CN110211025A - For the bundle adjustment method of image mosaic, storage medium and calculate equipment - Google Patents
For the bundle adjustment method of image mosaic, storage medium and calculate equipment Download PDFInfo
- Publication number
- CN110211025A CN110211025A CN201910337080.2A CN201910337080A CN110211025A CN 110211025 A CN110211025 A CN 110211025A CN 201910337080 A CN201910337080 A CN 201910337080A CN 110211025 A CN110211025 A CN 110211025A
- Authority
- CN
- China
- Prior art keywords
- transformation matrix
- gradient
- virtual target
- adjustment method
- error function
- 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 65
- 239000011159 matrix material Substances 0.000 claims abstract description 161
- 230000009466 transformation Effects 0.000 claims abstract description 156
- 238000009825 accumulation Methods 0.000 claims abstract description 43
- 238000009434 installation Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 66
- 230000007704 transition Effects 0.000 abstract description 2
- 238000011478 gradient descent method Methods 0.000 description 5
- 230000010355 oscillation Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 239000000463 material Substances 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 235000013399 edible fruits Nutrition 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005303 weighing 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to for the bundle adjustment method of image mosaic, storage medium and calculating equipment, method includes: S1: Initialize installation, it include: the transformation matrix between the corresponding points pair on preset image to be spliced, initialization accumulation gradient is zero, the threshold value of preset error；S2: instrument error function, to calculate the error function value of transformation matrix, and the error function value of tectonic transition matrix is to the gradient function of virtual target transformation matrix；S3: calculating virtual target transformation matrix, and gradient of the error function value to virtual target transformation matrix of transformation matrix is calculated according to gradient function；S4: it according to the gradient of virtual target transformation matrix, updates accumulation gradient and updates transformation matrix；S5: the error function value of updated transformation matrix is calculated；S6: when the error function value of updated transformation matrix is less than threshold value, exporting updated transformation matrix, otherwise, updates the number of iterations and repeats step S3~S6 with updated transformation matrix.
Description
Technical field
The present invention relates to technical field of computer vision more particularly to a kind of bundle adjustment method for image mosaic,
Storage medium and calculating equipment.
Background technique
Image mosaic belongs to computer vision field.It, often can not be same when being shot using camera to actual scene
When reach wider visual field and higher resolution ratio, therefore produce image mosaic technology, i.e., using multiple cameras to actual field
The different zones of scape are shot, or are repeatedly shot using different zones of the same camera to actual scene, are recycled
Computer carries out the synthesis in later period, to meet wide visual field and high-resolution requirement simultaneously.
In later period synthesis, the transformation that the key of splicing is to find between two images adjacent in actual scene is closed
System, and clue is certain corresponding points pair in the lap of two images.Fig. 1 diagrammatically illustrates two images to be spliced
On corresponding points to and corresponding points pair coordinate.For the ease of showing, 5 corresponding points pair are only marked in Fig. 1.Such as Fig. 1 institute
Show, for sequence number 1~5 any one corresponding points for, for each corresponding points to there are two point, the two points have phase
Same picture material, the coordinate of the point in left figure is with (x1, y1) indicate, the coordinate of corresponding points is in right figure with (x2, y2) table
Show.In image mosaic, the coordinate of this two o'clock follows the following rule of the affine transformation as shown in formula (1) and formula (2):
x2=(h1x1+h2y1+h3)/(h7x1+h8y1+h9) (1)
y2=(h4x1+h5y1+h6)/(h7x1+h8y1+h9) (2)
In formula (1) and formula (2), transformation matrix H=[h1,h2,h3,h4,h5,h6,h7,h8,h9] it is unknown parameter, therefore,
It needs to solve it using certain calculation method, bundle adjustment method is referred to as the method for calculating transformation matrix H.Bundle
The purpose for tying up adjusting method is to find transformation matrix H appropriate, so that first image be after transformation matrix H transformation, with second
The error opened between image is minimum.In other words, bundle adjustment method is the process of the optimal solution of a searching transformation matrix H.
