CN109829939A - A method of it reducing multi-view images and matches corresponding image points search range - Google Patents
A method of it reducing multi-view images and matches corresponding image points search range Download PDFInfo
- Publication number
- CN109829939A CN109829939A CN201910048674.1A CN201910048674A CN109829939A CN 109829939 A CN109829939 A CN 109829939A CN 201910048674 A CN201910048674 A CN 201910048674A CN 109829939 A CN109829939 A CN 109829939A
- Authority
- CN
- China
- Prior art keywords
- corresponding image
- image points
- candidate
- picture point
- point
- 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 45
- 238000012545 processing Methods 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000003384 imaging method Methods 0.000 claims description 16
- 238000012937 correction Methods 0.000 claims description 4
- 238000012804 iterative process Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000012892 rational function Methods 0.000 claims 1
- 230000003247 decreasing effect Effects 0.000 abstract description 3
- 238000011002 quantification Methods 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 description 6
- 238000011160 research Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000012505 colouration Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 2
- 238000007670 refining Methods 0.000 description 2
- 230000035939 shock Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 230000000284 resting effect Effects 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses a kind of method of diminution multi-view images matching corresponding image points search range, this method determines initial candidate corresponding image points set first;Secondly it is observation data with all candidate points in picture point to be matched and its initial candidate corresponding image points set, carries out regarding forward intersection compensating computation, and calculate the standardized residual of each group candidate corresponding image points according to adjustment result more;Finally according to the standardized residual distribution map of candidate picture point and the practical corresponding image points position of picture point to be matched, refined processing is iterated to initial candidate corresponding image points set, obtain the corresponding image points search range that the quantification numerical indication that can reflect corresponding image points position feature in multi-view images and siding-to-siding block length are substantially reduced.The method of the present invention can be used for the matched corresponding image points precise search of multi-view images, so that the search number of candidate corresponding image points during Image Matching be greatly decreased, improve the computational efficiency and accuracy rate of multi-view images dense Stereo Matching.
Description
Technical field
The invention belongs to digital photogrammetry, GIS-Geographic Information System and computer vision techniques, and in particular to a kind of diminution
The method of multi-view images matching corresponding image points search range.
Background technique
It is photogrammetric as utilizing the image research shape of subject, position, size, characteristic and mutual alignment relation
A science, is always the technical way of three-dimensional geographic space acquisition of information, and is widely used in base surveying, geography
It is numerous in the developments of the national economy and social development such as national conditions monitoring, resource environment investigation, urban planning and smart city construction
Field.In photogrammetric Data Post, multi-view images matching is a committed step, and it is same that the purpose is to Automatic-searchings
Corresponding image points of the atural object on several images.And multi-view images matching obtain corresponding image points be then automatic triangulation,
The significant data basis of the photogrammetric datas processing links such as geographical information collection, 3 D scene rebuilding.In addition, multi-view images
It fits over and is also played an important role in the application such as video data processing, multisource data fusion, robot visual guidance.Therefore, shadow
As matching is always the hot research problem of digital photogrammetry and computer vision field.
Multi-view images matching includes two basic problems: matching calculating and the corresponding image points search range of similarity measure
It determines.Match measure be judge different picture points on several images whether be corresponding image points foundation, multi-view images can be matched
Accuracy rate and robustness generate important influence.Corresponding image points search range determines the candidate's picture of the same name for participating in matching primitives
The size of the quantity of point, value can generate important influence to the matched reliability of multi-view images and computational efficiency.Currently, existing
The research emphasis of image matching method be how using in local matching window image space information (gray value, feature vector etc.) and
Introduce various constraint conditions (core line, plane shock wave, the syntople etc. between picture point), the robustness of Lai Tigao match measure
With the search range of constraint corresponding image points.But image space information does not have uniqueness and invariance feature, can not construct can be unique
The specific characteristic of ground expression corresponding image points.In addition, due to blocking, the influence of the factors such as geometry deformation, the picture point in match window
Also the constraint conditions such as plane shock wave, spatial relationship holding are usually unsatisfactory for.Therefore, existing image matching method can not overcome shadow
As in " the different spectrum of jljl " or " same object different images " phenomenon bring match measure ambiguity problem and atural object block with perspective distortion because
Spatial relationship variation issue between picture point caused by element;Texture repetitions, weak texture, parallax be discontinuous in processing image, geometry change
When shape etc. matches difficult region, the defects of that there is matching rates is low, matching result poor reliability.
So, the corresponding image points position in multi-view images whether there is certain unique rule? whether there is certain numerical value to refer to
Can mark uniquely indicate the position of corresponding image points? how quantitative description corresponding image points this position rule? whether this rule
Can be used for further refining the corresponding image points in multi-view images matching process search range or uniquely determine corresponding image points?
The research of these problems has great importance for the defect for solving existing image matching method.
