CN106023298B - Point cloud Rigid Registration method based on local Poisson curve reestablishing - Google Patents
Point cloud Rigid Registration method based on local Poisson curve reestablishing Download PDFInfo
- Publication number
- CN106023298B CN106023298B CN201610456738.8A CN201610456738A CN106023298B CN 106023298 B CN106023298 B CN 106023298B CN 201610456738 A CN201610456738 A CN 201610456738A CN 106023298 B CN106023298 B CN 106023298B
- Authority
- CN
- China
- Prior art keywords
- point
- curved surface
- poisson
- registration
- sampling point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 74
- 230000009466 transformation Effects 0.000 claims abstract description 17
- 238000013459 approach Methods 0.000 claims abstract description 12
- 230000003993 interaction Effects 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 55
- 230000000007 visual effect Effects 0.000 claims description 8
- 239000013256 coordination polymer Substances 0.000 claims description 7
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000000926 separation method Methods 0.000 claims description 4
- 230000007547 defect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 230000002452 interceptive effect Effects 0.000 claims description 2
- 238000010845 search algorithm Methods 0.000 claims 1
- 238000007667 floating Methods 0.000 abstract description 5
- 238000013461 design Methods 0.000 abstract description 2
- 239000000686 essence Substances 0.000 description 18
- 230000008569 process Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 9
- 230000000694 effects Effects 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000003203 everyday effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 241000283973 Oryctolagus cuniculus Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 239000009951 qiqi Substances 0.000 description 1
- 238000012372 quality testing Methods 0.000 description 1
- 238000007634 remodeling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/10—Geometric effects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2215/00—Indexing scheme for image rendering
- G06T2215/06—Curved planar reformation of 3D line structures
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Computer Graphics (AREA)
- Processing Or Creating Images (AREA)
Abstract
Provided herein is a kind of point cloud Rigid Registration methods that Poisson curved surface searching corresponding points are rebuild based on fractional sample, belong to digitized design platform field, it is characterised in that：Interaction selected characteristic point pair in floating point cloud and fixed point cloud is just registered in, Poisson curved surface is built based on fixed point cloud feature neighborhood of a point point set；The KD trees of curved surface are established, closest approach of the inquiry floating point cloud sample point in KD trees is as a reference point, using sample point to the closest approach of reference point annulus dough sheet as corresponding points, based on corresponding points to establishing measure function and solving transformation parameter using SVD methods；On the basis of first registration, essence registration adaptively obtains characteristic point pair based on public domain, establishes error metric using the minimum distance of point to Poisson curved surface, so as to calculate transformation parameter, further improves registration accuracy.Just registration of the invention can obtain higher registration accuracy, and essence, which is matched, will definitely quickly converge on global optimum and with higher robustness.
Description
Technical field
The present invention provides the point cloud Rigid Registration method based on local Poisson curve reestablishing, available for being sampled to surface in kind
The registration of data various visual angles point cloud data, belongs to digitized design platform field.
Background technology
In fields such as reverse-engineering, computer graphics, quality testings, need to obtain three-dimensional point cloud number from surface in kind
According to.Due to object surface block and the limitation of the measurement range of scanning device, the laser measuring equipment and light of mainstream at present
Grid projection measuring apparatus must could obtain object being measured from multiple angles, subregion scanning the point of body surface whole
Cloud data.But in angle scanning process is converted, the coordinate system where the point cloud data that scanning obtains every time is different, it is therefore desirable to
It is a complete surface three dimension point cloud data that point cloud data under different coordinates, which is converted into the same coordinate system, and is exported,
I.e. three-dimensional point cloud is registrated.The precision of three-dimensional point cloud registration determines the precision of the subsequently various processing to sampled data, particularly pair
The precision of the post processings such as point cloud segmentation, feature recognition, Curvature Estimation, normal estimation, curve reestablishing in reverse-engineering has important
It influences.
Three-dimensional point cloud registration process is generally divided into just registration and essence two stages of registration.Just the registration stage, fix one and regard
The point cloud data (fixed point cloud) at angle, other visual angle point cloud datas that float (float point cloud), by matching fixed point cloud with floating
The correspondence geometric properties of point cloud public domain calculate rigid transformation parameters, and the point cloud that most floats at last is with fixed point cloud point cloud registration and integration
One complete point cloud data；Essence is registered in iteration registration process on the basis of just registration, until error convergence, so as to further
Just registration accuracy is improved, error is made to reach minimum.
