CN109493372A - The product point cloud data Fast global optimization method for registering of big data quantity, few feature - Google Patents
The product point cloud data Fast global optimization method for registering of big data quantity, few feature Download PDFInfo
- Publication number
- CN109493372A CN109493372A CN201811242567.4A CN201811242567A CN109493372A CN 109493372 A CN109493372 A CN 109493372A CN 201811242567 A CN201811242567 A CN 201811242567A CN 109493372 A CN109493372 A CN 109493372A
- Authority
- CN
- China
- Prior art keywords
- point
- center
- mass
- cloud
- overlapping region
- 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
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
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- 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
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of big data quantities, the product point cloud data Fast global optimization method for registering of few feature, the problems such as, initial position big for there are data volumes in registration process is inaccurately, global registration speed is slow, propose the key point global optimization method for registering based on overlapping region;This method carries out piecemeal to cloud using super voxel clustering method, finds overlapping region according to the mass center of each piecemeal and its parameter, and the FPFH of mass center is calculated in overlapping region and obtains corresponding corresponding center of mass point, to complete rough registration;The key point that point cloud in overlapping region is extracted using intrinsic shape characteristic method (ISS) completes essence registration in conjunction with global optimization approach;Show that the present invention can quickly and effectively complete the registration work of big data quantity, few characteristic product with quasi-experiment, solves the problems, such as initial position, further improve the precision and efficiency of registration.
Description
Technical field
The invention belongs to technical field of machine vision, in particular to the product point cloud data of a kind of big data quantity, few feature
Fast global optimization method for registering.
Background technique
Three-dimensional point cloud registration is a hot issue in computer vision, so-called point cloud registering gives different seats
The point set of two different postures under two groups of point sets or same coordinate system in mark system, by this two groups of point sets by rotation,
Translation is transformed under the same coordinate system or two groups of point sets overlaps.Three-dimensional point cloud is registered in field of machine vision and answers
With very extensive, including 3D modeling, Object identifying, Attitude estimation, robot navigation etc..Point cloud registering is broadly divided into two classes: one
It is the registration of rigid body translation, second is that the registration of non-rigid transformation, the present invention is directed the registration of rigid body translation.Since data are adopted
The partial dot cloud of object can only be collected by collecting reasons, the single measurements such as the visual field limitation of equipment, the eclipse phenomena of object itself, be needed
Required complete model could be obtained by being constantly registrated, in registration process, there are single measurement point clouds and most of point
Cloud or the registration problems for completely putting cloud (in the case of known CAD model), there are data volumes for registration of this part to entirety greatly, just
Value position deviation is big, global optimization is registrated slow-footed problem.
Point cloud registering main flow may be generally divided into two stages, point cloud rough registration and point cloud essence registration.
The point cloud rough registration stage is believed by solution point cloud or the local shape geometry of key point under normal conditions
Euclidean distance, point cloud depth angle value between breath, such as normal vector angle between points, the curvature for putting cloud, point etc.,
Then local feature description's for using such information for constituting point cloud, is registrated, therefore office using local feature description's
Portion's Feature Descriptor plays a key role in registration.One good Feature Descriptor should have height descriptive,
In order to provide comprehensive, specific local geometric shape.In order to ensure accurate and efficient point cloud registering, Feature Descriptor should also
With computational efficiency, compactedness and robustness.Many scholars propose many local feature description's at present, including quickly put spy
Sign histogram description sub (FPFH), point signature direction histogram description (SHOT), rotation projection statistics description (RoPS), office
Portion's characteristic statistics figure (LFSH) etc..These both provide local shape geometric description in considerable detail.Therefore, they have higher-dimension
Feature vector, accuracy are high but computationally intensive.In contrast the information Expressive Features such as curvature, normal direction have low descriptive but are easy to
It calculates.
Point cloud essence is registrated the stage, and presently best known method is iteration closest approach (ICP) algorithm, and this algorithm speed is slower
And it is more demanding to the initial position of cloud, so being typically all the essence registration for point cloud.Many scholars carry out ICP algorithm
It improves, Jost and Hugli improve ICP algorithm using multiresolution, improve the speed of ICP algorithm;Dai et al.
ICP algorithm is improved using two new structural constraints.In this way, insecure matching double points will be removed, high quality
Point to will stay for calculating transformational relation;Et al. realized by nesting BNB algorithm and ICP algorithm and convert square
The global convergence of battle array, but this method speed is slow, and only carries out global optimization to spin matrix.