Summary of the invention
Goal of the invention: the corresponding image points precise search problem that the purpose of the present invention is be directed in multi-view images matching process,
Based on the reliability theory of measurement adjustment system, a kind of method of diminution multi-view images matching corresponding image points search range is provided,
To obtain the numerical indication of corresponding image points position feature in reflection multi-view images, thus for searching for candidate corresponding image points is greatly reduced
The support of rope number providing method.
Technical solution: a method of it reducing multi-view images and matches corresponding image points search range, include the following steps:
(1) pass through the geometry imaging model and object space elevation information of image for the picture point to be matched in reference images, really
Its fixed initial candidate corresponding image points set with unified marshalling rule on search image;
(2) it is observation data with all candidate points in picture point to be matched and its initial candidate corresponding image points set, is based on
More view forward intersection principles calculate for the least square adjustment of the unknown object space three-dimensional coordinate of corresponding image points, and according to more
Depending on the adjustment result of forward intersection, the standardized residual of each group candidate corresponding image points is calculated;
(3) according to the standardized residual distribution map of candidate picture point and the practical corresponding image points position of picture point to be matched, to first
The candidate corresponding image points set that begins is iterated refined processing;
(4) according to step (3) each time iterative calculation as a result, determining corresponding image points position feature in reflection multi-view images
The corresponding image points search range that quantification numerical indication and siding-to-siding block length substantially reduce.
Further, reference images described in step (1) are any one width image in image set, described search image
To remove the image after reference images in image set, the picture point to be matched is any picture point in benchmark image.
Step (1) includes the following steps:
(11) input have elements of exterior orientation N width multi-view images (if linear array push-broom type imaging space photography image,
Then visual rational polynominal coefficient be image elements of exterior orientation) and image overlay area object space outline elevation range, it is described
Object space outline elevation Range Representation are as follows: [highest elevation value Zmax, minimum height value Zmin];
(12) for reference images I0Upper picture point p to be matched utilizes image according to the outline elevation range of imagery zone
Geometry imaging model determines it in remaining each width search image IkCorresponding epipolar line on MkA candidate's corresponding image points qk i;Described
Geometry imaging model expression formula is as follows:
(r, c)=F (X, Y, Z)
Wherein, r, c are line number and row number of the picture point in image, and X, Y, Z are the object space three-dimensional seat that picture point corresponds to culture point
Mark;F is the function for expressing image geometry imaging model, k=1,2 ..., N-1;I=1,2 ..., Mk;Perspective is projected
Aviation or close-range image, using collinearity condition equation model, for the space satellite image of multi-thread battle array push-broom type, use is reasonable
Function model;
(13) it is constrained based on object coordinates, the candidate corresponding image points quantity of different sizes on each width search image is carried out
Each width is searched for mutually independent each candidate corresponding image points on image by object coordinates and is uniformly organized as M by unification processingzGroup
Initial picture point set, MzFor MkIn maximum value, the initial picture point set are as follows:
{[b1 1, b2 1..., bN-1 1], [b1 2, b2 2..., bN-1 2] ..., [b1 j, b2 j..., bN-1 j] ..., [b1 Mz,
b2 Mz..., bN-1 Mz] (j ∈ { 1,2 ..., Mz})。
Further, step (2) includes the following steps:
(21) for M determined by step (1)zThe candidate corresponding image points set of group, according to each candidate's picture of the same name therein
The line number and row number numerical value of point reject the repetition picture point on every width search image;
It (22) is observation picture point with picture point to be matched and its all candidate corresponding image points rejected after repetition, based on image
Geometry imaging model, building solve the unknown object space three-dimensional coordinate approximation correction X of picture point3×1=[dX dY dZ]TError
Equation matrix V2M×1=A2M×3X3×1-L2M×1;
(23) it is based on the principle of least square, compensating computation is carried out to error equation matrix above-mentioned, and in adjustment result
On the basis of, calculate MzThe standardized residual value of jth group candidate's picture point in groupWherein, ω (bk j) it is that jth group is waited
Kth width in reconnaissance searches for image IkOn candidate corresponding image points bk jStandardized residual value.
Further, step (3) includes the following steps:
(31) M is drawnzThe scatter chart of the standardized residual of the candidate corresponding image points of group, further according to the reality of picture point to be matched
The n comprising real corresponding image points is selected from initial candidate corresponding image points set in border corresponding image points positionzGroup picture point set (nz
≤Mz): { [b1 1, b2 1..., bN-1 1], [b1 2, b2 2..., bN-1 2] ..., [b1 j, b2 j..., bN-1 j] ..., [b1 nz,
b2 nz..., bN-1 nz] (j ∈ { 1,2 ..., nz), as new candidate corresponding image points set;
(32) if nz≠Mz, then with nzObservation picture point of the new candidate corresponding image points set of group as a new compensating computation, weight
The standardized residual of the new candidate point set for carrying out step (2) calculates, and it is new candidate to select the t group comprising real corresponding image points again
Point set;This process iteration carries out, until the picture point quantity in new candidate corresponding image points set no longer changes;
(33) after above-mentioned iterative process, the t value of output is 2.
The utility model has the advantages that compared with prior art, the method for the invention obtain in reflection multi-view images matching process to
The unique error criterion for matching distribution characteristics between the real corresponding image points of picture point and other candidate picture points, is very beneficial for of the same name
The further accurate determination of picture point search range, another aspect this method are expected in algorithm principle level be diminution multi-view images
The corresponding image points search range matched provides a kind of new Research Thinking, thus candidate picture of the same name during Image Matching is greatly decreased
The search number of point, improves the computational efficiency and accuracy rate of Image Matching.
Detailed description of the invention
Fig. 1 is the method frame figure of the embodiment of the present invention;
Fig. 2 (a) is that the multi-view images initial candidate corresponding image points determination based on object space constraint tended under horizontal configuration is shown
It is intended to;
Fig. 2 (b) is that the multi-view images initial candidate corresponding image points determination based on object space constraint tended under vertical form is shown
It is intended to;
Fig. 3 is the unified marshalling signal of the multi-view images initial candidate corresponding image points based on object space constraint of the embodiment of the present invention
Figure;
Fig. 4 (a) is the picture point to be matched in embodiment on the ground of reference images, and its on two width search image
The practical corresponding image points location drawing;
Fig. 4 (b) is the picture point to be matched on the roof of reference images, and its practical same on two width search image
Name image point position figure;
Fig. 4 (c) is the picture point to be matched on the house wall surface of reference images, and its reality on two width search image
The border corresponding image points location drawing;
Fig. 5 is for the picture point to be matched in Fig. 4 (a), and the error criterion of all candidate corresponding image points of calculating is distributed bent
Line chart;
Fig. 6 is for the picture point to be matched in Fig. 4 (b), and the error criterion of all candidate corresponding image points of calculating is distributed bent
Line chart;
Fig. 7 is for the picture point to be matched in Fig. 4 (c), and the error criterion of all candidate corresponding image points of calculating is distributed bent
Line chart.
Specific embodiment
In order to which technical solution disclosed in this invention is described in detail, with reference to the accompanying drawings of the specification and specific embodiment is done
It is further elucidated above.
Disclosed in this invention is a kind of method of diminution multi-view images matching corresponding image points search range, and this method is first
Using the imaging model of image and the object space elevation information of priori, determine that the picture point to be matched in reference images searches for shadow in each width
As the upper initial candidate corresponding image points set with unified marshalling rule;Secondly of the same name with picture point to be matched and its initial candidate
All candidate points in picture point set are observation data, hand in front of more views for the unknown object space three-dimensional coordinate of corresponding image points
Meeting compensating computation, and according to the standardized residual of adjustment result calculating each group candidate corresponding image points;Finally according to candidate picture point
The practical corresponding image points position of standardized residual distribution map and picture point to be matched, changes to initial candidate corresponding image points set
For refined processing, to can reflect corresponding image points position feature in multi-view images as a result, obtaining according to each iterative calculation
Quantification numerical indication and the corresponding image points search range that is substantially reduced of siding-to-siding block length.The method of the present invention can be used for regard more
The corresponding image points precise search of Image Matching, so that the search number of candidate corresponding image points during Image Matching is greatly decreased,
Improve the computational efficiency and accuracy rate of multi-view images dense Stereo Matching.
Basic calculation process of the invention are as follows:
(1) it is regarded in aviation image from the N width of known internal and external orientation more, it is artificial to determine a width as reference images I0,
Remaining N-1 width is as search image I1、I2、…、IN-1, and a picture point to be matched is selected from reference images and search image
And its corresponding N-1 corresponding image points;
(2) for given picture point to be matched, it is based on core line and object space outline elevation range constraint, determines that it is searched in each width
Candidate corresponding image points search range on rope image;And unification processing is carried out to these different search ranges, so that different
The discrete candidate corresponding image points searched on image is unified to organize into groups, to form the candidate corresponding image points set of initial m group;
(3) using picture point to be matched and its image space of the candidate corresponding image points of m group as plane coordinates, based on more view forward intersections
Principle calculate for the least square adjustment of the unknown object space three-dimensional coordinate of picture point;Further according to adjustment result, each group is calculated
The standardized residual of candidate corresponding image points, and draw candidate picture point standardized residual distribution map;
(4) candidate of the same name from m group according to standardized residual distribution map and the practical corresponding image points position of picture point to be matched
The candidate corresponding image points set of new n group comprising real corresponding image points is selected in picture point set;
(5) it if n ≠ m, using the candidate corresponding image points of new n group as initial candidate corresponding image points set, is walked again
Suddenly the standardized residual of (3) calculates and the new candidate corresponding image points set of step (4) determines;
(6) iteration carries out the calculating process of step (5), until the picture point quantity in new candidate corresponding image points set no longer becomes
It turns to only;The candidate corresponding image points of the t group exported at the end of iteration, is the candidate corresponding image points after being refined according to standardized residual
Set, and t can be far smaller than m.
