CN107507127A - The global registration method and system of multiple views three-dimensional point cloud - Google Patents
The global registration method and system of multiple views three-dimensional point cloud Download PDFInfo
- Publication number
- CN107507127A CN107507127A CN201710660439.0A CN201710660439A CN107507127A CN 107507127 A CN107507127 A CN 107507127A CN 201710660439 A CN201710660439 A CN 201710660439A CN 107507127 A CN107507127 A CN 107507127A
- Authority
- CN
- China
- Prior art keywords
- cloud
- datum mark
- matched
- resampling
- list
- 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
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000012952 Resampling Methods 0.000 claims abstract description 106
- 238000005070 sampling Methods 0.000 claims abstract description 85
- 238000013459 approach Methods 0.000 claims abstract description 81
- 239000011159 matrix material Substances 0.000 claims abstract description 50
- 238000013519 translation Methods 0.000 claims abstract description 48
- 230000009466 transformation Effects 0.000 claims description 30
- 230000011218 segmentation Effects 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012216 screening 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
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The present invention relates to a kind of global registration method and system of multiple views three-dimensional point cloud, including：Obtain dimensional structured cloud of testee multiple views；By sampling grid, resampling single view depth data is obtained to the dimensional structured cloud progress resampling of each viewpoint；The proximity pair list and normal vector list of the to be matched cloud are obtained using the datum mark cloud and to be matched cloud of the resampling single view depth data, rigid body translation matrix is obtained using the datum mark cloud, the proximity pair list and normal vector list.The global registration method of above-mentioned multiple views three-dimensional point cloud, resampling single view depth data is divided by sampling grid, accelerates the lookup speed of to be matched cloud closest approach, and then improve the speed of global registration, even the large-scale point cloud of multiple views, it can also realize high efficiency global registration.
Description
Technical field
The present invention relates to three-dimensional imaging and modeling technology area, more particularly to global of a kind of multiple views three-dimensional point cloud
Method of completing the square and system.
Background technology
In the three-dimensional reconstruction of view-based access control model method, due to being limited and testee screening in itself by sensor field of view
Gear relation can just obtain its complete three-dimensional depth information, multiple views, it is necessary to carry out three-dimensional imaging to object from multiple viewpoints
With being inevitable key link in three-dimensional reconstruction.
Generally utilize the technology such as machine control unit, camera calibration to obtain the initial value of global registration, recycle iteration nearest
Point methods (Iterative closest point, ICP) carry out smart matching.In ICP algorithm, K_D tree etc. are generally utilized
Data structure carries out corresponding points lookup between point of destination cloud and source point cloud, and larger in a cloud, three-dimensional point number is more susceptible
Under condition, corresponding points search efficiency have impact on the speed of whole global registration.
The content of the invention
Based on this, it is necessary to provide a kind of overall situation for the efficient multiple views three-dimensional point cloud that can improve global registration speed
Matching process and system.
A kind of global registration method of multiple views three-dimensional point cloud, including：
Obtain dimensional structured cloud of testee multiple views；
By sampling grid, resampling haplopia is obtained to the dimensional structured cloud progress resampling of each viewpoint
Point depth data；
The to be matched cloud is obtained using the datum mark cloud and to be matched cloud of the resampling single view depth data
Proximity pair list and normal vector list, obtained using the datum mark cloud, the proximity pair list and normal vector list
Rigid body translation matrix.
In one of the embodiments, it is described by sampling grid, the dimensional structured cloud of each viewpoint is entered
Row resampling, which obtains the step of resampling single view depth data, to be included：
The coordinate system on the basis of camera coordinates system, the x/y plane of the equally spaced segmentation dimensional structured cloud, is formed
Uniform sampling grid, with the minimum and maximum coordinate of the x/y plane of the dimensional structured cloud respectively in the x and y direction
As sample range；
The z coordinate of the sampling grid apex is obtained according to the dimensional structured cloud；
The resampling single view depth data is obtained according to sampled result.
In one of the embodiments, it is described that the sampling grid apex is obtained according to the dimensional structured cloud
The step of z coordinate is specially：
According to four adjacent available points in the dimensional structured cloud, the efficiently sampling net of the sampling grid is determined
Case is put；
According to the efficiently sampling grid position, the z coordinate of the bilinear interpolation calculating sampling grid apex is utilized.
In one of the embodiments, the datum mark cloud and to be matched cloud of the resampling single view depth data are utilized
The proximity pair list and normal vector list of the to be matched cloud are obtained, is arranged using the datum mark cloud, the proximity pair
The step of table and normal vector list obtain rigid body translation matrix includes：
Obtain the first datum mark cloud and first to be matched cloud of the resampling single view depth data；
Using initial transformation, the local coordinate system by described first to be matched Cloud transform to the first datum mark cloud
In；
According to each point in the first datum mark cloud and first to be matched cloud in the first datum mark cloud
Position in sampling grid, using bilinear interpolation, obtain each in described first to be matched cloud put in first benchmark
The closest approach position of point cloud, the closest approach position includes z coordinate；
The closest approach position of the first datum mark cloud and first to be matched cloud is all transformed into global coordinate system
In；
The closest approach position of the first datum mark cloud and first to be matched cloud is obtained according to the transformation results
Point to list and normal vector list；
According to the first datum mark cloud and the point of the closest approach position of first to be matched cloud to list and normal direction
List is measured, the first rigid body translation matrix is obtained using least square method.
