature degree are extracted according to the normal vector features of the point cloud, because fiIn most cases, the temperature is mainly distributed between 0 and 60 degrees, and the invention leads f toiIs divided into [0 degree, 20 degree ]]、(20°，40°]、(40°，60°]And (60 DEG,180°]equal 4 intervals, only the first three spaced points are retained.
And step five, the essence of point cloud registration is to convert the point cloud to be registered and the target point cloud into the same coordinate system through three-dimensional coordinates. The robust constraint least square algorithm is linear about parameters, an ideal initial value can be obtained by adopting indirect adjustment, and the conversion parameters can be better solved by utilizing the initial value.
An equivalent weight robust principle is introduced into a constraint least square algorithm, and a robust estimation algorithm is adopted for parameter estimation, namely, an IGG III weight function is adopted for robust processing. The robust constraint least square algorithm is a twelve-parameter model observation equation based on five constraint conditions, compared with a traditional nonlinear seven-parameter model, the twelve-parameter model observation equation is linear about parameters, an ideal initial value can be obtained by adopting indirect adjustment, and the twelve-parameter model avoids complex operation of expressing a rotation matrix by a trigonometric function. The method specifically comprises the following steps:
(1) and carrying out barycentric processing on the coordinates of the point cloud to be registered and the target point cloud, and establishing an error equation after barycentric processing.
Setting the corresponding point pair obtained by searching in the step four as PiAnd Qi，PiAnd QiRepresenting the corresponding points searched in the point cloud to be registered and the target point cloud.
The mathematical model of common points in the universal three-dimensional coordinate transformation is as follows:
wherein [ x ]pi ypi zpi]TAnd [ x ]qi yqi zqi]TRespectively represent the ith point at the point to be registerediCoordinates and target point cloud QiCoordinates under a coordinate system; [ e ] axqi eyqi ezqi]TRepresenting the random error vector in the coordinates of the target point cloud. T [. DELTA.x.DELTA.y.DELTA.z]TRepresenting a translation vector, R is a 3 x 3 rotation matrix.
In point cloud coordinate P to be registeredi(Xpi,Ypi,Zpi) And target point cloud coordinates Qi(Xqi,Yqi,Zqi) The center of gravity treatment is carried out according to the following formula:
wherein (X'pi,Y'pi,Z'pi) And (X'qi,Y'qi,Z'qi) Represents the coordinates of the center of gravity (X)pg,Ypg,Zpg) And (X)qg,Yqg,Zqg) Respectively representing cloud coordinates P of points to be registerediAnd target point cloud coordinates QiAfter the determination, the barycentric coordinates of (a) can be processed as parameters.
In the formula, N represents the number of points included in the searched point cloud, and an error equation can be established by the above formula as follows:
v′i＝Bix′-Li
in formula (II), v'iIndicates the number of corrections, BiRepresenting coefficient terms, x' representing the correction of the approximation of the parameter to be determined, LiRepresenting an observation vector.
In the above formula:
x′＝(△x'△y'△z'(vec(RT))T)T
wherein T '[. DELTA.x'. DELTA.y '. DELTA.z']T，(X'pi,Y'pi,Z'pi) And (X'qi,Y'qi,Z'qi) Representing barycentric coordinates, "vec" represents a column vectorization operator, which stacks the columns of the matrix together in left-to-right order.
(2) Obtaining parameter approximate value by using basic formula of indirect adjustment as initial value of parameter to be estimated
(3) 5 constraint conditions of the robust constraint least square algorithm form a constraint condition equation, and W can be obtained from the initial values of the parametersxAnd a C matrix.
The five constraint conditions of the robust constraint least square algorithm are as follows:
in the formula, xi11，ξ12，ξ13，ξ21，ξ22，ξ23，ξ31，ξ32，ξ33Representing the constraint parameters.
The five constraints were linearized and written in the standard form:
Cx+Wx＝0
in the formula, C represents a coefficient matrix of a constraint equation, x represents a correction number of an approximate value of a parameter to be determined, and WxA closed difference constant vector representing a constraint equation.
Wherein:
(4) and solving the updated parameter value by using indirect adjustment with a limiting condition.
The adjustment model is as follows:
wherein V represents the observation error, l represents the constant term of the error equation, B represents the coefficient matrix of the error equation of the observation value,
representing the correction of the approximation of the parameter to be determined, C representing the coefficient matrix of the constraint equation, WxThe closed difference constant vector representing the constraint equation, D the covariance matrix, σ0Representing the variance of unit weight, and Q and P respectively represent the corresponding covariance matrix and observation weight matrix.
