CN105654422A - Point cloud registration method and system - Google Patents
Point cloud registration method and system Download PDFInfo
- Publication number
- CN105654422A CN105654422A CN201510977586.1A CN201510977586A CN105654422A CN 105654422 A CN105654422 A CN 105654422A CN 201510977586 A CN201510977586 A CN 201510977586A CN 105654422 A CN105654422 A CN 105654422A
- Authority
- CN
- China
- Prior art keywords
- registration
- point
- cloud
- same place
- subject
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 239000011159 matrix material Substances 0.000 claims description 64
- 230000009466 transformation Effects 0.000 claims description 46
- 238000013519 translation Methods 0.000 claims description 23
- 238000009825 accumulation Methods 0.000 claims description 22
- 238000001514 detection method Methods 0.000 claims description 13
- 230000008878 coupling Effects 0.000 claims description 9
- 238000010168 coupling process Methods 0.000 claims description 9
- 238000005859 coupling reaction Methods 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 7
- 230000005484 gravity Effects 0.000 claims description 5
- 230000008569 process Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 230000002452 interceptive effect Effects 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001143 conditioned effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
Classifications
-
- 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
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a point cloud registration method comprising the following steps that multiple same name points are selected from a point cloud under registration through an SIFT algorithm, and global registration is performed on the point cloud under registration by utilizing multiple same name points according to a preset module; and multiple same name points act as index points, and local registration is performed on the point cloud under registration through global registration by adopting a point cloud registration algorithm. According to the point cloud registration method, precision and efficiency of point cloud registration can be substantially enhanced. The invention also provides a point cloud registration system.
Description
Technical field
The present invention relates to airborne laser radar data post-processing technology field, particularly relate to a kind of point cloud registration method and system.
Background technology
In recent years, the extensive concern in the fields such as reverse-engineering, three-dimensional animation, computer vision it is subject to based on the three-dimensional reconstruction of a cloud. It is limited to the field range of scanner itself, scanning body form, sweep limits and the problem such as blocks, the scanning process of objective is generally required and carries out from multi-site or many air strips, then the data of different websites or air strips are carried out splicing registration, make data unified to identical coordinate system. Meanwhile, in the data fusion and application relevant issues of multidate point cloud, owing to the acquisition condition of different time phases cloud data is different, the error features of some cloud is also different, and therefore point cloud registering is a committed step.
Determining some cloud initial error subject to registration based on the point cloud registering of ICP (IterativeClosestPoint, iterative closest point) algorithm can not be too big, otherwise will greatly affect registration efficiency and precision. Needing a rough registration process to provide good initial value for registration point cloud, this process is called global registration. The process that ICP algorithm carries out accuracy registration is adopted to be called local registration. Current global registration method is generally divided into two classes: a class obtains corresponding point based on geometric properties, then calculates position orientation relation; Another kind of employing voting mechanism, such as RANSAC (RandomSampleConsensus, stochastic sampling is consistent) algorithm etc. But, no matter it is based on the algorithm of geometric properties, is also based on the algorithm of voting mechanism, the problem that all there is inefficiency, and also stability and computational accuracy are not high. Additionally, when between two subject to registration somes clouds, deviation is very big, existing point cloud registering mathematical model cannot realize a cloud is carried out accurate registration. And in the local registration stage based on ICP algorithm, mostly adopt stochastical sampling mode to search for closest approach, cause that algorithm efficiency over time and space is all not high.
Summary of the invention
Based on this, it is necessary to provide point cloud registration method and system that a kind of efficiency is higher.
A kind of point cloud registration method, comprises the following steps:
From subject to registration some cloud, choose multiple same place by SIFT algorithm, and utilize multiple described same place, according to preset model, described subject to registration some cloud is carried out global registration;
Using multiple described same places as index, adopt point cloud registration algorithm that the described subject to registration some cloud through described global registration is carried out local registration.
Wherein in an embodiment, described from described subject to registration some cloud, chosen multiple same place by SIFT algorithm, and utilize multiple described same place, according to preset model, described subject to registration some cloud is carried out global registration step to include:
Described subject to registration some cloud is carried out rasterizing, is converted into digital surface model;
Utilize SIFT algorithm that described digital surface model carries out feature detection, generate Feature Descriptor;
Feature Descriptor described in two groups is carried out characteristic matching, generates initial same place pair;
Same place is chosen from described initial same place centering;
Utilize described same place, according to preset model, described subject to registration some cloud is carried out global registration.
Wherein in an embodiment, described choose same place step from described initial same place centering and include:
Utilize the actual three-dimensional distance constraints between same place to described initial same place to screening;
From the described initial same place centering filtered out, interactively mode is adopted to choose same place.
Wherein in an embodiment, described preset model is:
Sample point cloud in described subject to registration some cloud is tentatively translated so that the center of gravity of the same place of two described subject to registration some clouds overlaps;
To described subject to registration some cloud after described preliminary translation, solve spin matrix and the translation vector of translation further.
