CN109186550A - A kind of coding and decoding and measurement method of codified close-range photogrammetry mark - Google Patents
A kind of coding and decoding and measurement method of codified close-range photogrammetry mark Download PDFInfo
- Publication number
- CN109186550A CN109186550A CN201810803118.6A CN201810803118A CN109186550A CN 109186550 A CN109186550 A CN 109186550A CN 201810803118 A CN201810803118 A CN 201810803118A CN 109186550 A CN109186550 A CN 109186550A
- Authority
- CN
- China
- Prior art keywords
- coding
- image
- decoding
- matrix
- mark
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
- G01C11/04—Interpretation of pictures
- G01C11/06—Interpretation of pictures by comparison of two or more pictures of the same area
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to the coding and decodings and measurement method of a kind of codified close-range photogrammetry mark, comprising the following steps: 1) cataloged procedure: 11) encoding the setting of multiplying power;12) generation of encoder dictionary;13) generation of coding maker;2) decoding process: the 21) pretreatment of coded image;22) extraction of central point;23) conversion and decoding of one-dimensional bar code;3) measurement process: after completing decoding, the rotation scaling relation occurred on the position mark point for encoding member in two images is resolved with translational movement using improved Fourier-Mellin transform algorithm.Compared with prior art, the present invention has many advantages, such as quick decoding, capacity controllable, registration, promotes wheel measuring precision.
Description
Technical field
The present invention relates to digital close range photogrammetry fields, more particularly, to a kind of codified close-range photogrammetry mark
Coding and decoding and measurement method.
Background technique
Because of its good scalability and powerful function, oneself is increasingly becoming current reverse-engineering to digital close range photogrammetry
In be best suited for a kind of technology of in-site measurement, be widely used in that field scene is deployed to ensure effective monitoring and control of illegal activities, large scale structure monitoring, industry produce in recent years
In the dimensional measurement and Design and manufacturing process of product.
With increasing for large scale and structure is complicated object measurement task, existing measuring system coding maker is also referred to as encoded
The deficiency of mark sum has constrained its performance for integrally measuring advantage.Extended coding number improves data-handling capacity,
Through become guarantee measurement accuracy and working efficiency there is an urgent need to.Meanwhile the location information of traditional photography measurement index point is external
Portion deploys to ensure effective monitoring and control of illegal activities information there are higher dependence, needs additionally to lay control net, in such a way that multi-group data measures adjustment, come
Obtain the geometry location information of index point.But coded target itself is used only, traditional coded target is difficult directly to obtain
High-precision relative displacement needed for obtaining geometry location information, especially large scale structure monitoring and industrial products measurement and rotation letter
Breath.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of codified close shots to take the photograph
The coding and decoding and measurement method of shadow measurement mark.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of coding and decoding and measurement method of codified close-range photogrammetry mark, comprising the following steps:
1) cataloged procedure:
11) setting of multiplying power: the coding member of the multiple fan-shaped one-dimension codes of building is encoded, and sets it and encodes multiplying power;
12) encoder dictionary the generation of encoder dictionary: is constituted by the coding member of multiple corresponding different coding information;
13) generation of coding maker;Complete coding maker is constituted by the combination of multiple coding members;
2) decoding process:
21) pretreatment of coded image obtains binary map only containing marginal information from the coding maker image of input
Picture;
22) extraction of central point: flat using the ellipse fitting of least square according to the bianry image only containing marginal information
The central point of difference algorithm acquisition coding maker;
23) conversion and decoding of one-dimensional bar code: using the arc searching method of given radius, i.e., with the central point of extraction
Centered on, it is searched for the black and white gray value that given radius carries out 360 degree, coding maker is converted to the coding of one-dimensional bar code
Afterwards, according to the corresponding angle of each coding member of encoder dictionary, when monochrome pixels value carries out traversal matching, obtains complete coding
The value of information completes corresponding decoding;
3) measurement process:
After completing decoding, using improved Fourier-Mellin transform algorithm to the positioning mark for encoding member in two images
The rotation scaling relation occurred on will point is resolved with translational movement.
In the step 11), the first fan by position mark point and four kinds of different angles of the coding of each sector one-dimension code
Shape figure spot permutation and combination is formed, and the position mark point is made of multiple and chequered with black and white concentric circles, preliminary fixed to realize
Position, the fan-shaped figure spot include wide figure spot and narrow figure spot, and specially narrow black, narrow white, wide black and width white are described
Encode the ratio for the corresponding central angle size that multiplying power δ is wide figure spot and narrow figure spot.
The fan-shaped figure spot of each coding member is made of three wide figure spots and three narrow figure spots, the information that each coding member includes
It is determined by the distribution of width figure spot and the configuration of black-white colors.
In the step 12), the information of encoder dictionary include number, letter and and additional character, the number
It is used to generate encoded information with letter, the additional character is the white space for encoding first upper left side fourth quadrant position, is used
In the start-stop of coding and the mark of calibration information.