Just common corresponding geometric properties have corresponding points, corresponding line and corresponding surface in registration.Chua etc. exists《3D human
face recognition using point signature》(IEEE International Conference on
Automatic Face and Gesture Recognition,2000.Proceedings.IEEE,2000:233-238) it is
Each point defined feature describes son, right with it in the Feature Descriptor searching fixed point cloud each put in the point cloud that floats by calculating
The Feature Descriptor answered, this method are computationally intensive and to noise sensitivity.Papazov etc. exists《An efficient ransac
for 3d object recognition in noisy and occluded scenes》(Computer Vision–ACCV
2010.Springer Berlin Heidelberg,2011:135-148.) using the method for RANSAC, sought in two groups of point clouds
Look for three points pair, if between any two points in each pair of point apart from approximately equal if think these three points to being corresponding points pair,
Transformation parameter is calculated to information based on point, this method is appropriate only for the point cloud registering of small data quantity.Bucksch etc. exists
《Localized registration of point clouds of botanic trees》(Geoscience and
Remote Sensing Letters,IEEE,2013,10(3):631-635.) it is registrated using the corresponding mode of point-line,
By finding arbitrary point in the point cloud that floats rigid transformation parameters, this method pair are calculated in the closest Eigenvector of fixed point cloud
Point cloud contour shape is more demanding and can not solve the problems, such as local convergence.Above method is based on the matching pair of entire point cloud data
The geometric properties answered, and then transformation parameter is calculated, such method calculation amount is very big, and distinct methods are to a cloud initial data
There are flatness, uniformity, without the different requirements such as noise.In addition there is scholar using Algebraic Surfaces modelling, the principal curvatures estimation technique
Calculating is waited to correspond to geometric properties information.In addition to this kind of relatively easy, effective just method for registering --- man-machine friendship can be used
Mutual method --- corresponding points pair are chosen manually, and transformation parameter can fast and efficiently be calculated by least square method, but this method
Greatest problem is that corresponding points are not high to the precision of selection, causes first registration accuracy relatively low or even converges on local optimum.
The main stream approach of current essence registration have focused largely on to closest approach iteration (Iterative Closet Point,
ICP) the improvement of method, this method are existed by Besl etc.《Method for registration of 3-D shapes》
(Robotics-DL tentative.International Society for Optics and Photonics,1992:
It 586-606.) proposes, ICP elaborates the basic theories frame of essence registration：To having two groups of good initial relative position information
Point cloud data can be used all geometric graphic elements (point, line, surface, body etc.) and be iterated registration and seek globally optimal solution.In ICP
In, author is calculated corresponding points pair by iteration closest approach and is calculated based on Quaternion Method using geometric graphic element of the point as registration
Rigid transformation parameters, but this method can not solve the problems, such as because initial positional deviation greatly caused by local convergence, and iteration time
Number is more.Chen etc. exists《Object modeling by registration of multiple range images》
(Robotics and Automation,1991.Proceedings.,1991IEEE International Conference
on.IEEE,1991:It 2724-2729.) proposes a kind of estimation error criterion based on set point normal direction to target point tangent plane, adds
Fast convergence rate, but this method Model registration big to Curvature varying does not apply to, and it is enough that two models is required to have
Overlapping region.Rusu etc.《Semantic 3d object maps for everyday manipulation in human
living environments》(KI-Künstliche Intelligenz,2010,24(4):345-348.) propose to be based on
The Feature Descriptor of FPFH carries out Feature Points Matching, and adds in characteristic point criterion, and one is clicked through for the candidate feature of identification
Step is excluded by geological informations such as curvature, normal direction, and incongruent characteristic point is rejected, and whole process uses KD trees
Carry out acceleration searching, this method achieves extraordinary registration effect, therefore this method cloud algorithms library PCL that increased income is adopted
With, but this method does not still have the drawbacks of solving to depend on initial relative position unduly, as long as the initial phase pair of two point cloud models
Position is undesirable, may result in registration failure.Various improved methods improve registration essence by finding more accurate geometric graphic element
The problem of spending and establish more efficient estimation error rule and improve convergent speed, but bringing is to calculate to complicate, the time
Cost is high, and often there are particularity for the point cloud data of each method processing, do not have universality.
Invention content
It is an object of the invention to improve in registration process corresponding points in the matched accuracy of public domain, make just to be registrated quick
It restrains and improves the precision being just registrated, essence matches the robustness that will definitely further reduce registration error and improve essence registration, technical side
Case is realized as follows：
A kind of point cloud Rigid Registration method based on local Poisson curve reestablishing, it is characterised in that step is followed successively by：First, it sets
Surface sampling point set P and Q in kind under two different visual angles to be registered, using P as the point cloud that floats, Q leads to as fixed point cloud
The mode for crossing man-machine interactively chooses subset C from P and Q respectivelyPWith CQ；2nd, F is setpFor empty set, in CpInterior interactive selected part is special
Point is levied, and selected characteristic point is sequentially added into Fp；3rd, F is setQFor empty set, for FpInterior each sampling point, in CQInterior interaction is chosen
Matched sampling point, and institute's sampling point is sequentially added into FQ；4th, based on FQNeighborhood point set of the interior sampling point in Q is FQ
Interior each sampling point structure Poisson curved surface, the specific steps are：(1) to the addition auxiliary magnet structure closing of each sample neighborhood of a point point set
Point set；(2) normal estimation is carried out to closing point set；(3) complete Poisson curved surface weight is carried out based on the closing point set after normal estimation
It builds；(4) by neighborhood point set corresponding to local Poisson curved surface separated from complete Poisson curved surface, with local Poisson curved surface make
Poisson curved surface for sampling point；5th, with FPThe distance of interior sampling point to corresponding Poisson curved surface, which is used as, to be estimated, and is solved and is caused
Point set is registrated the rigid transformation matrix of function minimization, so as to complete the preliminary registration of P and Q, i.e., so that CPWith CQIn same coordinate
It is fully overlapped under system；6th, by FpEmpty set is reset to, in CPFor interior selected part sampling point as new characteristic point, selection rule is to ensure
Institute's sampling point is in CPWith complete neighborhood point set, i.e., the sampling point is in CgInterior neighborhood point set can be uniformly distributed in the sampling point
Institute's sampling point is added in F by surrounding adjacent regionsp；7th, by FQEmpty set is reset to, for FpInterior each sampling point, CQIt is interior selection with
Nearest sampling point, and institute's sampling point is sequentially added into FQ；8th, method described in applying step four is FpInterior each sampling point structure
Build Poisson curved surface；9th, with FpThe distance of interior sampling point to corresponding Poisson curved surface, which is used as, to be estimated, and is solved so that point set is matched
The rigid transformation matrix that quasi-function minimizes so as to complete the accuracy registration of P and Q, as a result exports the P after rigid transformation with being registrated
Error.