Currently, bundle adjustment method includes: gradient descent method, Newton method and quasi-Newton method in the prior art.Generally use
The optimization methods such as gradient descent method, Newton method, quasi-Newton method to transformation matrix H are iterated solution, but these methods are all deposited
In respective drawback.
Gradient descent method only needs to calculate the gradient of parameter to be asked, that is, first derivative, because of the direction of gradient
Constantly iteration can find optimal solution to the negative direction for being exactly that function rises most fast direction, therefore only needing towards gradient.But
The shortcomings that being gradient descent method is: iteration speed is lower, and the problems such as be easy to produce the local oscillation at optimum point.
Newton method needs to calculate the first derivative and second dervative of parameter to be asked, and has spy for gradient descent method
Not fast iteration speed.But the shortcomings that Newton method is: the second dervative of parameter to be asked needs to expend very big calculation amount.
Quasi-Newton method is a kind of improvement to Newton method.The second order that quasi-Newton method does not need directly to calculate parameter to be asked is led
Number, but the iteration by treating the first derivative for seeking parameter, approximatively to obtain second dervative, to reduce certain calculating
Amount.But the shortcomings that quasi-Newton method is: the iterative process for the second dervative that approximation obtains parameter to be asked is still sufficiently complex.
Summary of the invention
In order to solve, existing bundle adjustment method calculating is complicated, iteration speed is low and easily occurs in local at optimum point
The technical issues of oscillation, the present invention provides a kind of bundle adjustments for image mosaic based on Nesterov speedup gradient method
Method, storage medium and calculating equipment not only allow for preceding iteration pair several times when iteration seeks the optimal solution of transformation matrix H
The influence of current iteration, it is also contemplated that the influence that next iteration will generate will not stop to ensure that the reliability of iteration
It stays in locally optimal solution, and globally optimal solution can be converged to, while also ensuring the validity of iteration, i.e., it will not be in optimal solution
It oscillates about, but efficiently iteration is to optimal solution.
According to an aspect of the invention, there is provided a kind of bundle adjustment method for image mosaic, comprising:
Step S1: Initialize installation, comprising: the transformation matrix between corresponding points pair on preset image to be spliced, initially
Changing accumulation gradient is zero, the threshold value of preset error；
Step S2: instrument error function to calculate the error function value of the transformation matrix, and constructs the transformation matrix
Error function value to the gradient function of virtual target transformation matrix；
Step S3: calculating virtual target transformation matrix, and the error of the transformation matrix is calculated according to the gradient function
Gradient of the functional value to the virtual target transformation matrix；
Step S4: it according to the error function value of the transformation matrix to the gradient of the virtual target transformation matrix, updates
Accumulation gradient simultaneously updates transformation matrix；
Step S5: the error function value of updated transformation matrix is calculated；And
Step S6: when the error function value of the updated transformation matrix is less than the threshold value of the error, institute is exported
Updated transformation matrix is stated, otherwise, the number of iterations is updated and step S3~S6 is repeated with the updated transformation matrix.
Preferably, for the bundle adjustment method of image mosaic further include:
In step sl, initialization the number of iterations is zero, and preset maximum number of iterations；And
In step s 6, when the number of iterations is greater than or equal to the maximum number of iterations, output last time iteration is more
Transformation matrix after new.
Preferably, in step s 2, by calculating the corresponding points on image to be spliced to transformed by transformation matrix
The sum of Euclidean distance constructs the error function of the transformation matrix.
Preferably, for the bundle adjustment method of image mosaic further include:
In step sl, preset accumulation gradient weight；And
In step s3, calculate the product of accumulation gradient and accumulation gradient weight, and by calculate the transformation matrix with
The difference of the product calculates the virtual target transformation matrix.
Preferably, for the bundle adjustment method of image mosaic further include:
In step sl, the gradient weights of preset virtual target transformation matrix；And
In step s 4, pass through the error letter by the product of accumulation gradient and accumulation gradient weight plus the transformation matrix
Numerical value updates the product of the gradient of the virtual target transformation matrix and the gradient weights of virtual target transformation matrix
Accumulation gradient.