Summary of the invention
That there are data volumes it is an object of the invention to overcoming the cloud registration process of the prior art is big, initial position is inaccurate,
The slow-footed problem of global registration provides the product point cloud data Fast global optimization registration side of a kind of big data quantity, few feature
Method.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of product point cloud data Fast global optimization method for registering of big data quantity, few feature, comprising:
The extraction of overlapping region is carried out to filtered CAD model point cloud based on super voxel clustering algorithm;To each cloud
The center of mass point of block extracts corresponding FPFH Feature Descriptor, carries out rough registration using these cloud Feature Descriptors;
The key point and combination global optimization approach obtained based on ISS algorithm realizes essence registration.
Preferably, the extraction of overlapping region is carried out to filtered CAD model point cloud based on super voxel clustering algorithm;To every
The center of mass point of a cloud mass extracts corresponding FPFH Feature Descriptor, carries out rough registration using these cloud Feature Descriptors,
It specifically includes:
A1, realize that the overlapping region of point cloud is extracted based on super voxel clustering algorithm, i.e., to CAD model data reduction and measurement
The region that point cloud is substantially overlapped;It include: that cloud piecemeal will be put based on super voxel clustering algorithm, and extract the mass center of each piecemeal point cloud
Point;Find out the principal curvatures of CAD model point cloud center of mass point and measurement pointcloud center of mass point, the normal vector for inquiring center of mass point and inquiry mass center
Point arrives the angle of its field point center of gravity line, and extracts criterion using the two elements as overlapping region similitude, obtains
To the corresponding center of mass point in overlapping region, is indexed using center of mass point and come out overlapping region data reduction;By CAD model overlapping region
Mass center point set and measurement pointcloud mass center point set are saved, the input point cloud as cloud rough registration;
A2, rough registration is carried out using mass center point set;It include: CALCULATION CAD model overlapping region mass center point set and measurement pointcloud
The quick point feature FPFH of mass center point set describes son, as the point feature of cloud;Based on the foundation of FPFH Feature Descriptor
With point pair, error matching points pair are removed using stochastical sampling consistency RANSAC algorithm;Calculate the covariance matrix of corresponding points pair
And singular value decomposition SVD is carried out, initial conversion matrix is obtained, the rough registration of point cloud is completed.
Preferably, described to realize that the overlapping region of point cloud is extracted based on super voxel clustering algorithm, it specifically includes:
A1-1, measurement pointcloud and CAD model point cloud are subjected to piecemeal first with super voxel clustering algorithm, seed point is equal
It is even to be distributed in a cloud space clouddiagonalIn, then to distance, that is, Voxel resolution ratio Rvoxel of voxel and seed point
Resolution ratio Rseed is defined, and the minimum bounding box of a cloud is acquired using AABB bounding box, recycles the diagonal of minimum bounding box
The factor of the half of wire length as voxel resolution and seed point resolution realizes that adaptive resolution extracts:
Rvoxel=0.01 × clouddiagonal+0.8 (1)
Rseed=0.037 × clouddiagonal+3.0 (2)
Formula (1) is the value formula of voxel resolution, and formula (2) is the value formula of seed point resolution;
A1-2, its center of mass point is extracted to each point cloud mass after super voxel clustering block, obtains each point cloud mass of CAD model
The mass center point set Q { n1, n2...nn } of mass center point set P { m1, m2...mn } and measurement pointcloud;
A1-3, judgement screening is carried out to the center of mass point extracted in A1-2, realizes the extraction of overlapping region;Judging criterion is
The normal vector of each center of mass point and inquiry center of mass point in the principal curvatures of each center of mass point and mass center point set P, Q in mass center point set P, Q
Angle between the center of gravity line in the field k;
A1-4, the principal curvatures of each center of mass point in mass center point set P, Q in A1-3 is calculated, is estimated using discrete curvature
Algorithm calculates principal curvatures, including minimum curvature and maximum curvature, using the method estimate vector based on local least square method;
A1-5, normal vector and the weight of inquiry center of mass point to the field k to each center of mass point in mass center point set P, Q in A1-3
Angle between heart line is solved;The focus point centerP of K neighborhood is calculated firstσ(i)And centerQσ(j), acquire inquiry
Line of the point center of mass point to its center of gravity | | centerPσ(i)-Pi| | and | | centerQσ(j)-Qj||;Enable nP(i)And nQ(j)Respectively point
Collect the normal vector of each query point of P, Q, then the angle between them is respectively α=accos < nP(i), | | centerPσ(i)-Pσ(i)||
>, β=accos < nQ(j), | | centerQσ(j)-PQσ(j)| | >;
The maximum principal curvatures of each center of mass point is denoted as P respectively in A1-6, mass center point set P, Qik1And Qjk1, minimum principal curvatures
It is denoted as P respectivelyik2And Qjk2;The center of gravity in the normal vector of each center of mass point and inquiry center of mass point to the field k connects in mass center point set P, Q
Angle between line is denoted as α and β;Criterion inequality are as follows: | | Pik1-Qjk1| | < ε 1, | | Pik2-Qjk2| | < ε 2, | | alpha-beta |
| < ε 3, wherein ε 1, ε 2,3 value of ε are respectively 0.03,0.03,0.9.
