CN101350101A - Method for auto-registration of multi-amplitude deepness image - Google Patents
Method for auto-registration of multi-amplitude deepness image Download PDFInfo
- Publication number
- CN101350101A CN101350101A CNA2008102220956A CN200810222095A CN101350101A CN 101350101 A CN101350101 A CN 101350101A CN A2008102220956 A CNA2008102220956 A CN A2008102220956A CN 200810222095 A CN200810222095 A CN 200810222095A CN 101350101 A CN101350101 A CN 101350101A
- Authority
- CN
- China
- Prior art keywords
- registration
- image
- consistent
- amplitude deepness
- cycle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to a multi-frame range image automatic registration method. (1) Any two frames of range images are registered through using SIFT characteristics, and the correctness of a result is judged. First, the SIFT characteristics of the two frames of the range images are calculated, corresponding points are bidirectionally crossly matched, then polar line constraint is calculated by a RANSAC algorithm, error matching is filtered, then the two frames of the range images are accurately registered through using an ICP algorithm, and the correctness of a result is judged. (2) a cycle space of a model diagram is researched, and a global consistent registration result is calculated. First, a derived cycle basis of the model diagram is calculated, an adjacency relational graph of the derived cycle basis is built, then a group basis of the consistent cycle space is calculated, and further a consistent registration result is obtained. The method can effectively increase the research speed of the cycle space, and an exponential time complexity can be increased to a line time complexity in an ideal condition. (3) the consistency of a cycle is judged, and the consistency of the registration result in the cycle is judged by a judgment method of relative differences. The invention can reliably automatically register multi-frame range images, through researching consistent cycles, error registration is taken out, thereby obtaining consistent registration results.
Description
Technical field
The invention belongs to the computer virtual reality technology field, specifically relate to auto-registration of multi-amplitude deepness image, obtain complete model, this method can be used for the Geometric Modeling of three-dimensional model.
Background technology
In the last few years, the fast development of 3-D scanning technology made the autoregistration technology become more and more important.Depth image is the image that has depth information that obtains behind the 3-D scanning device scan.Registration is meant that the depth image that will scan from diverse location, different angles changes to the process of the same coordinate system by rigid transformation.By registration, multi-amplitude deepness image is stitched together, thereby obtain complete three-dimensional model.Registration computation process mainly can be divided into two amplitude deepness image registrations and multi-amplitude deepness image registration.Two amplitude deepness image registrations are that the relative position relation that calculates between two width of cloth images carries out registration; The multi-amplitude deepness image registration claims that also global registration is on the basis of two amplitude deepness image registrations, removes wrong registration results in twos, calculates the consistent registration results of the overall situation, optimizes the global error of global registration then.
Two amplitude deepness image registrations mainly comprise roughly in twos registration and accurate two steps of registration in twos.Accurately registration mainly adopts ICP (Iterative Closest Point) algorithm (referring to P.J.Besl in twos, N.D.McKay.A Method for Registration of 3DShapes.IEEE Trans.Pattern Anal[J] .and Machine Intell.Vol.14, pp.239-256,1992 and Y.Chen, G.Medioni.Object Modeling by Registration of Multiple Range Images[J] .IEEE Conference onRobotics and Automation, pp.2724-2729,1991. [Rusinkiewicz 2001] Rusinkiewicz, S., Levoy, M.Efficient Variants of the ICP Algorithm[J] .In International Conference on 3D Digital Imagingand Modeling, 2001.), obtained good effect, but, because the ICP algorithm is a local convergence, so before adopting the ICP algorithm, to carry out rough in twos registration earlier, to obtain initial position preferably, avoid being absorbed in local optimum.Rough in twos registration mainly utilizes feature descriptor, as image rotating, integrating sphere etc., mates at least three pairs or more corresponding point, and then tries to achieve rigid transformation.
Global registration is on registration basis in twos, judges the correctness of registration in twos, removes error result, obtains whole consistent registration results.Huber (referring to Huber D.Automatic Three-dimensional Modeling from Reality.CarnegieMellon University, 2002) has proposed first global registration algorithm.Whether he utilizes visibility criteria, judge to exist between the depth image to block, and then judge whether registration is correct, and utilize Bayesian probability to obtain the reliability coefficient of each registration, then, uses greedy method, generates the minimum spanning tree of a whole model of registration.But all will optimize global error before blocking judgement at every turn, and block computation complexity, so counting yield is lower also than higher.And calculating the generation tree incipient stage, owing to have only less depth image to be aggregated to together, so the reliability of blocking property judgement is also lower.Hunag is (referring to HuangQ X, Flory S, Gelfand N, Hofer M, Pottmann H.Reassembling fractured objects by geometricmatching.ACM Transactions on Graphics (SIGGRAPH), 2006:569-578) improved the method for Huber, he notices between the submodel that two multi-amplitude deepness images merge may exist a plurality of relations of registrations in twos, this is more reliable than single-relation, therefore he preferentially is associated with the subgraph of more heterogeneous appearance registration relation, rather than only rely on single registration results and ask minimum spanning tree, but this method still will repeatedly be optimized total error, the computation complexity height, it is not high to merge same reliability of initial stage at model.Notice a plurality of compatible registration relation between submodel, come down to have constituted the loop, and registration results will satisfy consistency constraint in the loop in twos.