To realize goal of the invention, the point cloud Rigid Registration method based on local Poisson curve reestablishing, feature exists
In：In step 4, point set is closed to sample neighborhood of a point point set addition auxiliary magnet construction, the specific steps are：(1) sampling point is set as qi
∈FQ, qiNeighborhood point set be N (qi), by N (qi) project to plane, it is denoted as M (qi)；(2) known by boundary characteristic recognition methods
Do not go out M (qi) boundary point set BM；(3) according to the correspondence of projection, N (q are obtainedi) boundary point set BN；(4) it is based on
BM, BNDiscrete point is inserted between corresponding boundary point, you can structure closing point set.
To realize goal of the invention, the point cloud Rigid Registration method based on local Poisson curve reestablishing, feature exists
In：In step 4, the Poisson curved surface corresponding to neighborhood point set is detached by building dynamically spatial-data index, specific steps
For：(1) the complete Poisson curved surface that is closed constructed by setting establishes the spatial index-KD trees based on Poisson curved surface Λ and one side of something as Λ
The index structure that structure is combined；(2) arest neighbors of sampling point in the index in neighborhood point set is inquired；(3) net after separation is set
Lattice curved surface is Λ ', and an annulus of arest neighbors and its dough sheet information are stored in Λ ' by the topology information based on Half-edge Structure；(4) it is defeated
Go out local Poisson curved surface Λ ', ensure the topological integrity of the original data point face information of Λ ' holdings.
To realize goal of the invention, the point cloud Rigid Registration method based on local Poisson curve reestablishing, feature exists
In：In step 5, with FpThe minimum distance of interior sampling point to Poisson curved surface, which is used as, to be estimated, the specific steps are：(1) based on part
Poisson curved surface Λ ' establishes grid index structure-KD trees；(2) to pi∈Fp, p is searched by the K-NN search of KD treesiTo Λ's '
Nearest grid vertex a；(3) the topological neighborhood information based on index leaf node storage obtains an annulus tri patch of a；(4) it is sharp
P is calculated with point-to-plane distance formulaiTo the closest approach b of annulus tri patch, using b as piCorresponding points；(5) p is calculatediWith b away from
From using the distance as measure value.
To realize goal of the invention, the point cloud Rigid Registration method based on local Poisson curve reestablishing, feature exists
In：In step 6, in CPInterior selected part sampling point as new characteristic point, selection rule the specific steps are：(1) point is calculated
Collect the span ζ of P；(2) based on pi∈CpNeighborhood point set N (pi), calculate N (pi) span(2) τ is set as selection parameter, such as
FruitThen N (pi) there are hole defects, reject pi, conversely, by piAs characteristic point.
To realize goal of the invention, the point cloud Rigid Registration method based on local Poisson curve reestablishing, feature exists
In：In the step (1) of step 6 selected characteristic point rule, the span ζ of point set P is calculated, the specific steps are：A) from surface in kind
M sampling point is obtained in sampling point set P at random, forms point set Pm；B) D is set as empty set, for PmIn each sampling point, calculate it to P
Its k nearest sampling point of middle-range apart from mean value and be added into set D；C) using the mean value of all elements in D as span
ζ。
To realize goal of the invention, the point cloud Rigid Registration method based on local Poisson curve reestablishing, feature exists
In in step 9, using the minimum range progress estimation error based on one Poisson curved surface of sampling point, with pi∈FpTo local Poisson
The minimum range of curved surface Λ ' as error ξ, the specific steps are：(1) i ← 1, wherein ξ ← 0, i=1,2 ..., n；(2) p is calculatedi
To the closest approach q of Λ 'i；(3) p is calculatedi, qiEuclidean distance di；(4)ξ←ξ+di；(5) it is straight to repeat step (2)-(4) by i ← i+1
To i ＞ n；(6)ξ←ξ/n；(8) return error ξ.