As shown in Fig. 1, a kind of method reducing multi-view images matching corresponding image points search range mainly includes three portions
Point:
(a) multiple groups with unified marshalling rule of the picture point to be matched in reference images on each width search image are determined
Candidate corresponding image points set;
(b) standardized residual of each group candidate's corresponding image points based on measurement adjustment principle calculates;
(c) the iterative refinement processing of the candidate point set based on standardized residual.Specific implementation step are as follows:
Step 1: determining the multiple groups candidate of the same name picture with unified marshalling rule of the point to be matched on each width search image
Point set.
It is constrained based on the object spaces such as core line and object space outline elevation range, determines picture point to be matched on each width search image
The detailed process of multiple groups candidate's corresponding image points set with unified marshalling rule is as follows:
(1) it is regarded in aviation image from the N width of known internal and external orientation more, it is artificial to determine a width as reference images I0,
Remaining N-1 width is as search image I1、I2、…Ik、…、IN-1(k=1,2 ..., N-1), reference images and each width search for image
Photo centre be respectively S0、Sk, corresponding elements of exterior orientation is respectivelyAnd a picture to be matched is selected from reference images and search image
Point p and its corresponding N-1 corresponding image points (the image space ranks number of picture point p to be matched are denoted as (r, c), as plane coordinates is denoted as (x,
y))。
(2) according to priori knowledge, it may be determined that the approximate elevation range of reference images institute covering area: minimum elevation ZminWith
Highest elevation Zmax(this elevation range does not need very accurately, as long as can include the practical elevation of picture point to be matched),
The corresponding culture point P mono- of picture point p is positioned at photography light S0Line segment P on PminPmaxBetween.Utilize image geometry imaging model
The inverse operator formula (see formula (1), by taking the co-colouration effect of central projection image as an example) of (r, c)=F (X, Y, Z), by picture point
As plane coordinates and object space altimeter calculate point Pmin、PmaxObject space plane coordinates (Xmin,Ymin)、(Xmax,Ymax)。
In formula, f be phase owner in elements of interior orientation away from,It is by image I0Foreign side parallactic angle member
Element Nine direction cosines in determining spin matrix.
(3) according to the geometry imaging model of image (see formula (2), by taking the co-colouration effect of central projection image as an example),
It will point Pmin(Xmin,Ymin,Zmin)、Pmax(Xmax,Ymax,Zmax) toward search image IkOn projected, obtain corresponding picture pointAnd its as plane coordinatesWithThen point p to be matched is in image IkOn corresponding image points
pkOne is positioned at line segmentIt is upper that (this line segment is actually the corresponding epipolar line for utilizing elements of exterior orientation to determine, such as attached drawing
Shown in 2), all M on this corresponding epipolar linekA picture point is picture point p to be matched in image IkOn initial candidate corresponding image points.
In formula,It is by search image IkExterior orientation angle elementDetermining rotation
Nine direction cosines in matrix.
Using the principal point coordinate and pixel dimension in elements of interior orientation, picture point can be calculatedImage space row
Row number WithTo obtain corresponding epipolar lineLinear equation, and thus core line equation can
Calculate i-th of candidate point on core lineRanks numberThe linear equation of corresponding epipolar line is by as follows
Two kinds of situations calculate:
1. whenWhen, core line is intended to horizontal configuration (such as attached drawing 2 (a)), straight
Line equation isWherein, slopeInterceptIndependent variableValue range beThe sum of candidate corresponding image pointsInt () is bracket function, and abs () is the function that takes absolute value, and min () is to be minimized function,
Max () is to be maximized function.
2. whenWhen, core line is intended to vertical form (such as attached drawing 2 (b)), straight
Line equation isWherein, slopeInterceptIndependent variableValue range beThe sum of candidate corresponding image points
(4) in step (3), initial candidate corresponding image points quantity M of the point p to be matched on each width search imagekIt is not
With, and the candidate point on different search images is also mutual independence, it is difficult to connection is established according to the sequence number between them.For
Convenient for analyzing the regularity of distribution of candidate corresponding image points, using following method to the candidate corresponding image points on each width search image
Quantity carries out unification processing, so that mutually independent various discrete candidate corresponding image points be organized into groups by object coordinates rule is unified
For different candidate point sets:
1. with candidate corresponding image points quantity Mz=max (Mk| k=1,2 ..., N-1) it searches based on maximum width search image
Rope image (the image I in such as attached drawing 31), remaining N-2 width image is secondary search image.
2. for corresponding epipolar line b in main search image1 1b1 MzOn jth (j ∈ 1,2 ..., Mz) a candidate same placeFirst with double image forward intersection method, by the light S in reference images0Light on p and main search image
Calculate corresponding culture point PjObject space three-dimensional coordinate (Xj,Yj,Zj);Collinearity equation shown in recycling formula (2) and t width pair
Search for image ItThe linear equation of the corresponding epipolar line of (t ∈ { 2 ..., N-1 }), by point PjIt is projected toward each width pair search image
(the dotted line S in such as attached drawing 4tP1、…、StPj、…、StPMz), in image ItOn obtain and pointCorresponding candidate's same place
Then,For jth group candidate's corresponding image points corresponding with point to be matched on all search images.
3. when all candidate corresponding image points on main search image have all carried out the unification processing of step 2., then to be matched
Candidate corresponding image points number of the picture point p on N-1 width search image is all unified for Mz, and the candidate point of same sequence number belongs to together
One group.That is, the different candidate point of quantity is unified into following M on original each width search imagezGroup: { [b1 1,b2 1,…,
bN-1 1],[b1 2,b2 2,…,bN-1 2],…,
It is to be noted that through unification treated MzThe candidate corresponding image points of group, other than main search image, it is other certain
Width pair searches for image ItOn MzA candidate pointIt is to exist to repeat point.
Step 2: the standardized residual of each group candidate's corresponding image points based on measurement adjustment principle calculates.
To each group candidate's corresponding image points set { [b after the picture point p to be matched in reference images, and its unified marshalling1 i,
b2 i,…,bN-1 i]|i∈{1,2,…,Mz, each group candidate's corresponding image points based on measurement adjustment principle is carried out according to the following procedure
Standardized residual calculate, to obtain the standardized residual value ω of i-th group of candidate's corresponding image pointsi:
(1) duplicate candidate corresponding image points is removed.It can by the determination principle of candidate corresponding image points on search image above-mentioned
Know, kth width searches for M on imagezA candidate pointIt is middle to there is repetition point, and repeat candidate point and be only possible to occur
In the continuous consecutive points of sequence number.So since the 2nd candidate point, judge one by one its with adjacent previous point whether
It repeats, if repeating, is marked as repeating point.As all MzAfter a candidate point all carries out differentiation label, can it not weighed
Multiple MkA candidate pointIn addition, original repetition point set also can be obtained after repeated point differentiates
Midpoint sequence number j (j ∈ 1,2 ..., Mz) with duplicate removal complex point after put centrostigma sequence number j ' (j ' ∈ 1,2 ..., Mk)
Corresponding relationship: (function Cor indicates at j-th point to j '=Cor (j)With jth ' a pointIt is the same point).At this point, benchmark shadow
The sum of corresponding image points on picture and N-1 width search image is
(2) error equation of all picture points is constructed.According to the geometry imaging model of image, with the picture plane of picture point on image
Coordinate is observation, carries out Taylor series linearisation to the unknown object space three-dimensional coordinate (X, Y, Z) of picture point, can list formula (3) institute
The error equation (for the collinearity equation of the central projection image shown in formula (2)) of the solution object space three-dimensional coordinate shown:
In formula, (x, y) is the picture plane coordinates observation of picture point, (vx, vy) be observation residual error, ((x)0, (y)0) be
By the approximate object coordinates (X of picture point0, Y0, Z0) formula (2) approximation calculated is brought into as plane coordinates, (dX, dY, dZ) is wait ask
The correction of the object space three-dimensional coordinate approximation of solution.And 6 coefficient (a of error equation11, a12, a13, a21, a22, a23) press
Formula calculates:
In formula, (a1, a2, a3;b1, b2, b3;c1, c2, c3) it is nine direction cosines in the spin matrix of image, (Xs, Ys,
Zs) be image photographic center object space three-dimensional coordinate.
According to formula (3), for all M corresponding image points in reference images and each width search image, by first reference images
The sequence of image is searched again for, lists the error equation of picture point thereon one by one, forms the error equation matrix as shown in formula (5).
V2M×1=A2M×3X3×1-L2M×1 (5)
In formula, V is 2M as the residual matrix (dimension is 2M row × 1 column) of plane coordinates observation, and X is object to be solved
The approximation correction matrix number (dimension is the column of 3 rows × 1) of square three-dimensional coordinate, A, L are respectively the coefficient matrix and often of error equation
Several matrixes (dimension is respectively 2M row × 3 column, 2M row × 1 column).The concrete form of these matrixes is shown in formula (6).
In formula, the subscript of each element indicates that video number, subscript indicate the serial number of data item and point in matrix.
(3) reliability matrix is calculated.According to the power battle array P of factor arrays A, constant matrix L and observation in error equationll(herein
Observation be mutually indepedent observation, this matrix is unit matrix E), following reliability matrix R can be calculated:
(4) resolution error equation.Using the principle of least square, error equation matrix shown in formula (5) is solved, is obtained unknown
Error σ in matrix number X, the residual matrix V and weight unit of all observations0:
(5) the i-th " standardized residual ω (i ") of a point in M corresponding image points is calculated.Since each picture point has x, y two
As plane coordinates observation, the average value of the standardized residual of x, y of capture point as the point standardized residual (see formula
(9))。
(6) it calculates kth width and searches for image IkOn j-th candidates corresponding image pointsStandardized residual valueDue to
PointSequence number j correspond to duplicate removal complex point after sequence number j ', and from reference images to -1 width of kth search for image, deduplication
The sum of corresponding image points afterwards isSoCalculation it is as follows:
(7) the standardized residual value ω of i-th group of candidate's corresponding image points is calculatedi.Due to i-th group of candidate's corresponding image points [b1 i,
b2 i,…,bN-1 i] be made of N-1 candidate picture point on N-1 width search image, so, N-1 candidate picture point is taken as the following formula
Standardized residual average value as ωi:
Step 3: the iterative refinement of the candidate point set based on standardized residual is handled.