Preferably, using center of mass point, not all point converges carry out rough registration in step A2.
Preferably, the key point obtained based on ISS algorithm and combination global optimization approach realize essence registration, specific packet
It includes:
A3, the key point that filtered CAD model point cloud and measurement pointcloud are extracted using ISS key point extraction algorithm are right
The key point extracted carries out the essence registration of a cloud using branch and bound method BNB nesting ICP algorithm, and global optimization can be obtained
Transition matrix.
Preferably, as matching object, branch and bound method BNB nesting ICP is calculated the key point extracted in step A3 using ISS
Method carries out the essence registration of a cloud.
The present invention has the advantage that compared with existing the relevant technologies
(1) present invention proposes one kind based on super voxel clustering block and extracts weight to part to whole point cloud registering
The method in folded region, and directly using the center of mass point of the every piece of point cloud extracted in overlapping region as point cloud rough registration data
Source greatly reduces calculation amount, effectively increases the efficiency of rough registration.
(2) present invention has carried out adaptive operation to the parameter during overlapping extracted region, improves the degree of automation,
Avoid the time consumed since parameter is manually entered;
(3) present invention extracts key point using ISS key point extraction algorithm, proposes that carrying out global optimization based on key point matches
Quasi- (BNB nesting ICP algorithm), on the basis of avoiding local optimum, improves the speed and precision of registration.
Invention is further described in detail with reference to the accompanying drawings and embodiments;But a kind of big data quantity of the invention,
The product point cloud data Fast global optimization method for registering of few feature is not limited to the embodiment.
Detailed description of the invention
Fig. 1 is flow chart of the invention;
Fig. 2 is original CAD model and measurement point cloud atlas;Wherein Fig. 2 (a) is original CAD model, and Fig. 2 (b) is measurement pointcloud
Figure;
Fig. 3 is the effect picture that overlapping region is extracted in the embodiment of the present invention;
Fig. 4 is the effect picture of rough registration in the embodiment of the present invention;
Fig. 5 is the effect picture of essence registration in the embodiment of the present invention.
Specific embodiment
Shown in Figure 1, the invention proposes a kind of big data quantities, the product point cloud Fast global optimization registration of few feature
Method includes the following steps:
A1, realize that the overlapping region of point cloud is extracted based on super voxel clustering algorithm, i.e., to CAD model data reduction and measurement
The region that point cloud is substantially overlapped.Cloud piecemeal will be put by being primarily based on super voxel clustering algorithm, and extract the mass center of each piecemeal point cloud
Point;Find out the principal curvatures of CAD model point cloud center of mass point and measurement pointcloud center of mass point, the normal vector for inquiring center of mass point and inquiry mass center
Point arrives the angle of its field point center of gravity line, and extracts criterion using the two elements as overlapping region similitude, obtains
To the corresponding center of mass point in overlapping region, is indexed using center of mass point and come out overlapping region data reduction;By CAD model overlapping region
Mass center point set and measurement pointcloud mass center point set are saved, the input point cloud as cloud rough registration;
A2, rough registration is carried out using mass center point set.CALCULATION CAD model overlapping region mass center point set and measurement pointcloud center of mass point
The quick point feature (FPFH) of collection describes son, as the point feature of cloud;Match point is established based on FPFH Feature Descriptor
It is right, error matching points pair are removed using stochastical sampling consistency (RANSAC) algorithm;Calculate the covariance matrix of corresponding points pair simultaneously
It carries out singular value decomposition (SVD), obtains initial conversion matrix, complete the rough registration of point cloud;
A3, the key point that filtered CAD model point cloud and measurement pointcloud are extracted using ISS key point extraction algorithm are right
The key point extracted carries out the essence registration of a cloud using branch and bound method (BNB) nested ICP algorithm, and global optimization can be obtained
Transition matrix.
In the present embodiment, before step A1 further include: statistical filtering is carried out to CAD model point cloud and measurement pointcloud, it can be with
Abnormal point and noise spot are removed, influence of the noise spot to point cloud registering is eliminated.