In one of the embodiments, the datum mark cloud using the resampling single view depth data and to be matched
Point cloud obtains the proximity pair list and normal vector list of the to be matched cloud, utilizes the datum mark cloud, the closest approach
The step of obtaining rigid body translation matrix to list and normal vector list also includes：
Using any one single view depth data of the resampling single view depth data as second to be matched cloud,
Any one other single view depth data of the resampling single view depth data are all used as the second datum mark cloud；
Using initial transformation, described second to be matched Cloud transform is sat to the local of the second datum mark cloud each described
In mark system；
According to each point in each described second datum mark cloud and second to be matched cloud in second benchmark
Position in the sampling grid of point cloud, using bilinear interpolation, obtain each in described second to be matched cloud put at each
The closest approach position of the second datum mark cloud, the closest approach position include z coordinate；
The closest approach position of each described second datum mark cloud and second to be matched cloud is transformed into the overall situation
In coordinate system；
The nearest of each described second datum mark cloud and second to be matched cloud is obtained according to the transformation results
The point of point position is to list and normal vector list；
According to the point to list and normal vector list, the second rigid body translation matrix is obtained using least square method.
In one of the embodiments, it is described according to the point to list and normal vector list, obtained using least square method
The step of also comprising determining whether to meet default matching condition to after the step of the second rigid body translation matrix, if otherwise returning
Described any one single view depth data using the resampling single view depth data is described as second to be matched cloud
Any one other single view depth data of resampling single view depth data, which are all used as the step of the second datum mark cloud, to be continued
Carry out, until meeting the default matching condition.
On the other hand, the present invention also proposes a kind of global registration system of multiple views three-dimensional point cloud, including：
Dimensional structured cloud acquisition module, for obtaining dimensional structured cloud of testee multiple views；
Resampling single view depth data acquisition module, for by sampling grid, being tied to the three-dimensional of each viewpoint
Structure point cloud carries out resampling and obtains resampling single view depth data；
Rigid body translation matrix acquisition module, for the datum mark cloud using the resampling single view depth data and treat
Obtain the proximity pair list and normal vector list of the to be matched cloud with a cloud, using the datum mark cloud, it is described recently
Point obtains rigid body translation matrix to list and normal vector list.
In one of the embodiments, the rigid body translation matrix acquisition module includes：
First to be matched cloud acquisition module, for obtaining the first datum mark cloud of the resampling single view depth data
With first to be matched cloud；
First to be matched Cloud transform module, for using initial transformation, by described first to be matched Cloud transform to institute
In the local coordinate system for stating the first datum mark cloud；
First to be matched cloud closest approach acquisition module, for according to the first datum mark cloud and described first to be matched
Position of each point in the sampling grid of the first datum mark cloud, using bilinear interpolation, obtains described first in point cloud
Each point includes z coordinate in the closest approach position of the first datum mark cloud, the closest approach position in be matched cloud；
First datum mark Cloud transform module, for by the nearest of the first datum mark cloud and first to be matched cloud
Point position is all transformed in global coordinate system；
First datum mark cloud list acquisition module, for obtaining the first datum mark cloud and institute according to the transformation results
The point of the closest approach position of first to be matched cloud is stated to list and normal vector list；
First rigid body translation matrix acquisition module, for according to the first datum mark cloud and first to be matched cloud
Closest approach position point to list and normal vector list, obtain the first rigid body translation matrix using least square method.
In one of the embodiments, the rigid body translation matrix acquisition module also includes：
Second to be matched cloud acquisition module, for by any one single view of the resampling single view depth data
Depth data is as second to be matched cloud, any one other single view depth number of the resampling single view depth data
According to being all used as the second datum mark cloud；
Second to be matched Cloud transform module, for using initial transformation, by described second to be matched Cloud transform to every
In the local coordinate system of one the second datum mark cloud；
Second to be matched cloud closest approach acquisition module, for according to each the second datum mark cloud and described second
Position of each point in the sampling grid of the second datum mark cloud, using bilinear interpolation, obtains institute in be matched cloud
Each point in second to be matched cloud is stated to wrap in the closest approach position of each the second datum mark cloud, the closest approach position
Include z coordinate；
Second datum mark Cloud transform module, for by each described second datum mark cloud and second to be matched cloud
Closest approach position all transform in global coordinate system；
Second datum mark cloud list acquisition module, for obtaining each described second datum mark according to the transformation results
The point of the closest approach position of cloud and second to be matched cloud is to list and normal vector list；
Second rigid body translation matrix acquisition module, for, to list and normal vector list, utilizing a most young waiter in a wineshop or an inn according to the point
Multiplication obtains the second rigid body translation matrix.
In one of the embodiments, the resampling single view depth data acquisition module includes：
Sampling grid module is divided, for the coordinate system on the basis of camera coordinates system, the equally spaced segmentation three-dimensional knot
The x/y plane of structure point cloud, forms uniform sampling grid, with the x/y plane of the dimensional structured cloud respectively in x and y side
Upward minimum and maximum coordinate is as sample range；
Sampling grid summit z coordinate acquisition module, for obtaining the sampling grid according to the dimensional structured cloud
The z coordinate of apex；
Sampled result acquisition module, for obtaining the resampling single view depth data according to sampled result.
The global registration method of above-mentioned multiple views three-dimensional point cloud, by sampling grid, weight is carried out to dimensional structured cloud
New sampling obtains resampling single view depth data；Utilize the datum mark cloud and to be matched cloud of resampling single view depth data
The proximity pair list and normal vector list of to be matched cloud are obtained, is arranged using datum mark cloud and proximity pair list and normal vector
Table obtains rigid body translation matrix.The method divides resampling single view depth data by sampling grid, accelerates point to be matched
The lookup speed of cloud closest approach, and then the speed of global registration is improved, even the large-scale point cloud of multiple views, can also realize
High efficiency global registration.
Brief description of the drawings
Fig. 1 is the flow chart of the global registration method of multiple views three-dimensional point cloud in an embodiment；
Fig. 2 is the flow chart of the global registration method of multiple views three-dimensional point cloud in another embodiment；
Fig. 3 is the flow chart of the global registration method of multiple views three-dimensional point cloud in another embodiment；
Fig. 4 is the flow chart of the global registration method of multiple views three-dimensional point cloud in an embodiment.