The composition method equation is as follows:
in the formula, NbbA matrix of coefficients representing a normal equation,
expressing the correction of the approximate value of the undetermined parameter, C expressing the coefficient matrix of the constraint equation, KsA joint coefficient vector representing a constraint equation, and W represents a closure difference vector.
Wherein:
the solution to the equation is:
solving xi from the solution of the equation11，ξ12，ξ13，ξ21，ξ22，ξ23，ξ31，ξ32，ξ33The rotation matrix R is obtained:
(5) judging whether the convergence condition is met according to the size of the parameter, and if the two norms of the parameter to be estimated are met
The parameters are considered to satisfy convergence, and the parameter estimation values are updated; if the condition is not met, repeating the iteration process;
(6) and restoring the translation parameters, and calculating the coordinates of the point cloud P to be registered in the new coordinate system according to the obtained conversion parameters.
And because of the parameters between the two sets of coordinate systems after the center of gravity is changed, the ill-posed property of the normal equation is greatly weakened. Relative to two sets of coordinates without center of gravity, only the translation vector T [ [ delta ] x [ delta ] y [ delta ] z ] is changed]T
In the formula, [ X ]qg Yqg Zqg]T，[Xpg Ypg Zpg]TRespectively representing target point cloud coordinates QiAnd the cloud coordinate P of the point to be registerediThe coordinates after the center of gravity is changed,
respectively representing target point cloud coordinates QiAnd the cloud coordinate P of the point to be registerediIs determined from T '═ Δ x'. DELTA.y '. DELTA.z']TThe translation vector T [. DELTA.x.DELTA.y.DELTA.z ] can be restored]T:
In practical application, arbitrary angle three-dimensional coordinate conversion can be performed by using a constrained least squares method.
The robust estimation is to apply an estimation criterion with certain robust to the observation data directly for parameter estimation, and the robust processing is performed by adopting an IGG III weight function for calculating an equivalent weight matrix P. For the weight factor function of IGG III, the reciprocal is calculated to obtain the corresponding co-factor function:
in the formula (I), the compound is shown in the specification,
normalized residual, R, representing the ith point of the observationiiWeight factor, k, representing the ith point of the observed value0And k1Are all constants, k0Generally 1 to 1.5, k1Generally, it is about 3.
Since the third segment of the IGG III function is 0, the theoretical value of the corresponding co-factor is infinite, and a large number (10) is used for convenient actual calculation10) Instead, this is completely satisfactory in numerical calculation.
The addition of robust estimation can effectively carry out layering weighting processing on the point cloud, and effectively reduce the influence of small gross errors of the eliminated section on point cloud registration through a plurality of iterative processes. The conversion precision of the constraint least square algorithm is higher, and the influence of gross errors on data is effectively avoided. And (4) solving the rotation matrix R and the translational vector T of the corresponding point pair obtained in the step four by using an robust constraint least square algorithm, and registering the cloud of the point to be registered to the target point cloud through rotation and translation until the distance between the two point clouds meets a distance threshold (generally taking a value of 1 e-6). And if not, continuously executing the third step to the fifth step on the converted cloud of the point to be registered, and outputting a result after the threshold value is met.
Claims (6)
1. A point cloud accurate registration method based on robust constraint least square algorithm is characterized by comprising the following steps:
(1) acquiring a cloud of points to be registered and a target point cloud which are subjected to denoising and filtering;
(2) roughly registering the point cloud to be registered and the target point cloud by a PCA algorithm;
(3) searching corresponding points of the point cloud to be registered in the target point cloud by using a kd-tree, setting a search distance threshold eta, and deleting wrong point pairs;
(4) extracting points with larger feature degree according to the normal vector features of the point cloud;
(5) and registering the cloud of the point to be registered to the target point cloud by utilizing an anti-difference constraint least square algorithm to the obtained corresponding point pair until the two point clouds meet a distance threshold.
2. The method for accurately registering point clouds based on robust constrained least squares algorithm according to claim 1, wherein the step (1) is realized by the following steps:
denoising and filtering the two-point cloud by adopting an SOA outlier noise removing algorithm, and setting the point cloud to be processed as P ═ Pi∈R31,2,3, …, n, where n is the number of sampling points in the point cloud to be processed, and for any sampling point piEstablishing piK neighborhood of (a), calculating a sampling point piAnd (3) average distance to k neighborhood points of the point is obtained, then the average distance is judged, if the distance is within a set range, the point is judged as a main point cloud, otherwise, the point is judged as a noise point, and then the noise point is removed.