In the present embodiment, the step of overlapping region for putting cloud is extracted is realized in the A1 based on super voxel clustering algorithm
Are as follows:
A1-1, measurement pointcloud and CAD model point cloud are subjected to piecemeal first with super voxel clustering algorithm.Seed point is equal
It is even to be distributed in a cloud space (clouddiagonal) in, then to the distance, that is, Voxel resolution ratio (Rvoxel) and seed of voxel
The resolution ratio (Rseed) of point is defined, and the minimum bounding box of a cloud is acquired using AABB bounding box, recycles minimum bounding box
Diagonal line length the factor of the half as voxel resolution and seed point resolution, realize that adaptive resolution extracts:
Rvoxel=0.01 × clouddiagonal+0.8 (1)
Rseed=0.037 × clouddiagonal+3.0 (2)
Formula (1) is the value formula of voxel resolution, and formula (2) is the value formula of seed point resolution.
A1-2, its center of mass point is extracted to each point cloud mass after super voxel clustering block, obtains each point cloud mass of CAD model
The mass center point set Q { n1, n2...nn } of mass center point set P { m1, m2...mn } and measurement pointcloud;
A1-3, judgement screening is carried out to the center of mass point extracted in A1-2, realizes the extraction of overlapping region.Judging criterion is
The normal vector of each center of mass point and inquiry center of mass point in the principal curvatures of each center of mass point and mass center point set P, Q in mass center point set P, Q
Angle between the center of gravity line in the field k;
A1-4, the principal curvatures of each center of mass point in mass center point set P, Q in A1-3 is calculated, is estimated using discrete curvature
Algorithm calculates principal curvatures, including minimum curvature and maximum curvature, using the method estimate vector based on local least square method.
A1-5, normal vector and the weight of inquiry center of mass point to the field k to each center of mass point in mass center point set P, Q in A1-3
Angle between heart line is solved.The focus point centerP of K neighborhood is calculated firstσ(i), centerσ(j), acquire query point
Line of the center of mass point to its center of gravity | | centerPσ(i)-Pi| |, | | centerQσ(j)-Qj||.Enable nP(i)nQ(j)It is respectively looked into for point set P, Q
Ask the normal vector of point.Then the angle between them is α=accos < ηP(i), | | centerPσ(i)-Pσ(i)| | >, β=accos <
nQ(j), | | centerQσ(j)-PQσ(j)||>。
The maximum principal curvatures of each center of mass point is denoted as P respectively in A1-6, mass center point set P, Qik1And Qjk1, minimum principal curvatures
It is denoted as P respectivelyik2And Qjk2;The center of gravity in the normal vector of each center of mass point and inquiry center of mass point to the field k connects in mass center point set P, Q
Angle between line is denoted as α and β;Criterion inequality are as follows: | | Pik1-Qjk1| | < ε 1, | | Pik2-Qjk2| | < ε 2, | | alpha-beta |
| < ε 3, wherein ε 1, ε 2,3 value of ε are respectively 0.03,0.03,0.9.
Using center of mass point, not all point converges carry out rough registration in the A2.
The key point extracted in the A3 using ISS is as matching object, branch and bound method (BNB) nesting ICP algorithm
Carry out the essence registration of a cloud.
It will be as follows that input is registrated with measurement pointcloud with specific CAD model point cloud, specifically include:
Step 1: input CAD model point cloud and measurement pointcloud first.Shown in Figure 2, CAD model point cloud includes about
298599 points, measurement pointcloud include about 145813 points;Statistics filter is carried out to CAD model point cloud and measurement pointcloud respectively again
Wave.
Step 2: the overlapping region of point cloud, which is extracted, to be realized based on super voxel clustering algorithm to filtered cloud.Base first
Cloud piecemeal will be put in super voxel clustering algorithm, the center of mass point of each piecemeal point cloud is then extracted, then by acquiring CAD model point
The normal vector and inquiry center of mass point to its field point of the principal curvatures and inquiry center of mass point of cloud center of mass point and measurement pointcloud center of mass point
The angle of center of gravity line.Criterion is extracted using the two elements as overlapping region similitude, it is corresponding to obtain overlapping region
Center of mass point, then using center of mass point index overlapping region data reduction is come out, effect is as shown in Figure 3.Finally by CAD model
Overlapping region mass center point set and measurement pointcloud mass center point set are saved, and the input point cloud as cloud rough registration;
The principal curvatures of the point cloud center of mass point and the normal vector and inquiry center of mass point to its field point for inquiring center of mass point
The calculation method of the two factors of the angle of center of gravity line are as follows:
(1) using the most common analytic method in discrete curvature estimation algorithm, principal curvatures includes most for the extraction of principal curvatures
Small curvature and maximum curvature, they represent the extreme value of normal curvature.It is required that principal curve value, must first acquire the normal direction of center of mass point
Amount.Point cloud data is all along with noise under normal conditions, using the normal estimation based on local least square method.
(2) line of the normal vector of each center of mass point and the center of gravity of inquiry center of mass point to the field k in mass center point set P, Q is calculated
Between angle.The focus point centerP of K neighborhood is calculated firstσ(i), centerQσ(j), query point center of mass point is acquired to its center of gravity
Line | | centerPσ(i)-Pi| |, | | centerQσ(j)-Qj||.Enable nP(i)nQ(j)For the normal vector of each query point of point set P, Q.
Then the angle between them is α=accos < nP(i), | | centerPσ(i)-Pσ(i)| | >, β=accos < nQ(j), | |
centerQσ(j)-PQσ(j)||>。
(3) principal curvatures of the point cloud center of mass point acquired by above (1), (2) step and the normal vector for inquiring center of mass point
The differentiation inequality for extracting similar overlapping region is established to the angle of the line of inquiry center of mass point to its field point center of gravity:
||Pik1-Qjk1| | < ε 1
||Pik2-Qjk2| | < ε 2
| | alpha-beta | | < ε 3
Wherein, Pik1And Qjk1For the maximum principal curvatures of two o'clock cloud, Pik2And Qjk2For the minimum principal curvatures of two o'clock cloud.Wherein ε
1, ε 2,3 value of ε are respectively 0.03,0.03,0.9.
Step 3: rough registration is carried out using mass center point set.CALCULATION CAD model overlapping region mass center point set and measurement pointcloud matter
The quick point feature (FPFH) of heart point set describes son, describes son as the point feature of point cloud using FPFH.
Step 4: matching double points are established based on FPFH Feature Descriptor, then reapply stochastical sampling consistency
(RANSAC) algorithm removes error matching points pair, finally obtains accurate matching double points;
Step 5: calculating the covariance matrix of corresponding points pair and carries out singular value decomposition (SVD) to obtain initial conversion
Matrix.So far, the rough registration of a cloud is completed.Effect is as shown in Figure 4;
Step 6: the key of filtered CAD model point cloud and measurement pointcloud is extracted using ISS key point extraction algorithm
Point is registrated the key point extracted using the essence that branch and bound method (BNB) nested ICP algorithm carries out a cloud, available complete
The transition matrix of office's optimization.
The position of measurement pointcloud and CAD model point cloud after conversion is analyzed, available mean square deviation is
0.652mm.The entire registration time is 53.258s.
It is 3.457mm according to the mean square deviation that the method for document obtains.Overall time is 80.233s.
By the experimental data of above-described embodiment 1 it is found that big data quantity proposed by the present invention lacks the point cloud quick global of feature
Optimize method for registering from data volume is substantially reduced, to improve the speed of registration.
The above is only a preferable embodiments in present example.But the present invention is not limited to above-mentioned embodiment party
Case, it is all by the present invention any equivalent change and modification done, generated function without departing from this programme range when,
It belongs to the scope of protection of the present invention.
Claims (6)
1. the product point cloud data Fast global optimization method for registering of a kind of big data quantity, few feature characterized by comprising
The extraction of overlapping region is carried out to filtered CAD model point cloud based on super voxel clustering algorithm;To each cloud mass
Center of mass point extracts corresponding FPFH Feature Descriptor, carries out rough registration using these cloud Feature Descriptors;
The key point and combination global optimization approach obtained based on ISS algorithm realizes essence registration.
2. the product point cloud data Fast global optimization method for registering of big data quantity according to claim 1, few feature,
It is characterized in that, carries out the extraction of overlapping region to filtered CAD model point cloud based on super voxel clustering algorithm;To each cloud
The center of mass point of block extracts corresponding FPFH Feature Descriptor, carries out rough registration using these cloud Feature Descriptors, specific to wrap
It includes:
A1, realize that the overlapping region of point cloud is extracted based on super voxel clustering algorithm, i.e., to CAD model data reduction and measurement pointcloud
The region being substantially overlapped;It include: that cloud piecemeal will be put based on super voxel clustering algorithm, and extract the center of mass point of each piecemeal point cloud;
Find out the principal curvatures of CAD model point cloud center of mass point and measurement pointcloud center of mass point, the normal vector for inquiring center of mass point and inquiry center of mass point
Criterion is extracted to the angle of its field point center of gravity line, and using the two elements as overlapping region similitude, is obtained
The corresponding center of mass point in overlapping region is indexed using center of mass point and comes out overlapping region data reduction;By CAD model overlapping region matter
Heart point set and measurement pointcloud mass center point set are saved, the input point cloud as cloud rough registration;
A2, rough registration is carried out using mass center point set;It include: CALCULATION CAD model overlapping region mass center point set and measurement pointcloud mass center
The quick point feature FPFH of point set describes son, as the point feature of cloud;Match point is established based on FPFH Feature Descriptor
It is right, error matching points pair are removed using stochastical sampling consistency RANSAC algorithm;The covariance matrix for calculating corresponding points pair is gone forward side by side
Row singular value decomposition SVD obtains initial conversion matrix, completes the rough registration of point cloud.