Embodiment
For the ease of understanding the present invention, the present invention is described more fully below with reference to relevant drawings.In accompanying drawing
Give the preferred embodiment of the present invention.But the present invention can realize in many different forms, however it is not limited to this paper institutes
The embodiment of description.On the contrary, the purpose for providing these embodiments is made to the disclosure more thorough and comprehensive.
Unless otherwise defined, all of technologies and scientific terms used here by the article is with belonging to technical field of the invention
The implication that technical staff is generally understood that is identical.Term used in the description of the invention herein is intended merely to description tool
The purpose of the embodiment of body, it is not intended that in the limitation present invention.Term as used herein " and/or " include one or more phases
The arbitrary and all combination of the Listed Items of pass.
Referring to Fig. 1, Fig. 1 is the flow chart of the global registration method of multiple views three-dimensional point cloud in an embodiment.
In the present embodiment, the global registration method of the multiple views three-dimensional point cloud includes：
S100, obtain dimensional structured cloud of testee multiple views.
The prototype structure put cloud and possess three dimensional depth picture of the multiple views obtained in the collection of view-based access control model method, referred to as three
Tie up structuring point cloud.The distribution of dimensional structured cloud is similar to the arrangement of image pixel, and each location of pixels is corresponding one
Three-dimensional vertices.The point cloud for the single viewpoint that view-based access control model method is gathered is referred to as single view depth data, single view depth data
Acquiescence is located in camera coordinates system.In one embodiment, each location of pixels also corresponds to a flag bit, is used to specify this
Whether the three-dimensional vertices of position are effective.
S200, by sampling grid, resampling list is obtained to the dimensional structured cloud progress resampling of each viewpoint
Viewpoint depth data.
In camera coordinates system, by sampling grid, resampling is carried out to dimensional structured cloud and obtains resampling list
Viewpoint depth data.
S300, to be matched cloud is obtained most using the datum mark cloud and to be matched cloud of resampling single view depth data
Near point obtains rigid body translation matrix to list and normal vector list using datum mark cloud, proximity pair list and normal vector list.
Utilize the resampling single view depth data obtained in step S200, the base of setting resampling single view depth data
Cloud and to be matched cloud on schedule, closest approach of the every bit of to be matched cloud in datum mark cloud is then found by interpolation, and
The z coordinate of the closest approach is calculated, and then obtains the datum mark cloud and to be matched cloud of resampling single view depth data
Proximity pair list and normal vector list.Using datum mark cloud, proximity pair list and normal vector list, pass through certain algorithm
Obtain rigid body translation matrix.
The global registration method of above-mentioned multiple views three-dimensional point cloud, by obtaining dimensional structured cloud of testee, profit
With sampling grid, resampling is carried out to dimensional structured cloud and obtains resampling single view depth data, utilizes resampling list
The datum mark cloud of viewpoint depth data and to be matched cloud obtain the proximity pair list and normal vector list of to be matched cloud, profit
Rigid body translation matrix is obtained with datum mark cloud, proximity pair list and normal vector list.By being carried out to dimensional structured cloud
Resampling, sampling grid division resampling single view depth data, accelerates the lookup speed of to be matched cloud closest approach, enters
And the speed of global registration is improved, even the large-scale point cloud of multiple views, it can also realize high efficiency global registration.
Referring to Fig. 2, Fig. 2 is the flow chart of the global registration method of multiple views three-dimensional point cloud in another embodiment.
In the present embodiment, the global registration method of the multiple views three-dimensional point cloud includes：
S101, obtain dimensional structured cloud of testee multiple views.
S102, the coordinate system on the basis of camera coordinates system, the x/y plane of dimensional structured cloud of equally spaced segmentation, formed
Uniform sampling grid.
The coordinate system on the basis of camera coordinates system, the x/y plane of dimensional structured cloud of equally spaced segmentation, in x/y plane
Uniform sampling grid is formed, with the minimum and maximum coordinate of the x/y plane of dimensional structured cloud respectively in the x and y direction
As sample range.Sampling interval can be used for the density of control sampling.In one embodiment, due to the structure of sampling grid
Simply, the form of matrix can be used to preserve the result of sampling, each matrix element represents a sampling location.
S103, according to four adjacent available points in dimensional structured cloud, determine the efficiently sampling grid of sampling grid
Position.
In sampling grid and not all sampling location is all effective, adjacent is had according to four in dimensional structured cloud
Imitate point, it may be determined that the efficiently sampling grid position of sampling grid.
S104, according to efficiently sampling grid position, utilize the z coordinate of bilinear interpolation calculating sampling grid apex.
S105, resampling single view depth data is obtained according to sampled result.
The z of sampling grid apex is calculated by the uniform sampling grid of x/y plane in camera coordinates system and bilinear interpolation
Coordinate, obtain the three-dimensional data of resampling single view depth data.In one embodiment, resampling is in whole global registration mistake
Only need to carry out once in journey.
S106, to be matched cloud is obtained most using the datum mark cloud and to be matched cloud of resampling single view depth data
Near point obtains rigid body translation matrix to list and normal vector list using datum mark cloud, proximity pair list and normal vector list.
Resampling single view depth data is obtained using uniform sampling grid, setting resampling single view depth data
Datum mark cloud and to be matched cloud, closest approach of the every bit of to be matched cloud in datum mark cloud is then found by interpolation,
And the z coordinate of the closest approach is calculated, and then obtain the datum mark cloud and to be matched cloud of resampling single view depth data
Proximity pair list and normal vector list.Using datum mark cloud, proximity pair list and normal vector list, pass through certain calculation
Method obtains rigid body translation matrix.
The global registration method of above-mentioned multiple views three-dimensional point cloud, by obtaining dimensional structured cloud of testee, profit
With uniform efficiently sampling grid, resampling is carried out to dimensional structured cloud and obtains resampling single view depth data, profit
Proximity pair list and the method for to be matched cloud are obtained with the datum mark cloud and to be matched cloud of resampling single view depth data
Vector lists, rigid body translation matrix is obtained using datum mark cloud, proximity pair list and normal vector list.By to three-dimensional structure
Change point cloud and carry out resampling, uniform sampling grid divides resampling single view depth data.According to dimensional structured cloud
In four adjacent available points, determine the efficiently sampling grid position of sampling grid, accelerate looking into for be matched cloud closest approach
Speed is looked for, and then improves the speed of global registration, even the large-scale point cloud of multiple views, can also realize global of high efficiency
Match somebody with somebody.
Referring to Fig. 3, Fig. 3 is the flow chart of the global registration method of multiple views three-dimensional point cloud in another embodiment.
In the present embodiment, the global registration method of the multiple views three-dimensional point cloud includes：
S201, obtain dimensional structured cloud of testee multiple views.
S202, by sampling grid, resampling list is obtained to the dimensional structured cloud progress resampling of each viewpoint
Viewpoint depth data.
S203, obtain the first datum mark cloud and first to be matched cloud of resampling single view depth data.
As usual way, the initial value of global registration is obtained by technologies such as machine control unit, camera calibrations.Step
The single view depth data after a pair of resamplings in rapid S202, one of single view depth data is as the first datum mark
Cloud, another single view depth data is as first to be matched cloud.
S204, using initial transformation, by the local coordinate system of first to be matched Cloud transform to the first datum mark cloud.
Coordinate system (local coordinate system) on the basis of camera coordinates system, the first datum mark cloud are located in the frame of reference.Profit
With initial transformation, by the local coordinate system of first to be matched Cloud transform to the first datum mark cloud.If put cloud on the basis of M1 to arrive
The conversion of global coordinate system, M1.inverse are M1 inverse transformation；M2 be to be matched cloud to the conversion of global coordinate system, then treat
The conversion M of match point cloud to the local coordinate system of datum mark cloud is：M=M1.inverse*M2.
S205, according to each o'clock sampling grid in the first datum mark cloud in the first datum mark cloud and first to be matched cloud
In position, using bilinear interpolation, obtain in first to be matched cloud at each o'clock in the closest approach position of the first datum mark cloud,
Closest approach position includes z coordinate.
After the local coordinate system of first to be matched Cloud transform to the first datum mark cloud, first to be matched cloud is calculated
In each o'clock position in the sampling grid of the first datum mark cloud, it is evident that only positioned at the first datum mark cloud sampling grid model
Position in enclosing is only active position, in that context it may be convenient to rejects the point outside scope.Using any point of first to be matched cloud as
Example, using position of this o'clock in the sampling grid of the first datum mark cloud, obtain a new point conduct with reference to bilinear interpolation and treat
One closest approach of match point cloud, while the normal vector of the new point is calculated.It so can be obtained by first to be matched cloud
In include z coordinate in the closest approach position of the first datum mark cloud, closest approach position at each o'clock.In one embodiment, apart from threshold
Value is sentenced away from can be used to judge whether the closest approach position is effective with normal vector angle.
S206, the closest approach position of the first datum mark cloud and first to be matched cloud is all transformed in global coordinate system.
After the closest approach position of first to be matched cloud all obtains, by the first datum mark cloud and first to be matched cloud
Closest approach position all transform in global coordinate system.
S207, the point of closest approach position of the first datum mark cloud and first to be matched cloud is obtained according to transformation results to row
Table and normal vector list.
S208, list and normal vector are arranged according to the first datum mark cloud and the point of the closest approach position of first to be matched cloud
Table, the first rigid body translation matrix is obtained using least square method.In one embodiment, resampling single view depth data and right
The rigid body translation answered is retained separately.
The global registration method of above-mentioned multiple views three-dimensional point cloud, by carrying out resampling to dimensional structured cloud,
Even sampling grid division resampling single view depth data.Part by first to be matched Cloud transform to the first datum mark cloud
In coordinate system, the lookup speed of to be matched cloud closest approach is accelerated, and then improves the speed of global registration, even regard more
The large-scale point cloud of point, can also realize high efficiency global registration.
Referring to Fig. 4, Fig. 4 is the flow chart of the global registration method of multiple views three-dimensional point cloud in an embodiment.
In the present embodiment, the global registration method of the multiple views three-dimensional point cloud includes：
S301, obtain dimensional structured cloud of testee multiple views.
S302, by sampling grid, resampling list is obtained to the dimensional structured cloud progress resampling of each viewpoint
Viewpoint depth data.
S303, using any one single view depth data of resampling single view depth data as the second point to be matched
Cloud, any one other single view depth data of resampling single view depth data are all used as the second datum mark cloud.
We regard the point cloud of all viewpoints of resampling single view depth data the engineering of an entirety as, comprising from 1
To n point cloud.Using any one resampling single view depth data as second to be matched cloud, any one other resampling
Single view depth data is all used as the second datum mark cloud.Specifically, the 1st resampling single view depth data is treated as second
Match point cloud, using other (the 2nd to n-th) all resampling single view depth datas all as the second datum mark cloud；By the 2nd
As second to be matched cloud, all by other (the 1st, the 3rd to n-th) heavy adopt individual resampling single view depth data
Sample single view depth data is all used as the second datum mark cloud；..., carry out successively, until by n-th of resampling single view depth
Data are as second to be matched cloud, using other (the 1st to (n-1)th) all resampling single view depth datas all as
Two datum mark clouds.
S304, using initial transformation, the local coordinate by second to be matched Cloud transform to each the second datum mark cloud
In system.
Coordinate system (local coordinate system) on the basis of camera coordinates system, the second datum mark cloud are located in the frame of reference.Profit
With initial transformation, by the local coordinate system of second to be matched Cloud transform to each the second datum mark cloud.If on the basis of M1
For point cloud to the conversion of global coordinate system, M1.inverse is M1 inverse transformation；M2 is change of the to be matched cloud to global coordinate system
Change, then the conversion M of to be matched cloud to the local coordinate system of datum mark cloud is：M=M1.inverse*M2.
S305, according to each o'clock adopting in the second datum mark cloud in each second datum mark cloud and second to be matched cloud
Position in sample grid, using bilinear interpolation, obtain each in second to be matched cloud put in each the second datum mark cloud
Closest approach position, closest approach position includes z coordinate.
After the local coordinate system of second to be matched Cloud transform to each the second datum mark cloud, calculate second and treat
With position of each point in the sampling grid of each the second datum mark cloud in a cloud, it is evident that only positioned at the second datum mark
Position in the range of cloud sampling grid is only active position, in that context it may be convenient to rejects the point outside scope.With second to be matched cloud
Any point exemplified by, using position of this o'clock in the sampling grid of the second datum mark cloud, one is obtained with reference to bilinear interpolation
A closest approach of the individual new point as to be matched cloud, while the normal vector of the new point is calculated.It so can be obtained by
Each point includes z coordinate in the closest approach position of each the second datum mark cloud, closest approach position in two to be matched clouds.One
In individual embodiment, distance threshold is sentenced away from can be used to judge whether the closest approach position is effective with normal vector angle.
S306, the closest approach position of each second datum mark cloud and second to be matched cloud is transformed into world coordinates
In system.
After second to be matched cloud all obtains in the closest approach position of each the second datum mark cloud, by each
The closest approach position of two datum mark clouds and second to be matched cloud is all transformed in global coordinate system.
S307, the closest approach position of each second datum mark cloud and second to be matched cloud is obtained according to transformation results
Point is to list and normal vector list.
According to the transformation results in step S306, each second datum mark cloud and corresponding second to be matched cloud are obtained
Closest approach position point to list and normal vector list.
S308, according to point to list and normal vector list, the second rigid body translation matrix is obtained using least square method.
According to being obtained in step S307 to list and normal vector list, the second rigid body translation is obtained using least square method
Matrix, into step S309.
S309, judge whether to meet default matching condition.
If meeting default matching condition, this global registration terminates, and otherwise return to step S303 continues, until
Meet default matching condition.Regard step S303 to step S308 as a whole matching, it is in one embodiment, default
Matching condition is maximum whole matching number.In another embodiment, default matching condition is to meet matching threshold condition,
Threshold condition can be that the distance between closest approach of the second datum mark cloud and second to be matched cloud meets default distance,
Can be the second rigid body translation matrix variation effect it is sufficiently small.
The global registration method of above-mentioned multiple views three-dimensional point cloud, by carrying out resampling to dimensional structured cloud,
Even sampling grid division resampling single view depth data, accelerate the lookup speed of to be matched cloud closest approach.By more
Whole matching is taken turns, can the effectively average distribution alignment between the point cloud of all viewpoints of resampling single view depth data
Error, and then the speed of global registration is improved, even the large-scale point cloud of multiple views, it can also realize global of high efficiency
Match somebody with somebody.
On the other hand, the present invention also proposes a kind of global registration system of multiple views three-dimensional point cloud, including：
Dimensional structured cloud acquisition module, for obtaining dimensional structured cloud of testee multiple views；
Resampling single view depth data acquisition module, for by sampling grid, to the dimensional structured of each viewpoint
Point cloud carries out resampling and obtains resampling single view depth data；
Rigid body translation matrix acquisition module, for the datum mark cloud using resampling single view depth data and point to be matched
Cloud obtains the proximity pair list and normal vector list of to be matched cloud, utilizes datum mark cloud, proximity pair list and normal vector
List obtains rigid body translation matrix.
In one embodiment, rigid body translation matrix acquisition module includes：
First to be matched cloud acquisition module, for obtaining the first datum mark cloud and of resampling single view depth data
One to be matched cloud；
First to be matched Cloud transform module, for using initial transformation, by first to be matched Cloud transform to the first base
On schedule in the local coordinate system of cloud；
First to be matched cloud closest approach acquisition module, for according to every in the first datum mark cloud and first to be matched cloud
The individual o'clock position in the sampling grid of the first datum mark cloud, using bilinear interpolation, obtain each in first to be matched cloud
O'clock include z coordinate in the closest approach position of the first datum mark cloud, the closest approach position；
First datum mark Cloud transform module, for by the closest approach position of the first datum mark cloud and first to be matched cloud all
Transform in global coordinate system；
First datum mark cloud list acquisition module, for obtaining the first datum mark cloud and first to be matched according to transformation results
The point of the closest approach position of point cloud is to list and normal vector list；
First rigid body translation matrix acquisition module, for the closest approach according to the first datum mark cloud and first to be matched cloud
The point of position obtains the first rigid body translation matrix to list and normal vector list using least square method.
In one embodiment, rigid body translation matrix acquisition module also includes：
Second to be matched cloud acquisition module, for by any one single view depth of resampling single view depth data
Data are as second to be matched cloud, any one other single view depth data all conduct of resampling single view depth data
Second datum mark cloud；
Second to be matched Cloud transform module, for using initial transformation, by second to be matched Cloud transform to each
In the local coordinate system of second datum mark cloud；
A second to be matched cloud closest approach acquisition module, for being treated according to each described second datum mark cloud and second
With each o'clock position in the sampling grid of the second datum mark cloud in a cloud, using bilinear interpolation, it is to be matched to obtain second
Each point includes z coordinate in the closest approach position of each the second datum mark cloud, closest approach position in point cloud；
Second datum mark Cloud transform module, for by the closest approach of each the second datum mark cloud and second to be matched cloud
Position is all transformed in global coordinate system；
Second datum mark cloud list acquisition module, for obtaining each second datum mark cloud and second according to transformation results
The point of the closest approach position of to be matched cloud is to list and normal vector list；
Second rigid body translation matrix acquisition module, for, to list and normal vector list, utilizing least square method according to point
Obtain the second rigid body translation matrix.