3. The method for accurately registering point clouds based on robust constrained least squares algorithm according to claim 1, wherein the step (2) is realized by the following steps:
(21) calculating the coordinates of the mass center points of the point cloud P to be registered and the target point cloud Q:
wherein N and M represent the number of point clouds P and Q to be registered, UPAnd UQRepresenting the coordinates of the centroid points of P and Q;
(22) the covariance of the two sets of point clouds was calculated:
covP＝(P-UP)(P-UP)T
covQ＝(Q-UQ)(Q-UQ)T
in the formula of UPAnd UQCentroid point coordinates representing P, Q, cov representing covariance, T representing transpose;
(23) respectively solving the characteristic vectors of the two groups of point clouds through the covariance matrix, arranging the characteristic vectors from large to small according to characteristic values, solving a rotation matrix R and a translation matrix T of the coarse registration, unifying coordinate axes formed by the two groups of point clouds to the same coordinate system, and finishing the coarse registration.
4. The method for accurately registering point clouds based on robust constrained least squares algorithm according to claim 1, wherein the search distance threshold η in step (3) is 1.5 times of the average distance of all corresponding points.
5. The method for accurately registering point cloud based on robust constrained least squares algorithm according to claim 1, wherein the step (4) is realized by the following steps:
(41) a certain point p in the point cloudiDegree of change of normal vector, i.e. its feature degree fiDefined as the arithmetic mean of the angles between its normal vector and its k neighbors:
wherein, thetaijIs a point piNormal vector of and its neighboring point pjThe included angle of the normal vector of (a);
(42) selecting proper threshold value1Removing the flatter portion of the point cloud and retaining fi>1For any p in the reserved pointsmIf it satisfies:
pmf(pm)＝max[f(pm1),f(pm2),…,f(pmk)]
p is to bemAs a feature point, wherein f (p)m1),f(pm2),…,f(pmk) Is a point pmK is the feature degree of the neighboring point;
(43) setting two point clouds as P and Q, wherein P is the point cloud to be registered, Q is the target point cloud, respectively extracting the characteristic points of the two point clouds to obtain a characteristic point set of P as Pt＝{ptl,pt2,pt3…,ptn′Q, the characteristic point set of Q is Qt＝{qtl,qt2,qt3,…,qtm′N 'and m' are the numbers of characteristic points of P and Q, respectively.
6. The method for accurately registering point cloud based on robust constrained least squares algorithm according to claim 1, wherein the step (5) is realized by the following steps:
(51) carrying out barycenter processing on the coordinates of corresponding points obtained by searching the point cloud to be registered and the target point cloud, and establishing an error equation after barycenter processing;
(52) obtaining a parameter approximate value by using a basic formula of indirect adjustment as an initial value of a parameter to be estimated;
(53) obtaining 5 constraint conditions to form a constraint condition equation, and obtaining W from the initial value of the parameter obtained in (52)xAnd a C matrix;
(54) using indirect adjustment with a limiting condition to update parameter values to obtain a rotation matrix R;
(55) judging whether the convergence condition is satisfied according to the size of the parameter, generally considering the two-norm of the parameter to be estimated
The parameters can be considered to satisfy convergence, and the parameter estimation values are updated; if the condition is not met, repeating the iteration process;
(56) and restoring the translation parameter T, and calculating the coordinates of the point cloud to be registered in the new coordinate system according to the obtained conversion parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010875627.7A CN112017220B (en) | 2020-08-27 | 2020-08-27 | Point cloud accurate registration method based on robust constraint least square algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010875627.7A CN112017220B (en) | 2020-08-27 | 2020-08-27 | Point cloud accurate registration method based on robust constraint least square algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112017220A true CN112017220A (en) | 2020-12-01 |
| CN112017220B CN112017220B (en) | 2023-07-28 |
Family
ID=73503661
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010875627.7A Active CN112017220B (en) | 2020-08-27 | 2020-08-27 | Point cloud accurate registration method based on robust constraint least square algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112017220B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112509018A (en) * | 2020-12-03 | 2021-03-16 | 湖南大学 | Quaternion space optimized three-dimensional image registration method |
| CN112700480A (en) * | 2020-12-29 | 2021-04-23 | 河北工业大学 | Point cloud rapid registration method for small-size object rotation scanning and application |
| CN112767457A (en) * | 2021-01-25 | 2021-05-07 | 北京影谱科技股份有限公司 | Principal component analysis-based plane point cloud matching method and device |
| CN112862844A (en) * | 2021-02-20 | 2021-05-28 | 苏州工业园区测绘地理信息有限公司 | Road boundary interactive extraction method based on vehicle-mounted point cloud data |
| CN112945198A (en) * | 2021-02-02 | 2021-06-11 | 贵州电网有限责任公司 | Automatic detection method for power transmission line iron tower inclination based on laser LIDAR point cloud |
| CN113124782A (en) * | 2021-04-14 | 2021-07-16 | 重庆市勘测院 | Construction perpendicularity detection method based on point cloud tolerance self-adaption |
| CN113192114A (en) * | 2021-07-01 | 2021-07-30 | 四川大学 | Blade multi-field point cloud registration method based on overlapping features and local distance constraint |
| CN113205547A (en) * | 2021-03-18 | 2021-08-03 | 北京长木谷医疗科技有限公司 | Point cloud registration method, bone registration method, device, equipment and storage medium |
| CN113706381A (en) * | 2021-08-26 | 2021-11-26 | 北京理工大学 | Three-dimensional point cloud data splicing method and device |
| CN114170279A (en) * | 2021-11-30 | 2022-03-11 | 哈尔滨工程大学 | Point cloud registration method based on laser scanning |
| CN115100254A (en) * | 2022-06-10 | 2022-09-23 | 兰州交通大学 | Point cloud registration method based on dual quaternion description under planar feature constraint |
| CN116883469A (en) * | 2023-07-20 | 2023-10-13 | 中国矿业大学 | Point cloud registration method based on EIV model description under plane feature constraint |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160171735A1 (en) * | 2014-12-16 | 2016-06-16 | Fu Tai Hua Industry (Shenzhen) Co., Ltd. | Computing device and method for joining point clouds |
| US20190362178A1 (en) * | 2017-11-21 | 2019-11-28 | Jiangnan University | Object Symmetry Axis Detection Method Based on RGB-D Camera |
| CN111145232A (en) * | 2019-12-17 | 2020-05-12 | 东南大学 | Three-dimensional point cloud automatic registration method based on characteristic information change degree |
| CN111563920A (en) * | 2020-04-15 | 2020-08-21 | 西安工程大学 | 3D color point cloud registration method based on global optimization and multi-constraint condition iteration |
-
2020
- 2020-08-27 CN CN202010875627.7A patent/CN112017220B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160171735A1 (en) * | 2014-12-16 | 2016-06-16 | Fu Tai Hua Industry (Shenzhen) Co., Ltd. | Computing device and method for joining point clouds |
| US20190362178A1 (en) * | 2017-11-21 | 2019-11-28 | Jiangnan University | Object Symmetry Axis Detection Method Based on RGB-D Camera |
| CN111145232A (en) * | 2019-12-17 | 2020-05-12 | 东南大学 | Three-dimensional point cloud automatic registration method based on characteristic information change degree |
| CN111563920A (en) * | 2020-04-15 | 2020-08-21 | 西安工程大学 | 3D color point cloud registration method based on global optimization and multi-constraint condition iteration |
Non-Patent Citations (1)
| Title |
|---|
| LIU RAB 等: "Normal Estimation Algorithm for Point Cloud using KD-Tree", IET INTERNATIONAL CONFERENCE ON SMART AND SUSTAINABLE CITY 2013 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112509018A (en) * | 2020-12-03 | 2021-03-16 | 湖南大学 | Quaternion space optimized three-dimensional image registration method |
| CN112509018B (en) * | 2020-12-03 | 2023-02-17 | 湖南大学 | Quaternion space optimized three-dimensional image registration method |
| CN112700480A (en) * | 2020-12-29 | 2021-04-23 | 河北工业大学 | Point cloud rapid registration method for small-size object rotation scanning and application |
| CN112767457A (en) * | 2021-01-25 | 2021-05-07 | 北京影谱科技股份有限公司 | Principal component analysis-based plane point cloud matching method and device |
| CN112945198B (en) * | 2021-02-02 | 2023-01-31 | 贵州电网有限责任公司 | Automatic detection method for inclination of power transmission line iron tower based on laser LIDAR point cloud |
| CN112945198A (en) * | 2021-02-02 | 2021-06-11 | 贵州电网有限责任公司 | Automatic detection method for power transmission line iron tower inclination based on laser LIDAR point cloud |
| CN112862844B (en) * | 2021-02-20 | 2024-01-05 | 园测信息科技股份有限公司 | Road boundary interactive extraction method based on vehicle-mounted point cloud data |