Need to optimize global error after the global registration.With the ICP class of algorithms seemingly, reduce error gradually by iteration, but owing to relate to multi-amplitude deepness image, so calculate more complicated.The global registration algorithm mainly is divided into two classes, optimize Bergevin and Pulli in twos (referring to Bergevin R., Soucy M., Gagnon H., et al.Towards a General Multi-View RegistrationTechnique.IEEE Trans.Pattern Anal.and Machine Intell, 1996,18 (5): 540-547 and Pulli K.Multiview Registration for Large Data Sets.nternational Conference on 3D Digital Imaging andModeling) and optimize Krishnan and Neugebauer algorithm simultaneously (referring to Krishnan S., Lee P.Y., Moore J., etal.Global Registration of Multiple 3D Point Sets via Optimization-on-a-Manifold.Proc.ofSymposium on Geometry Processing, 2005:187-196 and Neugebauer P.J.Geometrical Cloning of3D Objects via Simultaneous Registration of Multiple Range Images[C] .Proceed-ings of the 1997International Conference on Shape Modeling and Applications (SMA ' 97), 1997,130).Optimize in twos and adopt ICP algorithm optimization two width of cloth images repeatedly, so speed of convergence is slow and possibly can't restrain, the method for You Huaing is directly optimized total error simultaneously, and speed of convergence is very fast, but complicated computation optimization less stable.
Summary of the invention
Technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, a kind of method for auto-registration of multi-amplitude deepness image is provided, it is fast that this method is carried out multi-amplitude deepness image registration speed, and coupling accurately.
Technical solution of the present invention: method for auto-registration of multi-amplitude deepness image, comprise two amplitude deepness image registrations and multi-amplitude deepness image global registration process, wherein:
Described two amplitude deepness image registrations are as follows:
(1) on two width of cloth images, calculates the SIFT feature respectively, obtain corresponding one 128 dimensional feature vector of each unique point;
(2) according to the described proper vector of step (1), the coupling corresponding point;
(3) determine to adopt the RANSAC algorithm to find the solution basis matrix after the coupling corresponding point, adopt basis matrix to reject the corresponding point of matching error then;
(4) corresponding point that will reject behind the matching error are pressed the ordering of SIFT characteristic distance, remove 20% maximum corresponding point of distance between the proper vector of two width of cloth images;
(5) corresponding point according to two width of cloth images adopt the hypercomplex number method to try to achieve transformation matrix, thereby obtain rough in twos registration;
(6) behind the rough in twos registration, use accurate registration two amplitude deepness images of ICP algorithm;
Described multi-amplitude deepness image global registration process is as follows:
(7) accurate two amplitude deepness images of registration in twos that obtain according to step (6) are set up the illustraton of model of multiple image;
(8) try to achieve the derivation cycle basis of illustraton of model, set up the syntople figure Г that derives cycle basis, described derivation cycle basis is meant the base that does not have the derivation of string circle to constitute;
(9) search turn space, described cycle space is the set in all loops in the illustraton of model, when search, only search for the circle that the connection summit among the syntople figure Г that derives cycle basis generates, and when obtaining consistent a circle, just delete a cycle basis, till the remaining number of vertices of Г is less than the number of vertices that will search for, obtain the global registration of multiple image.
Adopt bidirectional crossed method coupling corresponding point in the described step (2), when the steps include: to mate corresponding point, for the unique point p on first width of cloth image, it has a proper vector, in all proper vectors of second amplitude deepness image, find the nearest unique point p ' of proper vector of distance feature point p, conversely for p ', in all proper vectors of first amplitude deepness image, find the nearest unique point p of proper vector of distance feature point p ' "; if p=p ", then p and p ' are the coupling corresponding point.
To judge also after the described step (6) whether registration results is correct in twos, and determining step is: at first, according to SIFT characteristic matching corresponding point the time,, judge that then registration is incorrect if the correspondence that the match is successful is counted less than setting threshold; Secondly, when using the accurate registration of ICP algorithm,,, judge that then registration is incorrect greater than setting threshold if error is excessive; If above two conditions all can't be judged the registration results mistake, think that then the result is correct.
Illustraton of model G=＜V described in the described step (7), E〉be a non-directed graph, V is the set on summit, E is the set on limit, each vertex v
i∈ V represents an amplitude deepness image, each bar limit e
Ij=(v
i, v
j) ∈ E, and if only if depth image v
iAnd v
jThe T of registration results in twos
IjExist, and T
IjAs limit e
IjAttribute, v wherein
iAnd v
jBe two summits of illustraton of model.
In the described step (8) for obtaining a consistent determination methods of enclosing: from circle, appoint and get an amplitude deepness image, try to achieve its diameter d, with this depth image along the circle in one week of the conversion of conversion in twos, obtain the new depth image of a width of cloth, calculate the error e between new depth image and the initial depth image corresponding point, with the error of e/d, if the error of circle is consistent less than the setting threshold circle as circle, otherwise, be inconsistent.
The present invention's advantage compared with prior art is:
(1) it is fast to the present invention is based on the feature registration computing velocity of SIFT, and cross-matched corresponding point reliability height filters corresponding point with the RANSAC algorithm, has at utmost improved the reliability of corresponding point.