Wherein in an embodiment, the number of described subject to registration some cloud is two, and using one of them described subject to registration some cloud as template, another described subject to registration some cloud is as sample; Described point cloud registration algorithm is ICP algorithm;
Described using multiple described same places as index, adopt point cloud registration algorithm to carry out local registration step and include:
Choose the point being arranged in described template according to described index point, and be designated as point set S1;
According to index point, choose the point being positioned at described index point position pre-set radius scope in described sample, be designated as point set S2, and described point set S2 is set up KD tree;
Based on described point set S1, choose the nearest point of point set S1 Euclidean distance described in described point set S2 middle-range as corresponding point;
According to described corresponding point, calculate interim transformation matrix M1 based on described preset model, if registration error restrains or reaches maximum iteration time, then adopt accumulation transformation matrix M to treat registration point cloud and convert; Otherwise, described accumulation transformation matrix M is updated based on described interim transformation matrix M1; Wherein, update described accumulation transformation matrix M by M=M1*M, and the initial value of described accumulation transformation matrix M is unit matrix;
According to described interim transformation matrix M1, described point set S2 is converted, generate new point set S2, and return and described described point set S2 is set up KD tree step.
A kind of point cloud registering system, including global registration module and local registration module;
Described global registration module for selecting multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple described same place, according to preset model, described subject to registration some cloud is carried out global registration;
Described local registration module is used for multiple described same places as index, adopts point cloud registration algorithm that the described subject to registration some cloud through described global registration is carried out local registration.
Wherein in an embodiment, described global registration module includes rasterizing unit, characteristic detection unit, characteristic matching unit, same place choose unit and global registration unit;
Described rasterizing unit, for described subject to registration some cloud is carried out rasterizing, is converted into digital surface model;
Described characteristic detection unit is used for utilizing SIFT algorithm that described digital surface model carries out feature detection, generates Feature Descriptor;
Described characteristic matching unit, for Feature Descriptor described in two groups is carried out characteristic matching, generates initial same place pair;
Described same place chooses unit for choosing same place from described initial same place centering;
Described global registration unit is used for utilizing described same place, according to preset model, described subject to registration some cloud is carried out global registration.
Wherein in an embodiment, described same place is chosen unit and is utilized the actual three-dimensional distance constraints between same place to described initial same place to screening, and from the described initial same place centering filtered out, adopts interactively mode to choose same place.
Wherein in an embodiment, described preset model is:
Sample point cloud in described subject to registration some cloud is tentatively translated so that the center of gravity of the same place of two described subject to registration some clouds overlaps;
To described subject to registration some cloud after described preliminary translation, solve spin matrix and the translation vector of translation further.
Wherein in an embodiment, the number of described subject to registration some cloud is two, and using one of them described subject to registration some cloud as template, another described subject to registration some cloud is as sample; Described point cloud registration algorithm is ICP algorithm;
Described local registration module includes that template point chooses unit, sample point chooses unit, corresponding point search unit, coupling performance element and sample point updating block;
Described template point chooses unit for choosing the point being arranged in described template according to described index point, and is designated as point set S1;
Described sample point chooses unit for according to index point, choosing the point being positioned at described index point position pre-set radius scope in described sample, is designated as point set S2, and described point set S2 is set up KD tree;
Described corresponding point search unit is for based on described point set S1, choosing the nearest point of point set S1 Euclidean distance described in described point set S2 middle-range as corresponding point;
Described coupling performance element is for according to described corresponding point, calculating interim transformation matrix M1 based on described preset model, if registration error restrains or reaches maximum iteration time, then adopts accumulation transformation matrix M to treat registration point cloud and converts; Otherwise, call described sample point updating block based on described interim transformation matrix M1, and update described accumulation transformation matrix M; Wherein, update described accumulation transformation matrix M by M=M1*M, and the initial value of described accumulation transformation matrix M is unit matrix;
Described Sample Refreshment unit updates described point set S2 for the described interim transformation matrix M1 drawn according to described coupling performance element, and described point set S2 is set up KD tree, as the input of described corresponding point search unit.
Above-mentioned point cloud registration method and system, choose multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple same place to treat registration point cloud according to preset model to carry out global registration; Then using multiple same places as index, adopt point cloud registration algorithm that the subject to registration some cloud through global registration is carried out local registration, it is possible to significantly increase precision and the efficiency of point cloud registering.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of one embodiment of point cloud registration method of the present invention;
Fig. 2 is the schematic flow sheet of the step S100 in Fig. 1;
Fig. 3 is the schematic flow sheet of the step S140 in Fig. 2;
Fig. 4 is the schematic flow sheet of the step S200 in Fig. 1;
Fig. 5 is the structural representation of one embodiment of point cloud registering system of the present invention;
Fig. 6 is the structural representation of the global registration module in one embodiment of point cloud registering system of the present invention;
Fig. 7 is the structural representation of the local registration module in one embodiment of point cloud registering system of the present invention;
Fig. 8 is two phase point cloud whole structure and partial enlargement effect before registration;
Fig. 9 is the match point adopting interactive mode to obtain after being extracted by SIFT algorithm
Figure 10 puts cloud whole structure and partial enlargement effect after global registration;
Figure 11 is several places profile of final effect after local registration.