Compared with prior art, the present invention it has the following advantages：
(1) it carries out boundary characteristic identification based on fractional sample data and builds closing point set, improve Poisson reconstruction process
The middle robustness for minimizing scale indicator function；
(2) just registration by manual selected characteristic point, using the searching strategy of the corresponding points of point to Poisson curved surface, improves
First registration accuracy；
(3) essence registration can solve the problems, such as that registration is caused to fail due to first registration position deviation is big, improve registration
Robustness avoids registration from being absorbed in local optimum in an iterative process；
(4) the radial error estimation criterion based on point-Poisson curved surface, can significantly improve the precision of registration, reduce and match
The number of quasi- iteration convergence.
Description of the drawings
Fig. 1 is the flow chart of the point cloud Rigid Registration method the present invention is based on local Poisson curve reestablishing；
Fig. 2 is the schematic diagram of fractional sample span estimation；
Fig. 3 is the organigram for closing point set；
Fig. 4 is Poisson curve reestablishing and its schematic diagram of separation；
Fig. 5 is based on point-matched schematic diagram of Poisson curved surface closest approach；
Fig. 6 is registration error analysis schematic diagram；
Fig. 7 is the mathematical model figure in embodiment；
Fig. 8 is to carry out just registration effect comparison diagram using the present invention and Geomagic Studio；
Fig. 9 is to carry out just registration accuracy comparison diagram using the present invention and Geomagic Studio；
Figure 10 is influence schematic diagram of the feature point number to smart registration accuracy；
Figure 11 is using the present invention and ICP methods (Besl P J, McKay N D.Method for registration
of 3-D shapes[C].Robotics-DL tentative.International Society for Optics and
Photonics,1992:586-606.), ICP methods (Rusu R B.Semantic 3d object maps for are improved
everyday manipulation in human living environments[J].KI-Künstliche
Intelligenz,2010,24(4):345-348.) carry out essence registration comparison diagram；
Figure 12 is three visual angle registration effect figures of Hood models；
Figure 13 is six visual angle registration effect figures of Bunny models.
Specific embodiment
Below in conjunction with the accompanying drawings and example the invention will be further described.
Fig. 1 is the present invention is based on the flow chart of the Rigid Registration method of local Poisson curve reestablishing, in the point Yun Hegu that floats
The public domain for pinpointing cloud selects initial corresponding points pair, and corresponding points, Jin Erxiu are recalculated according to point-Poisson SURFACES MATCHING rule
Just initial corresponding points pair calculate rigid transformation parameters and registration error based on revised point to information.In initial corresponding points pair
Selection course in, just registration by interaction choose, essence registration adaptively chosen by public domain.Initial corresponding points are to choosing
Later, using the neighborhood point of point pair as the fractional sample data of registration, boundary characteristic is extracted to local sample data and builds pool
Loose curved surface calculates sample point to the closest approach of Poisson curved surface, corrects initial corresponding points pair.Based on revised corresponding points to letter
Breath, establishes the measure function of least square and estimates rigid transformation parameters, and the point cloud that floats is made rigid transformation and calculates registration to miss
Difference, iterated transform process is until convergence.
Fractional sample is randomly selected to a cloud, calculates the span of each fractional sample data, and then estimates entire point cloud
Average span.In fractional sample data shown in Fig. 2, point cloud span is d1,d2,……,d7Etc. the arithmetic average of each distance
Value.
The boundary point projected by extracting fractional sample after being projected to fractional sample respectively with it, corresponding boundary point it
Between be inserted into discrete point, and then build closing point set, as shown in figure 3, closing point set structure the specific steps are：(1) i ← 1, if from
Scatterplot set H；(2) to fractional sample N (qi) project to plane point set M (qi), projector distance is 5 times of ζ；(3) based on boundary spy
Sign recognizer identifies M (qi) boundary set BMAnd N (qi) boundary point set BM；(4) based on BM, BNPoint is calculated to correspond to
The distance between boundary point Γi；(5) the number s ← Γ for being inserted into discrete point is calculated according to the average span of cloudi/ζ；(6) basis
The vector property of vector, calculates the coordinate of discrete point and discrete point is stored in H；(7) i ← i+1 repeats step (5)-(6) until i ＞
s；(8) B is traversedM, BNIn all corresponding boundary points, repeat step (4)-(7) and obtain discrete point set H.
The purpose of structure closing point set is to improve the quality of Poisson curve reestablishing, from Fig. 4 Poissons curve reestablishing and its separation
Schematic diagram in can be seen that rebuild Poisson curved surface uniformly, it is fine and close, there is not the defects of cavity, deformity, this hair can be met
Bright registration needs, and retains the topological structure integrality of mask data, can find corresponding points for registration and provide more accurate ginseng
Examination point.
Local Poisson curved surface based on structure can establish the matching rule of one Poisson curved surface of sampling point, with sample point piTo part
The closest approach b of Poisson curved surface Λ ' is corresponding points, as shown in Figure 5.
It is relative error result and actual registration as a result, as shown in fig. 6, figure after the partial enlargement registration result of figure end
Middle light color solid dot represents the point cloud characteristic point p that floatsiNeighborhood point set N (pi), dark solid dot represents fixed point cloud characteristic point
qiNeighborhood point set N (qi), curve represents the curved surface Λ ' rebuild.As can be seen from the figure N (qi) and Poisson curved surface Λ ' is not
It is completely superposed, reason is that Poisson curve reestablishing is to N (qi) most preferably approaching for scaling function is established, and Poisson's equation is
The differential form of elliptical equation, so the deformation of elliposoidal can occur for curved surface Λ '.Curved surface deformation introduces Poisson reconstruction error,
Estimation error need to consider the position of Λ ' to N (p at this timei), N (qi) influence.If dp, dqN (p are represented respectivelyi), N (qi) in institute
The mean value of Λ ' minimum ranges is a little arrived, in the result of actual registration, there are two kinds of situations for the position of Λ '：One kind is Λ ' positions
In N (pi), N (qi) between (Fig. 6-a), the error calculated at this time is then
ξ=dp+dq (1)
Another situation is exactly N (pi), N (qi) positioned at the same side of Λ ' (shown in Fig. 6-b), the error calculated at this time is practical
For
ξ=| dp-dq| (2)
Therefore it needs effectively to judge the position of Λ ' when calculating error, normal direction auxiliary law can be used to determine：
To pi∈N(pi), qi∈N(qi), if pi, qiIt is a pair of of corresponding points pair, m piThe nearest grid vertex in Λ ',
Respectively pi, qiThe normal direction at place then judges that factor Δ is：
Δ=(npi·pim)×(nqiqim)
If Δ ＜ 0 belongs to the first situation, if Δ ＞ 0, belongs to the second situation.According to above-mentioned analysis, pass through meter
Calculate piMinimum range to Λ ' is inaccurate as error, and Poisson curve reestablishing can cause reconstruction error dq, therefore carry out error and estimate
It needs to consider d when meterqInfluence, by dqIt is substituted into formula (1) and (2) as compensation error, just obtains the result of estimation error
Actually sample point to ∑ ' minimum range, i.e.,
ξ=dp
Final purpose using error criterion is to judge whether registration is effective according to error amount, however error size and point cloud
Data have much relations in itself, if point cloud span is big, it is rough that Poisson rebuilds curved surface so that registration error accordingly becomes larger；If point
Cloud span is small, and point cloud is relatively intensive, and Poisson rebuilds rear curved surface relative smooth, and registration error can become smaller.To eliminate point cloud data sheet
The influence that body judges registration error, present invention definition registration efficiency factor κ are built by the factor and error ξ, point cloud span ζ
It is vertical registration whether effective Rule of judgment：If κ ＞ 1, just registration failure is judged；If κ ＞ 0.2, essence registration failure is judged,
The calculation formula of middle κ is
It is tested and is analyzed with models such as Hood models shown in Fig. 7, Rabbit, wherein Hood models pass through CPC light
Spatial digitizer acquisition is learned, point cloud span is 3；Bunny models use the point cloud data of Stanford University's acquisition, and point cloud span is
0.001。
Embodiment one：In to registration at the beginning of Hood, Bunny, intersection of the present invention in the two not common regions of width view is matched
It is accurate more round and smooth, although Geomagic Studio are registrated the process of no iteration convergence, take very short, Geomagic
There is larger deviation and occur being substantially misaligned in Studio registrations, as shown in Figure 8.The first method for registering of the present invention passes through
N(pi) matching corresponding points are carried out, during corresponding points are found, ρ can suitably take small (ρ ＜ 2), expand Poisson surface mesh
The number on vertex, increase find range, reach better registration accuracy；It, can be because of two point cloud model initial positions if p acquirements are excessive
Deviation, which crosses conference, to be caused to judge by accident.It is unknown due to being just registrated two point cloud model initial positions, it prevents from causing with quasi-divisor value is improper
Registration failure can add in ρ into Mobile state tune based on the point cloud data completed after converting for the first time during iterated transform later
Section.Although there are large errors for the characteristic point and its corresponding points of primary election, during iteration, which can be gradually reduced.
Embodiment two：In first registration process, Hood models only need 5 iteration that can restrain, for relative complex
Bunny models, registration is convergent slow, and taking around 10 times could restrain, as shown in Figure 9.Efficiency factor is registrated by calculating
The validity that judgement is just registrated, for Hood models, κ=0325, for Bunny models, κ=0268, two the first of model are matched
Accurate efficiency factor is no more than 0.5, therefore just registration has reached desired effect.It can be seen that according to κ value sizes, for relatively simple
The relative error that Hood models for list, Geomagic Studio and the method for the invention calculate is than Bunny model
Greatly, therefore the method for registering of the present invention will definitely obtain effect more better than Geomagic Studio to matching at the beginning of complex model.Due to
The characteristic point of first registration process selection is insufficient, makes its precision that can not meet the follow-up studies such as point cloud data normal estimation, reconstruction
Requirement, to further improve the precision being just registrated, by calculate just registration after two point cloud models public domain it is adaptive
Increase feature points and realize essence registration.
Embodiment three：To compare influence and then determining two model essences registration of the different characteristic point number to error precision
Best features point number carries out two models after being just registrated in Fig. 7 essence registration test under different characteristic point number.From figure
10 as can be seen that when feature point number is 20, and the increase of feature point number is very little for the precision influence of registration,
Registration process about reaches convergence at 10 times or so.But feature points further increase can cause Poisson rebuild number,
Inquiring the time of corresponding points increases, and the efficiency and precision being registrated in order to balance, smart registration features point number of the invention take 20.
Example IV：Using method for registering of the present invention, ICP methods, improve ICP methods in Fig. 8 just registration after Hood,
Bunny models carry out essence registration contrast test, and statistics essence is with not Tongfang after the iterations after quasi-convergence and each iteration
Registration error value is as shown in figure 11 acquired by method.Compared to point-matching criterior, the matching criterior of the point used to face can be with
Convergence faster, as can be seen from Figure 11, original I CP methods are convergent most slow, and the error after convergence is maximum, and the present invention is registered in
It can restrain for 10 times or so.Although smaller error can be converged to by improving ICP methods, the two differs in registration accuracy
And less, but improving ICP methods needs iteration that can just restrain for 20 times, therefore the method for the invention is improving registration accuracy
Under the premise of, it can more quickly restrain.
Embodiment five：Hood, Bunny model are completely registrated using 20 characteristic points using the method for the invention
Process, Hood models totally 3 views, using the 1st view as fixed point cloud data, other two is matched for floating point cloud data
Standard, complete registration effect is as shown in figure 12 (wherein a-c is the different views point cloud data at 3 visual angles of Hood).
Embodiment six：Since 2-5 view and the 1st view have public intersecting area, the 6th regards Bunny models
Figure and the 1st no public intersecting area, so following registration strategies is taken to obtain complete Bunny point cloud models：With the 1st
View is fixed point cloud data, and the 2nd, 3,4,5 view is floating point cloud model, is just registrated with the 1st view progress successively, essence
Registration obtains new point cloud model 1 ', then carries out just registration, essence registration with the 6th view with 1 ' and finally obtain complete point cloud number
According to effect is as shown in figure 13 (wherein a-f is the different views point cloud data at 6 visual angles of Bunny).
The above is only the preferred embodiments of the present invention, is not the limitation for making other forms to the present invention, any
Those skilled in the art are changed or are modified as to change on an equal basis equivalent possibly also with the technology contents of the disclosure above
Embodiment.But it is every without departing from technical solution of the present invention content, technical spirit according to the present invention makees above example
Any simple modification, equivalent variations and remodeling, still fall within the protection content of technical solution of the present invention.
Claims (7)
- A kind of 1. point cloud Rigid Registration method based on local Poisson curve reestablishing, it is characterised in that step is followed successively by：First, it sets and treats Surface sampling point set P and Q in kind under two different visual angles of registration, using P as the point cloud that floats, Q passes through as fixed point cloud The mode of man-machine interactively chooses subset C from P and Q respectivelyPWith CQ；2nd, F is setPFor empty set, in CPInterior interactive selected part feature Point, and selected characteristic point is sequentially added into FP；3rd, F is setQFor empty set, for FPInterior each sampling point, in CQIt is interior interaction choose with The sampling point to match, and institute's sampling point is sequentially added into FQ；4th, based on FQNeighborhood point set of the interior sampling point in Q is FQIt is interior Each sampling point structure Poisson curved surface, the specific steps are：(1) to each sample neighborhood of a point point set addition auxiliary magnet structure enclosed point Collection；(2) normal estimation is carried out to closing point set；(3) complete Poisson curve reestablishing is carried out based on the closing point set after normal estimation； (4) by neighborhood point set corresponding to local Poisson curved surface separated from complete Poisson curved surface, using local Poisson curved surface as The Poisson curved surface of sampling point；5th, with FPThe distance of interior sampling point to corresponding Poisson curved surface, which is used as, to be estimated, and solution makes invocation point The rigid transformation matrix of collection registration function minimization, so as to complete the preliminary registration of P and Q, i.e., so that CPWith FQIn the same coordinate system Under be fully overlapped；6th, by FPEmpty set is reset to, in CpFor interior selected part sampling point as new characteristic point, selection rule is to ensure institute Sampling point is in CpWith complete neighborhood point set, i.e., the sampling point is in CQInterior neighborhood point set can be uniformly distributed in the week of the sampling point Adjacent domain is enclosed, institute's sampling point is added in into Fp；7th, by FQEmpty set is reset to, for FPInterior each sampling point, CQInterior selection is therewith Nearest sampling point, and institute's sampling point is sequentially added into FQ；8th, method described in applying step four is FPInterior each sampling point structure Poisson curved surface；9th, with FPThe distance of interior sampling point to corresponding Poisson curved surface, which is used as, to be estimated, and is solved so that point set registration The rigid transformation matrix of function minimization so as to complete the accuracy registration of P and Q, as a result exports the P after rigid transformation and is missed with being registrated Difference.
- 2. the point cloud Rigid Registration method according to claim 1 based on local Poisson curve reestablishing, it is characterised in that： In step 4, point set is closed to sample neighborhood of a point point set addition auxiliary magnet construction, the specific steps are：(1) sampling point is set as qi∈FQ, qiNeighborhood point set be N (qi), by N (qi) project to plane, it is denoted as M (qi)；(2) M is identified by boundary characteristic recognizer (qi) boundary point set BM；(3) according to the correspondence of projection, N (q are obtainedi) boundary point set BN；(4) based on BM,BN Discrete point is inserted between corresponding boundary point, you can structure closing point set.
- 3. the point cloud Rigid Registration method according to claim 1 based on local Poisson curve reestablishing, it is characterised in that： In step 4, the Poisson curved surface corresponding to neighborhood point set is detached by building dynamically spatial-data index, the specific steps are： (1) the constructed complete Poisson curved surface that is closed is set as Λ, is established the spatial index based on Poisson curved surface Λ-KD trees and is tied with half of The index structure that structure is combined；(2) arest neighbors of sampling point in the index in neighborhood point set is inquired；(3) grid after separation is set Curved surface is Λ ', and an annulus of arest neighbors and its dough sheet information are stored in Λ ' by the topology information based on Half-edge Structure；(4) it exports Local Poisson curved surface Λ ' ensures the topological integrity of the original data point face information of Λ ' holdings.
- 4. the point cloud Rigid Registration method according to claim 1 based on local Poisson curve reestablishing, it is characterised in that： In step 5, with FPThe minimum distance of interior sampling point to Poisson curved surface, which is used as, to be estimated, the specific steps are：(1) based on local Poisson Curved surface Λ ' establishes grid index structure-KD trees；(2) to pi∈Fp, p is searched by the K-NN search algorithm of KD treesiTo Λ ' Nearest grid vertex a；(3) the topological neighborhood information based on index leaf node storage obtains an annulus tri patch of a；(4) P is calculated using point-to-plane distance formulaiTo the closest approach b of annulus tri patch, using b as piCorresponding points；(5) p is calculatediWith b's Distance, using the distance as measure value.
- 5. the point cloud Rigid Registration method according to claim 1 based on local Poisson curve reestablishing, it is characterised in that： In step 6, in CpInterior selected part sampling point as new characteristic point, selection rule the specific steps are：(1) calculate point set P's Span ζ；(2) based on pi∈CpNeighborhood point set N (pi), calculate N (pi) span ζpd；(2) τ is set as selection parameter, if ζpd＜ τ ζ, then N (pi) there are hole defects, reject pi, conversely, by piAs characteristic point.
- 6. the point cloud Rigid Registration method according to claim 5 based on local Poisson curve reestablishing, it is characterised in that： The span ζ of point set P is calculated in step (1), the specific steps are：A) m sampling point is obtained at random from sampling point set P in surface in kind, Form point set Pm；B) D is set as empty set, for PmIn each sampling point, the distance for calculating it to P middle-ranges its k nearest sampling points is equal It is worth and is added into set D；C) using the mean value of all elements in D as span ζ.
- 7. the point cloud Rigid Registration method according to claim 1 based on local Poisson curve reestablishing, which is characterized in that In step 9, estimation error is carried out using the minimum range based on sampling point-Poisson curved surface, with pi∈FpTo local Poisson curved surface The minimum range of Λ ' as error ξ, the specific steps are：(1) i ← 1, wherein ξ ← 0, i=1,2 ..., n；(2) p is calculatediIt arrives The closest approach q of Λ 'i；(3) p is calculatedi, qiEuclidean distance di；(4)(5) it is straight to repeat step (2)-(4) by i ← i+1 To i ＞ n；(6)(8) return error
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610456738.8A CN106023298B (en) | 2016-06-22 | 2016-06-22 | Point cloud Rigid Registration method based on local Poisson curve reestablishing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610456738.8A CN106023298B (en) | 2016-06-22 | 2016-06-22 | Point cloud Rigid Registration method based on local Poisson curve reestablishing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106023298A CN106023298A (en) | 2016-10-12 |
| CN106023298B true CN106023298B (en) | 2018-06-19 |
Family
ID=57086364
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610456738.8A Expired - Fee Related CN106023298B (en) | 2016-06-22 | 2016-06-22 | Point cloud Rigid Registration method based on local Poisson curve reestablishing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106023298B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107610228A (en) * | 2017-07-05 | 2018-01-19 | 山东理工大学 | Curved surface increment topology rebuilding method based on massive point cloud |
| CN108151712B (en) * | 2017-11-13 | 2020-08-07 | 合肥阿巴赛信息科技有限公司 | Human body three-dimensional modeling and measuring method and system |
| CN108230377B (en) * | 2017-12-19 | 2020-08-21 | 武汉国安智能装备有限公司 | Point cloud data fitting method and system |
| WO2019161558A1 (en) * | 2018-02-26 | 2019-08-29 | Intel Corporation | Method and system of point cloud registration for image processing |
| CN108520550B (en) * | 2018-04-10 | 2021-09-24 | 南京信息工程大学 | Point cloud data Poisson surface reconstruction method based on noise classification and MLS |
| CN109191583B (en) * | 2018-08-10 | 2022-09-23 | 广东工业大学 | Curved surface accurate alignment method based on anisotropic MLS |
| CN109685841B (en) * | 2019-01-03 | 2020-09-18 | 上海狮迈科技有限公司 | Registration method and system of three-dimensional model and point cloud |
