CN107945113B - The antidote of topography's splicing dislocation - Google Patents
The antidote of topography's splicing dislocation Download PDFInfo
- Publication number
- CN107945113B CN107945113B CN201711153716.5A CN201711153716A CN107945113B CN 107945113 B CN107945113 B CN 107945113B CN 201711153716 A CN201711153716 A CN 201711153716A CN 107945113 B CN107945113 B CN 107945113B
- Authority
- CN
- China
- Prior art keywords
- point
- splicing
- subgraph
- corrected zone
- width
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000000729 antidote Substances 0.000 title claims abstract description 14
- 238000012876 topography Methods 0.000 title claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims abstract description 57
- 230000009466 transformation Effects 0.000 claims abstract description 54
- 238000012937 correction Methods 0.000 claims abstract description 21
- 230000007704 transition Effects 0.000 claims abstract description 3
- 238000000034 method Methods 0.000 claims description 46
- 238000001514 detection method Methods 0.000 claims description 14
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000012790 confirmation Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims description 2
- 229940075522 antidotes Drugs 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 241000196324 Embryophyta Species 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000009738 saturating Methods 0.000 description 1
- 238000001228 spectrum 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
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of antidotes of topography splicing dislocation, including step 1: determining the region of splicing dislocation in the panoramic mosaic image that there is splicing dislocation, the two width splicing subgraph for splicing the region is determined, using the region of splicing dislocation as corrected zone；Step 2: the two width splicing subgraph in corrected zone accurately being matched, acquisition can realize two width splicing subgraph accurately matched transformation matrix in corrected zone, and calculate the initial coordinate offset for correcting dislocation accordingly；Step 3: being weighted amendment, correction transformation is carried out to weight revised coordinate shift amount, value of the offset weight coefficient in corrected zone central area is 1, is 0 in the value of corrected zone marginal position, the gradually transition between the region that value is 1 and the region that value is 0.The present invention guarantees the correction transformation that specified region is completed while the image linking in corrected zone and outside corrected zone, the panoramic mosaic image for the dislocation that is eliminated.
Description
Technical field
The present invention relates to a kind of antidotes of topography splicing dislocation, especially for the Local map under panoramic mosaic
As the antidote of splicing dislocation, belong to panoramic mosaic, image registration and feature identification technique field.
Background technique
Image mosaic technology is exactly the seamless high-definition picture that the image that several have lap is combined into width large size
Technology, a kind of widely used connecting method is exactly panoramic mosaic in image mosaic technology, and panoramic mosaic is by several figures
After being registrated on re-projection a to common surface, then all adjacent images are merged, ultimately generates a width panorama sketch
Process, therefore the core technology of panoramic mosaic is exactly image registration techniques.
Current image registration algorithm can be divided into two major classes substantially: similar with based on gray level based on the method for frequency domain
Method, wherein method i.e. phase correlation based on frequency domain, the two major classes method is specific as follows:
1) based on the method for frequency domain: two images subject to registration being transformed to frequency domain first with Fourier transformation, are then led to
It crosses their crosspower spectrum and directly calculates translation vector between two images, to realize the registration of image.
2) it is based on the similar method of gray level: being with the similitude of the pixel grayscale of two images lap for registration
Criterion, the registration position of Automatic-searching image.The thought that this method uses is more intuitive, and most image registration is calculated at present
Method can be classified as this kind, and according to the specific implementation of its registration Algorithm, these algorithms can be divided into direct method and search method again
Two major classes.Direct method mainly includes transformation optimization, it initially sets up the transformation model between two images to be spliced, then uses
Nonlinear iteration minimizes the transformation parameter that algorithm directly calculates model, so that it is determined that the registration position of image；Search method master
If searching for optimal registration position, common search method in another piece image using certain features in piece image as foundation
There are ratio matching method, block matching method and mesh fitting method.
Method for registering images based on frequency domain, have the characteristics that it is simple and accurate, but the method generally require it is bigger
Overlap proportion, usually requiring that between registration image has 50% overlap proportion to be easy to cause translation if overlap proportion is too small
The erroneous estimation of vector, thus the registration of image relatively difficult to achieve.Under the engineering demand of extensive panoramic mosaic, such as to reach this
Need the condition of bigger overlap proportion, it is desirable to provide number of cameras almost double rise, this cost for hardware
With maintenance aspect have very high requirement, while excessive number of cameras the rate and real-time of splicing will also result in it is larger
It influences.
The shortcomings that overcoming the above-mentioned method for registering images based on frequency domain based on the similar method of gray level, uses
Thought is intuitive, and calculation amount and precision can meet engineering demand, however the method still had in practical application it is certain
Problem.Since physical condition limits, it is difficult to ensure that without distance interval, in video camera between the video camera of acquisition splicing sub-video image
Clipping room is away from larger, i.e., when video camera clipping room is away from > 2m, two video cameras are larger for the supervision angle difference of the same area, lead
Causing can be there is a situation where attending to one thing and lose sight of another, to generate obvious in the gray level similitude matching for carrying out sub- stitching image
Splicing dislocation, as shown in Figure 1, in Fig. 1 left hand view be respectively between video camera 1 and two video cameras of video camera 2 spacing compared with
Hour, video camera 1 and video camera 2 in Fig. 1 left hand view are directed to the shooting of A point or B point in Fig. 1 left hand view being represented by dashed line
Angle is without significant difference；And be respectively when spacing is larger between video camera 1 and two video cameras of video camera 2 in Fig. 1 right part of flg,
Video camera 1 and video camera 2 in Fig. 1 right part of flg are directed to the shooting angle of A point or B point in Fig. 1 right part of flg being represented by dashed line
It differs very big, obvious splicing dislocation will be generated under the conditions of shooting angle differs very big.