Preferably, for the bundle adjustment method of image mosaic further include:
In step s 4, by calculating the difference of the accumulation gradient of current transformation matrix and update, the change updated
Change matrix.
Preferably, preset accumulation gradient weight is 0.9, and preset virtual target gradient weights are 0.8.
Preferably, preset maximum number of iterations is 1000.
According to another aspect of the present invention, provide a kind of storage medium, be stored thereon with executable code, it is described can
Code is executed when being executed by processor, the processor is made to execute above-mentioned bundle adjustment method.
According to another aspect of the present invention, a kind of calculating equipment is additionally provided, comprising:
Processor；And
Memory is stored thereon with executable code, and the executable code makes described when being executed by the processor
Processor executes above-mentioned bundle adjustment method.
Compared with prior art, one or more embodiments in above scheme can have following advantage or beneficial to effect
Fruit:
It is set using bundle adjustment method, storage medium and the calculating for image mosaic that one embodiment of the invention provides
It is standby, it overcomes existing bundle adjustment method and calculates that complicated, iteration speed is low and easily occurs in local oscillation at optimum point
Problem, the present invention can predict the gradient of next iteration in-position while considering accumulation gradient, and pre- using this
Gradient is surveyed to correct the gradient of current location, so that transformation matrix H is in the negative direction update along gradient, Neng Gougeng
Quickly and steadily iteration is to optimal solution, while generation the problems such as also avoid local oscillation.
Other features and advantages of the present invention will be illustrated in the following description, and partly becomes from specification
It is clear that understand through the implementation of the invention.The objectives and other advantages of the invention can be by wanting in specification, right
Specifically noted structure is sought in book and Figure of description to be achieved and obtained.
Detailed description of the invention
Attached drawing is used to provide further understanding of the present invention, and constitutes part of specification, with reality of the invention
It applies example and is used together to explain the present invention, be not construed as limiting the invention.
Fig. 1 diagrammatically illustrate the corresponding points on two images to be spliced to and corresponding points pair coordinate.
Fig. 2 is the flow chart according to the bundle adjustment method for image mosaic of one embodiment of the invention.
Fig. 3 is the flow chart according to the bundle adjustment method for image mosaic of another embodiment of the present invention.
Fig. 4 is the update transformation matrix H according to the bundle adjustment method for image mosaic of one embodiment of the invention
Schematic diagram.
Specific embodiment
Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings and examples, how to apply to the present invention whereby
Technological means solves technical problem, and the realization process for reaching technical effect can fully understand and implement.It needs to illustrate
As long as not constituting conflict, each feature in each embodiment and each embodiment in the present invention can be combined with each other,
It is within the scope of the present invention to be formed by technical solution.
Meanwhile in the following description, for illustrative purposes and numerous specific details are set forth, to provide to of the invention real
Apply the thorough understanding of example.It will be apparent, however, to one skilled in the art, that the present invention can not have to tool here
Body details or described ad hoc fashion are implemented.
In order to solve in image mosaic, existing bundle adjustment method calculating is complicated, iteration speed is low and easily occurs in
At optimum point the technical issues of local oscillation, one embodiment of the invention provides a kind of bundle adjustment side for image mosaic
Method, its object is to: bundle adjustment computational problem when image mosaic is solved by using Nesterov speedup gradient method,
Guarantee under the premise of calculating accuracy, the efficiency of calculating is improved, to improve the validity and reliability of image mosaic.
The present invention is specifically described with reference to the accompanying drawings and embodiments, but the invention is not limited in detail below
Embodiment.
As shown in Figure 1, in an embodiment of the present invention, left figure being defined as first image, right figure is defined as second
Image is opened, the purpose of the embodiment is to calculate optimal transformation matrix to splice this two images.
If the corresponding points pair between this two images to be spliced are as follows:
{(x1,i,y1,i),(x2,i,y2,i) | i=1,2 ..., m },
Wherein, i indicates the sequence number of corresponding points pair, and i=1,2 ..., m, m is positive integer；x1,iAnd y1,iIndicate i-th pair
One o'clock actual transverse and longitudinal coordinate on first image of centering should be put；x2,iAnd y2,iIndicate i-th of corresponding points centering another
Point (corresponding points) actual transverse and longitudinal coordinate on second image.
For the corresponding points between two images to be spliced to { (x1,i,y1,i),(x2,i,y2,i) | i=1,2 ..., m }, lead to
The following bundle adjustment method for image mosaic is crossed to calculate the optimal solution of transformation matrix H.
Fig. 2 is the flow chart according to the bundle adjustment method for image mosaic of one embodiment of the invention.Such as Fig. 2 institute
Show, this method comprises:
Step S1: Initialize installation, comprising: the transformation matrix between corresponding points pair on preset image to be spliced, initially
Changing accumulation gradient is zero, the threshold value of preset error；
Step S2: instrument error function to calculate the error function value of the transformation matrix, and constructs the transformation matrix
Error function value to the gradient function of virtual target transformation matrix；
Step S3: calculating virtual target transformation matrix, and the error of the transformation matrix is calculated according to the gradient function
Gradient of the functional value to the virtual target transformation matrix；
Step S4: it according to the error function value of the transformation matrix to the gradient of the virtual target transformation matrix, updates
Accumulation gradient simultaneously updates transformation matrix；
Step S5: the error function value of updated transformation matrix is calculated；And
Step S6: when the error function value of the updated transformation matrix is less than the threshold value of the error, institute is exported
Updated transformation matrix is stated, otherwise, the number of iterations is updated and step S3~S6 is repeated with the updated transformation matrix.
Specifically, step S1: Initialize installation, comprising: a random preset initial transformation matrix is set as H0, wherein become
Change matrix H0In element h1~h9Be be randomly provided and it is known after being arranged；Initialize accumulation gradient θn=[0,0,0,0,0,0,
0,0,0]；The gradient weights β of preset accumulation gradient weight α and virtual target transformation matrix；The threshold tau of preset error；Initialization
The number of iterations n=0；Preset maximum number of iterations N.
Preferably, in an embodiment of the present invention, preset accumulation gradient weight α is 0.9, preset virtual target gradient weights
β is 0.8, and preset maximum number of iterations N is 1000.But above-mentioned parameter is also possible to other reasonable numerical value, the present invention is unlimited
In this.
Next, executing step S2: instrument error function calculates the error function value of the transformation matrix, and constructs institute
State gradient function of the error function value to virtual target transformation matrix of transformation matrix.In step s 2, to be spliced by calculating
Corresponding points on image are to by the sum of the transformed Euclidean distance of transformation matrix instrument error function.The error function table of construction
It is as follows up to formula:
Wherein, i indicates the sequence number of corresponding points pair, i=1,2 ..., m；N indicates the number of iterations；HnIndicate nth iteration
Transformation matrix；x2,iAnd y2,iIndicate i-th of corresponding points centering, one o'clock actual transverse and longitudinal coordinate on second image；x'1,i
And y'1,iIndicate another transverse and longitudinal coordinate after transformation matrix converts on second image of i-th of corresponding points centering.
Error function can indicate two images in spliced error, and the smaller then splicing effect of error is better.
In step s 2, next, calculating the error function value of transformation matrix according to error function.It changes for the first time
When generation, i.e. the number of iterations n=0, by transformation matrix H preset in step S10Error function is substituted into, transformation matrix H is obtained0Mistake
Difference function values J (H0)。
In step s 2, next, the error function value of tectonic transition matrix is to the gradient letter of virtual target transformation matrix
Number, expression formula are as follows:
Wherein,For local derviation numerical symbol, H'nFor the virtual target transformation matrix of nth iteration, h1'~h9' is virtual mesh
Mark the element in transformation matrix.
Gradient function of the error function value of transformation matrix to virtual target transformation matrix it can be seen from above-mentioned expression formula
It is one and seeks partial derivative, J (H in formulan) indicate be nth iteration transformation matrix error function value, be a numerical value；
The gradient function is exactly to ask each element in virtual target transformation matrix to the local derviation of this numerical value.
The purpose of construction gradient function is: so in each iteration, it is only necessary to become the virtual target of nth iteration
The error function value for changing the transformation matrix of matrix and nth iteration substitutes into the gradient function, can obtain the transformation of nth iteration
Gradient of the error function value of matrix to virtual target transformation matrix.
Next, executing step S3: calculating virtual target transformation matrix, and calculate the transformation according to the gradient function
Gradient of the error function value of matrix to the virtual target transformation matrix.
Firstly, calculating virtual target transformation matrix H'n, expression formula is as follows:
H'n=Hn-αθn,
Wherein, α indicates accumulation gradient weight, θnIndicate accumulation gradient.
It can be seen from above-mentioned expression formula in step s3, virtual target transformation matrix is equal to transformation matrix HnWith accumulation
The difference of gradient and the product of accumulation gradient weight.
In step s3, next, being become according to the error function that gradient function calculates the transformation matrix to virtual target
Change matrix H 'nGradient, specifically: the virtual target transformation matrix H' that will be obtainednWith the error function value of obtained transformation matrix
J(Hn) be updated to step S2 transformation matrix error function to the gradient function of virtual target transformation matrix among, and then
To the error function value J (H of transformation matrixn) to virtual target transformation matrix H'nGradient.
For example, calculating transformation matrix H in the number of iterations n=00Error function value J (H0) square is converted to virtual target
Battle array H'nGradient.
Next, executing step S4: according to the error function of the transformation matrix to the virtual target transformation matrix
Gradient updates accumulation gradient and updates transformation matrix.Specifically, in step s 4, by weighing accumulation gradient and accumulation gradient
The product of value adds gradient and virtual target transformation matrix of the error function value to virtual target transformation matrix of transformation matrix
The product of gradient weights, the accumulation gradient updated.
Update accumulation gradient is θn+1, expression formula is as follows:
Wherein, β indicates the gradient weights of virtual target.
Specifically: calculate accumulation gradient θnWith the product of accumulation gradient weight α, the transformation matrix obtained along with step S3
Error function value to the product of the gradient of virtual target transformation matrix and the gradient weights β of virtual target transformation matrix, will tie
Fruit is assigned to updated accumulation gradient θn+1。
In step s 4, next, being obtained more by the difference for the accumulation gradient for calculating current transformation matrix and update
New transformation matrix.
Update current transform matrix is Hn+1, expression formula is as follows:
Hn+1=Hn-θn+1。
Specifically: calculate transformation matrix HnWith accumulation gradient θn+1Difference, result, which is assigned to updated transformation matrix, is
Hn+1。
Next, executing step S5: calculating updated transformation matrix Hn+1Error function value J (Hn+1)。
Specifically: the updated transformation matrix H that will be obtained in step S4n+1It is brought into the error function of step S2, obtains
To updated transformation matrix Hn+1Error function value J (Hn+1)。
Next, executing step S6: when the error function value of the updated transformation matrix is less than the threshold of the error
When value, the updated transformation matrix is exported, otherwise, the number of iterations is updated and repeats to walk with the updated transformation matrix
Rapid S3~S6.
Specifically: judge updated transformation matrix Hn+1Error function value J (Hn+1) and the preset error threshold τ
Between size relation,
If J (Hn+1)≤τ, then end step S1~S6, and export updated transformation matrix Hn+1；
Otherwise, the number of iterations n=n+1 is updated, and step S3~S6 is repeated with updated transformation matrix.
It in an embodiment of the present invention, can be according to pair between two images to be spliced by step S1 to step S6
It should put pair, complete the optimal solution H of transformation matrix Hn+1Calculating.
Fig. 3 is the flow chart according to the bundle adjustment method for image mosaic of another embodiment of the present invention.Such as Fig. 3 institute
Show, in an alternative embodiment of the invention, this method further include: in step s 6: when the number of iterations is greater than or equal to the maximum
When the number of iterations, the updated transformation matrix is exported.