M of the picture point p to be matched on N-1 width search image is calculated in second stepzCandidate corresponding image points the set { [b of group1 1,
b2 1,…,bN-1 1],[b1 2,b2 2,…,bN-1 2],…,[b1 i,b2 i,…,bN-1 i],…,In every group
The standardized residual value ω of picture pointi(i∈{1,2,…,Mz) after, according to the following procedure at the iterative refinement of progress candidate point set
Reason:
(1) according to the standardized residual value ω of every group of picture pointi, using group number i as horizontal axis, with standardized residual value ωiIt is vertical
Axis draws out MzThe standardized residual scatter chart of the candidate corresponding image points of group.
(2) according to MzThe features of shape and picture point p to be matched of the standardized residual scatter chart of the candidate corresponding image points of group
Practical corresponding image points position on N-1 width search image, if the standardized residual distribution curve of candidate picture point present it is minimum
The approximate V-arrangement distribution shape of point, then from MzThe n comprising real corresponding image points is selected in the candidate corresponding image points set of groupzGroup picture point
Gather (nz≤Mz): { [b1 1,b2 1,…,bN-1 1],[b1 2,b2 2,…,bN-1 2],…,[b1 nz,
b2 nz,…,bN-1 nz]}(j∈{1,2,…,nz), as new candidate corresponding image points set.
(3) if nz≠Mz, then to nzThe new candidate corresponding image points set of group, re-starts the standard of the candidate point set of second step
Change residual computations, and selects the new candidate point set of t group comprising real corresponding image points again;If t ≠ nz, then continue candidate
The standardized residual of point set calculates, and selects the new candidate point set of t group next time;Above procedure iteration carries out, until new wait
Until selecting the picture point quantity in corresponding image points set no longer to change.
(4) the candidate corresponding image points of the t group exported at the end of above-mentioned iterative process, is according to standardized residual iteration
Candidate corresponding image points set after refining.
According to above step, the experiment for having selected the experimental data of 4 (a)-Fig. 4 (c) of attached drawing to carry out the method for the present invention is tested
Card.Attached drawing 4 illustrates 3 picture points to be matched resting on the ground, on top of building, building wall facade, and its searches in 2 width
Corresponding image points position on rope image (straight line on search image is corresponding epipolar line);To 3 picture points to be matched in attached drawing 4 into
The iterative refinement of row candidate corresponding image points set of the invention calculates, and the standardization of the candidate corresponding image points of each iterative calculation is residual
Poor distribution curve is as shown in attached drawing 5,6,7.
With reference to the accompanying drawings 5,6,7 as a result, the present invention sums up the position feature of corresponding image points in multi-view images are as follows: 1. for
Picture point to be matched in multi-view images on a certain width image, it is all in the region of search on remaining each width image corresponding epipolar line
" V " font Distribution Phenomena is presented in the standardized residual value of candidate corresponding image points.2. residual in the standardization of all candidate corresponding image points
On distribution of the difference curve, Minimum Residual difference point is substantially at the middle position of all candidate points, moreover, really corresponding image points position
It is not good enough between starting point (or terminal) in Minimum Residual.3. from " V " font distribution curve, by the minimum comprising real corresponding image points
For the local point set of left (or right) side of residue points as new candidate point set, iteration carries out the adjustment of the standardized residual of new point set
It calculates, until the picture point quantity in new candidate point set no longer changes.Then in each iterative process, each candidate's corresponding image points
Standardized residual value still obey rule Distribution Phenomena 1., 2.;Moreover, the candidate after iteration, on every width search image
The quantity of corresponding image points can be reduced to 2.
Therefore, from the results, it was seen that method of the invention, which can get, distinguishes picture to be matched in multi-view images matching process
The unique error criterion and its distribution characteristics of the real corresponding image points of point and other candidate picture points.By the limited of the method for the present invention
The quantity of the iterative calculation of number, candidate corresponding image points of the picture point to be matched on each width search image in reference images is final
2 can be reduced to, the search number of candidate corresponding image points in multi-view images matching process is greatly reduced in this, is very beneficial for mentioning
The computational efficiency and accuracy rate of high multi-view images matching (especially to regard dense Stereo Matching) more.
Claims (6)
1. a kind of method for reducing multi-view images matching corresponding image points search range, characterized by the following steps:
(1) the geometry imaging model and object space elevation information for passing through image for the picture point to be matched in reference images, determine it
The initial candidate corresponding image points set with unified marshalling rule on search image;
It (2) is observation data with all candidate points in picture point to be matched and its initial candidate corresponding image points set, based on more views
Forward intersection principle calculate for the least square adjustment of the unknown object space three-dimensional coordinate of corresponding image points, and according to more views before
The adjustment result just intersected calculates the standardized residual of each group candidate corresponding image points;
(3) it according to the standardized residual distribution map of candidate picture point and the practical corresponding image points position of picture point to be matched, is waited to initial
Corresponding image points set is selected to be iterated refined processing;
(4) according to iterating to calculate as a result, determining that corresponding image points position feature quantifies in reflection multi-view images step (3) each time
Change the corresponding image points search range that numerical indication and siding-to-siding block length substantially reduce.
2. a kind of method for reducing multi-view images matching corresponding image points search range according to claim 1, feature exist
In: reference images described in step (1) are any one width image in image set, and described search image is to remove in image set
Other images outside reference images, the picture point to be matched are the arbitrary point in benchmark image.
3. a kind of method for reducing multi-view images matching corresponding image points search range according to claim 1, feature exist
In: step (1) includes the following steps:
(11) input has the N width multi-view images of elements of exterior orientation and the object space outline elevation range of image overlay area, described
Object space outline elevation Range Representation are as follows: [highest elevation value Zmax, minimum height value Zmin];
(12) for reference images I0Upper picture point p to be matched utilizes the geometry of image according to the outline elevation range of imagery zone
Imaging model determines it in remaining each width search image IkCorresponding epipolar line on MkA candidate's corresponding image points qk i;The geometry
Imaging model expression formula is as follows:
(r, c)=F (X, Y, Z)
Wherein, r, c are line number and row number of the picture point in image, and X, Y, Z are the object space three-dimensional coordinate that picture point corresponds to culture point;F
For the function for expressing image geometry imaging model, k=1,2 ..., N-1;I=1,2 ..., Mk;For the boat of perspective projection
Empty or close-range image, uses rational function for the space satellite image of multi-thread battle array push-broom type using collinearity condition equation model
Model;
(13) it is constrained, the candidate corresponding image points quantity of different sizes on each width search image is carried out consistent based on object coordinates
Each width is searched for mutually independent each candidate corresponding image points on image by object coordinates and is uniformly organized as M by change processingzGroup is initial
Picture point set, MzFor MkIn maximum value, the initial picture point set are as follows:
4. a kind of method for reducing multi-view images matching corresponding image points search range according to claim 3, feature exist
In: if the space photography image of linear array push-broom type imaging in step (11), then visual rational polynominal coefficient is the outer of image
The element of orientation.
5. a kind of method for reducing multi-view images matching corresponding image points search range according to claim 1, feature exist
In: step (2) includes the following steps:
(21) for M determined by step (1)zThe candidate corresponding image points set of group, according to the row of each candidate corresponding image points therein
Number with row number numerical value, reject the repetition picture point on every width search image;
It (22) is observation picture point with picture point to be matched and its all candidate corresponding image points rejected after repetition, based on the several of image
What imaging model, building solve the unknown object space three-dimensional coordinate approximation correction X of picture point3×1=[dX dY dZ]TError equation
Matrix V2M×1=A2M×3X3×1-L2M×1;
(23) it is based on the principle of least square, compensating computation is carried out to error equation matrix above-mentioned, and on the basis of adjustment result
On, calculate MzThe standardized residual value of jth group candidate's picture point in groupWherein,For jth group candidate point
In kth width search for image IkOn candidate corresponding image pointsStandardized residual value.
6. a kind of method for reducing multi-view images matching corresponding image points search range according to claim 1, feature exist
In: step (3) includes the following steps:
(31) M is drawnzThe scatter chart of the standardized residual of the candidate corresponding image points of group, further according to the practical same of picture point to be matched
Name image point position, selects the n comprising real corresponding image points from initial candidate corresponding image points setzGroup picture point set (nz≤
Mz): As new candidate corresponding image points set;
(32) if nz≠Mz, then with nzObservation picture point of the new candidate corresponding image points set of group as a new compensating computation, again into
The standardized residual of the candidate point set of row step (2) calculates, and selects the new candidate point set of the t group comprising real corresponding image points again
It closes;This process iteration carries out, until the picture point quantity in new candidate corresponding image points set no longer changes;
(33) after above-mentioned iterative process, the t value of output is 2.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910048674.1A CN109829939B (en) | 2019-01-18 | 2019-01-18 | Method for narrowing search range of multi-view image matching same-name image points |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910048674.1A CN109829939B (en) | 2019-01-18 | 2019-01-18 | Method for narrowing search range of multi-view image matching same-name image points |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109829939A true CN109829939A (en) | 2019-05-31 |
CN109829939B CN109829939B (en) | 2023-03-24 |
Family
ID=66860940
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910048674.1A Active CN109829939B (en) | 2019-01-18 | 2019-01-18 | Method for narrowing search range of multi-view image matching same-name image points |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109829939B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117664088A (en) * | 2024-01-31 | 2024-03-08 | 中国人民解放军战略支援部队航天工程大学 | Method, system and equipment for determining homonymy point by ultra-wide vertical orbit circular scanning satellite image |
CN117664087A (en) * | 2024-01-31 | 2024-03-08 | 中国人民解放军战略支援部队航天工程大学 | Method, system and equipment for generating vertical orbit circular scanning type satellite image epipolar line |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103606151A (en) * | 2013-11-15 | 2014-02-26 | 南京师范大学 | A wide-range virtual geographical scene automatic construction method based on image point clouds |
CN103604417A (en) * | 2013-11-15 | 2014-02-26 | 南京师范大学 | Multi-view image bidirectional matching strategy with constrained object information |
CN104318566A (en) * | 2014-10-24 | 2015-01-28 | 南京师范大学 | Novel multi-image plumb line track matching method capable of returning multiple elevation values |
-
2019
- 2019-01-18 CN CN201910048674.1A patent/CN109829939B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103606151A (en) * | 2013-11-15 | 2014-02-26 | 南京师范大学 | A wide-range virtual geographical scene automatic construction method based on image point clouds |
CN103604417A (en) * | 2013-11-15 | 2014-02-26 | 南京师范大学 | Multi-view image bidirectional matching strategy with constrained object information |
CN104318566A (en) * | 2014-10-24 | 2015-01-28 | 南京师范大学 | Novel multi-image plumb line track matching method capable of returning multiple elevation values |
Non-Patent Citations (1)
Title |
---|
张卡等: "基于数字视差模型和改进SIFT特征的数字近景立体影像匹配", 《测绘学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117664088A (en) * | 2024-01-31 | 2024-03-08 | 中国人民解放军战略支援部队航天工程大学 | Method, system and equipment for determining homonymy point by ultra-wide vertical orbit circular scanning satellite image |
CN117664087A (en) * | 2024-01-31 | 2024-03-08 | 中国人民解放军战略支援部队航天工程大学 | Method, system and equipment for generating vertical orbit circular scanning type satellite image epipolar line |
CN117664088B (en) * | 2024-01-31 | 2024-04-02 | 中国人民解放军战略支援部队航天工程大学 | Method, system and equipment for determining homonymy point by ultra-wide vertical orbit circular scanning satellite image |
CN117664087B (en) * | 2024-01-31 | 2024-04-02 | 中国人民解放军战略支援部队航天工程大学 | Method, system and equipment for generating vertical orbit circular scanning type satellite image epipolar line |
Also Published As
Publication number | Publication date |
---|---|
CN109829939B (en) | 2023-03-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106127771B (en) | Tunnel orthography system and method is obtained based on laser radar LIDAR point cloud datas | |
CN105792353B (en) | Crowd sensing type WiFi signal fingerprint assisted image matching indoor positioning method | |
CN103604417B (en) | The multi-view images bi-directional matching strategy that object space is information constrained | |
CN113592989B (en) | Three-dimensional scene reconstruction system, method, equipment and storage medium | |
CN103337052B (en) | Automatic geometric correcting method towards wide cut remote sensing image | |
CN106529538A (en) | Method and device for positioning aircraft | |
CN110084785B (en) | Power transmission line vertical arc measuring method and system based on aerial images | |
CN113916130B (en) | Building position measuring method based on least square method | |
CN110889899A (en) | Method and device for generating digital earth surface model | |
CN114241464A (en) | Cross-view image real-time matching geographic positioning method and system based on deep learning | |
CN104318566B (en) | Can return to the new multi-view images plumb line path matching method of multiple height values | |
CN112946679B (en) | Unmanned aerial vehicle mapping jelly effect detection method and system based on artificial intelligence | |
CN112270698A (en) | Non-rigid geometric registration method based on nearest curved surface | |
CN117315146B (en) | Reconstruction method and storage method of three-dimensional model based on trans-scale multi-source data | |
CN112489099A (en) | Point cloud registration method and device, storage medium and electronic equipment | |
Hong et al. | Rapid three-dimensional detection approach for building damage due to earthquakes by the use of parallel processing of unmanned aerial vehicle imagery | |
CN114140539A (en) | Method and device for acquiring position of indoor object | |
CN109829939A (en) | A method of it reducing multi-view images and matches corresponding image points search range | |
CN117218201A (en) | Unmanned aerial vehicle image positioning precision improving method and system under GNSS refusing condition | |
CN106875449B (en) | A kind of non-scalability camera calibration method of unmanned plane based on flying quality | |
CN114140700A (en) | Step-by-step heterogeneous image template matching method based on cascade network | |
CN114137564A (en) | Automatic indoor object identification and positioning method and device | |
CN112634447B (en) | Outcrop stratum layering method, device, equipment and storage medium | |
CN112132950B (en) | Three-dimensional point cloud scene updating method based on crowdsourcing image | |
CN113487741B (en) | Dense three-dimensional map updating method and device |
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 |