Summary of the invention
To solve the above problems, the present invention provides a kind of antidote of topography splicing dislocation, using for office
The antidote that the behavioral characteristics in portion region are converted splices to avoid in the biggish situation of camera arrangements spacing
Dislocation, enables spliced panoramic picture preferably to meet practical application request.
The technical scheme is that a kind of antidote of topography's splicing dislocation comprising following steps:
Step 1: determining to have the region of splicing dislocation in the panoramic mosaic image that there is splicing dislocation and be made with this
For corrected zone, the corresponding two width splicing subgraph of corrected zone is determined；
Step 2: to the two width splicing subgraph in corrected zone, (i.e. two width, which splice, corresponds to corrected zone in subgraph
Part) accurately matched, it obtains and realizes two width splicing subgraph accurately matched transformation relation in corrected zone, such as convert
Matrix, and the coordinate shift amount (can be described as initial coordinate offset) that can accurately correct dislocation is calculated accordingly；
Step 3: introducing offset weight coefficient and amendment is weighted to initial coordinate offset, to weight revised coordinate
Offset carries out correction transformation, and then the panoramic mosaic image for the dislocation that has been eliminated, and the offset weight coefficient is in correction area
The value of domain central area is 1, is 0 in the value of corrected zone marginal position, in the region that value is 1 and the area that value is 0
Gradually transition between domain.
Further, the panoramic mosaic image that there is splicing dislocation can be used by multiple image is based on gray level phase
As any ways such as method be spliced, the two width splicing subgraph in corrected zone carries out accurate matched method
It can be matching process or other proper methods based on ORB feature.
In the step 1, the region that splicing dislocation occurs can be selected by manual frame in the panoramic picture of splicing dislocation, by
This realizes the determination to the corrected zone.
In the step 2, the two width splicing subgraph in corrected zone can be carried out using ORB feature extracting method special
The registration of two width splicing subgraph is realized in sign detection accordingly.
Preferably, described that feature inspection is carried out to the two width splicing subgraph in corrected zone using ORB feature extracting method
The step of survey may include:
Step 2-1 (feature point extraction and matching step): angle is carried out to two width splicing subgraph in corrected zone
Point detection, the point using angle point as matching characteristic, the angle point constitutive characteristic point detected are alternatively gathered, and determine the master of matching characteristic point
Description of matching characteristic point is extracted in direction in conjunction with the principal direction of matching characteristic point, determines two width splicing subgraph in correction area
Alternative features matching double points in domain；
Step 2-2 (transformation matrix, which calculates, determines step): Mismatching point pair is rejected from alternative features matching double points, is obtained
For the characteristic matching point pair of transformation, accurately matched according to the characteristic matching point for transformation to two width splicing subgraph is calculated
Transformation matrix, obtain transformation matrix in each matrix element value.
Preferably, in step 2-1,
1) angle point of each splicing subgraph is detected using FAST algorithm:
To each pixel P of the splicing subgraph in corrected zone, is drawn using pixel P as the center of circle and have 16 on a circumference
The circle of a pixel, if pixel value is all larger than X on the circlepThe continuous image vegetarian refreshments or respectively less than X of+TpThe number of the continuous image vegetarian refreshments of-T
Amount is no less than N number of, then pixel P is an angle point, is alternatively gathered using all angle points detected as characteristic point, wherein XpFor
The pixel value of pixel P, N are the positive integer of setting, such as under normal conditions can be 9 or 12, and T is the first threshold of setting,
The setting of first threshold T will affect the amount detection of characteristic point, can set according to actual needs, for example, usual situation can be set
It is 40；
The principal direction of characteristic point is determined by the Gray Moment of Image neighborhood, shown in the calculating of the Gray Moment such as formula (1):
Wherein mpqIt indicates that the p+q rank Gray Moment of feature vertex neighborhood, p, q are nonnegative integer, indicates the order of Gray Moment, I (x,
Y) gray value being characterized in vertex neighborhood at (x, y), wherein x indicates that abscissa, y indicate ordinate, the mass center of feature vertex neighborhood
Are as follows:
C=(Cx,Cy), wherein
Thus the principal direction θ of this feature point is shown in formula (2):
θ=arctan2 (m01,m10) (2)；
2) extraction of feature point description is carried out using BRIEF as character description method:
Coordinate coordinates matrix S shown in formula (3) of point pair around characteristic point is indicated:
Wherein, n is the number of point pair around this feature point, is positive integer, that is, 2n point is shared around characteristic point,
(xi,yi) it is i-th point of the coordinate around this feature point, xiFor i-th point of abscissa, yiFor i-th point of ordinate；
Use spin matrix RθConstruct the correction matrix S of coordinates matrix Sθ, the correction matrix SθAs shown in formula (4):
Wherein, spin matrix RθAs shown in formula (5):
θ is the principal direction of this feature point, xi' and yi' be respectively the above-mentioned i-th point of abscissa after rotational correction and
Ordinate,
Pass through correction matrix SθFollowing feature point description can be generated:
I=1,2 ... ... 2n, for i-th point around expressing feature point,
J=1,2 ... ... 2n, for the jth point around expressing feature point,
Pi' and Pj' it is P respectivelyiAnd PjThe postrotational point of point, for example, P '2k-1、P′2kRespectively P2k-1、P2kPoint is postrotational
Point,
I (P) indicates the gray value of image of P point, for example, I (Pi') and I (Pj') it is respectively point Pi' and Pj' image grayscale
Value,
The Feature Descriptor g of generationn(1, θ) is two-value word string, therefore can pass through the phase between Hamming distance characteristic feature point
Like property namely ORB characteristic distance.
3) alternative features matching double points are determined according to ORB characteristic distance:
According to feature point description, two width of measurement splice the ORB characteristic distance between subgraph characteristic point, obtain each feature
The arest neighbors characteristic point of point and time neighbour's characteristic point calculate nearest neighbor distance (the ORB spy between arest neighbors characteristic point of characteristic point
Levy distance) ratio with time nearest neighbor distance (with the ORB characteristic distance between secondary neighbour's characteristic point), when the ratio is less than the be arranged
When two threshold values, confirmation this feature point and arest neighbors characteristic point are alternative features matching double points, and second threshold can be set in usual situation
It is 0.6, it is excessive to be easy to cause wrong matching, it is too small, cause leakage to match.
Preferably, the step 2-2 may include:
Step 2-2-1:
1) from alternative features matching double points, 4 pairs is randomly selected and is used as initial characteristics matching double points, using following equation
(6) two splicing subgraph transformation of coordinates matrix Hs are calculated, 4 selected alternative features matching double points are in each splicing subgraph
In any 3 points it is not conllinear,
Wherein, (xi′,yi', 1) and (xi,yi, 1) and it is respectively that i-th of alternative features match point is constituted in two splicing subgraphs
Pair two alternative features match points homogeneous coordinates, H is transformation matrix,
2) with transformation matrix H, to remaining each alternative features matching double points (when alternative features matching double points sum is L,
The quantity of remaining alternative features matching double points is L-4), with transformation matrix H to the alternative spy in one of splicing subgraph
The homogeneous coordinates of sign match point are converted, and the homogeneous coordinates formed after calculating conversion are spliced corresponding in subgraph with another
The distance between the homogeneous coordinates of alternative features match point, specific formula are as follows:
Dv=d (A 'l,HAl) (7)
Wherein AlWith A 'lTwo in remaining first of alternative features matching double points are constituted in respectively two splicing subgraphs
Characteristic matching point (xl, yl)、(x′l, y 'l) homogeneous coordinates matrix
If the distance is less than the third threshold value of setting, then it is assumed that the alternative features matching double points are interior point pair, and otherwise this is standby
Select characteristic matching point to for exterior point, usual situation, third threshold value is set as 3, i.e., less than 3 pixels of transformed coordinate distance are
It is interior, the alternative features matching double points for exterior point are rejected, point number in current (point sum in this) is obtained；
Step 2-2-2: repeating step 2-2-1 several times, such as 5 times, and the transformation matrix for selecting interior point number most is real
Existing two width splice subgraph accurately matched transformation matrix.
The initial coordinate offset can be calculated according to following equation:
Wherein x, y are the coordinate of the point (x, y) on a splicing subgraph, and x ', y ' are the point (x, y) by transformed
Coordinate, f (x, y) and g (x, y) are respectively that two width splicing subgraph accurately matched corresponding coordinate transformation letter is realized in corrected zone
Number, can be determined by transformation matrix, and Δ x and Δ y divide for the transformed x coordinate offset of point (x, y) and y-coordinate offset.
Preferably, amendment can be weighted to Δ x and Δ y according to following equation, it is modified corresponding inclined obtains weighting
Shifting amount Δ x ' and Δ y ':
Δ x'=ωx·ωy·Δx (9)
Δ y'=ωx·ωyΔ y (10),
Wherein, ωxAnd ωyWeight coefficient respectively relevant to x and y coordinates position.
Weight coefficient ω can be determined using following mannerxAnd ωy:
If x1<x<x2, ωx=1；
Otherwise, ωx=1-3*min (︱ x-x1︱, ︱ x-x2︱)/width；
If y1<y<y2, ωy=1；
Otherwise, ωy=1-3*min (︱ y-y1︱, ︱ y-y2︱)/height,
Wherein, corrected zone is the rectangle that each side is respectively parallel to x and y-axis direction, and width and height are respectively to correct
Region width in the x and y direction and height,
x1And x2Corrected zone is respectively divided into the x coordinate of two parallel divisional lines of three equal parts, y in the x direction1With
y2The respectively corrected zone y-coordinate that is divided into two parallel divisional lines of three equal parts in y-direction.
The invention has the benefit that dislocation zone can be positioned and carry out correction process to it, dislocation is effectively eliminated, together
When, so that non-dislocation zone is kept original state, avoid generating minus effect to it, makes image and the outer image of corrected zone in corrected zone
Linking is kept, does not introduce new dislocation.
Detailed description of the invention
Fig. 1 is the effect picture of the shooting angle between video camera under different spacing, and wherein left hand view is expressed as between video camera
The effect picture of shooting angle when spacing is smaller, right part of flg be expressed as spacing between video camera it is larger when shooting angle effect
Figure；
Fig. 2 is flow diagram of the invention；
Fig. 3 is that the present invention relates to the schematic diagrames of corrected zone and splicing subgraph relationship；
Fig. 4 is the flow chart the present invention relates to Corner Detection and transformation matrices；
Fig. 5 is that the present invention relates to the schematic diagrames in corrected zone according to weight coefficient demarcation interval.
Specific embodiment
The present invention is described further below in conjunction with drawings and examples.
As Figure 2-Figure 5, due in practical engineering application, often there is a situation where between video camera spacing it is larger, because
This, the shortcomings that splicing dislocation in the panoramic picture for needing to obtain for gray level similar method, provides a kind of for partial zones
The antidote that the behavioral characteristics in domain are converted makes spliced panoramic picture meet practical application need to eliminate dislocation
It asks.
The key technical problem that the present invention solves is: how to guarantee the image in corrected zone and the figure outside corrected zone
In the case where as linking, the dislocation of corrected zone is eliminated.
The key technical problem is solved, the present invention needs to complete dislocation zone positioning, eliminates in local misalignment and region
Outer image is connected three steps, overall flow as shown in Fig. 2, i.e. described topography's splicing dislocation antidote, including
Following steps:
Step 1: determining the region of splicing dislocation in the panoramic picture of splicing dislocation, and determine for being spliced into the area
Two width in domain splice subgraph, using the region of the splicing dislocation as corrected zone；The panoramic picture of the splicing dislocation can
To be by being obtained based on the similar image split-joint method of gray level.
Step 2: transformation matrix calculating being carried out to the two width splicing subgraph determined in step 1, correction area is eliminated with this
The splicing in domain misplaces；The splicing dislocation of the corrected zone i.e. local misalignment.
Step 3: introduce offset weight and calculate final coordinate shift amount, with this realize image in corrected zone and
Image outside corrected zone completes correction transformation while linking, the panoramic mosaic image for the dislocation that has been eliminated.
In the step 1, it can be through interactive approach and determine the region of splicing dislocation, the interactive approach
Are as follows: manual frame selects the region of splicing dislocation in the panoramic picture of splicing dislocation, using this region as corrected zone.
As shown in figure 3, black surround indicates one to corrected zone, the corrected zone is by warp_image2 and warp_image3
Subgraph obtain, wherein image1 --- image4 indicates splicing input picture, warp_image1 --- warp_image4
Indicate image1 --- image of the image4 after coordinate transform, they are under the same coordinate system, can complete image spelling
It connects.
The transformation matrix in step 2 calculates the splicing dislocation that corrected zone is eliminated for this, can use ORB feature
Extracting method carries out feature detection to the two width splicing subgraph in the corrected zone determined in step 1, realizes and is used to splice
The accurate matching of the two width splicing subgraph of the corrected zone out.
The step of feature detection is carried out to the two width splicing subgraph determined in step 1 using ORB feature extracting method
It is as follows:
Step 2-1: it first passes through headed by feature point extraction and matching, the feature point extraction and matched mode to described two
Width splices subgraph and carries out Corner Detection, then the angle point that the Corner Detection obtains determines characteristic point as set of characteristic points
The principal direction of characteristic point in set, then the principal direction in conjunction with the characteristic point extracts description of characteristic point, further according to
Description of the characteristic point primarily determines out characteristic matching point pair to two width splicing subgraph；
Step 2-2: calculating transformation matrix, deletes the mistake of the characteristic matching point centering primarily determined out described in step 2-1
Matching double points, then according to characteristic matching point of the deletion Mismatching point to after between saturating the image for being calculated as transformation matrix parameter
Penetrate the parameter of transformation.
Further, in the step 2-1, FAST is all made of to every width splicing subgraph in two width splicing subgraph and is calculated
Method detects that the angle point of the two width splicing subgraph executes as flowed down each pixel P specially in corrected zone
Journey:
The circle that 16 pixels are had on a circumference is drawn as the center of circle using pixel P, if there is N number of continuous picture on the circle
The pixel value of vegetarian refreshments, N number of continuous pixel is all larger than Xp+ T or respectively less than Xp- T, then pixel P is width splicing
One angle point of subgraph, characteristic point of the angle point as width splicing subgraph, all angle points constitute width splicing subgraph
The set of characteristic points of picture, wherein XpIt is expressed as the pixel value of pixel P, N is positive integer, and T is the first threshold of setting；
The principal direction of each characteristic point, the calculating of the Gray Moment such as formula are determined by the Gray Moment of Image neighborhood
(1) shown in:
Wherein mpqIt indicates that p+q rank Gray Moment, p, q are nonnegative number, indicates that the order of Gray Moment, I (x, y) are characterized vertex neighborhood
Gray value at interior (x, y), wherein x indicates that abscissa, y indicate ordinate.Mass center C=(the C of feature vertex neighborhood can be obtainedx,Cy),Thus the principal direction θ of this feature point is shown in formula (2):
θ=arctan2 (m01,m10) (2)；
The mode that extracts description of characteristic point is that ORB algorithm is used to extract each described characteristic point with BRIEF
Description son, wherein BRIEF is a kind of Feature Descriptor extraction algorithm.
Specifically: coordinate coordinates matrix S shown in formula (3) of point pair around this feature point is indicated first:
Wherein, n is the number of point pair and n is positive integer around this feature point, that is, 2n is shared around characteristic point
Point, (xi,yi) it is i-th point of the coordinate around this feature point, xiFor i-th point around this feature point of abscissa, yiFor this
I-th point of ordinate around characteristic point, i is positive integer and its value range is 1 to n.
Then, using spin matrix RθConstruct the correction matrix S of coordinates matrix Sθ, the correction matrix SθSuch as formula (4) institute
Show:
Wherein, spin matrix RθAs shown in formula (5):
Wherein, θ is the principal direction of this feature point.Pass through correction matrix SθFollowing feature point description can be generated:
Wherein, I (P) indicates the gray value of image of P point.
The Feature Descriptor g of generationn(1, θ) is two-value word string, therefore can pass through the phase between Hamming distance characteristic feature point
Like property namely ORB characteristic distance.
The side of characteristic matching point pair is primarily determined out to two width splicing subgraph according to description of the characteristic point
Formula is to obtain the ORB characteristic distance for each characteristic point that two width splices in subgraph, then root by description of characteristic point
According to the ORB characteristic distance of each characteristic point in each splicing subgraph, the nearest neighbor distance of this feature point, secondary neighbour are obtained
Distance, arest neighbors characteristic point and time neighbour's characteristic point, that is, compare each of this feature point and another splicing subgraph
ORB characteristic distance between characteristic point, the obtained shortest distance of ORB characteristic distance are the nearest neighbor distance of this feature point, only
That ORB characteristic distance longer than nearest neighbor distance is the secondary nearest neighbor distance of this feature point, another described splicing subgraph
On meet nearest neighbor distance that characteristic point be this feature point arest neighbors characteristic point, that for meeting time nearest neighbor distance be special
Sign point is secondary neighbour's characteristic point of this feature point, then calculates the nearest neighbor distance of this feature point and the ratio of time nearest neighbor distance,
When the ratio is less than the second threshold of setting, it is confirmed that this feature point and arest neighbors characteristic point are characterized matching double points, the spy
Levy matching double points alternately characteristic matching point pair.
The mode of the calculating transformation matrix in the step 2-2 are as follows: in obtained two width splicing subgraph
All characteristic matching points primarily determined out to being set as M (x, y) and M ' (x, y), the correspondence between M (x, y) and M ' (x, y)
Relationship can be described with the projective transformation model of 8 parameters of formula (6), i.e., are as follows:
Wherein, (xi′,yi', 1) and (xi,yi, 1) and it is respectively that M (x, y) and i-th point of M ' (x, y) of homogeneous coordinates indicate,
H is transformation matrix, and the freedom degree of such H is 8, can be estimated by least 4 pairs of characteristic points.Due to being obtained by ORB algorithm
Characteristic matching point to be possible to generate Mismatching point, need to optimize it, the confidence level of Lai Tigao matching result.
The high robustness of RANSAC is applied in the Feature Points Matching of image, Mismatching point is rejected using RANSAC method
It is right: to set initial best interior points NiIt is 0, NiFor positive integer, following steps are executed:
Step 2-2-1:
1) from n to randomly selecting 4 alternative features matching double points in the alternative features matching double points as initial characteristics
Matching double points, the parameter of transformation matrix H can be according to 4 alternative features matching double points of the selection between two planes (splicing subgraph)
Be linearly calculated, specific calculate can be found in formula (6)；
2) remaining L-4 alternative features matching double points are calculated one by one using formula (7) and passes through the transformed seat of transformation matrix
The distance between scale value and alternative features matching double points:
Dv=d (A 'l,HAl) (7)
If distance is less than the third threshold value of setting, then it is assumed that the alternative features matching double points are interior point, otherwise the alternative spy
Sign matching double points are exterior point, and the alternative features match point as exterior point just weeds out, and count interior after calculating one by one and put number
As current interior point number；
3) if current interior point number is greater than Ni, then transformation matrix H is current optimal transformation matrix, and NiValue update
For current interior point number；
Step 2-2-2: repeating in return step 2-3-1, after repeating several times, selects interior point number most
Mostly and transformation matrix of the smallest transformation matrix parameter of error function between image, the change between more accurate image is so far obtained
Matrix is changed to obtain transformed image, the splicing dislocation of corrected zone can be effectively eliminated.
Offset weight is introduced in the step 3, and figure in corrected zone is realized calculating final coordinate shift amount with this
Image outside picture and corrected zone completes correction transformation while linking, include the following steps:
Step 3-1: according to obtaining converting between described image in 2-2-2, a splicing subgraph is obtained in corrected zone
Each pixel coordinate is to the transformed coordinate of another splicing subgraph；
Step 3-2: coordinate shift amount is determined；
Step 3-3: offset weight is calculated；
Step 3-4: image flame detection transformation is carried out according to coordinate shift amount.
Further, the coordinate transform in the step 3-1 and 3-2 and coordinate shift such as formula (8) are shown:
The mode of offset weight is calculated in the step 3-2 according to the coordinate to be corrected of input are as follows:
Δ x and Δ y are weighted, ωxAnd ωyDistance according to coordinate points away from regional center determines (referring to Fig. 5), area
Domain center weight is 1, and zone boundary weight is 0；
If x1<x<x2, ωx=1；
Otherwise, ωx=1-3*min (︱ x-x1︱, ︱ x-x2︱)/width；
If y1<y<y2, ωy=1；
Otherwise, ωy=1-3*min (︱ y-y1︱, ︱ y-y2︱)/height；
The concrete mode such as formula of coordinate shift amount is determined according to the transformation matrix between described image obtained in step 3-3
(9) and shown in formula (10):
Δ x'=ωx·ωy·Δx (9)
Δ y'=ωx·ωy·Δy (10)。
By being weighted to offset, it is ensured that transformed image in corrected zone, closer to corrected zone side
Edge, deformation is smaller, and corrected zone edge is undeformed, guarantees that the image in corrected zone and outside corrected zone keeps linking with this,
New dislocation will not be introduced.
In the step 3-4, the mode of image flame detection transformation is carried out according to coordinate shift amount are as follows:.
I (x+ Δ x ', y+ Δ y ')=I (x, y)
The present invention is described in a manner of Detailed description of the invention above, it will be understood by those of skill in the art that the disclosure
It is not limited to embodiments described above, in the case of without departing from the scope of the present invention, can make a variety of changes, change and replace
It changes.
Claims (4)
1. a kind of antidote of topography's splicing dislocation, it is characterised in that include the following steps:
Step 1: determining there is the region of splicing dislocation and in this, as strong in there is the panoramic mosaic image for splicing dislocation
Positive region determines the corresponding two width splicing subgraph of corrected zone；
Step 2: the two width splicing subgraph in corrected zone accurately being matched, obtains and realizes the splicing of two width in corrected zone
The accurate matched transformation relation of subgraph, and the initial coordinate offset that can accurately correct dislocation is calculated accordingly；
Step 3: introducing offset weight coefficient and amendment is weighted to initial coordinate offset, to weight revised coordinate shift
Amount carries out correction transformation, and then the panoramic mosaic image for the dislocation that has been eliminated, and the offset weight coefficient is in corrected zone
The value in heart district domain is 1, is 0 in the value of corrected zone marginal position, the region that the region that value is 1 and value are 0 it
Between gradually transition,
The initial coordinate offset is calculated according to following equation:
X'=f (x, y), Δ x=x'-x
Y'=g (x, y), Δ y=y'-y
Wherein x, y are the coordinate of the point (x, y) on a splicing subgraph, and x ', y ' are that the point (x, y) passes through transformed seat
Mark, f (x, y) and g (x, y) are respectively that two width splicing subgraph accurately matched corresponding coordinate transformation letter is realized in corrected zone
Number, Δ x and Δ y points are the transformed x coordinate offset of point (x, y) and y-coordinate offset,
Amendment is weighted to Δ x and Δ y according to following equation, obtains and weights modified respective offsets amount Δ x ' and Δ y ':
Δ x'=ωx·ωy·Δx
Δ y'=ωx·ωy·Δy
Wherein, ωxAnd ωyWeight coefficient respectively relevant to x and y coordinates position,
Weight coefficient ω is determined using following mannerxAnd ωy:
If x1< x < x2, ωx=1；
Otherwise, ωx=1-3*min (︱ x-x1︱, ︱ x-x2︱)/width；
If y1< y < y2, ωy=1；
Otherwise, ωy=1-3*min (︱ y-y1︱, ︱ y-y2︱)/height,
Wherein, corrected zone is the rectangle that each side is respectively parallel to x and y-axis direction, and width and height are respectively corrected zone
Width and height in the x and y direction,
x1And x2Corrected zone is respectively divided into the x coordinate of two parallel divisional lines of three equal parts, y in the x direction1And y2Respectively
It is divided into the y-coordinate of two parallel divisional lines of three equal parts in y-direction for corrected zone,
In the step 2, feature detection is carried out to the two width splicing subgraph in corrected zone using ORB feature extracting method,
The registration of two width splicing subgraph is realized accordingly, it is described that son is spliced to two width in corrected zone using ORB feature extracting method
Image carry out feature detection the step of include:
Step 2-1: Corner Detection is carried out to two width splicing subgraph in corrected zone, using angle point as matching characteristic
Point determines the principal direction of matching characteristic point, and description of matching characteristic point is extracted in conjunction with the principal direction of matching characteristic point, determines two
Width splices alternative features matching double points of the subgraph in corrected zone；
Step 2-2: rejecting Mismatching point pair from alternative features matching double points, obtain the characteristic matching point pair for transformation, according to
Splice subgraph accurately matched transformation matrix to two width are calculated according to the characteristic matching point for transformation,
The step 2-2 includes:
Step 2-2-1:
1) from alternative features matching double points, 4 pairs is randomly selected and is used as initial characteristics matching double points, calculate two using following equation
Splice subgraph transformation of coordinates matrix H, 4 selected alternative features matching double points are in each any 3 spliced in subgraph
Point is not conllinear:
Wherein, (xi′,yi', 1) and (xi,yi, 1) and it is respectively that i-th of alternative features matching double points is constituted in two splicing subgraphs
The homogeneous coordinates of two alternative features match points, H are transformation matrix,
2) with transformation matrix H, characteristic point is alternatively gathered in remaining each alternative features matching double points, with transformation matrix H to it
In the homogeneous coordinates of alternative features match point in a splicing subgraph converted, calculate the homogeneous coordinates formed after conversion
With the distance between the homogeneous coordinates of the corresponding alternative features match point in another splicing subgraph, specific formula are as follows:
Dv=d (A 'l,HAl)
Wherein AlWith A 'lTwo features in remaining first of alternative features matching double points are constituted in respectively two splicing subgraphs
Match point (xl, yl)、(x′l, y 'l) homogeneous coordinates matrix
If the distance is less than the third threshold value of setting, then it is assumed that the alternative features matching double points are interior point pair, otherwise the alternative spy
Sign matching double points are exterior point, reject the alternative features matching double points for exterior point, obtain and put number in current；
Step 2-2-2: repeating step 2-2-1 several times, and the transformation matrix for selecting interior point number most is to realize two width splicing
The transformation matrix of accurate matching of image.
2. the antidote of topography's splicing dislocation according to claim 1, it is characterised in that described to there is splicing mistake
The panoramic mosaic image of position is used by multiple image and is spliced based on the similar method of gray level, described in corrected zone
It is the matching process based on ORB feature that two width, which splice subgraph to carry out accurate matched method,.
3. the antidote of topography's splicing dislocation according to claim 1, it is characterised in that in the step 1,
Splice manual frame in the panoramic picture of dislocation and select the region that splicing dislocation occurs, is achieved in the corrected zone really
It is fixed.
4. the antidote of topography's splicing dislocation according to claim 1, it is characterised in that in step 2-1,
1) using the angle point of FAST algorithm detection splicing subgraph:
To each pixel P of the splicing subgraph in corrected zone, is drawn using pixel P as the center of circle and have 16 pictures on a circumference
The circle of vegetarian refreshments, if pixel value is all larger than X on the circlepThe continuous image vegetarian refreshments or pixel value of+T is respectively less than XpThe continuous image vegetarian refreshments of-T
Quantity is no less than N number of, then pixel P is an angle point, is alternatively gathered using all angle points detected as characteristic point, wherein Xp
For the pixel value of pixel P, N is the positive integer of setting, and T is the first threshold of setting；
The principal direction of characteristic point is determined by the Gray Moment of Image neighborhood, shown in the calculating of the Gray Moment such as formula (1):
Wherein mpqIt indicates that the p+q rank Gray Moment of feature vertex neighborhood, p, q are nonnegative integer, indicates that the order of Gray Moment, I (x, y) are
Gray value in feature vertex neighborhood at (x, y), wherein x indicates that abscissa, y indicate ordinate, the mass center C=of feature vertex neighborhood
(Cx,Cy),
Wherein,
Thus the principal direction θ of this feature point is shown in formula (2):
θ=arctan2 (m01,m10) (2)；
2) extraction of feature point description is carried out using BRIEF as character description method:
Coordinate coordinates matrix S shown in formula (3) of point pair around characteristic point is indicated:
Wherein, n is the number of point pair around this feature point, is positive integer, (xi,yi) it is i-th point of the seat around this feature point
Mark, xiFor i-th point of abscissa, yiFor i-th point of ordinate；
Use spin matrix RθConstruct the correction matrix S of coordinates matrix Sθ, the correction matrix SθAs shown in formula (4):
Wherein, spin matrix RθAs shown in formula (5):
Wherein, θ is the principal direction of this feature point, xi' and yi' be respectively the above-mentioned i-th point of abscissa after rotational correction and
Ordinate,
Pass through correction matrix SθFollowing feature point description can be generated:
I=1,2 ... ... 2n, for i-th point around expressing feature point,
J=1,2 ... ... 2n, for the jth point around expressing feature point,
Pi' and Pj' it is P respectivelyiAnd PjThe postrotational point of point,
I (P) indicates the gray value of image of P point,
3) alternative features matching double points are determined according to ORB characteristic distance:
According to feature point description, two width of measurement splice the ORB characteristic distance between subgraph characteristic point, obtain each characteristic point
Arest neighbors characteristic point and time neighbour's characteristic point, calculate the nearest neighbor distance of characteristic point and the ratio of time nearest neighbor distance, when the ratio
Less than setting second threshold when, confirmation this feature point and arest neighbors characteristic point be alternative features matching double points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711153716.5A CN107945113B (en) | 2017-11-17 | 2017-11-17 | The antidote of topography's splicing dislocation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711153716.5A CN107945113B (en) | 2017-11-17 | 2017-11-17 | The antidote of topography's splicing dislocation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107945113A CN107945113A (en) | 2018-04-20 |
| CN107945113B true CN107945113B (en) | 2019-08-30 |
Family
ID=61933154
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711153716.5A Active CN107945113B (en) | 2017-11-17 | 2017-11-17 | The antidote of topography's splicing dislocation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107945113B (en) |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596867A (en) * | 2018-05-09 | 2018-09-28 | 五邑大学 | A kind of picture bearing calibration and system based on ORB algorithms |
| CN109544447B (en) * | 2018-10-26 | 2022-10-21 | 广西师范大学 | Image splicing method and device and storage medium |
| CN109598675B (en) * | 2018-11-13 | 2023-03-10 | 北京交通大学 | Splicing method of multiple repeated texture images |
| CN109697705B (en) * | 2018-12-24 | 2019-09-03 | 北京天睿空间科技股份有限公司 | Chromatic aberration correction method suitable for video-splicing |
| CN110020995B (en) * | 2019-03-06 | 2023-02-07 | 沈阳理工大学 | Image splicing method for complex images |
| CN110400266B (en) * | 2019-06-13 | 2021-12-28 | 北京小米移动软件有限公司 | Image correction method and device and storage medium |
| CN110458870B (en) * | 2019-07-05 | 2020-06-02 | 北京迈格威科技有限公司 | Image registration, fusion and occlusion detection method and device and electronic equipment |
| CN110930301B (en) * | 2019-12-09 | 2023-08-11 | Oppo广东移动通信有限公司 | Image processing method, device, storage medium and electronic equipment |
| CN111127361B (en) * | 2019-12-24 | 2023-07-28 | 中山大学 | Perspective distortion correction method for video stitching |
| CN113538252B (en) * | 2020-04-17 | 2024-03-26 | 嘉楠明芯(北京)科技有限公司 | Image correction method and device |
| CN111768337B (en) * | 2020-06-01 | 2024-05-14 | 中国科学院空天信息创新研究院 | Image processing method and device and electronic equipment |
| CN112017114B (en) * | 2020-06-08 | 2023-08-04 | 武汉精视遥测科技有限公司 | Method and system for splicing full images of half images in tunnel detection |
| CN111915521A (en) * | 2020-07-31 | 2020-11-10 | 北京卓立汉光仪器有限公司 | Spliced image correction method and device |
| CN112017117A (en) * | 2020-08-10 | 2020-12-01 | 武汉威盛通科技有限公司 | Panoramic image acquisition method and system based on thermal infrared imager |
| CN113052119B (en) * | 2021-04-07 | 2024-03-15 | 兴体(广州)智能科技有限公司 | Ball game tracking camera shooting method and system |
| CN114841862B (en) * | 2022-06-07 | 2023-02-03 | 北京拙河科技有限公司 | Image splicing method and system based on hundred million pixel array type camera |
| CN115393196B (en) * | 2022-10-25 | 2023-03-24 | 之江实验室 | Infrared multi-sequence image seamless splicing method for unmanned aerial vehicle area array swinging |
| CN115546072B (en) * | 2022-11-28 | 2023-03-10 | 南京航空航天大学 | Image distortion correction method |
| CN115620154B (en) * | 2022-12-19 | 2023-03-07 | 江苏星湖科技有限公司 | Panoramic map superposition replacement method and system |
| CN116188024B (en) * | 2023-04-24 | 2023-07-04 | 山东蓝客信息科技有限公司 | Medical safety payment system |
| CN117437122B (en) * | 2023-12-21 | 2024-03-29 | 宁波港信息通信有限公司 | Method and system for splicing panoramic images of container |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101556692A (en) * | 2008-04-09 | 2009-10-14 | 西安盛泽电子有限公司 | Image mosaic method based on neighborhood Zernike pseudo-matrix of characteristic points |
| CN103247029B (en) * | 2013-03-26 | 2015-11-25 | 中国科学院上海技术物理研究所 | A kind of high spectrum image geometrical registration method generated for spliced detector |
| CN105335948B (en) * | 2014-08-08 | 2018-06-29 | 富士通株式会社 | Splicing apparatus, method and the scanner of file and picture |
| CN104599247B (en) * | 2015-01-04 | 2018-04-24 | 清华大学 | Method for correcting image and device |
-
2017
- 2017-11-17 CN CN201711153716.5A patent/CN107945113B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107945113A (en) | 2018-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107945113B (en) | The antidote of topography's splicing dislocation | |
| CN105245841B (en) | A kind of panoramic video monitoring system based on CUDA | |
| CN105631880B (en) | Lane line dividing method and device | |
| CN110288657B (en) | Augmented reality three-dimensional registration method based on Kinect | |
| CN110111248A (en) | A kind of image split-joint method based on characteristic point, virtual reality system, camera | |
| CN104392435B (en) | Fisheye camera scaling method and caliberating device | |
| CN109064409B (en) | Visual image splicing system and method for mobile robot | |
| CN106157304A (en) | A kind of Panoramagram montage method based on multiple cameras and system | |
| CN106485751B (en) | Unmanned aerial vehicle photographic imaging and data processing method and system applied to foundation pile detection | |
| CN104574347A (en) | On-orbit satellite image geometric positioning accuracy evaluation method on basis of multi-source remote sensing data | |
| CN109544447A (en) | A kind of image split-joint method, device and storage medium | |
| CN112163622B (en) | Global and local fusion constrained aviation wide-baseline stereopair line segment matching method | |
| CN107945221A (en) | A kind of three-dimensional scenic feature representation based on RGB D images and high-precision matching process | |
| CN106910208A (en) | A kind of scene image joining method that there is moving target | |
| CN109697696B (en) | Benefit blind method for panoramic video | |
| CN109117851A (en) | A kind of video image matching process based on lattice statistical constraint | |
| CN106651897A (en) | Parallax correction method based on super pixel segmentation | |
| EP0895189A1 (en) | Method for recovering radial distortion parameters from a single camera image | |
| CN116883464A (en) | Registration method for large-viewing-angle difference optics and SAR remote sensing image | |
| CN110276717B (en) | Image stitching method and terminal | |
| CN105787464A (en) | A viewpoint calibration method of a large number of pictures in a three-dimensional scene | |
| CN109993718A (en) | A kind of multi-channel projection image interfusion method and device | |
| CN103714544B (en) | A kind of optimization method based on SIFT feature Point matching | |
| CN107517366A (en) | Projector's image information method for automatic measurement based on RGBD | |
| CN106878628A (en) | A kind of method that video-splicing is carried out by camera |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Wu Gang Inventor after: Lin Shuhan Inventor after: Hou Wenjing Inventor after: Zheng Wentao Inventor after: Wang Guofu Inventor before: Wu Gang Inventor before: Hou Wenjing Inventor before: Zheng Wentao Inventor before: Wang Guofu |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |