CN107133986A - A kind of camera calibration method based on two-dimensional calibrations thing - Google Patents
A kind of camera calibration method based on two-dimensional calibrations thing Download PDFInfo
- Publication number
- CN107133986A CN107133986A CN201710284292.XA CN201710284292A CN107133986A CN 107133986 A CN107133986 A CN 107133986A CN 201710284292 A CN201710284292 A CN 201710284292A CN 107133986 A CN107133986 A CN 107133986A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msup
- msub
- image
- iac
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
Abstract
The present invention proposes a kind of camera calibration method based on two-dimensional calibrations thing, carrying out realizing camera calibration on the basis of uncalibrated image is preferred, including preparing plane reference plate, with the image of each angle of camera acquisition plane scaling board, angle point grid is carried out to each image, calculate the corresponding homography matrix of each image, the parameter of uniformity is randomly selected in initialization, homography matrix is randomly selected from all homography matrixes, calculate IAC, and judge the distance between the corresponding homography matrix of each image and IAC, the summation of interior point set and interior points is obtained according to judgement, and carry out local optimum；Finally according to all interior point estimation IAC, the corresponding image set of exterior point is removed, uncalibrated image is completed preferred；Calibration result is further obtained according to uncalibrated image preferred result.The present invention can lift the scaling method of Zhang Zhengyou two dimensional surfaces, obtain more stable reliable stated accuracy.
Description
Technical field
The present invention relates to camera calibration technical field, more particularly to the camera calibration technology based on two-dimensional calibrations thing, specifically
It is related to based on the preferred camera calibration method of uncalibrated image.
Background technology
Camera calibration technology refers to finding the parameter of camera model.Parameter generally comprises the parameter from scene to image
(internal reference), and it is tied to from reference coordinate the parameter (outer ginseng) of camera coordinates system.Camera lens are comprised in addition because manufacture is missed
Difference and produce nonlinear distortion variable element.The accuracy of camera calibration directly affects three-dimensional reconstruction, vision-based detection and vision and led
The precision of boat.The scholar for having many is studied camera calibration, is broadly divided into self-calibration and the demarcation based on demarcation thing.From
Demarcation is without demarcating thing, directly according to the corresponding points on the image of collection static scene come calibration for cameras, but its corresponding points
Extraction accuracy is limited, and the scene having does not have corresponding points, and it does not account for the distortion [2] of camera lens, therefore its demarcation essence
Degree is limited [1].In the high occasion of required precision frequently with the scaling method based on demarcation thing.
Scaling method based on demarcation thing is broadly divided into the method based on three-dimensional scaling thing and based on two-dimensional calibrations thing, is based on
The method of three-dimensional scaling thing needs the accurately known three-dimensional scaling thing of coordinate, and the equipment that it needs is costly.It is most widely used
It is the scaling method based on two-dimensional calibrations thing, most classical is Zhang Zhengyou two-dimensional planar template scaling method [3].Zhang Zhengyou's
Need to gather the image of multiple different visual angles in two dimensional surface scaling method to be demarcated, the precision of demarcation is dependent on suitable
Visual angle [3] (when scaling board angle and camera imaging plane angle be 45 degree when, stated accuracy can be higher), angle point grid
Accuracy etc..Different image sets can cause different stated accuracies.In order to handle this problem, [2] are proposed with random
Uniformity [4] (RANSAC) is extracted to reject insecure image to reach preferably demarcation effect, but be due to standard
Some shortcomings of RANSAC are as follows：(1) the RANSAC run times of standard than longer [5] predicted in theory, (2) it is assumed that from interior
The model that point the inside is obtained is consistent with all interior points, and this assumes generally invalid, because interior point sometimes can also be made an uproar
The influence of sound.Therefore the interior points that RANSAC is obtained have certain randomness, are differed with theoretic interior point, resulting
Model is also got along well theoretical identical certainly.
Relevant references：
[1]S.Bougnoux,“From projective to euclidean space under any practical
situation,a criticism of self-calibration,”Proc.IEEE.International Conference
on Computer Vision(ICCV 98),IEEE.CS Pess.Dec.1998,pp.790–796,doi:10.1109/
ICCV.1998.710808.
[2]Y.Lv,J.Feng,Z.Li,W.Li and J.Cao.“A New Robust 2D Camera
Calibration Method using RANSAC,”Optik-International Journal for Light and
Electron Optics,vol.126,Dec.2015,pp.4910–4915.
[3]Z.Zhang,“A Flexible New Technique for Camera Calibration,”IEEE
Transactions on Pattern Analysis and Machine Intelligence,vol.22,Dec.2000,
pp.1330–1334,doi:10.1109/34.888718.
[4]M.A.Fischler and R.C.Bolles,“Random Sample Consensus:A Paradigm
for Model Fitting with Applications to Image Analysis and Automated
Cartography,”Communications of the ACM,vol.24,Dec.1981,pp.381–395.
[5]B.Tordoff and D.W.Murray,“Guided Sampling and Consensus for Motion
Estimation,”Proc.European Conference on Computer Vision(ECCV 02),Springer-
Verlag Press,Dec.2002,pp.82–96.doi:10.1007/3-540-47969-4_6.
[6]O.Chum,J.Matas and J.Kittler,“Locally optimized RANSAC,”Lecture
Notes in Computer Science,vol.2781,Dec.2003,pp.236–243.
[7]D.C.Brown,“Close-range Camera Calibration,”Photogramm Eng,vol.37,
Dec.2002,pp.855–866.
[8]J.Salvi,X.Armangue and J.Batlle,“A Comparative Review of Cam-era
Calibrating Methods with Accuracy Evaluation,”Pattern Recognition,vol.35,
Dec.2002,pp.1617–1635.
[9]R.Hartley and A.Zisserman,Multiple View Geometry in Computer
Vision,2nd ed.,vol.30.Cambridge University Press,2003,pp.1865–1872.
[10]http://www.vision.caltech.edu/bouguetj/calib_doc/.
The content of the invention
Need to take image from different angles for traditional scaling method based on two dimensional model, the precision of demarcation by
Acquisition angles, collection distance, the influence of the positioning of characteristic point, the invention provides a kind of consistent with randomly selecting for local optimum
Property is come the method that improves Zhang Zhengyou camera calibration method accuracys.
Technical solution of the present invention proposes a kind of camera calibration method based on two-dimensional calibrations thing, carries out uncalibrated image preferred
On the basis of realize camera calibration, comprise the following steps,
Step 1, plane reference plate is prepared；
Step 2, with the image of each angle of camera acquisition plane scaling board, if obtaining J width images；
Step 3, angle point grid is carried out to each image, calculates the corresponding homography matrix of each image, be designated as Hj, wherein
J is picture numbers, and value is 1,2,3 ..., J；
Step 4, the parameter of uniformity is randomly selected in initialization, including takes theoretical iterations k to be ∞, and initial reality changes
Generation number is i=0, and best interior point set is bestSub=[], and best interior points are bestNin=0；[] represents empty set；
Step 5, s homography matrix is randomly selected from homography matrix Hj all obtained by step 3, IAC is calculated, and
Judge the distance between the corresponding homography matrix Hj of each image and IAC Dj, judged whether according to default respective threshold thr
Dj<Thr, is to think that Hj is interior point, otherwise it is assumed that Hj is exterior point, obtains interior point set sub and the summation Nin of interior points；Wherein s
For minimum sampling number；The IAC represents K-TK-1, K is intrinsic parameter；I=i+1；
Step 6, Nin is judged whether>BestNin, if it is, carrying out local optimum, otherwise return to step 5 into step 7；
Step 7, local optimum, including following sub-step are carried out,
Step 7.1, the local at most interior points in the number of times d=10 of initialization local optimum, initialization current iteration
LobestNin=0, local preferably interior point set lobestSub=[]；
Step 7.2, n=4, Nin=0, sub=[] are made, point meter in slo is randomly selected from current interior point set sub
IAC is calculated, slo is more than minimum sampling number s, judges the distance between the corresponding homography matrix Hj of each image and IAC Dj；
Step 7.3, according to n times of the given threshold value thr judgement counted in IAC, Dj is comprised determining whether<n*
Thr, this thinks that Hj is interior point if, otherwise it is assumed that Hj is exterior point, after the completion of judging each image, obtains current interior point
Collect sub and the summation Nin of interior points, make n=n-1；
Step 7.4, n is judged whether>1, if otherwise entering step 7.5, if it is calculated by current interior point set sub
IAC, return to step 7.3；
Step 7.5, Nin is judged>LobestNin, if so, lobestNin=Nin, lobestSub=sub, into step
7.6, if otherwise entering step 7.6；
Step 7.6, d=d-1, judges whether d<1, if otherwise return to step 7.2, if it is return to final result
lobestSub,lobestNin；
Step 8, the result returned according to step 7, makes bestNin=lobestNin, betSub=lobestsub, i=i
+1；Show that interior point, than ε=bestNin/J, renewal theory iterations k, judges whether i according to bestNin>K, if otherwise returned
Step 5 is returned to, is then to enter step 9；
Step 9, all interior point estimation IAC in bestSub, remove the corresponding image set of exterior point, complete calibration maps
As preferably；Calibration result is further obtained according to uncalibrated image preferred result.
Moreover, the calculating IAC is carried out according to the following formula,
h1 TK-TK-1h2=0
h1 TK-TK-1h1=h2 TK-TK-1h2
Wherein, h1,h2It is homography matrix H column vector.
Moreover, described judge that the distance between the corresponding homography matrix Hj and IAC of each image Dj is carried out according to the following formula,
Wherein, d is distance, B=K-TK-1, h30=h1-h2, h40=h1+h2, BhiIt is B and hiProduct, i=1,2,30,40,
Bhi(1) Bh first element, Bh are representedi(2) Bh second element is represented.
Moreover, the renewal theory iterations k is carried out according to the following formula,
Wherein, ε is the ratio of interior point, and η is given respective threshold.
Moreover, thr is preferably 2*10-5。
Moreover, J is 20.
Moreover, s is 2, slo=min (4, Nin).
The present invention is proposed, carries out uncalibrated image first before camera calibration preferably, including proposes that defining one singly should
The distance between image of property matrix and absolute conic, then with the stochastical sampling uniformity (LO- of local optimum
RANSAC) reject insecure image, eliminate RANSAC randomness, it is so as to obtain reliable and stable image set, real
Now lift the scaling method of Zhang Zhengyou two dimensional surfaces, obtain more stable reliable stated accuracy, it is to avoid by acquisition angles, gather away from
From the influence of the factors such as, positioning feature point., can be with using technical solution of the present invention from the point of view of emulation experiment and true experimental data
More preferable calibration result is obtained, due to just accurately being screened to unreliable image before demarcation so that Zhang Zhengyou two dimension
Plane reference method performs precision and efficiency is improved, and system resources consumption is low, with important market value.
Brief description of the drawings
Fig. 1 is the process principle figure of the embodiment of the present invention.
Fig. 2 randomly selects uniformity process principle figure for the local optimum of the embodiment of the present invention.
Embodiment
Describe the embodiment of the present invention in detail below in conjunction with drawings and examples.
The model of camera is divided into linear model and nonlinear model, and the linear model based on two-dimensional calibrations plate is represented
It is as follows：
Wherein m is scale factor, and K is intrinsic parameter here, wherein, fx,fyThe focal length of camera is represented, is pixel unit；(u0,
v0) be picture centre pixel coordinate；r1,r2It is that the column vector and t of spin matrix are the coordinate systems that camera is tied to from reference coordinate
Translation vector；(X, Y) is the world coordinates of scaling board；(u, v) is that the pixel coordinate of the corresponding imaging point of scaling board assumes
(X, Y) and (u, v) are known, and the present invention is readily available the homography square between the world coordinates of scaling board and pixel coordinate
Battle array H, it is 3*3 matrix, wherein h11,h12...,h33Respectively H element.Each image can obtain a list
Answering property matrix.
The lens distortion of camera is divided into following three class:Radial distortion, decentering distortion and thin prism distortion, its central diameter
It is maximum to distortion accounting, and mainly influenceed [7] by the Section 1 of radial distortion, therefore the present invention only considers radial distortion
Section 1, is equally applicable for other distortion.And more distortion considers also improve the precision [8] of distortion.Distortion
Model is as follows:
Here r=(x+y)2, (x, y) is preferable image coordinate, (ud,vd) it is the pixel coordinate distorted.k1It is radially
The coefficient of distortion, nonlinear model is that the lens distortion in order to reject camera obtains correct linear model, then using linear
Model recovers the parameter of camera.
In machine vision, RANSAC is widely used in estimating a model from the data set polluted by gross error
The data set of a smallest subset is randomly selected first to estimate model parameter, then calculate data point to model distance, that
A little points for being less than given threshold value from modal distance are considered interior point, and the more models of interior point think that it is better.The mistake randomly selected
Cheng Yizhi is repeated, until the probability that more interior points are found than current best model is less than given threshold value.I.e. in k times is sampled
Lose an interior point set and be less than given respective threshold η for the probability of s sample.
Wherein ε is the ratio of interior point.
It should be noted that RANSAC methods assume the model obtained inside the interior point be with all interior points it is consistent,
This hypothesis is incorrect because interior point also can by influence of noise, the time that RANSAC finds correct model is greater than k
Secondary [5].Moreover, the minimum sample of RANSAC selections carrys out computation model, thus noise influence to the accuracy of the model of calculating
Influence can be bigger, and the interior points that this causes RANSAC to find are smaller than theoretical value and unstable.
In order to handle RANSAC this problem, document [6] proposes LO-RANSAC, and it passes through local optimum RANSAC
The best model reached in often walking, can eliminate the influence of noise of interior point, can obtain stabilization, more accurate model.Therefore
The present invention proposes to reject insecure image of camera collection using LO-RANSAC, so as to obtain stabilization, high-precision mark
Determine effect.In the present invention, when RANSAC reaches each current best model, a local optimization operations are carried out.
Local optimum is using an internal RANSAC, plus an iteration, and this is explained inside two following paragraphs.
Internal RANSAC：When RANSAC reaches best model in th iteration of kth, sample from best model
Interior point concentrates selection, therefore, and sample size of this sampling can not be minimum, and experiment shows bigger sample to estimate
Model, the precision of model is higher [6].
Internally after RANSAC, in order to obtain more reliable result, an iteration framework is reused：Use all ratios one
The small points of individual bigger threshold value n*thr carry out least square method to estimate model, and n is integer, then are sequentially reduced n and iteration is straight
It is thr to threshold value.This iteration is because when using least square method, the thick data point of an error will cause estimation
The mistake of model, in iteration, the distance of each sampled point is both less than given threshold value, therefore without error very thick data
Point.
Internal RANSAC can reduce the influence of the noise of interior point plus the iteration of least square method, therefore can obtain more
Accurate and stable model.
In the Zhang Zhengyou method based on plane reference, calibration result is limited to the quality of the image of collection, in [3]
It has been shown that, best calibration result is when the angle of image and scaling board plane is 45 degree, when angle increase, to have an X-rayed abnormal
Change can make angle point grid more inaccurate, and different distances also results in the accuracy of angle point grid.It is all these to influence to adopt
The quality of the image of sample, different sample graph image set is inconsistent by the accuracy for causing demarcation.It is insecure in order to reject these
Image, obtains optimal image set, and the present invention removes insecure image using LO-RANSAC.As aforesaid, LO-
RANSAC can reduce RANSAC randomness, can obtain more, more reliable image sets, and then obtain more accurate
Calibrating parameters.