| CN112862844A (en) * | 2021-02-20 | 2021-05-28 | 苏州工业园区测绘地理信息有限公司 | Road boundary interactive extraction method based on vehicle-mounted point cloud data |
| CN113205547A (en) * | 2021-03-18 | 2021-08-03 | 北京长木谷医疗科技有限公司 | Point cloud registration method, bone registration method, device, equipment and storage medium |
| CN113124782B (en) * | 2021-04-14 | 2022-09-20 | 重庆市勘测院 | Construction perpendicularity detection method based on point cloud tolerance self-adaption |
| CN113124782A (en) * | 2021-04-14 | 2021-07-16 | 重庆市勘测院 | Construction perpendicularity detection method based on point cloud tolerance self-adaption |
| CN113192114A (en) * | 2021-07-01 | 2021-07-30 | 四川大学 | Blade multi-field point cloud registration method based on overlapping features and local distance constraint |
| CN113706381A (en) * | 2021-08-26 | 2021-11-26 | 北京理工大学 | Three-dimensional point cloud data splicing method and device |
| CN114170279A (en) * | 2021-11-30 | 2022-03-11 | 哈尔滨工程大学 | Point cloud registration method based on laser scanning |
| CN115100254A (en) * | 2022-06-10 | 2022-09-23 | 兰州交通大学 | Point cloud registration method based on dual quaternion description under planar feature constraint |
| CN115100254B (en) * | 2022-06-10 | 2023-06-30 | 兰州交通大学 | Point cloud registration method for dual quaternion description under plane feature constraint |
| CN116883469A (en) * | 2023-07-20 | 2023-10-13 | 中国矿业大学 | Point cloud registration method based on EIV model description under plane feature constraint |
| CN116883469B (en) * | 2023-07-20 | 2024-01-19 | 中国矿业大学 | Point cloud registration method based on EIV model description under plane feature constraint |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112017220B (en) | 2023-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112017220A (en) | Point cloud accurate registration method based on robust constraint least square algorithm | |
| CN110853075B (en) | Visual tracking positioning method based on dense point cloud and synthetic view | |
| CN100559398C (en) | Automatic deepness image registration method | |
| CN110189399B (en) | Indoor three-dimensional layout reconstruction method and system | |
| CN109493372B (en) | Rapid global optimization registration method for product point cloud data with large data volume and few characteristics | |
| CN114677418B (en) | Registration method based on point cloud feature point extraction | |
| CN110599506B (en) | Point cloud segmentation method for three-dimensional measurement of complex special-shaped curved surface robot | |
| CN107818598B (en) | Three-dimensional point cloud map fusion method based on visual correction | |
| CN114972459B (en) | Point cloud registration method based on low-dimensional point cloud local feature descriptor | |
| CN110322492B (en) | Space three-dimensional point cloud registration method based on global optimization | |
| CN111145232A (en) | Three-dimensional point cloud automatic registration method based on characteristic information change degree | |
| CN112257722A (en) | Point cloud fitting method based on robust nonlinear Gaussian-Hummer model | |
| CN109766903B (en) | Point cloud model curved surface matching method based on curved surface features | |
| CN112396641A (en) | Point cloud global registration method based on congruent two-baseline matching | |
| CN104463953A (en) | Three-dimensional reconstruction method based on inertial measurement unit and RGB-D sensor | |
| CN115861397A (en) | Point cloud registration method based on improved FPFH-ICP | |
| CN111820545A (en) | Method for automatically generating sole glue spraying track by combining offline and online scanning | |
| CN111598995A (en) | Self-supervision multi-view three-dimensional human body posture estimation method based on prototype analysis | |
| CN108597016A (en) | Torr-M-Estimators basis matrix robust estimation methods based on joint entropy | |
| CN117132630A (en) | Point cloud registration method based on second-order spatial compatibility measurement | |
| CN117274339A (en) | Point cloud registration method based on improved ISS-3DSC characteristics combined with ICP | |
| CN116091727A (en) | Complex Qu Miandian cloud registration method based on multi-scale feature description, electronic equipment and storage medium | |
| CN113642397B (en) | Object length measurement method based on mobile phone video | |
| CN113313200B (en) | Point cloud precision matching method based on normal constraint | |
| CN112581511A (en) | Three-dimensional reconstruction method and system based on approximate vertical scanning point cloud rapid registration |
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 |