In one embodiment, resampling single view depth data acquisition module includes：
Sampling grid module is divided, for the coordinate system on the basis of camera coordinates system, equally spaced segmentation is dimensional structured
The x/y plane of cloud is put, forms uniform sampling grid, is distinguished in the x and y direction most with the x/y plane of dimensional structured cloud
Big and minimum coordinate is as sample range；
Sampling grid summit z coordinate acquisition module, for obtaining the z of sampling grid apex according to dimensional structured cloud
Coordinate；
Sampled result acquisition module, for obtaining resampling single view depth data according to sampled result.
Each technical characteristic of embodiment described above can be combined arbitrarily, to make description succinct, not to above-mentioned reality
Apply all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, the scope that this specification is recorded all is considered to be.
Embodiment described above only expresses the several embodiments of the present invention, and its description is more specific and detailed, but simultaneously
Can not therefore it be construed as limiting the scope of the patent.It should be pointed out that come for one of ordinary skill in the art
Say, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the protection of the present invention
Scope.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (10)
- A kind of 1. global registration method of multiple views three-dimensional point cloud, it is characterised in that including：Obtain dimensional structured cloud of testee multiple views；By sampling grid, resampling single view depth is obtained to the dimensional structured cloud progress resampling of each viewpoint Degrees of data；The to be matched cloud is obtained using the datum mark cloud and to be matched cloud of the resampling single view depth data most Near point obtains rigid body to list and normal vector list using the datum mark cloud, the proximity pair list and normal vector list Transformation matrix.
- 2. the global registration method of multiple views three-dimensional point cloud according to claim 1, it is characterised in that described to pass through sampling Grid, the step of resampling obtains resampling single view depth data is carried out to the dimensional structured cloud of each viewpoint Including：The coordinate system on the basis of camera coordinates system, the x/y plane of the equally spaced segmentation dimensional structured cloud, is formed uniform Sampling grid, using the minimum and maximum coordinate of the x/y plane of the dimensional structured cloud respectively in the x and y direction as Sample range；The z coordinate of the sampling grid apex is obtained according to the dimensional structured cloud；The resampling single view depth data is obtained according to sampled result.
- 3. the global registration method of multiple views three-dimensional point cloud according to claim 2, it is characterised in that described in the basis Dimensional structured cloud obtains the step of z coordinate of the sampling grid apex and is specially：According to four adjacent available points in the dimensional structured cloud, the efficiently sampling grid position of the sampling grid is determined Put；According to the efficiently sampling grid position, the z coordinate of the bilinear interpolation calculating sampling grid apex is utilized.
- 4. the global registration method of multiple views three-dimensional point cloud according to claim 1, it is characterised in that heavy adopted using described The datum mark cloud of sample single view depth data and to be matched cloud obtain proximity pair list and the normal direction of the to be matched cloud The step of measuring list, rigid body translation matrix is obtained using the datum mark cloud, the proximity pair list and normal vector list is wrapped Include：Obtain the first datum mark cloud and first to be matched cloud of the resampling single view depth data；Using initial transformation, by the local coordinate system of described first to be matched Cloud transform to the first datum mark cloud；According to each point in the first datum mark cloud and first to be matched cloud in the sampling of the first datum mark cloud Position in grid, using bilinear interpolation, obtain each in described first to be matched cloud put in the first datum mark cloud Closest approach position, the closest approach position includes z coordinate；The closest approach position of the first datum mark cloud and first to be matched cloud is all transformed in global coordinate system；The point of the closest approach position of the first datum mark cloud and first to be matched cloud is obtained according to the transformation results To list and normal vector list；List and normal vector are arranged according to the first datum mark cloud and the point of the closest approach position of first to be matched cloud Table, the first rigid body translation matrix is obtained using least square method.
- 5. the global registration method of multiple views three-dimensional point cloud according to claim 1, it is characterised in that described in the utilization The datum mark cloud of resampling single view depth data and to be matched cloud obtain the to be matched cloud proximity pair list and Normal vector list, the step of rigid body translation matrix is obtained using the datum mark cloud, the proximity pair list and normal vector list Suddenly also include：It is described using any one single view depth data of the resampling single view depth data as second to be matched cloud Any one other single view depth data of resampling single view depth data are all used as the second datum mark cloud；Using initial transformation, the local coordinate system by described second to be matched Cloud transform to the second datum mark cloud each described In；According to each point in each described second datum mark cloud and second to be matched cloud in the second datum mark cloud Sampling grid in position, using bilinear interpolation, obtain in described second to be matched cloud each point described in each The closest approach position of second datum mark cloud, the closest approach position include z coordinate；The closest approach position of each described second datum mark cloud and second to be matched cloud is transformed into world coordinates In system；The closest approach position of each described second datum mark cloud and second to be matched cloud is obtained according to the transformation results The point put is to list and normal vector list；According to the point to list and normal vector list, the second rigid body translation matrix is obtained using least square method.
- 6. the global registration method of multiple views three-dimensional point cloud according to claim 5, it is characterised in that described in the basis Point is also including judgement after the step of obtaining the second rigid body translation matrix using least square method to list and normal vector list No the step of meeting default matching condition, if otherwise returning to any one by the resampling single view depth data Single view depth data is as second to be matched cloud, any one other single view of the resampling single view depth data Depth data, which is all used as the step of the second datum mark cloud, to be continued, until meeting the default matching condition.
- A kind of 7. global registration system of multiple views three-dimensional point cloud, it is characterised in that including：Dimensional structured cloud acquisition module, for obtaining dimensional structured cloud of testee multiple views；Resampling single view depth data acquisition module, for by sampling grid, to the described dimensional structured of each viewpoint Point cloud carries out resampling and obtains resampling single view depth data；Rigid body translation matrix acquisition module, for the datum mark cloud using the resampling single view depth data and point to be matched Cloud obtains the proximity pair list and normal vector list of the to be matched cloud, utilizes the datum mark cloud, the proximity pair List and normal vector list obtain rigid body translation matrix.
- 8. the global registration system of multiple views three-dimensional point cloud according to claim 7, it is characterised in that the rigid body translation Matrix acquisition module includes：First to be matched cloud acquisition module, for obtaining the first datum mark cloud and of the resampling single view depth data One to be matched cloud；First to be matched Cloud transform module, for using initial transformation, by described first to be matched Cloud transform to described In the local coordinate system of one datum mark cloud；First to be matched cloud closest approach acquisition module, for according to the first datum mark cloud and first to be matched cloud In position of each point in the sampling grid of the first datum mark cloud, using bilinear interpolation, obtain described first and treat Include z coordinate in the closest approach position of the first datum mark cloud, the closest approach position with each point in a cloud；First datum mark Cloud transform module, for by the closest approach position of the first datum mark cloud and first to be matched cloud Put and all transform in global coordinate system；First datum mark cloud list acquisition module, for obtaining the first datum mark cloud and described according to the transformation results The point of the closest approach position of one to be matched cloud is to list and normal vector list；First rigid body translation matrix acquisition module, for according to the first datum mark cloud and first to be matched cloud most The point of near point position obtains the first rigid body translation matrix to list and normal vector list using least square method.
- 9. the global registration system of multiple views three-dimensional point cloud according to claim 7, it is characterised in that the rigid body translation Matrix acquisition module also includes：Second to be matched cloud acquisition module, for by any one single view depth of the resampling single view depth data Data are as second to be matched cloud, and any one other single view depth data of the resampling single view depth data are all As the second datum mark cloud；Second to be matched Cloud transform module, for using initial transformation, by described second to be matched Cloud transform to each In the local coordinate system of the second datum mark cloud；A second to be matched cloud closest approach acquisition module, for being treated according to each described second datum mark cloud and described second With position of each point in the sampling grid of the second datum mark cloud in a cloud, using bilinear interpolation, described the is obtained Each point includes z in the closest approach position of each the second datum mark cloud, the closest approach position in two to be matched clouds Coordinate；Second datum mark Cloud transform module, for by each described second datum mark cloud and second to be matched cloud most Near point position is all transformed in global coordinate system；Second datum mark cloud list acquisition module, for according to the transformation results obtain each described second datum mark cloud and The point of the closest approach position of second to be matched cloud is to list and normal vector list；Second rigid body translation matrix acquisition module, for, to list and normal vector list, utilizing least square method according to the point Obtain the second rigid body translation matrix.
- 10. the global registration system of multiple views three-dimensional point cloud according to claim 7, it is characterised in that the resampling Single view depth data acquisition module includes：Sampling grid module is divided, for the coordinate system on the basis of camera coordinates system, equally spaced segmentation is described dimensional structured The x/y plane of point cloud, forms uniform sampling grid, with the x/y plane difference of the dimensional structured cloud in the x and y direction Minimum and maximum coordinate as sample range；Sampling grid summit z coordinate acquisition module, for obtaining the sampling grid summit according to the dimensional structured cloud The z coordinate at place；Sampled result acquisition module, for obtaining the resampling single view depth data according to sampled result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710660439.0A CN107507127B (en) | 2017-08-04 | 2017-08-04 | Global matching method and system for multi-viewpoint three-dimensional point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710660439.0A CN107507127B (en) | 2017-08-04 | 2017-08-04 | Global matching method and system for multi-viewpoint three-dimensional point cloud |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107507127A true CN107507127A (en) | 2017-12-22 |
| CN107507127B CN107507127B (en) | 2021-01-22 |
Family
ID=60690422
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710660439.0A Active CN107507127B (en) | 2017-08-04 | 2017-08-04 | Global matching method and system for multi-viewpoint three-dimensional point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107507127B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053481A (en) * | 2017-12-26 | 2018-05-18 | 深圳市易尚展示股份有限公司 | Generation method, device and the storage medium of three-dimensional point cloud normal vector |
| CN108346165A (en) * | 2018-01-30 | 2018-07-31 | 深圳市易尚展示股份有限公司 | Robot and three-dimensional sensing components in combination scaling method and device |
| CN109493375A (en) * | 2018-10-24 | 2019-03-19 | 深圳市易尚展示股份有限公司 | The Data Matching and merging method of three-dimensional point cloud, device, readable medium |
| KR20190134030A (en) * | 2018-05-24 | 2019-12-04 | 주식회사 이누씨 | Method And Apparatus Creating for Avatar by using Multi-view Image Matching |