Specifically: judge the size relation between current iteration number and the preset maximum number of iterations, if n >=most
Big the number of iterations N, then end step S1~S6, and export the updated transformation matrix H of last time iterationn+1；
Otherwise, step S3~S6 is repeated with updated transformation matrix.
It in an alternative embodiment of the invention, can be according between two images to be spliced by step S1 to step S6
Corresponding points pair complete the optimal solution H of transformation matrix Hn+1Calculating.
Fig. 4 is according in the step S3 and step S4 for the bundle adjustment method of image mosaic of one embodiment of the invention
Update the schematic diagram of transformation matrix H.As shown in figure 4, in current transform matrix the working as to the n-th+2 times iteration of (n+1)th iteration
During preceding transformation matrix, if only considering the influence of accumulation gradient, iteration result can only obtain (n+1)th iteration
Virtual target transformation matrix can thus make each iteration result and actual iteration result deviation occur, so that
The number of iterations increases, and iteration time extends.In order to solve the above-mentioned technical problem, step S3 of the invention and step S4 are in iteration knot
On the basis of fruit is virtual target transformation matrix, it is contemplated that the gradient of virtual target transformation matrix, that is, accumulation gradient is being considered
While, it can predict the gradient of next iteration in-position, and using this prediction gradient come the gradient to current location
It is corrected, so that the iteration result of step S4 output is the current transform matrix of the n-th+2 times iteration, to improve iteration meter
The efficiency of calculation.
Correspondingly, one embodiment of the invention additionally provides a kind of storage medium, is stored thereon with executable code, it is described can
Code is executed when being executed by processor, the processor is made to execute the above-mentioned bundle adjustment method for image mosaic.
Correspondingly, one embodiment of the invention additionally provides a kind of calculating equipment, comprising:
Processor；And
Memory is stored thereon with executable code, and the executable code makes described when being executed by the processor
Processor executes the above-mentioned bundle adjustment method for image mosaic.
In conclusion the present invention provides a kind of binding tune for image mosaic based on Nesterov speedup gradient method
Adjusting method, storage medium and calculating equipment, when iteration seeks the optimal solution of transformation matrix H, due to considering preceding iteration several times
Influence to current iteration, and then ensure that the reliability of iteration, i.e., locally optimal solution will not be rested on, and can converge to complete
Office's optimal solution.It simultaneously as considering the influence that next iteration will generate, and then ensure that the validity of iteration, i.e., will not
It is oscillated about in optimal solution, but efficiently iteration is to optimal solution.
It should be understood that disclosed embodiment of this invention is not limited to particular procedure step disclosed herein or material
Material, and the equivalent substitute for these features that those of ordinary skill in the related art are understood should be extended to.It should also be understood that
It is that term as used herein is used only for the purpose of describing specific embodiments, and is not intended to limit.
" embodiment " mentioned in specification means that the special characteristic or characteristic described in conjunction with the embodiments is included in the present invention
At least one embodiment in.Therefore, the phrase or " embodiment " that specification various places throughout occurs might not refer both to together
One embodiment.
In addition, described feature or characteristic can be integrated to one or more embodiments in any other suitable manner
In.In the above description, some concrete details, such as thickness, quantity etc. are provided, to provide to the embodiment of the present invention
Comprehensive understanding.However, those skilled in the relevant art are readily apparent that, the present invention is not necessarily to said one or multiple concrete details
It realizes, or other methods, component, material can also be used etc. and realize.
Although above-mentioned example is used to illustrate principle of the present invention in one or more application, for the technology of this field
For personnel, without departing from the principles and ideas of the present invention, hence it is evident that can in form, the details of usage and implementation
It is upper that various modifications may be made and does not have to make the creative labor.Therefore, the present invention is defined by the appended claims.