Detailed description of the invention
For making the purpose of the present invention, technical scheme and advantage clearly understand, below in conjunction with accompanying drawing, the detailed description of the invention of point cloud registration method of the present invention and system is illustrated. Should be appreciated that specific embodiment described herein is only in order to explain the present invention, is not intended to limit the present invention.
Referring to Fig. 1, in an embodiment, point cloud registration method may comprise steps of:
S100, chooses multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple same place to treat registration point cloud according to preset model to carry out global registration.
Wherein, global registration can provide good initial value for next step local registration. For the problems such as poor stability, registration accuracy are not high in tradition global registration, introduce SIFT (Scale-invariantfeaturetransform, scale invariant feature is changed) algorithm. SIFT algorithm is feature extraction algorithm classical in computer vision.
Referring to Fig. 2, in an embodiment, step S100 can include following step:
S110, treats registration point cloud and carries out rasterizing, be converted into digital surface model.
Concrete, it is possible to according to certain mesh spacing, treat registration point cloud and carry out rasterizing. Subject to registration some cloud after rasterizing is converted into DSM (DigitalSurfaceModel, digital surface model). Wherein, mesh spacing can be configured according to actual needs. The situation that required precision is higher, it is smaller that mesh spacing can be arranged; The situation that required precision is relatively low, it is larger that mesh spacing can be arranged.
S120, utilizes SIFT algorithm that digital surface model carries out feature detection, generates Feature Descriptor.
After subject to registration some cloud is converted into DSM by step S110, utilizes SIFT algorithm that digital surface model is carried out feature detection, and generate Feature Descriptor. This step can be passed through this area routine techniques means and realize, therefore do not repeat them here.
Two stack features are described son and carry out characteristic matching, generate initial same place pair by S130.
Carry out characteristic matching according to pre-conditioned son that two stack features are described, and generate initial same place pair according to the Feature Descriptor matched.
S140, chooses same place from initial same place centering.
Referring to Fig. 3, in an embodiment, step S140 can be realized by following steps:
S141, utilizes the actual three-dimensional distance constraints between same place to initial same place to screening.
It should be understood that it is many to be likely to comparison by the quantity of the step S130 initial same place pair generated, therefore directly choose same place from the step S130 initial same place centering generated, it is necessary to the time of cost can be longer. Therefore, it can the actual three-dimensional distance constraints first passing through between same place, initial same place is screened. By the initial same place of the actual three-dimensional distance constraints met between same place to performing step S142.
S142, from the initial same place centering filtered out, adopts interactively mode to choose same place.
Through step S141 to initial same place to screening after, the initial same place logarithm quantitative change filtered out is few, it is possible to significantly improve efficiency and the precision of global registration.May then pass through the mode of man-machine interactive, choose same place from the initial same place centering filtered out.
S150, utilizes same place, treats registration point cloud according to preset model and carries out global registration.
In traditional point cloud registering model, first obtain spin matrix, then solve translation vector. But, when actual deviation between two subject to registration some clouds is very big, owing to center of rotation and reference center are apart from each other, rotary course are easily caused point cloud registering relatively large deviation occurs. For this, in this step, before obtaining spin matrix, the sample point cloud first treated in registration point cloud tentatively translates so that the center superposition of the same place of subject to registration some cloud. Then again to subject to registration some cloud after tentatively translation, spin matrix and translation vector are solved.
Concrete, it is assumed that for two given some cloud P, Q (P, Q �� R3), with a cloud P for template, some cloud Q is sample.
First, a cloud Q tentatively being translated, the vector of preliminary translation is:Wherein, N is the quantity of same place. Then the vector of each point in the some cloud Q after tentatively translation becomes q 'i=qi+t1. Calculate some cloud P and the covariance matrix of some cloud Q:
Wherein,
Then spin matrix R is calculated according to covariance matrix. And further translation vector is:Then the transformation relation between some cloud P and some cloud Q is: Q'=R (Q+t1)+t2=R Q+R t1+t2. Thus draw transformation matrix:
Wherein, T=R t1+t2��
Above-mentioned preset model may be used for subject to registration the bigger situation of cloud initial deviation, it is also possible to for the situation that subject to registration some cloud initial deviation is less.
S200, using multiple same places as index point, adopts point cloud registration algorithm that the subject to registration some cloud through global registration is carried out local registration.
Wherein, step S100 realizes treating the preliminary registration of registration point cloud, and step S200 realizes treating the accuracy registration of registration point cloud on the basis of step S100.
Referring to Fig. 4, in an embodiment, step S200 may comprise steps of:
S210, chooses the point being arranged in template, and is designated as point set S1 according to index point.
Wherein, the number of subject to registration some cloud is two, respectively first subject to registration some cloud and second subject to registration some cloud. Using first subject to registration some cloud as template, second subject to registration some cloud is as sample. The point being arranged in first subject to registration some cloud in initial same place can be designated as point set S1.
S220, according to index point, chooses the point being positioned at index point position pre-set radius scope in sample, is designated as point set S2, and point set S2 is set up KD tree.
Wherein it is possible to search for, initial same place is arranged in second subject to registration some cloud and the point in predeterminable range scope, using the point that searches as point set S2. Then point set S2 is set up KD (k-dimensional) tree.
S230, based on point set S1, chooses the nearest point of point set S2 middle-range point set S1 Euclidean distance as corresponding point.
Point set S1 can include multiple point. For a point in point set S1, from point set S2, choose this Euclidean distance nearest some corresponding point as this point based on KD tree. For each point in point set S1, all select corresponding point.
S240, according to corresponding point, calculates interim transformation matrix M1 based on preset model, carries out local registration; If registration error restrains or reaches maximum iteration time, then adopt accumulation transformation matrix M to treat registration point cloud and convert; Otherwise, accumulation transformation matrix M is updated based on interim transformation matrix M1.
Wherein it is possible to update accumulation transformation matrix M by M=M1*M. In the present embodiment, the initial value of accumulation transformation matrix M is unit matrix. Under normal circumstances, the initial value of transformation matrix M1 can be quadravalence unit matrix.
S250, converts point set S2 according to interim transformation matrix M1, generates new point set S2, and point set S2 is set up KD tree step by return.
It should be understood that when corresponding point reach certain quantity and distribution comparatively disperses, the quantity increasing corresponding point further can't significantly improve registration accuracy. And in order to improve registration speed, it is possible to use only a part of corresponding point participate in registration. On the other hand when some cloud as rigid body, using point cloud registering as linear transformation time, the change of corresponding point position can't interfere significantly on point cloud registering precision. Based on this, the same place position chosen by SIFT algorithm and man-machine interactive operation is directly adopted to index for the offer of choosing of corresponding point in local registration. Using the key point in same place centering template as known point, using the point within the scope of certain distance in sample as point set to be searched. In registration process, only update the coordinate of point in point set to be searched. Make use of known same place position so on the one hand, avoid the invalid search to Non-overlapping Domain in registration process, what achieve on the other hand a cloud is down-sampled, be conducive to accelerating closest approach search speed, improve registration efficiency, meanwhile, in point cloud registering of many phases, due to two issues according to and inconsistent, adopt increase closest approach index mode be conducive to improve point cloud registering precision.
Above-mentioned point cloud registration method, chooses multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple same place to treat registration point cloud according to preset model to carry out global registration; Then using multiple same places as index, adopt point cloud registration algorithm that the subject to registration some cloud through global registration is carried out local registration, it is possible to significantly increase precision and the efficiency of global registration. And the model of registration has been improved, so that model can be used in subject to registration the bigger situation of cloud deviation.
Based on same inventive concept, the present invention also proposes a kind of point cloud registering system, and this system can be implemented by above-mentioned point cloud registration method, therefore repeats part and repeat no more. Referring to Fig. 5, in an embodiment, point cloud registering system can include global registration module 100 and local registration module 200.
Global registration module 100 for selecting multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple same place to treat registration point cloud according to preset model to carry out global registration. Local registration module 200 is used for multiple same places as index, adopts point cloud registration algorithm that the subject to registration some cloud through global registration is carried out local registration.
Referring to Fig. 6, in an embodiment, global registration module 100 includes rasterizing unit 110, characteristic detection unit 120, characteristic matching unit 130, same place choose unit 140 and global registration unit 150.
Rasterizing unit 110 is used for treating registration point cloud and carries out rasterizing, is converted into digital surface model. Concrete, it is possible to according to certain mesh spacing, treat registration point cloud and carry out rasterizing. Subject to registration some cloud after rasterizing is converted into DSM. Wherein, mesh spacing can be configured according to actual needs. The situation that required precision is higher, it is smaller that mesh spacing can be arranged; The situation that required precision is relatively low, it is larger that mesh spacing can be arranged.
Characteristic detection unit 120 is used for utilizing SIFT algorithm that digital surface model carries out feature detection, generates Feature Descriptor. Characteristic matching unit 130 carries out characteristic matching for two stack features are described son, generates initial same place pair. Same place chooses unit 140 for choosing same place from initial same place centering. Global registration unit 150 is used for utilizing same place, treats registration point cloud according to preset model and carries out global registration.
In one embodiment, same place is chosen unit 140 and the actual three-dimensional distance constraints between same place can be utilized screening to initial same place, and from the initial same place centering filtered out, adopts interactively mode to choose same place. Understandable, it is likely to comparison many by the quantity of the initial same place pair of characteristic matching unit 130 generation, therefore same place is chosen in the initial same place centering directly generated from characteristic matching unit 130, the time needing cost can be longer, inefficient, also can affect the precision of global registration to a certain extent. Therefore, it can the actual three-dimensional distance constraints first passing through between same place, initial same place is screened. Same place is chosen from the initial same place centering of the actual three-dimensional distance constraints met between same place.
Further, the preset model that global registration unit 150 is used is:
The sample point cloud treated in registration point cloud tentatively translates so that the center of gravity of the same place of two subject to registration some clouds overlaps; To subject to registration some cloud after tentatively translation, solve spin matrix and translation vector.
Concrete, it is assumed that for two given some cloud P, Q (P, Q �� R3), with a cloud P for template, some cloud Q is sample.
First, a cloud Q tentatively being translated, the vector of preliminary translation is:Wherein, N is the quantity of same place. Then the vector of each point in the some cloud Q after tentatively translation becomes q 'i=qi+t1. Calculate some cloud P and the covariance matrix of some cloud Q:
Wherein,
Then spin matrix R is calculated according to covariance matrix. And further translation vector is:Then the transformation relation between some cloud P and some cloud Q is: Q'=R (Q+t1)+t2=R Q+R t1+t2. Thus draw transformation matrix:
Wherein, T=R t1+t2��
Above-mentioned preset model may be used for subject to registration the bigger situation of cloud initial deviation, it is also possible to for the situation that subject to registration some cloud initial deviation is less.
Further, point cloud registration algorithm can be ICP algorithm. The number of subject to registration some cloud is two, and using one of them subject to registration some cloud as template, another subject to registration some cloud is as sample. Referring to Fig. 7, in an embodiment, local registration module 200 can include that template point chooses unit 210, sample point chooses unit 220, corresponding point search unit 230, coupling performance element 240 and sample point updating block 250.
Template point chooses unit 210 for choosing the point being arranged in template according to index point, is designated as point set S1.
Sample point chooses unit 220 for according to index point, choosing and be positioned at the point of certain radius scope near index point position in sample, is designated as point set S2, and point set S2 is set up KD tree.
Corresponding point search unit 230 is for based on point set S1, choosing the nearest point of point set S2 middle-range point set S1 Euclidean distance as corresponding point.
Coupling performance element 240 is for according to corresponding point, calculating transformation matrix M1 based on the point cloud registering model proposed in this patent, if registration error restrains or reaches maximum iteration time, then utilizes accumulation transformation matrix M to perform the conversion to registration point cloud;Otherwise, call sample point updating block 250 based on interim transformation matrix M1, and to update accumulation transformation matrix M=M1*M, M initial value be unit matrix.
Sample point updating block 250, for the interim transformation matrix M1 newly obtained according to coupling performance element 240, updates point set S2, and point set S2 is set up KD tree, as the input of corresponding point search unit 230.
Hereinafter adopt test data that point cloud registering system is further described. Test data be by space flight UAS gather different time tilt image data. Fig. 8 to Figure 10 carries out, through three-dimensional reconstruction software, the some cloud that a cloud dense Stereo Matching derives, and data field is positioned at southern china city. Scope of data size respectively may be about 600 �� 600m2With 540 �� 540m2, data volume respectively may be about 480M and 395M, and overlay region area is about 450 �� 500m2, some cloud equalization point is spaced about 0.2m, and two phase point cloud horizontal errors are about 25m, and vertical error is about 12m.
Fig. 8 is whole structure and the partial enlargement effect of 2 cloud Overlapping displays. Fig. 9 illustrates after being extracted by SIFT algorithm, adopts the match point that interactive mode obtains. Having 36 same places pair in figure, the grid distance of DSM is 0.5m. Figure 10 is the result after adopting preset model in this paper to carry out global registration. Figure 11 carries out several places profile of final result after registration further with proposed local registration method. By Fig. 8 to Figure 11 it can be seen that the registration accuracy of above-mentioned point cloud registering system is higher.
Above-mentioned point cloud registering system, chooses multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple same place to treat registration point cloud according to preset model to carry out global registration; Then using multiple same places as index, adopt point cloud registration algorithm that the subject to registration some cloud through global registration is carried out local registration, it is possible to significantly increase precision and the efficiency of global registration. And the model of registration has been improved, so that model can be used in subject to registration the bigger situation of cloud deviation.
Embodiment described above only have expressed the several embodiments of the present invention, and it describes comparatively concrete and detailed, but therefore can not be interpreted as the restriction to the scope of the claims of the present invention. It should be pointed out that, for the person of ordinary skill of the art, without departing from the inventive concept of the premise, it is also possible to making some deformation and improvement, these broadly fall into protection scope of the present invention. Therefore, the protection domain of patent of the present invention should be as the criterion with claims.
Claims (10)
1. a point cloud registration method, it is characterised in that comprise the following steps:
From subject to registration some cloud, choose multiple same place by SIFT algorithm, and utilize multiple described same place, according to preset model, described subject to registration some cloud is carried out global registration;
Using multiple described same places as index point, adopt point cloud registration algorithm that the described subject to registration some cloud through described global registration is carried out local registration.
2. point cloud registration method according to claim 1, it is characterized in that, described from described subject to registration some cloud, chosen multiple same place by SIFT algorithm, and utilize multiple described same place, according to preset model, described subject to registration some cloud is carried out global registration step to include:
Described subject to registration some cloud is carried out rasterizing, is converted into digital surface model;
Utilize SIFT algorithm that described digital surface model carries out feature detection, generate Feature Descriptor;
Feature Descriptor described in two groups is carried out characteristic matching, generates initial same place pair;
Same place is chosen from described initial same place centering;
Utilize described same place, according to preset model, described subject to registration some cloud is carried out global registration.
3. point cloud registration method according to claim 2, it is characterised in that described choose same place step from described initial same place centering and include:
Utilize the actual three-dimensional distance constraints between same place to described initial same place to screening;
From the described initial same place centering filtered out, interactively mode is adopted to choose same place.
4. the point cloud registration method according to claims 1 to 3 any one, it is characterised in that described preset model is:
To described subject to registration in sample point cloud tentatively translate so that the center of gravity of the same place of two described subject to registration some clouds overlaps;
To described subject to registration some cloud after described preliminary translation, solve spin matrix and the translation vector of translation further.
5. point cloud registration method according to claim 4, it is characterised in that the number of described subject to registration some cloud is two, using one of them described subject to registration some cloud as template, another described subject to registration some cloud is as sample; Described point cloud registration algorithm is ICP algorithm;
Described using multiple described same places as index point, adopt point cloud registration algorithm to carry out local registration step and include:
Choose the point being arranged in described template according to described index point, and be designated as point set S1;
According to index point, choose the point being positioned at described index point position pre-set radius scope in described sample, be designated as point set S2, and described point set S2 is set up KD tree;
Based on described point set S1, choose the nearest point of point set S1 Euclidean distance described in described point set S2 middle-range as corresponding point;
According to described corresponding point, calculate interim transformation matrix M1 based on described preset model, if registration error restrains or reaches maximum iteration time, then adopt accumulation transformation matrix M to treat registration point cloud and convert; Otherwise, described accumulation transformation matrix M is updated based on described interim transformation matrix M1; Wherein, update described accumulation transformation matrix M by M=M1*M, and the initial value of described accumulation transformation matrix M is unit matrix;
According to described interim transformation matrix M1, described point set S2 is converted, generate new point set S2, and return and described described point set S2 is set up KD tree step.
6. a point cloud registering system, it is characterised in that include global registration module and local registration module;
Described global registration module for selecting multiple same place by SIFT algorithm from subject to registration some cloud, and utilizes multiple described same place, according to preset model, described subject to registration some cloud is carried out global registration;
Described local registration module is used for multiple described same places as index point, adopts point cloud registration algorithm that the described subject to registration some cloud through described global registration is carried out local registration.
7. point cloud registering system according to claim 6, it is characterised in that described global registration module includes rasterizing unit, characteristic detection unit, characteristic matching unit, same place choose unit and global registration unit;
Described rasterizing unit, for described subject to registration some cloud is carried out rasterizing, is converted into digital surface model;
Described characteristic detection unit is used for utilizing SIFT algorithm that described digital surface model carries out feature detection, generates Feature Descriptor;
Described characteristic matching unit, for Feature Descriptor described in two groups is carried out characteristic matching, generates initial same place pair;
Described same place chooses unit for choosing same place from described initial same place centering;
Described global registration unit is used for utilizing described same place, according to preset model, described subject to registration some cloud is carried out global registration.
8. point cloud registering system according to claim 7, it is characterized in that, described same place is chosen unit and is utilized the actual three-dimensional distance constraints between same place to described initial same place to screening, and from the described initial same place centering filtered out, adopt interactively mode to choose same place.
9. the point cloud registering system according to claim 6 to 8 any one, it is characterised in that described preset model is:
To described subject to registration in sample point cloud tentatively translate so that the center of gravity of the same place of two described subject to registration some clouds overlaps;
To described subject to registration some cloud after described preliminary translation, solve spin matrix and the translation vector of translation further.
10. point cloud registering system according to claim 9, it is characterised in that the number of described subject to registration some cloud is two, using one of them described subject to registration some cloud as template, another described subject to registration some cloud is as sample; Described point cloud registration algorithm is ICP algorithm;
Described local registration module includes that template point chooses unit, sample point chooses unit, corresponding point search unit, coupling performance element and sample point updating block;
Described template point chooses unit for choosing the point being arranged in described template according to described index point, and is designated as point set S1;
Described sample point chooses unit for according to index point, choosing the point being positioned at described index point position pre-set radius scope in described sample, is designated as point set S2, and described point set S2 is set up KD tree;
Described corresponding point search unit is for based on described point set S1, choosing the nearest point of point set S1 Euclidean distance described in described point set S2 middle-range as corresponding point;
Described coupling performance element is for according to described corresponding point, calculating interim transformation matrix M1 based on described preset model, if registration error restrains or reaches maximum iteration time, then adopts accumulation transformation matrix M to treat registration point cloud and converts; Otherwise, call described sample point updating block based on described interim transformation matrix M1, and update described accumulation transformation matrix M; Wherein, update described accumulation transformation matrix M by M=M1*M, and the initial value of described accumulation transformation matrix M is unit matrix;
Described Sample Refreshment unit updates described point set S2 for the described interim transformation matrix M1 drawn according to described coupling performance element, and described point set S2 is set up KD tree, as the input of described corresponding point search unit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510977586.1A CN105654422A (en) | 2015-12-23 | 2015-12-23 | Point cloud registration method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510977586.1A CN105654422A (en) | 2015-12-23 | 2015-12-23 | Point cloud registration method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105654422A true CN105654422A (en) | 2016-06-08 |
Family
ID=56477632
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510977586.1A Pending CN105654422A (en) | 2015-12-23 | 2015-12-23 | Point cloud registration method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105654422A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106204718A (en) * | 2016-06-28 | 2016-12-07 | 华南理工大学 | A kind of simple and efficient 3 D human body method for reconstructing based on single Kinect |
CN107015238A (en) * | 2017-04-27 | 2017-08-04 | 睿舆自动化(上海)有限公司 | Unmanned vehicle autonomic positioning method based on three-dimensional laser radar |
CN108053432A (en) * | 2017-11-14 | 2018-05-18 | 华南理工大学 | The method for registering of indoor sparse cloud scene based on local I CP |
TWI625700B (en) * | 2017-09-15 | 2018-06-01 | 財團法人金屬工業研究發展中心 | Redundant point detection method for point cloud data bonding |
CN108376408A (en) * | 2018-01-30 | 2018-08-07 | 清华大学深圳研究生院 | A kind of three dimensional point cloud based on curvature feature quickly weights method for registering |
CN108595373A (en) * | 2018-01-31 | 2018-09-28 | 中南林业科技大学 | It is a kind of without control DEM method for registering |
CN109146935A (en) * | 2018-07-13 | 2019-01-04 | 中国科学院深圳先进技术研究院 | A kind of point cloud registration method, device, electronic equipment and readable storage medium storing program for executing |
CN109670557A (en) * | 2019-01-03 | 2019-04-23 | 中国矿业大学 | A kind of highway point cloud autoegistration method based on rod-shaped structures |
CN110246184A (en) * | 2018-03-07 | 2019-09-17 | 杭州海康威视数字技术股份有限公司 | Join outside a kind of in-vehicle camera and determines method, apparatus, equipment and system |
CN110443836A (en) * | 2019-06-24 | 2019-11-12 | 中国人民解放军战略支援部队信息工程大学 | A kind of point cloud data autoegistration method and device based on plane characteristic |
CN110850810A (en) * | 2019-11-19 | 2020-02-28 | 中国航空制造技术研究院 | Finish machining registration method based on double-reference constraint |
CN112001955A (en) * | 2020-08-24 | 2020-11-27 | 深圳市建设综合勘察设计院有限公司 | Point cloud registration method and system based on two-dimensional projection plane matching constraint |
CN112446844A (en) * | 2020-11-27 | 2021-03-05 | 广东电网有限责任公司肇庆供电局 | Point cloud feature extraction and registration fusion method |
CN118071583A (en) * | 2024-04-17 | 2024-05-24 | 法奥意威(苏州)机器人***有限公司 | Scaling calculation method, scaling calculation device, electronic equipment and readable storage medium |
CN118071583B (en) * | 2024-04-17 | 2024-07-19 | 法奥意威(苏州)机器人***有限公司 | Scaling calculation method, scaling calculation device, electronic equipment and readable storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101887525A (en) * | 2010-07-09 | 2010-11-17 | 北京师范大学 | Grading-based positive and inverse reversing three-dimensional dense point set rapid registering method |
US20120194646A1 (en) * | 2011-02-02 | 2012-08-02 | National Tsing Hua University | Method of Enhancing 3D Image Information Density |
CN104112289A (en) * | 2014-01-29 | 2014-10-22 | 辽宁师范大学 | Three-dimensional object point cloud registration method based on parallel cascaded EM-ICP |
CN104794678A (en) * | 2015-05-04 | 2015-07-22 | 福建师范大学 | Automatic registration method for high-spatial-resolution remote-sensing images based on SIFI feature points |
-
2015
- 2015-12-23 CN CN201510977586.1A patent/CN105654422A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101887525A (en) * | 2010-07-09 | 2010-11-17 | 北京师范大学 | Grading-based positive and inverse reversing three-dimensional dense point set rapid registering method |
US20120194646A1 (en) * | 2011-02-02 | 2012-08-02 | National Tsing Hua University | Method of Enhancing 3D Image Information Density |
CN104112289A (en) * | 2014-01-29 | 2014-10-22 | 辽宁师范大学 | Three-dimensional object point cloud registration method based on parallel cascaded EM-ICP |
CN104794678A (en) * | 2015-05-04 | 2015-07-22 | 福建师范大学 | Automatic registration method for high-spatial-resolution remote-sensing images based on SIFI feature points |
Non-Patent Citations (1)
Title |
---|
李勇: "基于激光扫描点云数据的建筑物三维重建研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106204718A (en) * | 2016-06-28 | 2016-12-07 | 华南理工大学 | A kind of simple and efficient 3 D human body method for reconstructing based on single Kinect |
CN106204718B (en) * | 2016-06-28 | 2019-08-20 | 华南理工大学 | A kind of simple and efficient 3 D human body method for reconstructing based on single Kinect |
CN107015238A (en) * | 2017-04-27 | 2017-08-04 | 睿舆自动化(上海)有限公司 | Unmanned vehicle autonomic positioning method based on three-dimensional laser radar |
TWI625700B (en) * | 2017-09-15 | 2018-06-01 | 財團法人金屬工業研究發展中心 | Redundant point detection method for point cloud data bonding |
CN108053432B (en) * | 2017-11-14 | 2020-09-22 | 华南理工大学 | Registration method of indoor sparse point cloud scene based on local ICP |
CN108053432A (en) * | 2017-11-14 | 2018-05-18 | 华南理工大学 | The method for registering of indoor sparse cloud scene based on local I CP |
CN108376408A (en) * | 2018-01-30 | 2018-08-07 | 清华大学深圳研究生院 | A kind of three dimensional point cloud based on curvature feature quickly weights method for registering |
CN108595373A (en) * | 2018-01-31 | 2018-09-28 | 中南林业科技大学 | It is a kind of without control DEM method for registering |
CN108595373B (en) * | 2018-01-31 | 2021-07-30 | 中南林业科技大学 | Uncontrolled DEM registration method |
CN110246184B (en) * | 2018-03-07 | 2021-06-11 | 杭州海康威视数字技术股份有限公司 | Method, device, equipment and system for determining external parameters of vehicle-mounted camera |
CN110246184A (en) * | 2018-03-07 | 2019-09-17 | 杭州海康威视数字技术股份有限公司 | Join outside a kind of in-vehicle camera and determines method, apparatus, equipment and system |
CN109146935A (en) * | 2018-07-13 | 2019-01-04 | 中国科学院深圳先进技术研究院 | A kind of point cloud registration method, device, electronic equipment and readable storage medium storing program for executing |
CN109670557A (en) * | 2019-01-03 | 2019-04-23 | 中国矿业大学 | A kind of highway point cloud autoegistration method based on rod-shaped structures |
CN109670557B (en) * | 2019-01-03 | 2023-07-21 | 中国矿业大学 | Automatic highway point cloud registration method based on rod-shaped structures |
CN110443836A (en) * | 2019-06-24 | 2019-11-12 | 中国人民解放军战略支援部队信息工程大学 | A kind of point cloud data autoegistration method and device based on plane characteristic |
CN110850810A (en) * | 2019-11-19 | 2020-02-28 | 中国航空制造技术研究院 | Finish machining registration method based on double-reference constraint |
CN112001955A (en) * | 2020-08-24 | 2020-11-27 | 深圳市建设综合勘察设计院有限公司 | Point cloud registration method and system based on two-dimensional projection plane matching constraint |
CN112446844A (en) * | 2020-11-27 | 2021-03-05 | 广东电网有限责任公司肇庆供电局 | Point cloud feature extraction and registration fusion method |
CN112446844B (en) * | 2020-11-27 | 2024-01-23 | 广东电网有限责任公司肇庆供电局 | Point cloud feature extraction and registration fusion method |
CN118071583A (en) * | 2024-04-17 | 2024-05-24 | 法奥意威(苏州)机器人***有限公司 | Scaling calculation method, scaling calculation device, electronic equipment and readable storage medium |
CN118071583B (en) * | 2024-04-17 | 2024-07-19 | 法奥意威(苏州)机器人***有限公司 | Scaling calculation method, scaling calculation device, electronic equipment and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105654422A (en) | Point cloud registration method and system | |
CN104766084B (en) | A kind of nearly copy image detection method of multiple target matching | |
CN101477529B (en) | Three-dimensional object retrieval method and apparatus | |
Ye et al. | Drinet: A dual-representation iterative learning network for point cloud segmentation | |
CN108648256B (en) | Grayscale rock core three-dimensional reconstruction method based on super-dimension | |
Sahillioglu et al. | 3d shape correspondence by isometry-driven greedy optimization | |
CN106780458B (en) | Point cloud framework extraction method and device | |
CN104299260A (en) | Contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration | |
CN107292919B (en) | A kind of multiple dimensioned settlement place matching process of space-oriented subdivision | |
Yang et al. | An ensemble Wasserstein generative adversarial network method for road extraction from high resolution remote sensing images in rural areas | |
CN104038792B (en) | For the video content analysis method and apparatus of IPTV supervision | |
CN108052863A (en) | Electrical energy power quality disturbance recognition methods based on the maximum variance method of development | |
Zhou et al. | Octr: Octree-based transformer for 3d object detection | |
CN111027140B (en) | Airplane standard part model rapid reconstruction method based on multi-view point cloud data | |
CN104867137A (en) | Improved RANSAC algorithm-based image registration method | |
CN110349170B (en) | Full-connection CRF cascade FCN and K mean brain tumor segmentation algorithm | |
CN104331899A (en) | Registration method and device for SAR image | |
CN107316327A (en) | Knochenbruch section and knochenbruch Model registration method based on maximum public subgraph and bounding box | |
CN109299301A (en) | A kind of method for searching three-dimension model based on distribution of shapes and curvature | |
CN113327276A (en) | Method for registering point cloud data of large volume facing mobile measurement | |
CN109272458A (en) | A kind of point cloud filtering method based on prior information | |
CN102013106A (en) | Image sparse representation method based on Curvelet redundant dictionary | |
CN117274339A (en) | Point cloud registration method based on improved ISS-3DSC characteristics combined with ICP | |
CN106022359A (en) | Fuzzy entropy space clustering analysis method based on orderly information entropy | |
Zou et al. | Automatic 3D point cloud registration algorithm based on triangle similarity ratio consistency |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160608 |
|
RJ01 | Rejection of invention patent application after publication |