(2) the present invention sets up the illustraton of model of multiple image, try to achieve the derivation cycle basis of illustraton of model, set up the syntople figure Г that derives cycle basis, when the search turn space, only search for the circle that the connection summit among the syntople figure Г that derives cycle basis generates, and when obtaining consistent a circle, just delete a cycle basis, till the remaining number of vertices of Г is less than the number of vertices that will search for, obtain the global registration of multiple image, improved search speed greatly, thus fast and reliable obtain the consistent registration results of the multiple image overall situation.
(3) and the present invention judge by the consistance of circle, further improved the reliability of multiple image.
Description of drawings
Fig. 1 is the process flow diagram of the inventive method;
Fig. 2 is for setting up the synoptic diagram of illustraton of model among the present invention;
Fig. 3 is for adopting the error map of 64 circles among the present invention;
Fig. 4 is the partial enlarged drawing of Fig. 3;
Fig. 5 gives birth to through covert cave figure down for the Muller that adopts the inventive method to carry out autoregistration.
Embodiment
As shown in Figure 1, the present invention is divided into two amplitude deepness image registrations and multi-amplitude deepness image global registration process.
1, two amplitude deepness image registrations
The rough registration of two amplitude deepness images is the process to the rough registration of partly overlapping two amplitude deepness images that is in the optional position.Purpose is for accurate registration provides initial position preferably, to guarantee its convergence.Owing to comprise colouring information in the depth image, so the present invention adopts image corresponding point matching method to carry out registration, the steps include:
(1) at first, calculate the SIFT feature respectively on two width of cloth images, corresponding one 128 dimensional feature vector of each unique point adopts the distance between two SIFT features of euclidean distance metric; Calculate the SIFT feature and adopt Lowe in this step, D.Distinctive image features from scale-invariant keypoints.Int.J.of Computer Vision 60,2 has introduced method in the 91-110.2004. literary composition and has got final product.
(2) adopt bidirectional crossed coupling corresponding point, when promptly mating corresponding point, for the unique point p on first width of cloth image, it has a proper vector, in all proper vectors of second amplitude deepness image, find the nearest unique point p ' of proper vector of distance feature point p,, in all proper vectors of first amplitude deepness image, find the nearest unique point p of proper vector of distance feature point p ' conversely for p ' "; if p=p ", then p and p ' are the coupling corresponding point.
(3) after the coupling, adopt the RANSAC algorithm to find the solution basis matrix, to reject the error point, adopting the RANSAC algorithm to find the solution basis matrix can be referring to Hartley, R.I., AND ZISSERMAN, A.2004.Multiple View Geometry inComputer Vision.Cambridge University Press, Cambridge, UK.
(4) then, corresponding point are pressed SIFT characteristic distance ordering, remove 20% maximum corresponding point of distance.
(5) last, use the hypercomplex number method to try to achieve transformation matrix according to corresponding point, the process of finding the solution transformation matrix of use hypercomplex number method can be referring to Horn, B.Closed Form Solutions of Absolute Orientation Using Unit Quaternions.Journal of the Optical Society, Vol.4,629-642,1987.
(6) behind the rough in twos registration, use accurate registration two amplitude deepness images of ICP (Iterative Closest Point) algorithm.The ICP algorithm is referring to P.J.Besl, N.D.McKay.A Method for Registration of 3D Shapes.IEEE Trans.Pattern Anal[J] .and Machine Intell.Vol.14, pp.239-256,1992., each iteration is revised distance threshold according to error dynamics, has obtained the result of the accurate registrations of two amplitude deepness images like this.
(7) result to the accurate registration of two amplitude deepness images carries out the consistance judgement of registration in twos.Feature Points Matching mistake or two amplitude deepness images are not overlapped, and all may cause registration results mistake in twos.Therefore registration is to judge consequently that not correctly determining step is: at first, according to SIFT characteristic matching corresponding point the time, if the correspondence that the match is successful is counted less than setting threshold, judge that then registration is incorrect in twos; Secondly, when using the accurate registration of ICP algorithm,,, judge that then registration is incorrect greater than setting threshold if error is excessive.If above two conditions all can't be judged the registration results mistake, think that then the result is correct, and registration results joins in the illustraton of model of back multiple image global registration in twos.
2, multi-amplitude deepness image global registration
(1), sets up the illustraton of model of multiple image according to two amplitude deepness images of accurate registration in twos that obtain;
After the registration, obtain illustraton of model G in twos, but only rely on the consistance of registration in twos to judge that the result who obtains is insecure, for example since model from symmetry, cause the registration mistake, but be difficult to judge by registration in twos.Therefore, after the registration, carry out global coherency and judge in twos.Because consistency constraint all should be satisfied in any loop of illustraton of model G, otherwise is exactly incorrect, the registration results in a loop can self-verification like this, and the present invention proposes the global registration method based on the loop.
Illustraton of model G=＜V, E〉be a non-directed graph, each vertex v
i∈ V represents an amplitude deepness image, limit e
Ij=(v
i, v
j) ∈ E and if only if depth image v
iAnd v
jThe T of registration results in twos
IjExist, and T
IjAs limit e
IjAttribute.It should be noted that any loop all will satisfy consistency constraint, promptly total conversion should be unit transformation.As Fig. 2, in loop 16521, should satisfy T
12T
25T
56T
61=I.If one consistency constraint is satisfied in the loop, then be called consistent loop, otherwise be called inconsistent.
(2) try to achieve the derivation cycle basis of illustraton of model, set up the syntople figure Г that derives cycle basis;
Simply introduce earlier the notion of relevant figure, circle and cycle space.A figure G=＜V, E 〉, wherein V=v (G) represents vertex set, the set on E=ε (G) expression limit.The path that each summit occurs being no more than once is called elementary path (Path).Starting point and terminal point are that the elementary path on same summit is called circle (Cycle).Article one, the limit connects two summits in the circle, but it is not the limit in the circle, is called string (Chord).One does not have the circle of string to be called derivation circle (Induced Cycle).If figure is G
1And G
2There is not common edge, then
Be called figure G
1And G
2Bian Buchong also.Loop be meant the circle with the circle Bian Buchong also.Obviously circle is a loop, and circle is communicated with, but loop not necessarily is communicated with.Figure G
1And G
2Symmetric difference (Symmetric Difference) operation definition be
Operation of symmetric difference satisfies law of commutation and law of association.
Connected graph G=＜V, E〉all loops and the set formed of empty set be C={C
1, C
2, C
n), define operation of symmetric difference and number multiplication 0C thereon
i=Φ, 1C
i=C
i, then it constitutes number field
On one | E|-|V|+1 dimensional linear space, the note make C (G), be called cycle space (Cycle Space).Any cycle space can be by deriving cycle basis to generate whole cycle space.
(3) search turn space;
If the derivation cycle basis of illustraton of model G is
Whole cycle space C (G) by
Open into, so the size of C (G) is 2
| E|-|V|+1, be the index complexity, search efficiency is too low.All consistent circles have constituted linear subspaces in the illustraton of model, are called uniform subspace.The present invention proposes a kind of method of one group of base of the subspace that tries to achieve a consensus fast, has improved the search efficiency of cycle space greatly, degree of being mixed with when ideally being linear answering.
Notice that these two circles must have common edge so if the symmetric difference of two circles still is circle.For acceleration search, at first define the syntople figure of cycle basis
Limit (B
i, B
j) ∈ L and if only if ε (B
i) ⌒ ε (B
j) ≠ Φ.Therefore have to draw a conclusion: establish
If C is a circle, exist so
In
Be communicated with.The loop that attention is generated by the pairing cycle basis in the summit that is communicated with among the Г also may not be a circle, therefore also will check it whether to enclose.When the search turn space,, can significantly reduce searching times so like this as long as search by all circles of the base generation of any a plurality of connections, is judged their consistance.
In search during consistent the circle, at first basis in twos the illustraton of model G of registration obtain the derivation cycle basis
And generate
Syntople figure
Adopting and deriving cycle basis is because its length is less, thereby consistent possibility is bigger, helps to improve search efficiency.Ask one group of base of S (G) then, at first, judge each circle C=B
iWhether consistent, if consistent, then C joined in the base of S (G), and from Г, delete B
iJudge circle then by the cycle basis generation of any two connections
Whether consistent, if consistent, C joined in the base of S (G), and from Г, delete B
iJudge the circle that generates by the cycle basis of any three connections then, go down like this, in Г remaining summit less than the connection number of vertices that will search for till.Then, all consistent limits merging of enclosing are obtained consistent illustraton of model G ', detailed algorithm provides in algorithm 1.This algorithm is a linear complexity under best-case, compares with exponential complexity, has improved counting yield greatly.If G ' has only a connected subgraph, then global registration calculates and finishes.If G ' has a plurality of connected subgraphs, then no longer there is consistent loop between the connected subgraph, the present invention adopts the determination methods based on observability---and free sky asks that conflict (Free Space Violation) continues global registration [Huber 2002], generate the minimum spanning tree between the connected subgraph, to merge more subgraph.Because generally speaking, each connected subgraph all contains multi-amplitude deepness image, so for single image, observability judges to have higher reliability.
| Algorithm 1: |
| Input: the in twos illustraton of model G of registration output: the derivation cycle basis B that the consistent illustraton of model G ' Procedure of the overall situation obtains G sets up syntople figure Г＜B of B, L〉among consistent_cycle ← Φ N ← 1 while Г among number of vertex＞=N for each Г N be communicated with the consistent then of circle C if C that the summit generates and C be inserted into number of vertex＜Nthen break end if end if end for N ← N+1 end while G ' ← Φ among the If Г who deletes among the consistent_cycle in this N summit all limits that the summit among the G is added among the G ' all circles among the consistent_cycle are added among the G ' |
(4) consistance of circle is judged;
Try to achieve after the circle, judge whether it is consistent.A circle is consistent, is meant the total approximate unit of the conversion in twos conversion in the circle.If circle C=V
1V
2V
nV
1, appoint and get an amplitude deepness image V
i, try to achieve V
iTo V
iTotal conversion T=T in the loop
I, i+1T
N, 1T
I-1, i, T
I, jExpression V
jTo V
iTransformation matrix.Total mapping fault is in the calculating circle
P wherein
k∈ V
iBe depth image V
iIn point, n represents V
iThe number of mid point.
If depth image V
iThe encirclement bulb diameter be d.If ε＜＜d, it is consistent enclosing C so, otherwise circle C is inconsistent.Definition r=ε/d is the error in the circle, and r is that convergent-divergent is constant, and it has nothing to do with the scale transformation of depth image like this.If r is consistent less than the setting threshold circle, otherwise, be inconsistent.
(5) test findings;
When the consistance of circle was judged, the error profile in each loop was shown in Fig. 3,4, and there is obviously fracture in error profile between 0.03 and 0.3 as can be seen, and therefore, the present invention sets and judges that the consistent error threshold that encloses is 0.03.
As shown in Figure 5, the Dazu Rock Carvings Muller is given birth to down through covert cave model and is had 81 amplitude deepness images, comprises 22477248 points, about 27.7 ten thousand points of average every amplitude deepness image.Adopt the same parameter setting, carried out 3240 times altogether in twos behind the registration, obtain illustraton of model and comprise 217 limits, so the dimension of cycle space is 137.During the search turn space, finish search after having searched for 140 circles altogether, obtain 124 consistent circles, promptly uniform subspace is 124 dimensions, and the consistent illustraton of model that obtains comprises 184 limits and has been communicated with, 42.5 seconds search turn space times spent.Whole registration calculates about 35 minutes of time spent, and the visible overwhelming majority time all is used for registration in twos, the complete model that obtains at last.
In a word, the present invention is the autoregistration multi-amplitude deepness image reliably, by searching for consistent the circle, removes misregistration, obtains consistent registration results.
The content that is not described in detail in the instructions of the present invention belongs to this area professional and technical personnel's known prior art.
The above only is a preferred implementation of the present invention; should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.
Claims (6)
1, method for auto-registration of multi-amplitude deepness image, two amplitude deepness image registrations and multi-amplitude deepness image global registration process is characterized in that step is as follows:
Described two amplitude deepness image registrations are as follows:
(1) on two width of cloth images, calculates the SIFT feature respectively, obtain corresponding one 128 dimensional feature vector of each unique point;
(2) according to the described proper vector of step (1), the coupling corresponding point;
(3) determine to adopt the RANSAC algorithm to find the solution basis matrix after the coupling corresponding point, adopt basis matrix to reject the corresponding point of matching error then;
(4) corresponding point that will reject behind the matching error are pressed the ordering of SIFT characteristic distance, remove 20% maximum corresponding point of distance between the proper vector of two width of cloth images;
(5) corresponding point according to two width of cloth images adopt the hypercomplex number method to try to achieve transformation matrix, thereby obtain rough in twos registration;
(6) behind the rough in twos registration, use accurate registration two amplitude deepness images of ICP algorithm;
Described multi-amplitude deepness image global registration process is as follows:
(7) accurate two amplitude deepness images of registration in twos that obtain according to step (6) are set up the illustraton of model of multiple image;
(8) try to achieve the derivation cycle basis of illustraton of model, set up the syntople figure Γ that derives cycle basis, described derivation cycle basis is meant the base that does not have the derivation of string circle to constitute;
(9) search turn space, described cycle space is the set in all loops in the illustraton of model, when search, only search for the circle that the connection summit among the syntople figure Γ that derives cycle basis generates, and when obtaining consistent a circle, just delete a cycle basis, till the remaining number of vertices of Γ is less than the number of vertices that will search for, obtain the global registration of multiple image.
2, according to the described method for auto-registration of multi-amplitude deepness image of claim 1, it is characterized in that: adopt bidirectional crossed method coupling corresponding point in the described step (2), when the steps include: to mate corresponding point, for the unique point p on first width of cloth image, it has a proper vector, in all proper vectors of second amplitude deepness image, find the nearest unique point p ' of proper vector of distance feature point p, conversely for p ', in all proper vectors of first amplitude deepness image, find the nearest unique point p of proper vector of distance feature point p ' "; if p=p ", then p and p ' are the coupling corresponding point.
3, according to the described method for auto-registration of multi-amplitude deepness image of claim 1, it is characterized in that: will judge also after the described step (6) whether registration results is correct in twos, determining step is: at first, according to SIFT characteristic matching corresponding point the time, if the correspondence that the match is successful is counted less than setting threshold, judge that then registration is incorrect; Secondly, when using the accurate registration of ICP algorithm,,, judge that then registration is incorrect greater than setting threshold if error is excessive; If above two conditions all can't be judged the registration results mistake, think that then the result is correct.
4, according to the described method for auto-registration of multi-amplitude deepness image of claim 1, it is characterized in that: the illustraton of model G=＜V described in the described step (7), E be a non-directed graph, V is the set on summit, E is the set on limit, each vertex v
i∈ V represents an amplitude deepness image, each bar limit e
Ij=(v
i, v
j) ∈ E, and if only if depth image v
iAnd v
jThe T of registration results in twos
IjExist, and T
IjAs limit e
IjAttribute, v wherein
iAnd v
jBe two summits of illustraton of model.
5, according to the described method for auto-registration of multi-amplitude deepness image of claim 1, it is characterized in that: in the described step (8) for obtaining a consistent determination methods of enclosing: from circle, appoint and get an amplitude deepness image, try to achieve its diameter d, with this depth image along the circle in one week of the conversion of conversion in twos, obtain the new depth image of a width of cloth, calculate the error e between new depth image and the initial depth image corresponding point, with the error of e/d as circle, if the error of circle is consistent less than the setting threshold circle, otherwise, be inconsistent.