| CN110555085A (en) * | 2018-03-29 | 2019-12-10 | 中国石油化工股份有限公司 | Three-dimensional model loading method and device |
| WO2020017668A1 (en) * | 2018-07-16 | 2020-01-23 | 주식회사 이누씨 | Method and apparatus for generating avatar by using multi-view image matching |
| CN115222787A (en) * | 2022-09-20 | 2022-10-21 | 天津中科智能技术研究院有限公司 | Real-time point cloud registration method based on mixed retrieval |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101127123A (en) * | 2007-09-11 | 2008-02-20 | 东南大学 | Sign point hole filling method based on neural network in tri-D scanning point cloud |
| CN103093498A (en) * | 2013-01-25 | 2013-05-08 | 西南交通大学 | Three-dimensional human face automatic standardization method |
| CN104217458A (en) * | 2014-08-22 | 2014-12-17 | 长沙中科院文化创意与科技产业研究院 | Quick registration method for three-dimensional point clouds |
| CN105654483A (en) * | 2015-12-30 | 2016-06-08 | 四川川大智胜软件股份有限公司 | Three-dimensional point cloud full-automatic registration method |
-
2017
- 2017-08-04 CN CN201710660439.0A patent/CN107507127B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101127123A (en) * | 2007-09-11 | 2008-02-20 | 东南大学 | Sign point hole filling method based on neural network in tri-D scanning point cloud |
| CN103093498A (en) * | 2013-01-25 | 2013-05-08 | 西南交通大学 | Three-dimensional human face automatic standardization method |
| CN104217458A (en) * | 2014-08-22 | 2014-12-17 | 长沙中科院文化创意与科技产业研究院 | Quick registration method for three-dimensional point clouds |
| CN105654483A (en) * | 2015-12-30 | 2016-06-08 | 四川川大智胜软件股份有限公司 | Three-dimensional point cloud full-automatic registration method |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053481A (en) * | 2017-12-26 | 2018-05-18 | 深圳市易尚展示股份有限公司 | Generation method, device and the storage medium of three-dimensional point cloud normal vector |
| CN108053481B (en) * | 2017-12-26 | 2021-11-30 | 深圳市易尚展示股份有限公司 | Method and device for generating three-dimensional point cloud normal vector and storage medium |
| CN108346165B (en) * | 2018-01-30 | 2020-10-30 | 深圳市易尚展示股份有限公司 | Robot and three-dimensional sensing assembly combined calibration method and device |
| CN108346165A (en) * | 2018-01-30 | 2018-07-31 | 深圳市易尚展示股份有限公司 | Robot and three-dimensional sensing components in combination scaling method and device |
| CN110555085B (en) * | 2018-03-29 | 2022-01-14 | 中国石油化工股份有限公司 | Three-dimensional model loading method and device |
| CN110555085A (en) * | 2018-03-29 | 2019-12-10 | 中国石油化工股份有限公司 | Three-dimensional model loading method and device |
| KR20190134030A (en) * | 2018-05-24 | 2019-12-04 | 주식회사 이누씨 | Method And Apparatus Creating for Avatar by using Multi-view Image Matching |
| KR102068993B1 (en) * | 2018-05-24 | 2020-01-22 | 주식회사 이누씨 | Method And Apparatus Creating for Avatar by using Multi-view Image Matching |
| WO2020017668A1 (en) * | 2018-07-16 | 2020-01-23 | 주식회사 이누씨 | Method and apparatus for generating avatar by using multi-view image matching |
| CN109493375B (en) * | 2018-10-24 | 2021-01-12 | 深圳市易尚展示股份有限公司 | Data matching and merging method and device for three-dimensional point cloud and readable medium |
| CN109493375A (en) * | 2018-10-24 | 2019-03-19 | 深圳市易尚展示股份有限公司 | The Data Matching and merging method of three-dimensional point cloud, device, readable medium |
| CN115222787A (en) * | 2022-09-20 | 2022-10-21 | 天津中科智能技术研究院有限公司 | Real-time point cloud registration method based on mixed retrieval |
| CN115222787B (en) * | 2022-09-20 | 2023-01-10 | 宜科(天津)电子有限公司 | Real-time point cloud registration method based on hybrid retrieval |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107507127B (en) | 2021-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107507127A (en) | The global registration method and system of multiple views three-dimensional point cloud | |
| CN109447994B (en) | Remote sensing image segmentation method combining complete residual error and feature fusion | |
| Burns et al. | Comparison of commercial structure-from-motion photogrammety software used for underwater three-dimensional modeling of coral reef environments | |
| CN108875902A (en) | Neural network training method and device, vehicle detection estimation method and device, storage medium | |
| Patil et al. | Image classification tool for land use/land cover analysis: A comparative study of maximum likelihood and minimum distance method | |
| Kim et al. | Robust vehicle localization using entropy-weighted particle filter-based data fusion of vertical and road intensity information for a large scale urban area | |
| CN113052109A (en) | 3D target detection system and 3D target detection method thereof | |
| CN108401461A (en) | Three-dimensional mapping method, device and system, cloud platform, electronic equipment and computer program product | |
| CN103258345A (en) | Method for extracting parameters of tree branches based on ground laser radar three-dimensional scanning | |
| CN108280866A (en) | Road Processing Method of Point-clouds and system | |
| CN108010123A (en) | A kind of three-dimensional point cloud acquisition methods for retaining topology information | |
| Badías et al. | An augmented reality platform for interactive aerodynamic design and analysis | |
| CN114918918B (en) | Domain-containing self-adaptive robot disordered target pushing and grabbing method | |
| CN109902141A (en) | The method and autonomous agents of motion planning | |
| CN110998664A (en) | Computing representative shapes for a set of polygons | |
| CN107341837A (en) | Grid and vector data conversion and continuous yardstick expression based on image pyramid | |
| Naserentin et al. | Combining Open source and commercial tools in Digital Twin for cities generation | |
| CN106910218B (en) | Earth's surface surface area calculation method based on big region grid digital elevation model | |
| Liu et al. | A new multi-channel deep convolutional neural network for semantic segmentation of remote sensing image | |
| Verykokou et al. | A Comparative analysis of different software packages for 3D Modelling of complex geometries | |
| CN106210696A (en) | A kind of method and device of real-time virtual View Synthesis | |
| Dyken et al. | Real‐Time GPU Silhouette Refinement using Adaptively Blended Bézier Patches | |
| CN114868154A (en) | Map comprising covariance in multi-resolution voxels | |
| CN101271596A (en) | Method for establishing digital model of landform curved surface based on basic law of theory of surfaces | |
| KR102416714B1 (en) | System and method for city-scale tree mapping using 3-channel images and multiple deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |