CN104392426B - A kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation - Google Patents
A kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation Download PDFInfo
- Publication number
- CN104392426B CN104392426B CN201410571233.7A CN201410571233A CN104392426B CN 104392426 B CN104392426 B CN 104392426B CN 201410571233 A CN201410571233 A CN 201410571233A CN 104392426 B CN104392426 B CN 104392426B
- Authority
- CN
- China
- Prior art keywords
- point
- matrix
- geometric properties
- cloud
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to the Point-clouds Registration technology in three-dimensional measurement field, the no marks point three-dimensional point cloud method for automatically split-jointing of specially a kind of self adaptation, the present invention includes the lookup of geometric properties point, the lookup of image characteristic point, the foundation of registration Algorithm preference pattern, the geometric properties Point matching based on RANSAC excludes error hiding image characteristic point using RANSAC, rotation translation matrix RT is solved using svd algorithm, finally two panels point cloud is completed using RT matrixes.This method using object feature point instead of index point because spliced, available for the measurement occasion for being unable to binding mark point;The transformation matrix of multi-viewpoint cloud is calculated by character pair point simultaneously, the initial attitude without relying on point cloud, and the foundation of registration Algorithm preference pattern enables the system to adaptively selected suitable registration Algorithm, realizes the stable splicing of different testees.
Description
Technical field
The present invention relates to a kind of no marks of self adaptation point three-dimensional point cloud method for automatically split-jointing, it is three dimensional point cloud
A kind of method of processing, the Point-clouds Registration technology belonged in three-dimensional measurement field.
Background technology
Area-structure light three-dimensional measurement technology is (referring to document 1:Structural light three-dimensionals of the big based on digital fringe projection is surveyed in Lee
Amount technology and system research [D] [D] Wuhan:The Central China University of Science and Technology, 2009) limited by single measurement range, need never Tongfang
Position is taken multiple measurements to measured object to obtain complete geometrical model, and wherein multi-viewpoint cloud automatic Mosaic is crucial.
In order to realize multi-viewpoint cloud automatic Mosaic, conventional method has following two:(1) automatic Mosaic based on index point
Method, this method is by realizing auxiliary splicing in testee surface mount artificial target.Its splicing precision is higher, but meeting
Destroy the true three-dimension data on testee surface, while early stage patch point adds time of measuring, and can not measurement surface not
The object (such as rare cultural relics, human body) that can be marked, limits and uses scope.(2) method for automatically split-jointing of unmarked point.Common
Unmarked joining method includes iteration point closest approach algorithm (iterative closest point, ICP), based on geometry spy
The registration Algorithm levied and the registration Algorithm based on texture.Wherein, ICP algorithm is very high to a requirement of cloud initial attitude, it is impossible to right
The point cloud that initial position differs larger is spliced.Registration Algorithm based on geometric properties is only applicable to surface geometry pattern and compared
Complicated object, it is impossible to realize the point cloud registering of simple shape or symmetric objects.Registration Algorithm based on texture is only applicable to table
The object of face texture-rich, the less stable when the object single to texture is measured.
In summary:Existing three-dimensional point cloud method for automatically split-jointing all has some limitations, and can not still meet reality
The requirement of application.Therefore a kind of point cloud method effectively stablized is needed, source point cloud and target point cloud optimum can be realized
Registration.
The content of the invention
The present invention proposes a kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation, and this method can self adaptation
The no marks point three-dimensional point cloud automatic Mosaic of different objects is realized, and splicing has good stability.
A kind of no marks point three-dimensional point cloud method for automatically split-jointing for self adaptation that the present invention is provided, this method includes following steps
Suddenly:
1st step searches geometric properties point in source point cloud and target point cloud using point feature histogram method;Shot twice
Image characteristic point is searched in picture;
2nd step sets up registration Algorithm preference pattern using the geometric properties point and image characteristic point found out, calculates registration
Algorithm judges factor DrValue, automatically select suitable registration Algorithm according to the value:Work as Dr>When 0, into the 3rd step, work as Dr<When 0,
The 4th step is transferred to, works as DrWhen=0, user is pointed out to carry out point cloud by introducing index point;
The registration Algorithm preference pattern is:
Wherein, p1, p2Represent respectively geometric properties in two width sampled point clouds count out account for its sampled point points ratio;n1,
n2The number for shooting image characteristic point in picture twice is represented respectively;P is geometric properties threshold value, w1And w2It is two contrasts of description
The weight factor of difference;
The geometric properties point that 3rd step is found out for the 1st step, the matching of geometric properties point is carried out using RANSAC algorithms,
Obtain each and search the corresponding just matching point set Q ' of point, recycle SVD singular value decomposition methods to calculate spin matrix R and translation square
Battle array T, subsequently into the 5th step;
The image characteristic point that 4th step is found out for the 1st step, corresponding points are searched using RANSAC algorithms, recycle SVD strange
Different value decomposition method calculates spin matrix R and translation matrix T;
5th step carries out rotation translation using spin matrix R and translation matrix T to target point cloud, completes splicing.
The present invention sets up a registration Algorithm preference pattern using the geometric properties and texture information of body surface, according to this
Model system can be adaptive selected suitable registration Algorithm.When measuring splicing to different objects, body surface is utilized
Intrinsic geometric properties and texture information carry out the splicing of multi-viewpoint cloud data, can improve splicing stability.With existing method phase
Than, advantage of this approach is that:
1) spliced using object feature point instead of index point, available for the measurement occasion for being unable to binding mark point,
Reduce a workload for cloud post processing.
2) lookup of object feature point is realized using the geometric properties and texture information of body surface and matches to realize
Final splicing, independent of initial attitude of a cloud itself.
3) geometric properties and texture information are combined and set up registration Algorithm preference pattern and enable the system to adaptively selected conjunction
Suitable registration Algorithm, improves the stability of splicing.
Brief description of the drawings
Fig. 1 is the overall flow figure that multi-viewpoint cloud self adaptation of the present invention is spliced;
Fig. 2 is the point feature histogram calculation region of query point;
Fig. 3 is the fixation local coordinate system definition between 2 points.
Embodiment
The embodiment to the present invention is described further below in conjunction with the accompanying drawings.Herein it should be noted that for
The explanation of these embodiments is used to help understand the present invention, but does not constitute limitation of the invention.In addition, disclosed below
As long as each of the invention embodiment in involved technical characteristic do not constitute conflict each other and can just be mutually combined.
As shown in figure 1, a kind of adaptivity three-dimensional point cloud method for automatically split-jointing that the present invention is provided, this method includes following
Step:
S101 searches geometric properties point in source point cloud and target point cloud using point feature histogram method.
Point feature histogram represents for the statistic histogram of body surface point geometry feature.To a 6 DOF for cloud correspondence curved surface
It has consistency for attitude, and with robustness under the sound level of different sampling density or neighborhood.Point feature
Histogram method be based on point and its define relation and their estimation normal between neighborhood.Specific finding step is as follows:
1.1st step to shooting the two width multi-viewpoint clouds obtained, adopt in proportion by (usually always count 10~15%)
Sample, calculates the normal vector of each sampled point;Every bit in sampled point is considered as query point, it (is usually 3~4 times to define radius
Point spacing) determine inquiry neighborhood of a point, the sampled point in the neighborhood is referred to as neighborhood point.Fig. 2 illustrates query point pqAnd its
Neighborhood point pkPosition relationship, wherein query point pqFor any point in sampled point cloud.
1.2nd step in order to calculate the relative deviation between any two points s, t position relationship in neighborhood and correspondence normal,
A fixed local coordinate system is defined on one of point, as shown in figure 3, wherein:
ps, ptFor query point pqPoint s, t space coordinate, n in neighborhoods, ntFor the corresponding normal vector of point s, t, | | pt-ps||2
For 2 points of the space length, d represents the unit vector on 2 line directions.
The position relationship and Normal Error obtained using Formulas I between uvw coordinate systems as shown in Figure 3,2 points can use one
Organize angle to represent, such as formula (2):
Wherein, wntAnd untW, n are represented respectivelytAnd u, ntDot product, by α,θ span is respectively divided into b
Individual subinterval, then collectively generated b3Individual region, judges 2 points of α,The residing region of θ values, then the region count out plus
One;Calculate query point pqThe α of each pair point in neighborhood,θ values, judge its region, and statistics falls counting out in each region,
Generate point pqPoint feature histogram.
1.3rd step is utilized using the point feature histogram generation average characteristics histogram u-histogram of all sampled points
The point feature histogram of each point is compared the relative KL distances of acquisition with u-histogram by formula (3).