6, according to the described method for auto-registration of multi-amplitude deepness image of claim 1, it is characterized in that: the method in search turn space is as follows in the described step (9):
Input: the illustraton of model G of registration in twos
Output: the illustraton of model G ' that the overall situation is consistent
Procedure
Obtain the derivation cycle basis B of G
Set up syntople figure Г＜B of B, L 〉
consistent_cycle←Φ
N←1
Number of vertex＞=N among the while Γ
N is communicated with the circle C that the summit generates among for each Γ
The consistent then of ifC
C is inserted among the consistent_cycle
Delete in this N summit
Number of vertex among the If Γ＜N then
break
end?if
end?if
end?for
N←N+1
end?while
G’←Φ
Summit among the G is added among the G '
All limits of all circles among the consistent_cycle are added among the G '.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102220956A CN101350101B (en) | 2008-09-09 | 2008-09-09 | Method for auto-registration of multi-amplitude deepness image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008102220956A CN101350101B (en) | 2008-09-09 | 2008-09-09 | Method for auto-registration of multi-amplitude deepness image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101350101A true CN101350101A (en) | 2009-01-21 |
| CN101350101B CN101350101B (en) | 2011-12-07 |
Family
ID=40268880
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008102220956A Expired - Fee Related CN101350101B (en) | 2008-09-09 | 2008-09-09 | Method for auto-registration of multi-amplitude deepness image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101350101B (en) |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894272A (en) * | 2010-08-10 | 2010-11-24 | 福州展旭电子有限公司 | Automatic matching method of protein spots between two gel images |
| CN102034274A (en) * | 2010-12-10 | 2011-04-27 | 中国人民解放军国防科学技术大学 | Method for calculating corresponding relationship between frame and frame in 3D point cloud sequence |
| CN102034267A (en) * | 2010-11-30 | 2011-04-27 | 中国科学院自动化研究所 | Three-dimensional reconstruction method of target based on attention |
| CN102065313A (en) * | 2010-11-16 | 2011-05-18 | 上海大学 | Uncalibrated multi-viewpoint image correction method for parallel camera array |
| CN102208033A (en) * | 2011-07-05 | 2011-10-05 | 北京航空航天大学 | Data clustering-based robust scale invariant feature transform (SIFT) feature matching method |
| CN101697232B (en) * | 2009-09-18 | 2012-03-07 | 浙江大学 | SIFT characteristic reducing method facing close repeated image matching |
| CN102496183A (en) * | 2011-11-03 | 2012-06-13 | 北京航空航天大学 | Multi-view stereo reconstruction method based on Internet photo gallery |
| CN101807297B (en) * | 2009-02-16 | 2012-07-25 | 海信集团有限公司 | Medical ultrasonic image line detection method |
| CN101567051B (en) * | 2009-06-03 | 2012-08-22 | 复旦大学 | Image matching method based on characteristic points |
| CN103065311A (en) * | 2012-12-26 | 2013-04-24 | 中国土地勘测规划院 | Satellite image automatic registration method based on standard image database |
| CN103065329A (en) * | 2013-01-15 | 2013-04-24 | 西北工业大学 | Space rope tying robot camera automatic motion detection and compensation method |
| CN101989352B (en) * | 2009-08-06 | 2013-05-01 | 宋智礼 | Image registration method based on improved scale invariant feature transform (SIFT) algorithm and Lissajous figure track |
| CN103247040A (en) * | 2013-05-13 | 2013-08-14 | 北京工业大学 | Layered topological structure based map splicing method for multi-robot system |
| CN103390273A (en) * | 2013-07-19 | 2013-11-13 | 哈尔滨工程大学 | Multi-beam side-scan sonar image registration method based on GPS (global positioning system) positioning assistance |
| CN103927742A (en) * | 2014-03-21 | 2014-07-16 | 北京师范大学 | Global automatic registering and modeling method based on depth images |
| CN104217209A (en) * | 2013-06-03 | 2014-12-17 | 核工业北京地质研究院 | Method for automatically eliminating wrongly-matched registration points in remote sensing image |
| CN104680516A (en) * | 2015-01-08 | 2015-06-03 | 南京邮电大学 | Acquisition method for high-quality feature matching set of images |
| CN104778463A (en) * | 2015-05-04 | 2015-07-15 | 福建师范大学 | Remote sensing image urban area extracting method based on SIFT and multi-sub-graph matching |
| CN105139428A (en) * | 2015-08-11 | 2015-12-09 | 鲁东大学 | Quaternion based speeded up robust features (SURF) description method and system for color image |
| CN105787424A (en) * | 2014-12-26 | 2016-07-20 | 阿里巴巴集团控股有限公司 | Image processing method and image processing device |
| CN103927785B (en) * | 2014-04-22 | 2016-08-24 | 同济大学 | A kind of characteristic point matching method towards up short stereoscopic image data |
| CN106204530A (en) * | 2016-06-27 | 2016-12-07 | 西安交通大学 | Ancient times of based on image registration West Xia Dynasty's literary composition typography discrimination method |
| CN106355576A (en) * | 2016-09-08 | 2017-01-25 | 西安电子科技大学 | SAR image registration method based on MRF image segmentation algorithm |
| US9712808B2 (en) | 2015-01-05 | 2017-07-18 | Boe Technology Group Co., Ltd. | Image acquisition device and image processing method and system |
| CN107851303A (en) * | 2015-07-17 | 2018-03-27 | 皇家飞利浦有限公司 | The registration of histopathology image |
| CN110021025A (en) * | 2019-03-29 | 2019-07-16 | 上海联影智能医疗科技有限公司 | The matching of area-of-interest and display methods, device, equipment and storage medium |
| CN110697373A (en) * | 2019-07-31 | 2020-01-17 | 湖北凯瑞知行智能装备有限公司 | Conveying belt deviation fault detection method based on image recognition technology |
| CN113450391A (en) * | 2020-03-26 | 2021-09-28 | 华为技术有限公司 | Method and equipment for generating depth map |
| CN117911287A (en) * | 2024-03-20 | 2024-04-19 | 中国科学院西安光学精密机械研究所 | Interactive splicing and repairing method for large-amplitude wall painting images |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2798761B1 (en) * | 1999-09-17 | 2002-03-29 | Thomson Multimedia Sa | METHOD OF CONSTRUCTING A 3D SCENE MODEL BY ANALYSIS OF IMAGE SEQUENCE |
| CN100388319C (en) * | 2006-07-25 | 2008-05-14 | 深圳大学 | Multi-viewpoint attitude estimating and self-calibrating method for three-dimensional active vision sensor |
| CN100553347C (en) * | 2006-10-26 | 2009-10-21 | 上海交通大学 | The scene depth of stereo visual system recovers and three-dimensional rebuilding method |
-
2008
- 2008-09-09 CN CN2008102220956A patent/CN101350101B/en not_active Expired - Fee Related
Cited By (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101807297B (en) * | 2009-02-16 | 2012-07-25 | 海信集团有限公司 | Medical ultrasonic image line detection method |
| CN101567051B (en) * | 2009-06-03 | 2012-08-22 | 复旦大学 | Image matching method based on characteristic points |
| CN101989352B (en) * | 2009-08-06 | 2013-05-01 | 宋智礼 | Image registration method based on improved scale invariant feature transform (SIFT) algorithm and Lissajous figure track |
| CN101697232B (en) * | 2009-09-18 | 2012-03-07 | 浙江大学 | SIFT characteristic reducing method facing close repeated image matching |
| CN101894272B (en) * | 2010-08-10 | 2012-06-20 | 福州展旭电子有限公司 | Automatic matching method of protein spots between two gel images |
| CN101894272A (en) * | 2010-08-10 | 2010-11-24 | 福州展旭电子有限公司 | Automatic matching method of protein spots between two gel images |
| CN102065313A (en) * | 2010-11-16 | 2011-05-18 | 上海大学 | Uncalibrated multi-viewpoint image correction method for parallel camera array |
| CN102065313B (en) * | 2010-11-16 | 2012-10-31 | 上海大学 | Uncalibrated multi-viewpoint image correction method for parallel camera array |
| CN102034267A (en) * | 2010-11-30 | 2011-04-27 | 中国科学院自动化研究所 | Three-dimensional reconstruction method of target based on attention |
| CN102034274A (en) * | 2010-12-10 | 2011-04-27 | 中国人民解放军国防科学技术大学 | Method for calculating corresponding relationship between frame and frame in 3D point cloud sequence |
| CN102208033A (en) * | 2011-07-05 | 2011-10-05 | 北京航空航天大学 | Data clustering-based robust scale invariant feature transform (SIFT) feature matching method |
| CN102496183A (en) * | 2011-11-03 | 2012-06-13 | 北京航空航天大学 | Multi-view stereo reconstruction method based on Internet photo gallery |
| CN102496183B (en) * | 2011-11-03 | 2013-12-25 | 北京航空航天大学 | Multi-view stereo reconstruction method based on Internet photo gallery |
| CN103065311A (en) * | 2012-12-26 | 2013-04-24 | 中国土地勘测规划院 | Satellite image automatic registration method based on standard image database |
| CN103065329A (en) * | 2013-01-15 | 2013-04-24 | 西北工业大学 | Space rope tying robot camera automatic motion detection and compensation method |
| CN103065329B (en) * | 2013-01-15 | 2015-05-06 | 西北工业大学 | Space rope tying robot camera automatic motion detection and compensation method |
| CN103247040A (en) * | 2013-05-13 | 2013-08-14 | 北京工业大学 | Layered topological structure based map splicing method for multi-robot system |
| CN103247040B (en) * | 2013-05-13 | 2015-11-25 | 北京工业大学 | Based on the multi-robot system map joining method of hierarchical topology structure |
| CN104217209A (en) * | 2013-06-03 | 2014-12-17 | 核工业北京地质研究院 | Method for automatically eliminating wrongly-matched registration points in remote sensing image |
| CN104217209B (en) * | 2013-06-03 | 2017-06-20 | 核工业北京地质研究院 | A kind of Remote sensing image registration point erroneous matching removing method |
| CN103390273A (en) * | 2013-07-19 | 2013-11-13 | 哈尔滨工程大学 | Multi-beam side-scan sonar image registration method based on GPS (global positioning system) positioning assistance |
| CN103390273B (en) * | 2013-07-19 | 2015-12-02 | 哈尔滨工程大学 | A kind of multi-beam side-scan sonar image registration method auxiliary based on GPS location |
| CN103927742A (en) * | 2014-03-21 | 2014-07-16 | 北京师范大学 | Global automatic registering and modeling method based on depth images |
| CN103927785B (en) * | 2014-04-22 | 2016-08-24 | 同济大学 | A kind of characteristic point matching method towards up short stereoscopic image data |
| CN105787424B (en) * | 2014-12-26 | 2019-06-07 | 阿里巴巴集团控股有限公司 | A kind of image processing method and device |
| CN105787424A (en) * | 2014-12-26 | 2016-07-20 | 阿里巴巴集团控股有限公司 | Image processing method and image processing device |
| US9712808B2 (en) | 2015-01-05 | 2017-07-18 | Boe Technology Group Co., Ltd. | Image acquisition device and image processing method and system |
| CN104680516A (en) * | 2015-01-08 | 2015-06-03 | 南京邮电大学 | Acquisition method for high-quality feature matching set of images |
| CN104680516B (en) * | 2015-01-08 | 2017-09-29 | 南京邮电大学 | A kind of acquisition methods of image quality features set of matches |
| CN104778463A (en) * | 2015-05-04 | 2015-07-15 | 福建师范大学 | Remote sensing image urban area extracting method based on SIFT and multi-sub-graph matching |
| CN107851303A (en) * | 2015-07-17 | 2018-03-27 | 皇家飞利浦有限公司 | The registration of histopathology image |
| CN105139428A (en) * | 2015-08-11 | 2015-12-09 | 鲁东大学 | Quaternion based speeded up robust features (SURF) description method and system for color image |
| CN105139428B (en) * | 2015-08-11 | 2018-02-27 | 鲁东大学 | A kind of coloured image SURF character description methods and system based on quaternary number |
| CN106204530A (en) * | 2016-06-27 | 2016-12-07 | 西安交通大学 | Ancient times of based on image registration West Xia Dynasty's literary composition typography discrimination method |
| CN106355576B (en) * | 2016-09-08 | 2019-05-21 | 西安电子科技大学 | SAR image registration method based on MRF image segmentation algorithm |
| CN106355576A (en) * | 2016-09-08 | 2017-01-25 | 西安电子科技大学 | SAR image registration method based on MRF image segmentation algorithm |
| CN110021025A (en) * | 2019-03-29 | 2019-07-16 | 上海联影智能医疗科技有限公司 | The matching of area-of-interest and display methods, device, equipment and storage medium |
| CN110697373A (en) * | 2019-07-31 | 2020-01-17 | 湖北凯瑞知行智能装备有限公司 | Conveying belt deviation fault detection method based on image recognition technology |
| CN113450391A (en) * | 2020-03-26 | 2021-09-28 | 华为技术有限公司 | Method and equipment for generating depth map |
| CN117911287A (en) * | 2024-03-20 | 2024-04-19 | 中国科学院西安光学精密机械研究所 | Interactive splicing and repairing method for large-amplitude wall painting images |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101350101B (en) | 2011-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101350101B (en) | Method for auto-registration of multi-amplitude deepness image | |
| Guo et al. | View-volume network for semantic scene completion from a single depth image | |
| Mian et al. | Three-dimensional model-based object recognition and segmentation in cluttered scenes | |
| Kwitt et al. | Statistical topological data analysis-a kernel perspective | |
| CN101567051B (en) | Image matching method based on characteristic points | |
| Tevs et al. | Isometric registration of ambiguous and partial data | |
| CN104462163B (en) | A kind of threedimensional model characterizing method, search method and searching system | |
| CN104167000B (en) | Affine-invariant wide-baseline image dense matching method | |
| Varley et al. | A new algorithm for finding faces in wireframes | |
| CN105184786B (en) | A kind of floating type triangle character describes method | |
| Yang et al. | Sac-cot: Sample consensus by sampling compatibility triangles in graphs for 3-d point cloud registration | |
| Li et al. | 4FP-structure: A robust local region feature descriptor | |
| Sreevalsan-Nair et al. | Contour extraction in buildings in airborne lidar point clouds using multiscale local geometric descriptors and visual analytics | |
| Li et al. | GGM-net: Graph geometric moments convolution neural network for point cloud shape classification | |
| CN105574535B (en) | Graphic symbol recognition method based on indirect distance angle histogram space relational model | |
| Kustra et al. | Probabilistic view-based 3D curve skeleton computation on the GPU | |
| Chum et al. | Homography estimation from correspondences of local elliptical features | |
| Cao et al. | Stable image matching for 3D reconstruction in outdoor | |
| Biasotti et al. | Quantifying 3D Shape Similarity Using Maps: Recent Trends, Applications and Perspectives. | |
| Xu | An algorithm for comparing similarity between two trees | |
| CN103093011A (en) | CAD (computer aided design) model based feature recognition algorithm | |
| Sun et al. | Object detection, shape recovery, and 3d modelling by depth-encoded hough voting | |
| EP3063737B1 (en) | Method and device for constructing graph representation for a 3d object | |
| Gao et al. | Three Dimensional Reconstruction of Watermelon for Multimedia Traceability System | |
| Lv et al. | Optimisation of real‐scene 3D building models based on straight‐line constraints |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20111207 Termination date: 20140909 |
|
| EXPY | Termination of patent right or utility model |