Claims (10)
1. a kind of bundle adjustment method for image mosaic, comprising:
Step S1: Initialize installation, comprising: the transformation matrix between corresponding points pair on preset image to be spliced, initialization are tired
Product gradient is zero, the threshold value of preset error；
Step S2: instrument error function to calculate the error function value of the transformation matrix, and constructs the mistake of the transformation matrix
Gradient function of the difference function values to virtual target transformation matrix；
Step S3: calculating virtual target transformation matrix, and the error function of the transformation matrix is calculated according to the gradient function
It is worth the gradient to the virtual target transformation matrix；
Step S4: according to the error function value of the transformation matrix to the gradient of the virtual target transformation matrix, accumulation is updated
Gradient simultaneously updates transformation matrix；
Step S5: the error function value of updated transformation matrix is calculated；And
Step S6: when the error function value of the updated transformation matrix is less than the threshold value of the error, output is described more
Otherwise transformation matrix after new updates the number of iterations and repeats step S3~S6 with the updated transformation matrix.
2. bundle adjustment method according to claim 1, further includes:
In step sl, initialization the number of iterations is zero, and preset maximum number of iterations；And
In step s 6, when the number of iterations is greater than or equal to the maximum number of iterations, after output last time iteration updates
Transformation matrix.
3. bundle adjustment method according to claim 2, wherein in step s 2, by calculating on image to be spliced
Corresponding points construct the error function to by the sum of transformed Euclidean distance of transformation matrix.
4. bundle adjustment method according to claim 2, further includes:
In step sl, preset accumulation gradient weight；And
In step s3, calculate the product of accumulation gradient and accumulation gradient weight, and by calculate the transformation matrix with it is described
The difference of product calculates the virtual target transformation matrix.
5. bundle adjustment method according to claim 4, further includes:
In step sl, the gradient weights of preset virtual target transformation matrix；And
In step s 4, pass through the error function value by the product of accumulation gradient and accumulation gradient weight plus the transformation matrix
To the product of the gradient weights of the gradient and virtual target transformation matrix of the virtual target transformation matrix, the accumulation updated
Gradient.
6. bundle adjustment method according to claim 5, further includes:
In step s 4, by calculating the difference of the accumulation gradient of current transformation matrix and update, the transformation square updated
Battle array.
7. bundle adjustment method according to claim 5, wherein preset accumulation gradient weight is 0.9, preset virtual target
Gradient weights are 0.8.
8. bundle adjustment method according to claim 2, wherein preset maximum number of iterations is 1000.
9. a kind of storage medium is stored thereon with executable code, the executable code makes described when being executed by processor
Processor executes bundle adjustment method according to any one of claim 1 to 8.
10. a kind of calculating equipment, comprising:
Processor；And
Memory, is stored thereon with executable code, and the executable code makes the processing when being executed by the processor
Device executes bundle adjustment method according to any one of claim 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910337080.2A CN110211025B (en) | 2019-04-25 | 2019-04-25 | Binding adjustment method for image stitching, storage medium and computing device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910337080.2A CN110211025B (en) | 2019-04-25 | 2019-04-25 | Binding adjustment method for image stitching, storage medium and computing device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110211025A true CN110211025A (en) | 2019-09-06 |
| CN110211025B CN110211025B (en) | 2021-02-12 |
Family
ID=67786351
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910337080.2A Active CN110211025B (en) | 2019-04-25 | 2019-04-25 | Binding adjustment method for image stitching, storage medium and computing device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110211025B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318517A (en) * | 2014-11-19 | 2015-01-28 | 北京奇虎科技有限公司 | Image splicing method and device and client terminal |
| CN105303615A (en) * | 2015-11-06 | 2016-02-03 | 中国民航大学 | Combination method of two-dimensional stitching and three-dimensional surface reconstruction of image |
| CN105389787A (en) * | 2015-09-30 | 2016-03-09 | 华为技术有限公司 | Panorama image stitching method and device |
| CN106952225A (en) * | 2017-02-15 | 2017-07-14 | 山东科技大学 | A kind of panorama mosaic method towards forest fire protection |
| CN107918927A (en) * | 2017-11-30 | 2018-04-17 | 武汉理工大学 | A kind of matching strategy fusion and the fast image splicing method of low error |
| CN109002794A (en) * | 2018-07-13 | 2018-12-14 | 深圳大学 | A kind of non-linear Non-negative Matrix Factorization recognition of face construction method, system and storage medium |
| US10198872B2 (en) * | 2015-08-10 | 2019-02-05 | The Board Of Trustees Of The Leland Stanford Junior University | 3D reconstruction and registration of endoscopic data |
-
2019
- 2019-04-25 CN CN201910337080.2A patent/CN110211025B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104318517A (en) * | 2014-11-19 | 2015-01-28 | 北京奇虎科技有限公司 | Image splicing method and device and client terminal |
| US10198872B2 (en) * | 2015-08-10 | 2019-02-05 | The Board Of Trustees Of The Leland Stanford Junior University | 3D reconstruction and registration of endoscopic data |
| CN105389787A (en) * | 2015-09-30 | 2016-03-09 | 华为技术有限公司 | Panorama image stitching method and device |
| CN105303615A (en) * | 2015-11-06 | 2016-02-03 | 中国民航大学 | Combination method of two-dimensional stitching and three-dimensional surface reconstruction of image |
| CN106952225A (en) * | 2017-02-15 | 2017-07-14 | 山东科技大学 | A kind of panorama mosaic method towards forest fire protection |
| CN107918927A (en) * | 2017-11-30 | 2018-04-17 | 武汉理工大学 | A kind of matching strategy fusion and the fast image splicing method of low error |
| CN109002794A (en) * | 2018-07-13 | 2018-12-14 | 深圳大学 | A kind of non-linear Non-negative Matrix Factorization recognition of face construction method, system and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 秦绪佳等: "基于最佳缝合线的序列遥感图像拼接融合方法", 《计算机科学》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110211025B (en) | 2021-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Structure identification of uncertain general complex dynamical networks with time delay | |
| Lyth | The failure of cosmological perturbation theory in the new ekpyrotic and cyclic ekpyrotic scenarios | |
| CN104750731B (en) | A kind of method and device obtaining whole user portrait | |
| CN110889509B (en) | Gradient momentum acceleration-based joint learning method and device | |
| CN108090621B (en) | Short-term wind speed prediction method and system based on staged overall optimization | |
| CN110348571A (en) | A kind of neural network model training method, device, chip and system | |
| CN108446770B (en) | Distributed machine learning slow node processing system and method based on sampling | |
| CN103150424A (en) | Method for acquiring fine distribution of reactor core three dimensional neutron flux density of reactor | |
| CN109270946A (en) | The attitude control method of flexible spacecraft | |
| CN109583594A (en) | Deep learning training method, device, equipment and readable storage medium storing program for executing | |
| Erickson et al. | Cosmic perturbations through the cyclic ages | |
| CN108897716A (en) | By memory read/write operation come the data processing equipment and method of Reduction Computation amount | |
| CN104268869A (en) | Automatic multilevel-resolution remote-sensing image registration method based on particle swarm optimization | |
| Ferlaino et al. | Asymptotic symmetries and thermodynamics of higher spin black holes in AdS 3 | |
| Pargana et al. | Fully integrated design and analysis of Tensioned Fabric Structures: Finite elements and case studies | |
| Diab et al. | Optimum exact/approximate point synthesis of planar mechanisms | |
| CN110826208A (en) | Pneumatic numerical simulation accelerated convergence method | |
| CN110211025A (en) | For the bundle adjustment method of image mosaic, storage medium and calculate equipment | |
| CN111507904B (en) | Image stitching method and device for microscopic printing patterns | |
| CN109299531A (en) | Electromagnetical transient emulation method and device | |
| CN107194155A (en) | A kind of threat assessment modeling method based on small data set and Bayesian network | |
| CN106934453B (en) | Method for determining cubic material parent phase and child phase orientation relation | |
| CN109344963A (en) | Ultra-large hidden layer node fast selecting method in extreme learning machine | |
| Wang et al. | An improved energy method for form-finding of mesh reflectors | |
| Peng et al. | Kernel least mean square based on conjugate gradient |
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 |