The encoder dictionary adjusts the capacity of encoder dictionary by adjusting the size of coding multiplying power δ.
The step 21) specifically includes the following steps:
211) generation two generation of bianry image: is split to the coding maker image of input by given gray threshold
It is worth image;
212) side of coding maker the extraction of edge contour: is extracted from bianry image by Canny edge detection operator
Edge information.
In the step 23), given radius is 0.6 times of figure spot radius.
In the step 23), search process is started with any angle, according to the maximum white ash obtained in searching route
Mark of the angle value interval as judgement coding start-stop symbol, and using the angle of the wide-to-narrow ratio of one-dimensional bar code replacement coding member
Than.
The step 3) specifically includes the following steps:
31) the reference picture G of original logo point is obtained1With the measurement image for the index point for needing to measure rotation shift angle
G2Between translational movement, specifically:
311) respectively to the reference picture G of original logo point1With the measurement for the index point for needing to measure rotation shift angle
Image G2Discrete Fourier transform is carried out, and calculates crosspower spectrum matrix Q (u, v), the crosspower spectrum matrix Q's (u, v)
Calculating formula are as follows:
S2(u, v)=S1(u,v)exp{-i(ux0+vy0)}
Wherein, S2(u, v) is corresponding G2Coordinate is that the pixel of (x, y) obtains after discrete Fourier transform in image array
The matrix S arrived2Middle corresponding points as a result, S1 *(u, v) is G1Coordinate is the pixel of (x, y) in discrete fourier in image array
The matrix S obtained after transformation1Conjugate matrices S* 1Middle corresponding points as a result, u is the image obtained after discrete Fourier transform
The abscissa of corresponding points in fourier matrix S, v are in the fourier matrix S of image obtained after discrete Fourier transform
The ordinate of corresponding points, x0For the image array G before discrete Fourier transform1And G2Corresponding points abscissa direction displacement
Value, y0For the image array G obtained before discrete Fourier transform1And G2In corresponding points ordinate direction shift value;
312) polar coordinate transform is carried out to crosspower spectrum matrix Q (u, v) and obtains phase angle matrix ψ (u, v), it may be assumed that
ψ (u, v)=∠ Q (u, v)=ux0+vy0,
Plane fitting, the plane expression formula being fitted are carried out using stochastical sampling unification algorism are as follows:
a1u+b1v+c1+d1=0
Wherein, a1、b1、c1For the Slope Parameters of the plane equation, and c1=0, d1For the constant term of plane equation, and d1=
ψ(u,v);
313) the Slope Parameters a and b of plane equation are solved to get the shift value x of point image is marked to two width0=
A and y0=b;
32) the rotation scaling relation of image, including rotation angle, θ are obtained by log-polar transform0And zoom scale
γ, specifically:
321) respectively to the reference picture G of original logo point1With the measurement for the index point for needing to measure rotation shift angle
Image G2Carry out log-polar conversion;
322) Discrete Fourier Transform is carried out to the image after polar coordinates conversion and obtains the crosspower spectrum matrix Q under polar coordinates
(m, n), and obtain the phase angle matrix ψ (m, n) of matrix Q (m, n):
ψ (m, n)=∠ Q (m, n)=mln γ+n θ0;
323) plane equation is solved after carrying out plane fitting using stochastical sampling unification algorism to phase angle matrix ψ (m, n)
Slope Parameters a2And b2, the final rotation angle, θ for obtaining two width label point image0=b2And zoom scale
Solution before plane fitting twines step are as follows:
To eliminate influence of the crosspower spectrum data noise to phase unwrapping, adopted in the Fourier-Mellin transform algorithm
It is denoised with the method for vector filtering algorithm, the real and imaginary parts of crosspower spectrum is separated, point
Not Shi Yong the mean filter of wicket denoise, the phase angular moment that winds after arc tangent is denoised is carried out to filtered matrix
Battle array, and solution is carried out to phase angle matrix using the phase unwrapping algorithm based on minimum cost network stream and is twined.
Compared with prior art, the invention has the following advantages that
One, quickly decoding: proposing a kind of arc search strategy based on given radius in the present invention, realizes fan-shaped compile
Symbol is quickly converted to one-dimensional bar code, avoids the occurrence of unstable during lines detection in image, is realized and is compiled
The quick steady decoding of code.
Two, it capacity controllable: in the present invention using a kind of very useful concentric ring type index point as prototype, is added to and is based on
The coding element array of fan-shaped one-dimension code designs, and can carry out code Design.The dimension of the coded primary character allusion quotation can be according to reality
Border needs to be customized, to realize the reduction and expansion of code capacity.
Three, a kind of improved Fourier-Mellin transform algorithm complement mark point rotation registration: is used in the present invention
The accurate extraction of angle and translation distance, obtains the location information of photogrammetric index point.
Four, promote wheel measuring precision: consistent (RANSAC) algorithm of Cao Yong random sampling of the present invention is calculated instead of least square
Method carrys out the parameter a and b of robust iterative areal model au+bv-Q (u, v)=0, and only selection meets the number of the areal model after solution twines
According to progress phase pushing figure estimation, and elimination of rough difference is used as in robust iterative by the data of deviation effects, to weaken phase
Pass process large deviations are influenced caused by least-squares estimation, the precision and stability of deviant are improved, thus further
Promote the precision of wheel measuring.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention.
Fig. 2 is coded primary character allusion quotation schematic diagram.
Fig. 3 is two and uses identical encoder dictionary but the coding maker containing different coding information.
Fig. 4 is the flow chart of decoding process.
Fig. 5 is the Comparative result before and after image preprocessing, wherein figure (5a) is original image, and (5b) is binarization result,
(5c) is edge extracting result.
Fig. 6 is the ellipse fitting result example of edge image.
Fig. 7 is the one-dimensional coding transition diagram of the arc search strategy based on given radius.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.
Embodiment:
The present invention provides coding and decoding and measurement method including a kind of codified close-range photogrammetry mark, including three
The core content divided: the code Design of mark, the decoding design of mark, mark are displaced the measurement with rotation.
1, the code Design of close-range photogrammetry mark
As shown in Figure 1, the code Design of the novel close-range photogrammetry mark is divided into three key steps: determining coding times
Rate generates encoder dictionary, generates coding maker.
The setting of 1.1 coding multiplying powers
The core of entire code Design scheme is the coding member based on fan-shaped one-dimension code.Fig. 2 gives fan-shaped coding member
Sample.There are four types of the fan-shaped figure spot permutation and combination of different angles to obtain for each fan-shaped coding member.Fan-shaped figure spot includes four kinds: black
(narrow), white (narrow), black (width), white (width).Wherein, the size ratio of the angle value of wide figure spot and narrow figure spot, is referred to as compiled
The multiplying power δ of code.The multiplying power of coding will directly determine the number of the maximum coding member that can be accommodated of a coding maker.One coding
The number that member is encoded contained in mark is more, and the code capacity upper limit of the coding maker is also higher in principle.What Fig. 2 was provided
The size ratio of the angle value of coding multiplying power δ=3 of coding member, i.e., wide figure spot and narrow figure spot is three times.Under normal circumstances, often
A coding member can be made of three wide figure spots and three narrow figure spots.Wherein, the distribution of width figure spot and the configuration of black-white colors will
Determine the information that coding member is included.
The generation of 1.1 encoder dictionaries
The set of multiple coding members containing different coding information will constitute the dictionary (see Fig. 2) of coding member, in coding member
In dictionary, the different coding members encoded information that correspondence is different.Given in Fig. 2 go out encoder dictionary contain from digital 0-9,
Alphabetical A-F and four additional character % ,=,+,-.In the design of a whole set of encoder dictionary, number is mainly used for encoding with letter
The generation of information, and additional character will be mainly used for the mark of the start-stop and calibration information of coding.The encoder dictionary provided in Fig. 2
It altogether include 20 elements, but actually by the adjustment of coding multiplying power δ, making by oneself for encoder dictionary capacity is may be implemented in the present invention
Justice setting.
The generation of 1.1 coding makers
The combination of multiple coding members may be constructed complete coding maker.Fig. 3 gives two and uses identical encoder dictionary
But the coding maker containing different coding information.There is no rotation, the upper left side (the of each coding maker
Four-quadrant) position remains with certain white space, for identifying the start-stop and calibration code information of fan-shaped coding metasequence.It removes
Except this, at the center of coding maker, there are four position mark points composed by concentric circles.Position mark point is for encoding mark
In the decoding process of will, to the Primary Location of fan-shaped coding member.The extraction of circular edge and the determining minimum two by elliptic equation
Multiply fitting to obtain.Meanwhile chequered with black and white concentric design improves more error redundancies, so that the extraction of central point can be with
By modes such as multiple or total least square fittings, to overcome conventional sign dot center to extract vulnerable to noise and image light and shade
Spend the defect influenced.
2, the decoding of close-range photogrammetry mark
As shown in figure 4, decoding design of the invention is also classified into three key steps: the pretreatment of coded image, central point
Extraction, one-dimensional coding conversion and decoding.
The pretreatment of 2.1 coded images
The purpose of pretreated core of coded image is only to be contained side from Conventional color/gray level image of coding maker
The bianry image of edge information, in order to which subsequent central point extracts and decode operation.Particularly, entire pretreatment includes two
Key step: the generation of bianry image and the extraction of edge contour.
Bianry image is that only there are two the digital pictures of probable value for each pixel.Relative to original input picture, binary map
As eliminating adverse effect caused by picture contrast, the geological information of image is only remained.Bianry image in this programme
It generates using based on the dividing method for providing gray threshold.By given gray threshold θ, to the every of original input picture matrix
A pixel is traversed and is judged.Formula 1 gives the conversion formula by original image G (x, y) to bianry image B (x, y):
After obtaining bianry image, the present invention will use Canny edge detection operator to extract coding from bianry image
The marginal information of mark.Canny edge detection operator is the multistage edge that John F.Canny developed in 1986
Detection algorithm,
Its key step includes:
1. carrying out smoothed image using gaussian filtering, it is therefore an objective to remove noise.
2. looking for the intensity gradient (intensity gradients) of image.
3. application non-maximum suppression (non-maximum suppression) technology come eliminate side erroneous detection (originally be not but
Detected is).
4. determining possible (potential) boundary using the method for dual threshold.
5. tracking boundary using hysteresis techniques.Fig. 5 gives the result pair of original image, bianry image and edge extracting
Than.
The extraction of 2.2 central points
After the edge image for obtaining index point, edge image is also used for based on morphologic corrosion and expansive working
Processing, taken exercise with eliminating the edge occurred during edge extracting, burrs on edges phenomena such as.Then, based on elliptic equation
Least square fitting algorithm is applied in edge image.Formula group 2 gives the adjustment formula of ellipse fitting:
Ax′2+Bx′y′+Cy′2+ Dx '+Ey '+1=0 (2.1)
V=AX-L
X=[A B C D E]T (2.2)
Wherein the formula of the first row is elliptical parametric equation, and the formula of second and third row is elliptical least square adjustment
Equation, x0With y0For the initial value of adjustment, θ is the eccentricity of fitting result.Least square fitting is carried out using ellipse formula, and
Circle formula used in non-traditional index point carries out least square fitting, can effectively overcome as produced by projection error
Circular pattern problem on deformation, increase the redundancy of fitting.By adjusting different initial value x0With y0, can obtain different
Ellipse fitting result.According to the residual values V of different adjustment results, the validity of ellipse fitting can be judged.Ideal feelings
Under condition, residual values V is smaller, and the result of fitting is more reliable, is also more likely to become the Candidate Set of central point concentric circles.In addition, logical
The eccentricity θ that over-fitting result obtains can help to judge, the profile of fitting whether centered on put corresponding concentric circles.By right
The screening of fitting result under different parameters, the optimal fitting result of first four, corresponding common central coordinate of circle point (x ', y '),
The as central point of coded target.
The decoding of 2.3 coded targets
After the central point of coded target has been determined, a kind of arc search strategy based on given radius is applied to
In the decoding process of coding.Centered on determining index point, carried out with given radius (0.6 times of index point design radial)
The search of black and white gray value, and 360 ° of search result is recorded, get converted to the coding of one-dimensional bar code.Fig. 7 gives
The schematic diagram of this search process.The starting of search process is started with any angle, then according to obtained in searching route most
Mark of the big white gray value interval as judgement coding start-stop symbol.It is converted to the corresponding fan-shaped volume of wide-to-narrow ratio of one-dimensional bar code
The angle ratio of symbol.In decoding process, the angle ratio of fan-shaped coding member, Ke Yiyou are replaced using the wide-to-narrow ratio of one-dimensional bar code
Angle error caused by effect overcomes projection angle to convert, and constitute error caused by the line drawing of angle.
It, can when monochrome pixels value obtains one group of 4 vector tieed up by opposite width after being converted to one-dimensional bar code
Value carries out traversal matching using the corresponding angle of encoder dictionary each coding member when monochrome pixels value, it is available completely
Encoded information value completes corresponding decoding.
1.3 based on the rotation and displacement measurement for improving Fourier-Mellin transform
After the decoding for completing coding maker, the rotation that index point occurs is needed to resolve with translational movement.We
Case proposes a kind of improved Fourier-Mellin transform algorithm here, and improvement is the phase in Fourier-Mellin transform
In poor fit procedure, traditional least-squares algorithm is replaced to be fitted crosspower spectrum using stochastical sampling unification algorism, it can
Steadily to carry out antinoise estimation to the translation of index point and rotation amount.
The core of the algorithm is the displacement property of the Fourier transformation utilized: utilizing the principle of phase coherent techniques, figure
Displacement g (x, y)-g (x+x of each pixel g (x, y) on the direction x- and y- as between0,y+y0), size is equivalent to image Fu
In difference of the cross-power phase spectrum in the corresponding frequency spectrum s (u, v) of the pixel after leaf transformationWherein t0For in Fourier space
Phase difference.In phase space, phase difference is indicated by the angle difference of 0~2 π of phase.That is phase angle.Phase angle matrix
In, all elements have identical phase angle.That is phase difference t0Determine that the size of each element in the matrix of phase angle, the two are known
One can calculate another.By the estimation to phase difference, inverse Fourier transform can be used and find out position in theorem in Euclid space
Shifting amount (x0,y0)。(x0,y0) calculating refer to step (b).
Calculate two images G1With G2Between displacement and rotation the step of include:
(a) image G1With G2Discrete Fourier transform
Its corresponding relationship and crosspower spectrum matrix Q (u, v) can be expressed as in frequency domain after Fourier transformation:
S2(u, v)=S1(u,v)exp{-i(ux0+vy0)} (3.1)
S in formula1With S2For G1And G2Fourier transformation, S1WithTo be mutually conjugated.Wherein, G1For the reference of original logo point
Image, G2Measurement image for the index point for needing to measure displacement and rotating angle, S2(u, v) is corresponding G2It is sat in image array
It is designated as the matrix S that the pixel of (x, y) obtains after discrete Fourier transform2Middle corresponding points as a result, S1 *(u, v) is G1Image
The matrix S that coordinate obtains after discrete Fourier transform in matrix for the pixel of (x, y)1Conjugate matrices S* 1Middle corresponding points
As a result, abscissa of the u for the corresponding points in the fourier matrix S of the image obtained after discrete Fourier transform, v is direct computation of DFT
The ordinate of corresponding points in the fourier matrix S of the image obtained after leaf transformation, x0For the image before discrete Fourier transform
Matrix G1And G2Corresponding points abscissa (direction x-) shift value, y0For the image array obtained before discrete Fourier transform
G1And G2In corresponding points ordinate (direction y-) shift value;
(b) displacement relation of image is obtained by the estimation of phase difference
It can be obtained after being transformed under polar coordinates to above formula:
ψ (u, v)=∠ Q (u, v)=ux0+vy0 (4)
Here ψ (u, v) is the angle matrix of Q (u, v) matrix real and imaginary parts, i.e. phase angle matrix.Practical equation 4 above is corresponding
The expression formula of one plane equation:
Au+bv+c+d=0 (5)
Wherein a, b, c are the Slope Parameters of the plane equation, and d is the constant term of plane equation, i.e. ψ (u, v).Slope herein
Parameter c is constantly equal to 0.According to the relationship of Q (u, v) and u and v, the plane of formula 4 can be fitted, and then solve plane equation
Slope Parameters a and b, i.e., in the case where avoiding interpolation and inverse Fourier transform, obtain two width label point image shift value
x0=a and y0=b,.Specific plane fitting process refers to step (d)
(c) the rotation scaling relation of image is obtained by log-polar transform
If there is also rotation angle, θ and zoom scale γ other than shift value for two mark point images:
G2(x, y)=G1(a(x·cos(θ)+y·sin(θ),a(-x·sin(θ)+y·cos(θ)) (5.1)
To image array G1And G2Carry out log-polar conversion:
G (x, y)=K (λ, θ) (5.2)
K2(λ, θ)=K1(λ+lnγ,θ+θ0) (5.3)
Wherein K is the image array after polar coordinates expression.It can be found that its rotation and scaling relationship are in log-polar
Under also can be exchanged into displacement relation ln γ and θ0, then with step (b) to phase difference carry out estimating to obtain rotation angle, θ and
Zoom scale γ, i.e., to K1And K2Crosspower spectrum after carrying out Discrete Fourier Transform by formula 3.2 is calculated similar with formula 4
Expression formula:
ψ (m, n)=∠ Q (m, n)=mln γ+n θ0 (6)
M is the abscissa of the corresponding points in the fourier matrix that K is obtained after discrete Fourier transform, and n is K in discrete Fu
In corresponding points in the fourier matrix that obtains after leaf transformation ordinate.Here ψ (m, n) is Q (m, n) matrix real and imaginary parts
Angle matrix, i.e. phase angle matrix.The expression formula of the corresponding plane equation similar with formula 5 of practical equation 6 above:
Am+bn+c+d=0 (7)
Wherein a, b, c are the Slope Parameters of the plane equation, and d is the constant term of plane equation, i.e. ψ (m, n).Slope herein
Parameter c is constantly equal to 0.According to the relationship of Q (m, n) and m and n, the plane of formula 7 can be fitted, and then solve plane equation
Slope Parameters a and b, then two width label point image rotation θ0It can be obtained by following formula with scale γ:
γ=ea (8.1)
θ0=b (8.2)
Wherein e is the truth of a matter of natural logrithm function, i.e., scientific constant.Specific plane fitting process refers to step (d).
(d) phase angle is estimated by steady plane fitting
It obtains corresponding in the corresponding phase angle of crosspower spectrum (i.e. phase difference) matrix theory being one after converting in image Fu
The areal model that 2 π phases are wound on a ranks direction, conventional method are directly fitted two-dimensional surface using least-squares estimation,
Solution due to not being related to phase angle twines, and can only solve the deviant between -0.5 to 0.5 pixel, therefore, to assure that whole pixel does not occur
Mistake.Before carrying out plane fitting, need to carry out solution to phase angle matrix to twine.In order to guarantee crosspower spectrum data noise to phase
The influence that position solution twines, the method that this improved Fourier-Mellin transform scheme uses vector filtering algorithm
It is denoised, the real and imaginary parts of crosspower spectrum Q (u, v) is separated, denoised respectively using the mean filter of wicket, it is right
The phase angle matrix that filtered obtained matrix is wound after being denoised by arc tangent.This programme, which uses, is based on least cost net
The phase unwrapping algorithm of network stream (minimum cost network flow, MCNF) carries out solution to phase angle matrix and twines, Ke Yishi
Now preferably faster solution twines effect.For the deficiency of the phase related algorithm based on plane fitting, in conjunction with based on plane fitting
Phase correlation and RANSAC robust estimation algorithm, this programme are calculated using consistent (RANSAC) algorithm of random sampling instead of least square
Method carrys out the parameter a and b of robust iterative areal model au+bv-Q (u, v)=0, and only selection meets the number of the areal model after solution twines
According to progress phase pushing figure estimation, and elimination of rough difference is used as in robust iterative by the data of deviation effects, to weaken phase
Pass process large deviations are influenced caused by least-squares estimation, the precision and stability of deviant are improved, thus further
Promote the precision of wheel measuring.
On the basis of the present invention is the characteristics of analyzing existing bar coding and two class coded target of concentric ring type, propose
A kind of new coding maker design scheme.Scheme is added to base using a kind of very useful concentric ring type index point as prototype
It is designed in the coding element array of fan-shaped one-dimension code, code Design can be carried out.It is based in addition, this programme also proposed one kind
The arc search strategy of given radius realizes first being quickly converted to one-dimensional bar code of fan-shaped coding, avoids straight line in image
The occurrence of unstable in extraction process, realizes the quick steady decoding of coding.Finally, in order to obtain the phase of coded target
To geometry location information (i.e. rotation and displacement), this programme also proposed a kind of based on the displacement for improving Fourier-Mellin transform
With wheel measuring algorithm, the displacement and rotation of index point can be quickly and effectively directly obtained under different illumination and noise conditions
Transfering the letter breath.
Claims (10)
1. a kind of coding and decoding and measurement method of codified close-range photogrammetry mark, which comprises the following steps:
1) cataloged procedure:
11) setting of multiplying power: the coding member of the multiple fan-shaped one-dimension codes of building is encoded, and sets it and encodes multiplying power;
12) encoder dictionary the generation of encoder dictionary: is constituted by the coding member of multiple corresponding different coding information;
13) generation of coding maker;Complete coding maker is constituted by the combination of multiple coding members;
2) decoding process:
21) pretreatment of coded image obtains bianry image only containing marginal information from the coding maker image of input;
22) it the extraction of central point: is calculated according to the bianry image only containing marginal information using the ellipse fitting adjustment of least square
The central point of method acquisition coding maker;
23) conversion and decoding of one-dimensional bar code: using the arc searching method of given radius, i.e., it is with the central point of extraction
The heart is searched for the black and white gray value that given radius carries out 360 degree, after the coding that coding maker is converted to one-dimensional bar code,
According to the corresponding angle of each coding member of encoder dictionary, when monochrome pixels value carries out traversal matching, obtains complete encoded information
Value completes corresponding decoding;
3) measurement process:
After completing decoding, using improved Fourier-Mellin transform algorithm to the position mark point for encoding member in two images
Rotation scaling relation and the translational movement of upper generation are resolved.
2. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is, in the step 11), the first sector by position mark point and four kinds of different angles of the coding of each sector one-dimension code
Figure spot permutation and combination is formed, and the position mark point is made of multiple and chequered with black and white concentric circles, to realize Primary Location,
The fan-shaped figure spot includes wide figure spot and narrow figure spot, specially narrow black, narrow white, wide black and width white, the volume
Code multiplying power δ is the ratio of the corresponding central angle size of wide figure spot and narrow figure spot.
3. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 2, special
Sign is that the fan-shaped figure spot of each coding member is made of three wide figure spots and three narrow figure spots, the information that each coding member includes
It is determined by the distribution of width figure spot and the configuration of black-white colors.
4. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is, in the step 12), the information of encoder dictionary include number, letter and and additional character, the number with
Letter is the white space for encoding first upper left side fourth quadrant position for generating encoded information, the additional character, is used for
The start-stop of coding and the mark of calibration information.
5. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is that the encoder dictionary adjusts the capacity of encoder dictionary by adjusting the size of coding multiplying power δ.
6. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is, the step 21) specifically includes the following steps:
211) generation binary map the generation of bianry image: is split to the coding maker image of input by given gray threshold
Picture;
212) the edge letter of coding maker the extraction of edge contour: is extracted from bianry image by Canny edge detection operator
Breath.
7. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is, in the step 23), given radius is 0.6 times of figure spot radius.
8. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is, in the step 23), search process is started with any angle, according to the maximum white ash obtained in searching route
Mark of the angle value interval as judgement coding start-stop symbol, and using the angle of the wide-to-narrow ratio of one-dimensional bar code replacement coding member
Than.
9. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 1, special
Sign is, the step 3) specifically includes the following steps:
31) the reference picture G of original logo point is obtained1With the measurement image G for the index point for needing to measure rotation shift angle2It
Between translational movement, specifically:
311) respectively to the reference picture G of original logo point1With the measurement image G for the index point for needing to measure rotation shift angle2
Discrete Fourier transform is carried out, and calculates crosspower spectrum matrix Q (u, v), the calculating formula of the crosspower spectrum matrix Q (u, v)
Are as follows:
S2(u, v)=S1(u, v) exp {-i (ux0+vy0)}
Wherein, S2(u, v) is corresponding G2Coordinate is that the pixel of (x, y) obtains after discrete Fourier transform in image array
Matrix S2Middle corresponding points as a result, S1 *(u, v) is G1Coordinate is the pixel of (x, y) in discrete Fourier transform in image array
The matrix S obtained afterwards1Conjugate matrices S* 1Middle corresponding points as a result, u is in Fu of image obtained after discrete Fourier transform
The abscissa of corresponding points in leaf matrix S, v are the correspondence in the fourier matrix S of the image obtained after discrete Fourier transform
The ordinate of point, x0For the image array G before discrete Fourier transform1And G2Corresponding points abscissa direction shift value, y0
For the image array G obtained before discrete Fourier transform1And G2In corresponding points ordinate direction shift value;
312) polar coordinate transform is carried out to crosspower spectrum matrix Q (u, v) and obtains phase angle matrix ψ (u, v), it may be assumed that
ψ (u, v)=∠ Q (u, v)=ux0+vy0,
Plane fitting, the plane expression formula being fitted are carried out using stochastical sampling unification algorism are as follows:
a1u+b1v+c1+d1=0
Wherein, a0、b0、c0For the Slope Parameters of the plane equation, and c0=0, d1For the constant term of plane equation, and d1=ψ (u,
v);
313) the Slope Parameters a and b of plane equation are solved to get the shift value x of point image is marked to two width0=a and y0
=b;
32) the rotation scaling relation of image, including rotation angle, θ are obtained by log-polar transform0With zoom scale γ, tool
Body are as follows:
321) respectively to the reference picture G of original logo point1With the measurement image G for the index point for needing to measure rotation shift angle2
Carry out log-polar conversion;
322) to polar coordinates conversion after image carry out Discrete Fourier Transform obtain under polar coordinates crosspower spectrum matrix Q (m,
N), and the phase angle matrix ψ (m, n) of matrix Q (m, n) is obtained:
ψ (m, n)=∠ Q (m, n)=mln γ+n θ0;
323) the oblique of plane equation is solved after carrying out plane fitting using stochastical sampling unification algorism to phase angle matrix ψ (m, n)
Rate parameter alpha2And b2, the final rotation angle, θ for obtaining two width label point image0=b2And zoom scale
10. the coding and decoding and measurement method of a kind of codified close-range photogrammetry mark according to claim 9, special
Sign is that the solution before plane fitting twines step are as follows:
To eliminate influence of the crosspower spectrum data noise to phase unwrapping, used in the Fourier-Mellin transform algorithm
The method of vector filtering algorithm is denoised, and the real and imaginary parts of crosspower spectrum are separated, respectively
It is denoised using the mean filter of wicket, the phase angle matrix wound after arc tangent is denoised is carried out to filtered matrix,
And it carries out solution to phase angle matrix using the phase unwrapping algorithm based on minimum cost network stream to twine.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810803118.6A CN109186550B (en) | 2018-07-20 | 2018-07-20 | Coding decoding and measuring method for codable close-range photogrammetric mark |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810803118.6A CN109186550B (en) | 2018-07-20 | 2018-07-20 | Coding decoding and measuring method for codable close-range photogrammetric mark |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109186550A true CN109186550A (en) | 2019-01-11 |
CN109186550B CN109186550B (en) | 2021-03-12 |
Family
ID=64936559
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810803118.6A Active CN109186550B (en) | 2018-07-20 | 2018-07-20 | Coding decoding and measuring method for codable close-range photogrammetric mark |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109186550B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110210431A (en) * | 2019-06-06 | 2019-09-06 | 上海黑塞智能科技有限公司 | A kind of point cloud classifications method based on cloud semantic tagger and optimization |
CN112432612A (en) * | 2020-10-22 | 2021-03-02 | 中国计量科学研究院 | High-precision micro rotation angle measuring method based on monocular vision |
CN113129394A (en) * | 2020-12-23 | 2021-07-16 | 合肥工业大学 | Parallelogram coding mark based on region division coding and coding method thereof |
CN113538483A (en) * | 2021-06-28 | 2021-10-22 | 同济大学 | Coding and decoding method and measuring method of high-precision close-range photogrammetry mark |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE10347608A1 (en) * | 2003-10-09 | 2005-05-19 | Daimlerchrysler Ag | Reference mark for photogrammetry/metrophotography has elements for fixing the center of the mark and for variable encoding of data on a circle around the center of the mark |
CN102609979A (en) * | 2012-01-17 | 2012-07-25 | 北京工业大学 | Fourier-Mellin domain based two-dimensional/three-dimensional image registration method |
CN104154904A (en) * | 2014-07-18 | 2014-11-19 | 上海珞琪软件有限公司 | Circular coded mark based on graphic topology relationship and identification method of circular coded mark |
CN105760879A (en) * | 2016-01-14 | 2016-07-13 | 西安电子科技大学 | Fourier-Mellin transform-based image geometric matching method |
CN106682689A (en) * | 2016-12-16 | 2017-05-17 | 西安汇明光电技术有限公司 | Image matching method based on multiscale Fourier-Mellin transform |
-
2018
- 2018-07-20 CN CN201810803118.6A patent/CN109186550B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE10347608A1 (en) * | 2003-10-09 | 2005-05-19 | Daimlerchrysler Ag | Reference mark for photogrammetry/metrophotography has elements for fixing the center of the mark and for variable encoding of data on a circle around the center of the mark |
CN102609979A (en) * | 2012-01-17 | 2012-07-25 | 北京工业大学 | Fourier-Mellin domain based two-dimensional/three-dimensional image registration method |
CN104154904A (en) * | 2014-07-18 | 2014-11-19 | 上海珞琪软件有限公司 | Circular coded mark based on graphic topology relationship and identification method of circular coded mark |
CN105760879A (en) * | 2016-01-14 | 2016-07-13 | 西安电子科技大学 | Fourier-Mellin transform-based image geometric matching method |
CN106682689A (en) * | 2016-12-16 | 2017-05-17 | 西安汇明光电技术有限公司 | Image matching method based on multiscale Fourier-Mellin transform |
Non-Patent Citations (1)
Title |
---|
魏鹏: "近景摄影测量中标识点自动检测", 《计量与检测技术》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110210431A (en) * | 2019-06-06 | 2019-09-06 | 上海黑塞智能科技有限公司 | A kind of point cloud classifications method based on cloud semantic tagger and optimization |
CN112432612A (en) * | 2020-10-22 | 2021-03-02 | 中国计量科学研究院 | High-precision micro rotation angle measuring method based on monocular vision |
CN112432612B (en) * | 2020-10-22 | 2022-08-16 | 中国计量科学研究院 | High-precision micro rotation angle measuring method based on monocular vision |
CN113129394A (en) * | 2020-12-23 | 2021-07-16 | 合肥工业大学 | Parallelogram coding mark based on region division coding and coding method thereof |
CN113129394B (en) * | 2020-12-23 | 2022-09-06 | 合肥工业大学 | Parallelogram coding mark based on region division coding and coding method thereof |
CN113538483A (en) * | 2021-06-28 | 2021-10-22 | 同济大学 | Coding and decoding method and measuring method of high-precision close-range photogrammetry mark |
CN113538483B (en) * | 2021-06-28 | 2022-06-14 | 同济大学 | Coding and decoding method and measuring method of high-precision close-range photogrammetry mark |
Also Published As
Publication number | Publication date |
---|---|
CN109186550B (en) | 2021-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109186550A (en) | A kind of coding and decoding and measurement method of codified close-range photogrammetry mark | |
Wu et al. | High-speed and high-efficiency three-dimensional shape measurement based on Gray-coded light | |
CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
CN109961506A (en) | A kind of fusion improves the local scene three-dimensional reconstruction method of Census figure | |
El Helou et al. | NTIRE 2021 depth guided image relighting challenge | |
Hu et al. | Learning implicit feature alignment function for semantic segmentation | |
Chen et al. | 3-D shape measurement by composite pattern projection and hybrid processing | |
Yan et al. | Beyond the interference problem: hierarchical patterns for multiple-projector structured light system | |
Xu et al. | HA U-Net: Improved model for building extraction from high resolution remote sensing imagery | |
Li et al. | Absolute phase retrieval for a single-shot fringe projection profilometry based on deep learning | |
Spencer et al. | The monocular depth estimation challenge | |
Hurník et al. | Circular coded target system for industrial applications | |
Pedone et al. | Blur invariant translational image registration for $ N $-fold symmetric blurs | |
Song et al. | A robust structured light pattern decoding method for single-shot 3D reconstruction | |
Chen et al. | Robust local feature descriptor for multisource remote sensing image registration | |
Ji et al. | An evaluation of conventional and deep learning‐based image‐matching methods on diverse datasets | |
CN115908705A (en) | Three-dimensional imaging method and device based on special codes | |
Song et al. | Super-resolution phase retrieval network for single-pattern structured light 3D imaging | |
Wan et al. | Single-shot 3D measurement of highly reflective objects with deep learning | |
Farsangi et al. | Rectification based single-shot structured light for accurate and dense 3D reconstruction | |
CN116109772A (en) | Point cloud reconstruction method, device, equipment and readable storage medium | |
Kim et al. | Dense cross-modal correspondence estimation with the deep self-correlation descriptor | |
Hong et al. | A high-precision recognition method of circular marks based on CMNet within complex scenes | |
Shi et al. | Depth sensing with coding-free pattern based on topological constraint | |
Pan et al. | Multisource data registration based on NURBS description of contours |
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 |