| CN109887012B (en) * | 2019-01-09 | 2023-03-31 | 天津大学 | Point cloud registration method combined with self-adaptive search point set |
| CN113297340B (en) * | 2020-02-23 | 2023-12-19 | 北京初速度科技有限公司 | Vectorization method and device for point cloud map and method and device for converting vector map into point cloud map |
| CN111986308B (en) * | 2020-07-14 | 2023-09-01 | 武汉数字化设计与制造创新中心有限公司 | Point cloud normal and curvature change based double-constraint surface error salient point identification method |
| CN112506136B (en) * | 2020-12-10 | 2021-12-07 | 北方工业大学 | Positioning point set selection method based on statistical analysis of batch blade curved surface measurement data |
| CN113223067B (en) * | 2021-05-08 | 2021-11-19 | 广东三姆森科技股份有限公司 | Online registration method for incomplete three-dimensional scanning point cloud with plane reference |
| CN113344992B (en) * | 2021-05-31 | 2022-06-28 | 山东大学 | Global point cloud registration method, system, storage medium and equipment |
| CN113327275B (en) * | 2021-06-18 | 2022-11-08 | 哈尔滨工业大学 | Point cloud double-view-angle fine registration method based on multi-constraint point to local curved surface projection |
| CN114882087B (en) * | 2022-07-12 | 2022-10-04 | 武汉瀚迈科技有限公司 | Real-time registration method for three-dimensional scanning point cloud with incomplete basic graphic elements |
| CN117011128B (en) * | 2023-08-02 | 2024-02-13 | 广州知元科技有限责任公司 | Automatic three-dimensional model registration method and equipment based on region decomposition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8610706B2 (en) * | 2008-10-04 | 2013-12-17 | Microsoft Corporation | Parallel surface reconstruction |
| CN103489218A (en) * | 2013-09-17 | 2014-01-01 | 中国科学院深圳先进技术研究院 | Point cloud data quality automatic optimizing method and system |
| CN104299260A (en) * | 2014-09-10 | 2015-01-21 | 西南交通大学 | Contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration |
-
2016
- 2016-06-22 CN CN201610456738.8A patent/CN106023298B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8610706B2 (en) * | 2008-10-04 | 2013-12-17 | Microsoft Corporation | Parallel surface reconstruction |
| CN103489218A (en) * | 2013-09-17 | 2014-01-01 | 中国科学院深圳先进技术研究院 | Point cloud data quality automatic optimizing method and system |
| CN104299260A (en) * | 2014-09-10 | 2015-01-21 | 西南交通大学 | Contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration |
Non-Patent Citations (1)
| Title |
|---|
| 《基于邻域特征的点云配准算法研究》;贺永兴;《中国优秀硕士学位论文全文数据库 信息科技辑》;20130415;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106023298A (en) | 2016-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106023298B (en) | Point cloud Rigid Registration method based on local Poisson curve reestablishing | |
| Xu et al. | Reconstruction of scaffolds from a photogrammetric point cloud of construction sites using a novel 3D local feature descriptor | |
| CN109558879A (en) | A kind of vision SLAM method and apparatus based on dotted line feature | |
| CN110081888B (en) | Indoor positioning algorithm based on credibility-based inertial navigation and geomagnetic fusion | |
| CN108171780A (en) | A kind of method that indoor true three-dimension map is built based on laser radar | |
| CN109614935A (en) | Car damage identification method and device, storage medium and electronic equipment | |
| CN109903313A (en) | Real-time pose tracking method based on target three-dimensional model | |
| CN110222642A (en) | A kind of planar architectural component point cloud contour extraction method based on global figure cluster | |
| CN108550166B (en) | Spatial target image matching method | |
| CN111145228A (en) | Heterogeneous image registration method based on local contour point and shape feature fusion | |
| dos Santos et al. | Extraction of building roof boundaries from LiDAR data using an adaptive alpha-shape algorithm | |
| CN109523595A (en) | A kind of architectural engineering straight line corner angle spacing vision measuring method | |
| CN106780551B (en) | A kind of Three-Dimensional Moving Targets detection method and system | |
| Kroemer et al. | Point cloud completion using extrusions | |
| CN110047133A (en) | A kind of train boundary extraction method towards point cloud data | |
| CN112163588A (en) | Intelligent evolution-based heterogeneous image target detection method, storage medium and equipment | |
| CN113393524B (en) | Target pose estimation method combining deep learning and contour point cloud reconstruction | |
| CN108305277A (en) | A kind of heterologous image matching method based on straightway | |
| CN107610219A (en) | The thick densification method of Pixel-level point cloud that geometry clue perceives in a kind of three-dimensional scenic reconstruct | |
| CN108961385A (en) | A kind of SLAM patterning process and device | |
| CN109766903A (en) | A kind of point cloud model SURFACES MATCHING method based on curved surface features | |
| Wei et al. | Optimization of 3-D pose measurement method based on binocular vision | |
| CN117132737B (en) | Three-dimensional building model construction method, system and equipment | |
| Ward et al. | A model-based approach to recovering the structure of a plant from images | |
| Ma et al. | Complex buildings orientation recognition and description based on vector reconstruction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180619 Termination date: 20190622 |