Wherein, i represents the sequence number of point feature histogram regions, and value is 1~b3,Represent the point feature Nogata of sampled point
I-th of component of figure, uiRepresent histogrammic i-th of the component of average characteristics.Calculate the average value a and standard deviation sigma of KL distances, system
Three-dimensional point of the KL distances outside a ± σ is counted, is considered as geometric properties point.
S102 searches image characteristic point in picture is shot twice.
Image characteristic point can be searched using SIFT algorithms, in the case of not high to stability requirement, it would however also be possible to employ
SURF algorithm searches image characteristic point.
S103 sets up registration Algorithm preference pattern.Different measured surfaces should use different characteristic points when measuring splicing
Spliced.For the simple object of geometric properties, if being spliced using the geometric properties point in step S101, stability
Difference;For the simple object of texture, if being spliced using the image characteristic point in step S102, stability is poor.In order to improve
The point cloud stability of difference measurement occasion, proposes following selection model then:
The geometric properties point and image characteristic point found out using abovementioned steps sets up registration Algorithm preference pattern, the model
Registration Algorithm is generated on the basis of the geometry and texture information of comprehensive analysis body surface and judges the factor, and on this basis certainly
The dynamic suitable registration Algorithm of selection.Specific judgment criteria such as formula (4):
Wherein, DrIt is defined as registration Algorithm and judges the factor, its overall merit body surface geometric properties and textural characteristics
Relative component degree, system can select suitable feature vertex type be matched to realize final spelling according to the algorithm factor
Connect.p1, p2Represent respectively geometric properties in two width sampled point clouds count out account for its sampled point points ratio;n1, n2Represent respectively
The number of image characteristic point in picture is shot twice;P is geometric properties threshold value, represents to carry out stable splicing using geometric properties point
Required geometric properties point ratio, it is proposed that span is 5%~10%;N is characteristics of image threshold value, and expression utilizes characteristics of image
Point carries out the characteristics of image points needed for stable splicing, it is proposed that span is 250~300.w1And w2It is two contrast differences of description
The weight factor of value, and it is for 1, and both generally are taken as into 0.5.
When judging factor DrDuring more than 0, illustrate the geometric properties relative abundance of body surface, should select to be based on geometric properties
The registration Algorithm of point, enters back into step S104, obtains rotation translation matrix RT;Illustrate the texture information of body surface during less than 0
Relative abundance, should select the registration Algorithm based on image characteristic point, enter back into step S105 and obtain rotation translation matrix RT;If Dr
Equal to 0, it was demonstrated that the geometric properties and texture information of body surface are seldom, using the registration based on geometric properties and characteristics of image
Algorithm stability is poor, points out user to carry out point cloud by introducing index point.
Registration Algorithm preference pattern in the present invention can be adaptive selected base at the obvious surface of measure geometry feature
In the registration Algorithm of geometric properties point, it is to avoid splicing mistake based on image characteristic point;To geometric properties are few, textural characteristics
When the surface of relative abundance is spliced, adaptively using the registration Algorithm based on image characteristic point, eliminate unstable
Geometric properties registration Algorithm, finally improves the stability of splicing.
The geometric properties point that S104 is found out for step S101, that geometric properties point is carried out using RANSAC algorithms
Match somebody with somebody, obtain each and search the corresponding just matching point set Q ' of point, recycle SVD singular value decomposition methods to calculate spin matrix R peace
Move matrix T.
RANSAC algorithm principles are to carry out multi-times random sampling to first match point, and the match point needed for randomly selecting every time comes
Model parameter is determined, matching error is calculated further according to fixed model, the model of minimum match error will be possessed as final
Model.Concretely comprise the following steps:
4.1st step setting sampling number Snum, it is proposed that value 50~100.The geometric properties point of source point cloud is sampled,
Ensure that the mutual distance of these points is more than the minimum threshold of distance d of definitionmin, the geometric properties point after sampling is considered as lookup point,
Its space coordinate is designated as pτ(τ=1,2 ... r), wherein r is the geometric properties point number after sampled.dminIt is recommended that span is
10~15mm.
4.2nd step is p for each space coordinateτLookup point, using formula (5) target point cloud geometric properties point set
Q={ q1,q2…qnIn match point set Q '={ q at the beginning of similar with the lookup point point feature histogram point generation of search1′,q′2…
q′k, wherein n is the number of geometric properties point in target point cloud, the number of k match points at the beginning of each lookup point.Because source point cloud is several
There is r lookup point after what characteristic point is sampled, then match point set at the beginning of generating r.
Wherein i represents the sequence number in region,Representation space coordinate is pτLookup point histogrammic i-th minute of point feature
Amount,Represent xth point in target point cloud geometric properties point set Q (x=1, histogrammic i-th of the component of 2 ... point feature n),
Div represents 2 points of point feature histogram difference.Given threshold, it is proposed that span is 25~40, and Div is entered with taken threshold value
Row compares, and the point less than threshold value is considered as into just match point deposit just matches in point set Q '.
4.3rd step is p for space coordinateτLookup point, from it is each search point corresponding point set Q ' in randomly select one
Point as the lookup point corresponding points, remember the corresponding points space coordinate be q τ, using SVD decomposition methods calculate spin matrix R with
Translation matrix T, the range error d after rotation translation is calculated further according to formula (6)err, record current spin matrix R, translation matrix
T and error derr。
4.4th step repeats the 4.3rd step, by SnumBy error d after circulationerrMinimum spin matrix R and translation matrix T makees
For last spin matrix R and translation matrix T, then it is transferred to the splicing that S106 realizes two amplitude point clouds.
Spin matrix R and translation matrix T solution efficiency and accuracy can be improved using above-mentioned RANSAC algorithms, is reached most
The purpose of splicing stability is improved eventually.
The image characteristic point that S105 is found out for step S102, the lookup of corresponding points is realized with reference to RANSAC algorithms, really
After fixed three-dimensional corresponding points, the spin matrix R and translation matrix T between multi-viewpoint cloud are solved using SVD singular value decomposition methods, then
Into step S106.
The lookups of corresponding points is comprised the following steps that:
The image characteristic point that 5.1st step is found out for step S102, image characteristic point is realized using SIFT matching algorithms
First matching.Due to the influence of calculation error and measuring environment in first matching process, just there is error hiding in matching, influence is spelled
The stability connect, can utilize the RANSAC algorithms based on fundamental matrix to exclude error hiding, improve the spelling based on image characteristic point
Connect stability.
5.2nd step setting sampling number Snum, it is proposed that value 50~100.
5.3rd step is concentrated in first matched data and randomly selects 8 pairs of match points, and basic square is carried out using 8 algorithms of normalization
Battle array initial estimation, obtains fundamental matrix F.
In the case of not high to stability requirement, it would however also be possible to employ 5 algorithms replace 8 algorithms.
5.4th step calculates the error size E of every a pair first match points using formula (7)rr, m1、m2It is 2 points of the homogeneous seat
Mark, subscript T is transposition, sets ErrSpan (generally takes ± 0.3), and the point in span is considered as into optimal match point, note
Record current best match point and best match are counted out.
Err=m1 TFm2 (7)
5.5th step repeats the 5.3rd step and the 5.4th step, and best match is preserved all the time and is counted out most situations, until repeating
Number of times is equal to stochastical sampling number of times Snum。
5.6th step regard the optimal match point finally retained as correct corresponding points.Because using document【1】In method
Any three-dimensional point obtained in cloud data, the cloud data has unique images point to correspond to therewith, so passing through characteristics of image
The mapping relations one by one of point and three-dimensional point can realize the matching of three-dimensional point.
The Mismatching point in just matching is effectively eliminated using the RANSAC algorithms of the step of the 5.2nd step~the 5.5th, so as to improve
Spin matrix R and translation matrix T solution accuracy, reaches the final purpose for improving splicing stability.
In S104 and S105, the process that the SVD singular value decomposition methods solve spin matrix R and translation matrix T is:
6.1st step sets after obtained matching three-dimensional point as set G={ g1,g2…gNAnd G '={ g1′,g′2…g′N, profit
The barycenter of two panels point set is calculated with formula (8).Wherein N is match point logarithm, glAnd gl' the three-dimensional of matching three-dimensional point is sat to be any
Mark, g and g ' are 3 × 1 matrix.
6.2nd step is translated two panels point set G and G ' relative to respective barycenter using formula (9), obtains new point set J={ j1,
j2…jNAnd J '={ j1′,j2′…j′N}
jl=gl-g,jl'=gl'-g ' (l=1,2 ... N) (9)
6.3rd step calculates 3 × 3 matrix H using formula (10).
6.4th step carries out singular value decomposition to H-matrix and obtained:H=U Λ VT, wherein subscript T is matrix transposition, and U, V are 3 × 3
Unitary matrice, Λ is 3 × 3 diagonal matrix.Utilize the diagonal matrix A of formula (11) definition 3 × 3.
6.5th step calculates 3 × 3 spin matrix R and 3 × 1 translation matrix T using formula (12), and wherein subscript T is matrix
Transposition.
R=UAVT, T=g '-Rg (12)
S106 determines after rotation translation matrix RT that source point cloud is remained stationary as, and target point cloud is rotated using formula (13)
Translation completes splicing.
q′c=Rqc+T (13)
Wherein, qcFor the three-dimensional coordinate at any point in target point cloud, q 'cFor the three-dimensional coordinate of the point after transformed.
The above is presently preferred embodiments of the present invention, but the present invention should not be limited to the embodiment and accompanying drawing institute
Disclosure.So every do not depart from the lower equivalent or modification completed of spirit disclosed in this invention, guarantor of the present invention is both fallen within
The scope of shield.
Claims (6)
1. a kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation, this method comprises the steps:
1st step searches geometric properties point in source point cloud and target point cloud using point feature histogram method;Picture is being shot twice
Middle lookup image characteristic point;
2nd step sets up registration Algorithm preference pattern using the geometric properties point and image characteristic point found out, calculates registration Algorithm
Judge factor DrValue, automatically select suitable registration Algorithm according to the value:Work as Dr>When 0, into the 3rd step, work as Dr<When 0, it is transferred to
4th step, works as DrWhen=0, user is pointed out to carry out point cloud by introducing index point;
The registration Algorithm preference pattern is:
Wherein, p1, p2Represent respectively geometric properties in two width sampled point clouds count out account for its sampled point points ratio;n1, n2Point
The number for shooting image characteristic point in picture twice is not represented, and n is characteristics of image threshold value set in advance;P is geometric properties threshold
Value, w1And w2It is the weight factor for describing two contrast differences;
The geometric properties point that 3rd step is found out for the 1st step, the matching of geometric properties point is carried out using RANSAC algorithms, is obtained
Each searches the corresponding just matching point set Q ' of point, recycles SVD singular value decomposition methods to calculate spin matrix R and translation matrix T,
Subsequently into the 5th step;
The image characteristic point that 4th step is found out for the 1st step, corresponding points are searched using RANSAC algorithms, recycle SVD unusual
It is worth decomposition method and calculates spin matrix R and translation matrix T, subsequently into the 5th step;
5th step carries out rotation translation using spin matrix R and translation matrix T to target point cloud, completes splicing.
2. the no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation according to claim 1, it is characterised in that described
In 1 step, the process of the lookup geometric properties point is:
1.1st step is sampled in proportion to shooting the two width multi-viewpoint clouds obtained, calculates the normal vector of each sampled point;Will
Every bit in sampled point is considered as query point, and determines that the sampled point inquired about in neighborhood of a point, the neighborhood is referred to as neighborhood point;
1.2nd step defines a fixed local coordinate system, i.e., for any two points s, t in neighborhood on one of which point
Uvw coordinate systems:
ps, ptFor query point pqPoint s, t space coordinate, n in neighborhoods, ntFor the corresponding normal vector of point s, t, | | pt-ps||2For this
2 points of space length, d represents the unit vector on 2 line directions;
The uvw coordinate systems obtained using Formulas I, position relationship and Normal Error between 2 points are represented with one group of angle, such as formula
II:
Wherein, wntAnd untW, n are represented respectivelytAnd u, ntDot product, by α,θ span is respectively divided into b son
Interval, then collectively generated b3Individual region, the sequence number in each region is represented with i, judges 2 points of α,The residing region of θ values, then
Counting out for the region Jia one;Calculate query point pqThe α of each pair point in neighborhood,θ values, judge its region, and statistics falls
Counting out for each region, generates point pqPoint feature histogram;
1.3rd step utilizes formula using the point feature histogram generation average characteristics histogram u-histogram of all sampled points
The point feature histogram of each point is compared the relative KL distances of acquisition with u-histogram by III;
Wherein, i represents the sequence number of point feature histogram regions, and value is 1~b3,Represent that the point feature of sampled point is histogrammic
I-th of component, uiHistogrammic i-th of the component of average characteristics is represented, the average value a and standard deviation sigma of KL distances is calculated, counts KL
Three-dimensional point of the distance outside a ± σ, is considered as geometric properties point.
3. the no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation according to claim 2, it is characterised in that described
In 1 step, image characteristic point is searched using SIFT algorithms.
4. the no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation according to claim 1,2 or 3, it is characterised in that
The process that implements of 3rd step is:
3.1st step is sampled to the geometric properties point of source point cloud, and the geometric properties point mutual distance of each sampling is more than in advance
If minimum threshold of distance, the geometric properties point after sampling is considered as lookup point, its space coordinate is designated as pτ, τ=1,2 ... r, its
Middle r is the geometric properties point number after sampled;
3.2nd step is p for each space coordinateτLookup point, using Formula V target point cloud geometric properties point set Q=
{q1,q2…qnIn match point set Q '={ q ' at the beginning of similar with the lookup point point feature histogram point generation of search1,q′2…q
′k, wherein n is the number of geometric properties point in target point cloud, k for it is each search point at the beginning of match point number, at the beginning of generation r
With point set;
Wherein i represents the sequence number of point feature histogram regions, and value is 1~b3,Representation space coordinate is pτLookup point
Histogrammic i-th of the component of point feature,Represent the point feature histogrammic the of xth point in target point cloud geometric properties point set Q
I component, x=1,2 ... n, Div represents 2 points of point feature histogram difference;Div is compared with threshold value set in advance
Compared with match point deposit at the beginning of the point less than threshold value is considered as is just in matching point set Q ';
3.3rd step is p for space coordinateτLookup point, from it is each search point corresponding point set Q ' in randomly select a little make
For the corresponding points of the lookup point, the space coordinate for remembering the corresponding points is qτ, spin matrix R and translation are calculated using SVD decomposition methods
Matrix T, the range error d after rotation translation is calculated further according to Formula IVerr, record current spin matrix R, translation matrix T and mistake
Poor derr:
3.4th step repeats the 3.3rd step, by SnumBy error d after circulationerrMinimum spin matrix R and translation matrix T conducts
Last spin matrix R and translation matrix T, wherein SnumRepresent sampling number set in advance.
5. the no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation according to claim 1,2 or 3, it is characterised in that
In 4th step, the process of the lookup corresponding points is:
4.1st step realizes the first matching of image characteristic point using SIFT matching algorithms for the image characteristic point found out;
4.2nd step is concentrated in first matched data and randomly selects 8 pairs of match points, at the beginning of carrying out fundamental matrix using 8 algorithms of normalization
Begin to estimate, obtain fundamental matrix F;
4.3rd step calculates the error size E of every a pair first match points using Formula VIIrr, m1、m2It is 2 points of the homogeneous coordinates, on
Mark T is transposition, sets ErrSpan, optimal match point is considered as by the point in span, record current best match point and
Best match is counted out;
Err=m1 TFm2Formula VII
4.4th step repeats the 4.2nd step and the 4.3rd step, and best match is preserved all the time and is counted out most situations, until repeating time
Number is equal to sampling number S set in advancenum;
4.6th step regard the optimal match point finally retained as correct corresponding points.
6. the no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation according to claim 1,2 or 3, it is characterised in that
The process that the SVD singular value decomposition methods solve spin matrix R and translation matrix T is:
6.1st step sets after obtained matching three-dimensional point as set G={ g1,g2…gNAnd G '={ g '1,g′2…g′N, utilize
Formula VIII calculates the barycenter of two panels point set;Wherein N is match point logarithm, glWith g 'lThe three-dimensional of matching three-dimensional point is sat to be any
Mark, g and g ' are 3 × 1 matrix;
6.2nd step is translated two panels point set G and G ' relative to respective barycenter using Formula IX, obtains new point set J={ j1,j2…
jNAnd J '={ j '1,j2′…j′N}
jl=gl—g,j′l=g 'l - g ', l=1,2 ... N Formula IX
6.3rd step calculates 3 × 3 matrix H using Formula X;
6.4th step carries out singular value decomposition to H-matrix and obtained:H=U Λ VT, wherein subscript T is matrix transposition, and U, V are 3 × 3 tenth of the twelve Earthly Branches squares
Battle array, Λ is 3 × 3 diagonal matrix, and 3 × 3 diagonal matrix A is defined using Formula X I;
6.5th step calculates 3 × 3 spin matrix R and 3 × 1 translation matrix T using XII formulas:
R=UAVT, T=g '-Rg Formula X II
S106 determines after rotation translation matrix RT that source point cloud is remained stationary as, and rotation translation is carried out to target point cloud using Formula X III
Complete splicing:
q′c=Rqc+ T Formula X III
Wherein, qcFor the three-dimensional coordinate at any point in target point cloud, q 'cFor the three-dimensional coordinate of the point after transformed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410571233.7A CN104392426B (en) | 2014-10-23 | 2014-10-23 | A kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410571233.7A CN104392426B (en) | 2014-10-23 | 2014-10-23 | A kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104392426A CN104392426A (en) | 2015-03-04 |
CN104392426B true CN104392426B (en) | 2017-07-18 |
Family
ID=52610324
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410571233.7A Active CN104392426B (en) | 2014-10-23 | 2014-10-23 | A kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104392426B (en) |
Families Citing this family (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105719287B (en) * | 2016-01-19 | 2019-04-26 | 博众精工科技股份有限公司 | A kind of alignment method of irregular figure |
CN106137395B (en) * | 2016-07-22 | 2019-01-29 | 华南理工大学 | Full-automatic patient registry method applied to unmarked optical operation navigation system |
US10152786B2 (en) * | 2016-10-11 | 2018-12-11 | Biosense Webster (Israel) Ltd. | Registration of a magnetic tracking system with an imaging device |
CN108175500A (en) * | 2016-12-08 | 2018-06-19 | 复旦大学 | Surgical navigational spatial registration method based on handheld three-dimensional scanner |
CN106780459A (en) * | 2016-12-12 | 2017-05-31 | 华中科技大学 | A kind of three dimensional point cloud autoegistration method |
CN106886977B (en) * | 2017-02-08 | 2021-02-05 | 徐州工程学院 | Multi-image automatic registration and fusion splicing method |
CN106846387B (en) * | 2017-02-09 | 2019-07-12 | 中北大学 | Point cloud registration method based on neighborhood rotary volume |
CN108510439B (en) * | 2017-02-28 | 2019-08-16 | 贝壳找房(北京)科技有限公司 | Joining method, device and the terminal of point cloud data |
CN107702663B (en) * | 2017-09-29 | 2019-12-13 | 五邑大学 | Point cloud registration method based on rotating platform with mark points |
CN108171796A (en) * | 2017-12-25 | 2018-06-15 | 燕山大学 | A kind of inspection machine human visual system and control method based on three-dimensional point cloud |
CN111105435B (en) * | 2018-10-25 | 2023-09-29 | 北京嘀嘀无限科技发展有限公司 | Mark matching method and device and terminal equipment |
CN109584310B (en) * | 2018-11-26 | 2022-12-16 | 南昌航空大学 | Splicing method for large object surface shape measurement based on verticality constraint |
CN109754396B (en) * | 2018-12-29 | 2021-02-19 | 上海联影智能医疗科技有限公司 | Image registration method and device, computer equipment and storage medium |
CN109489553B (en) * | 2018-12-27 | 2020-10-16 | 中国科学院长春光学精密机械与物理研究所 | Method, device, equipment and storage medium for generating space marker point library |
CN110120013B (en) * | 2019-05-15 | 2023-10-20 | 深圳市凌云视迅科技有限责任公司 | Point cloud splicing method and device |
CN110533589A (en) * | 2019-07-18 | 2019-12-03 | 上海大学 | A kind of threedimensional model joining method based on zoom micro-image sequence |
CN111540063A (en) * | 2020-04-15 | 2020-08-14 | 富德康(北京)科技股份有限公司 | Full-automatic high-precision splicing method based on multi-station laser point cloud data |
CN112232319A (en) * | 2020-12-14 | 2021-01-15 | 成都飞机工业(集团)有限责任公司 | Scanning splicing method based on monocular vision positioning |
CN113140042B (en) * | 2021-04-19 | 2023-07-25 | 思看科技(杭州)股份有限公司 | Three-dimensional scanning splicing method and device, electronic device and computer equipment |
CN113204214B (en) * | 2021-04-28 | 2024-04-09 | 中国人民解放军陆军装甲兵学院 | Movable modularized multi-energy beam energy field material increasing and decreasing composite repairing equipment and method |
CN114170284B (en) * | 2022-02-09 | 2022-07-29 | 南京理工大学 | Multi-view point cloud registration method based on active landmark point projection assistance |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103150747A (en) * | 2012-11-16 | 2013-06-12 | 佳都新太科技股份有限公司 | Point cloud registration algorithm based on topological characteristic |
CN103236064A (en) * | 2013-05-06 | 2013-08-07 | 东南大学 | Point cloud automatic registration method based on normal vector |
CN103955939A (en) * | 2014-05-16 | 2014-07-30 | 重庆理工大学 | Boundary feature point registering method for point cloud splicing in three-dimensional scanning system |
-
2014
- 2014-10-23 CN CN201410571233.7A patent/CN104392426B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103150747A (en) * | 2012-11-16 | 2013-06-12 | 佳都新太科技股份有限公司 | Point cloud registration algorithm based on topological characteristic |
CN103236064A (en) * | 2013-05-06 | 2013-08-07 | 东南大学 | Point cloud automatic registration method based on normal vector |
CN103955939A (en) * | 2014-05-16 | 2014-07-30 | 重庆理工大学 | Boundary feature point registering method for point cloud splicing in three-dimensional scanning system |
Non-Patent Citations (2)
Title |
---|
《一种无标记点三维点云自动拼接技术》;郭进等;《计算机应用与软件》;20120430;第29卷(第4期);全文 * |
《中国海洋大学学报》;解则晓等;《三维点云数据拼接中ICP 及其改进算法综述》;20100131;第40卷(第1期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN104392426A (en) | 2015-03-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104392426B (en) | A kind of no marks point three-dimensional point cloud method for automatically split-jointing of self adaptation | |
CN109887015B (en) | Point cloud automatic registration method based on local curved surface feature histogram | |
CN106097348B (en) | A kind of fusion method of three-dimensional laser point cloud and two dimensional image | |
CN107192350B (en) | A kind of three-dimensional laser scanner intrinsic parameter scaling method and device | |
CN104867160B (en) | A kind of directionality demarcation target demarcated for camera interior and exterior parameter | |
CN107679537B (en) | A kind of texture-free spatial target posture algorithm for estimating based on profile point ORB characteristic matching | |
CN108133458A (en) | A kind of method for automatically split-jointing based on target object spatial point cloud feature | |
CN106548462B (en) | Non-linear SAR image geometric correction method based on thin-plate spline interpolation | |
CN109816724A (en) | Three-dimensional feature extracting method and device based on machine vision | |
CN111815686B (en) | Geometric feature-based coarse-to-fine point cloud registration method | |
CN105654483B (en) | The full-automatic method for registering of three-dimensional point cloud | |
CN109903319B (en) | Multi-resolution-based fast iteration closest point registration algorithm | |
CN104463894A (en) | Overall registering method for global optimization of multi-view three-dimensional laser point clouds | |
CN106067172B (en) | A method of slightly matching matches combination to the underwater topography image based on suitability analysis with essence | |
CN106023298A (en) | Point cloud rigid registration method based on local Poisson curved surface reconstruction | |
CN109658444A (en) | A kind of regular three-dimensional colour point clouds method for registering based on multi-modal feature | |
CN108830888A (en) | Thick matching process based on improved multiple dimensioned covariance matrix Feature Descriptor | |
CN109859256A (en) | A kind of three-dimensional point cloud method for registering based on automatic corresponding point matching | |
CN109544606A (en) | Fast automatic method for registering and system based on multiple Kinect | |
CN111415379A (en) | Three-dimensional point cloud data registration method based on cuckoo optimization | |
CN103700135B (en) | A kind of three-dimensional model local spherical mediation feature extracting method | |
CN110796691A (en) | Heterogeneous image registration method based on shape context and HOG characteristics | |
CN112381862A (en) | Full-automatic registration method and device for CAD (computer-aided design) model and triangular mesh | |
CN104318566B (en) | Can return to the new multi-view images plumb line path matching method of multiple height values | |
CN104952080B (en) | A kind of method for realizing remote sensing image coarse positioning |
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 |