3. the product point cloud data Fast global optimization method for registering of big data quantity according to claim 2, few feature,
It is characterized in that, it is described to realize that the overlapping region of point cloud is extracted based on super voxel clustering algorithm, it specifically includes:
A1-1, measurement pointcloud and CAD model point cloud are subjected to piecemeal first with super voxel clustering algorithm, seed point is uniformly divided
Cloth is in a cloud space clouddiagonalIn, then to distance, that is, Voxel resolution ratio Rvoxel of voxel and the resolution of seed point
Rate Rseed is defined, and the minimum bounding box of a cloud is acquired using AABB bounding box, recycles the diagonal line length of minimum bounding box
The factor of the half as voxel resolution and seed point resolution, realize that adaptive resolution extracts:
Rvoxel=0.01 × clouddiagonal+0.8 (1)
Rseed=0.037 × clouddiagonal+3.0 (2)
Formula (1) is the value formula of voxel resolution, and formula (2) is the value formula of seed point resolution;
A1-2, its center of mass point is extracted to each point cloud mass after super voxel clustering block, obtains the mass center of each point cloud mass of CAD model
The mass center point set Q { n1, n2 ... nn } of point set P { m1, m2 ... mn } and measurement pointcloud;
A1-3, judgement screening is carried out to the center of mass point extracted in A1-2, realizes the extraction of overlapping region;Judge criterion is mass center
In point set P, Q in the principal curvatures of each center of mass point and mass center point set P, Q the normal vector of each center of mass point and inquiry center of mass point to k neck
Angle between the center of gravity line in domain;
A1-4, the principal curvatures of each center of mass point in mass center point set P, Q in A1-3 is calculated, using discrete curvature estimation algorithm
Principal curvatures, including minimum curvature and maximum curvature are calculated, using the method estimate vector based on local least square method;
A1-5, the normal vector of each center of mass point in mass center point set P, Q in A1-3 and the center of gravity in inquiry center of mass point to the field k are connected
Angle between line is solved;The focus point centerP of K neighborhood is calculated firstσ(i)And centerQσ(j), acquire query point matter
Line of the heart point to its center of gravity | | centerPσ(i)-Pi| | and | | centerQσ(j)-Qj||;Enable nP(i)And nQ(j)Respectively point set P,
The normal vector of each query point of Q, then the angle between them is respectively α=accos < nP(i),||centerPσ(i)-Pσ(i)| | >, β=
accos<nQ(j),||centerQσ(j)-PQσ(j)||>;
The maximum principal curvatures of each center of mass point is denoted as P respectively in A1-6, mass center point set P, Qik1And Qjk1, minimum principal curvatures remembers respectively
For Pik2And Qjk2;The normal vector of each center of mass point and inquiry center of mass point are between the center of gravity line in the field k in mass center point set P, Q
Angle be denoted as α and β;Criterion inequality are as follows: | | Pik1-Qjk1||<ε1、||Pik2-Qjk2| | < ε 2, | | alpha-beta | | < ε 3,
Middle ε 1, ε 2,3 value of ε are respectively 0.03,0.03,0.9.
4. the product point cloud data Fast global optimization method for registering of big data quantity according to claim 2, few feature,
It is characterized in that, using center of mass point, not all point converges carry out rough registration in step A2.
5. the product point cloud data Fast global optimization method for registering of big data quantity according to claim 1, few feature,
It being characterized in that, the key point obtained based on ISS algorithm simultaneously combines global optimization approach to realize essence registration, it specifically includes:
A3, the key point that filtered CAD model point cloud and measurement pointcloud are extracted using ISS key point extraction algorithm, to extraction
To key point using branch and bound method BNB nesting ICP algorithm carry out a cloud essence registration, the conversion of global optimization can be obtained
Matrix.
6. the product point cloud data Fast global optimization method for registering of big data quantity according to claim 5, few feature,
It is characterized in that, as matching object, branch and bound method BNB nesting ICP algorithm carries out the key point extracted in step A3 using ISS
The essence registration of point cloud.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811242567.4A CN109493372B (en) | 2018-10-24 | 2018-10-24 | Rapid global optimization registration method for product point cloud data with large data volume and few characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811242567.4A CN109493372B (en) | 2018-10-24 | 2018-10-24 | Rapid global optimization registration method for product point cloud data with large data volume and few characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109493372A true CN109493372A (en) | 2019-03-19 |
CN109493372B CN109493372B (en) | 2021-10-01 |
Family
ID=65692706
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811242567.4A Active CN109493372B (en) | 2018-10-24 | 2018-10-24 | Rapid global optimization registration method for product point cloud data with large data volume and few characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109493372B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110276790A (en) * | 2019-06-28 | 2019-09-24 | 易思维(杭州)科技有限公司 | Point cloud registration method based on shape constraining |
CN110335297A (en) * | 2019-06-21 | 2019-10-15 | 华中科技大学 | A kind of point cloud registration method based on feature extraction |
CN111080653A (en) * | 2019-11-06 | 2020-04-28 | 广西大学 | Method for simplifying multi-view point cloud by using region segmentation and grouping random simplification method |
CN112200843A (en) * | 2020-10-09 | 2021-01-08 | 福州大学 | CBCT and laser scanning point cloud data tooth registration method based on hyper-voxels |
CN113139991A (en) * | 2021-05-13 | 2021-07-20 | 电子科技大学 | 3D point cloud registration method based on overlapping region mask prediction |
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 |
WO2022000127A1 (en) * | 2020-06-28 | 2022-01-06 | 华为技术有限公司 | Target tracking method and device therefor |
CN114255314A (en) * | 2022-02-28 | 2022-03-29 | 深圳大学 | Automatic texture mapping method, system and terminal for shielding avoidance three-dimensional model |
CN114511673A (en) * | 2022-01-26 | 2022-05-17 | 哈尔滨工程大学 | Improved ICP-based seabed local environment preliminary construction method |
CN117829381A (en) * | 2024-03-05 | 2024-04-05 | 成都农业科技职业学院 | Agricultural greenhouse data optimization acquisition system based on Internet of things |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101645170A (en) * | 2009-09-03 | 2010-02-10 | 北京信息科技大学 | Precise registration method of multilook point cloud |
US20140098199A1 (en) * | 2010-03-10 | 2014-04-10 | Shapequest, Inc. | Systems and methods for 2D image and spatial data capture for 3D stereo imaging |
CN104143210A (en) * | 2014-07-31 | 2014-11-12 | 哈尔滨工程大学 | Multi-scale normal feature point cloud registering method |
CN105701788A (en) * | 2014-11-24 | 2016-06-22 | 复旦大学 | Global optimization-based neural navigation automatic space registration method |
CN105761206A (en) * | 2014-12-16 | 2016-07-13 | 富泰华工业(深圳)有限公司 | Point cloud splicing method and system |
CN107545217A (en) * | 2016-06-28 | 2018-01-05 | 比亚迪股份有限公司 | Finger print matching method, fingerprint matching device and fingerprint recognition chip |
-
2018
- 2018-10-24 CN CN201811242567.4A patent/CN109493372B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101645170A (en) * | 2009-09-03 | 2010-02-10 | 北京信息科技大学 | Precise registration method of multilook point cloud |
US20140098199A1 (en) * | 2010-03-10 | 2014-04-10 | Shapequest, Inc. | Systems and methods for 2D image and spatial data capture for 3D stereo imaging |
US8867827B2 (en) * | 2010-03-10 | 2014-10-21 | Shapequest, Inc. | Systems and methods for 2D image and spatial data capture for 3D stereo imaging |
CN104143210A (en) * | 2014-07-31 | 2014-11-12 | 哈尔滨工程大学 | Multi-scale normal feature point cloud registering method |
CN105701788A (en) * | 2014-11-24 | 2016-06-22 | 复旦大学 | Global optimization-based neural navigation automatic space registration method |
CN105761206A (en) * | 2014-12-16 | 2016-07-13 | 富泰华工业(深圳)有限公司 | Point cloud splicing method and system |
CN107545217A (en) * | 2016-06-28 | 2018-01-05 | 比亚迪股份有限公司 | Finger print matching method, fingerprint matching device and fingerprint recognition chip |
Non-Patent Citations (2)
Title |
---|
Y. XU 等: "AUTOMATED COARSE REGISTRATION OF POINT CLOUDS IN 3D URBAN SCENES USING VOXEL BASED PLANE CONSTRAINT", 《ISPRS ANNALS OF THE PHOTOGRAMMETRY, REMOTE SENSING AND SPATIAL INFORMATION SCIENCES》 * |
沈萦华 等: "基于法向特征直方图的点云配准算法", 《光学精密工程》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110335297B (en) * | 2019-06-21 | 2021-10-08 | 华中科技大学 | Point cloud registration method based on feature extraction |
CN110335297A (en) * | 2019-06-21 | 2019-10-15 | 华中科技大学 | A kind of point cloud registration method based on feature extraction |
CN110276790A (en) * | 2019-06-28 | 2019-09-24 | 易思维(杭州)科技有限公司 | Point cloud registration method based on shape constraining |
CN111080653A (en) * | 2019-11-06 | 2020-04-28 | 广西大学 | Method for simplifying multi-view point cloud by using region segmentation and grouping random simplification method |
CN111080653B (en) * | 2019-11-06 | 2022-09-20 | 广西大学 | Method for simplifying multi-view point cloud by using region segmentation and grouping random simplification method |
WO2022000127A1 (en) * | 2020-06-28 | 2022-01-06 | 华为技术有限公司 | Target tracking method and device therefor |
CN112200843A (en) * | 2020-10-09 | 2021-01-08 | 福州大学 | CBCT and laser scanning point cloud data tooth registration method based on hyper-voxels |
CN113139991A (en) * | 2021-05-13 | 2021-07-20 | 电子科技大学 | 3D point cloud registration method based on overlapping region mask prediction |
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 |
CN114511673A (en) * | 2022-01-26 | 2022-05-17 | 哈尔滨工程大学 | Improved ICP-based seabed local environment preliminary construction method |
CN114511673B (en) * | 2022-01-26 | 2022-12-09 | 哈尔滨工程大学 | Improved ICP-based seabed local environment preliminary construction method |
CN114255314A (en) * | 2022-02-28 | 2022-03-29 | 深圳大学 | Automatic texture mapping method, system and terminal for shielding avoidance three-dimensional model |
CN117829381A (en) * | 2024-03-05 | 2024-04-05 | 成都农业科技职业学院 | Agricultural greenhouse data optimization acquisition system based on Internet of things |
CN117829381B (en) * | 2024-03-05 | 2024-05-14 | 成都农业科技职业学院 | Agricultural greenhouse data optimization acquisition system based on Internet of things |
Also Published As
Publication number | Publication date |
---|---|
CN109493372B (en) | 2021-10-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109493372A (en) | The product point cloud data Fast global optimization method for registering of big data quantity, few feature | |
CN107590825B (en) | Point cloud hole repairing method based on SFM | |
CN107886529B (en) | Point cloud registration method for three-dimensional reconstruction | |
CN109118574A (en) | A kind of fast reverse modeling method extracted based on three-dimensional feature | |
CN111724433B (en) | Crop phenotype parameter extraction method and system based on multi-view vision | |
CN109934855A (en) | A kind of livewire work scene power components three-dimensional rebuilding method based on cloud | |
CN109903319B (en) | Multi-resolution-based fast iteration closest point registration algorithm | |
CN102411779B (en) | Object model matching posture measuring method based on image | |
CN110363800A (en) | A kind of accurate rigid registration method blended based on point set data and characteristic information | |
CN114066953A (en) | Three-dimensional multi-modal image deformable registration method for rigid target | |
CN114677418A (en) | Registration method based on point cloud feature point extraction | |
CN108663026A (en) | A kind of vibration measurement method | |
CN108154176B (en) | 3D human body posture estimation algorithm aiming at single depth image | |
CN105069777A (en) | Automatic extracting method of neck-edge line of preparation body grid model | |
CN107909611A (en) | A kind of method using differential geometric theory extraction space curve curvature feature | |
CN108520550B (en) | Point cloud data Poisson surface reconstruction method based on noise classification and MLS | |
CN117132630A (en) | Point cloud registration method based on second-order spatial compatibility measurement | |
CN115578320A (en) | Full-automatic space registration method and system for orthopedic surgery robot | |
CN117541614B (en) | Space non-cooperative target close-range relative pose tracking method based on improved ICP algorithm | |
CN109064473B (en) | 2.5D ultrasonic panoramic image segmentation method | |
CN116763295B (en) | Livestock scale measuring method, electronic equipment and storage medium | |
CN117011299B (en) | Reference point cloud quality assessment method and system integrating graph resampling and gradient characteristics | |
CN115147433A (en) | Point cloud registration method | |
CN112419330A (en) | Temporal bone key anatomical structure automatic positioning method based on spatial relative position prior | |
CN111402221A (en) | Image processing method and device and electronic equipment |
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 |