In Zhang Zhengyou demarcation, 2 constraintss are as follows：
Wherein h1,h2It is homography matrix H column vector, i.e. h1=(h11 h21 h31)T,h2=(h12 h22 h32)TTherefore,
Two equations (5) are at least needed to calculate K-TK-1, in document [9], K-TK-1Picture (IAC, image referred to as absolute conic
Of the absolute conic), the present invention requires no knowledge about IAC concrete meaning, and the present invention only needs to know that it represents K- TK-1, after it is solved, the closing solution of equation (1) can be obtained.Finally, this closing solution is as initial value, and considers that camera lens is abnormal
Become, carry out nonlinear optimization therefore with row Weinberg's algorithm, the influence of this IAC solution to initial value is very big, it is irrational at the beginning of
Value is easily caused non-limiting optimization and is absorbed in locally optimal solution.Document [2] defines a distance between IAC and homography matrix
It is as follows：
Wherein, d is distance, B=K-TK-1, h30=h1-h2, h40=h1+h2, BhiIt is B and hiProduct, i=1,2,30,40.
And Bhi(1) Bh first element, Bh are representedi(2) represent Bh second element, during specific implementation threshold value be can be according to imitative
Very given, thr is preferably 2*10 in embodiments of the present invention-5.In order to reject insecure image, the present invention proposes to utilize this
Distance is screened.
The present invention is further designed, and the implementation principle of proposition is as follows：
1. enough photos of scaling board are gathered from different angles, it is preferred to use more than 20 width；
2. pair each image zooming-out angle point, and calculate the homography matrix of each image；
3. arrange parameter:S=2, ε=0, k=∞, thr=2*10-5, i=1；
4. randomly select s images and calculate IAC:K-TK-1；
Judge 5. being calculated according to given threshold value thr according to formula (6) between the corresponding homography matrix of each image and IAC
Distance, determine IAC interior points N in, i.e., image set consistent with IAC, update in point ratio epsilon.
6., if bigger interior points are found that, carry out local optimum.
7. the 8th step is gone to if updating required sampling number i ＞ k by formula (4)；Otherwise i=i+1, goes to the 4th step；
After 8.RANSAC terminates, the subset of the image for demarcating is determined that.Then demarcated with Zhang Zhengyou method
The parameter of camera.
Referring to Fig. 1, the idiographic flow of embodiment, which is realized, to be described as follows：
1. preparing plane reference plate first, embodiment is using 12*13 gridiron pattern, and each tessellated size is 30mm,
So each tessellated world coordinates (X, Y) is corresponding known.
2. gridiron pattern to be placed on to different angles, the image of each angle is gathered with camera.If obtaining J width images, implement
In example, 20 width images are collected.
3. pair each image carries out angle point grid, the pixel coordinate (u, v) of X-comers is drawn, according to each image
The corresponding pixel coordinate (u, v) of gridiron pattern and world coordinates (X, Y), the corresponding homography square of each image is calculated according to formula (2)
Battle array, is designated as Hj (j=1,2,3 ..., 20), and wherein j is picture numbers, and value is 1,2,3 ..., J.
4. the parameter of uniformity is randomly selected in initialization, including takes theoretical iterations k to be ∞, initial actual iteration time
Number is i=0, and best interior point set is bestSub=[], and best interior points are bestNin=0.[] represents empty set.
5. carry out randomly selecting the iteration of uniformity, from all corresponding homography matrix Hj of image set (j=1,2,
3 ..., 20) s=2 homography matrix Hj is randomly selected in, and (s is to remember j=r1, r2,1 herein in minimum sampling number, embodiment
<=r1<=20,1<=r2<=20, r1！=r2), IAC is calculated according to formula (5), and each image correspondence is judged according to formula (6)
The distance between homography matrix Hj and IAC Dj (j=1,2,3 ..., 20), judged whether according to predetermined respective threshold thr
Dj<thr.If it is, thinking that Hj is interior point, otherwise it is assumed that Hj is exterior point.Interior point set is sub=(Hj | Hj is interior point), calculate in
The summation Nin of points.And i=i+1.
IAC represents K described in this flow-TK-1, K is intrinsic parameter.
Specifically judge that flow may be designed as, initialize j=1, sub=[], then judge whether Dj<Thr, if otherwise Hj is
Exterior point, j=j+1, if then Nin=Nin+1, sub=(sub+Hj), Hj is interior point, and then j=j+1 judges whether j>20,
Judge whether Dj for current j if being otherwise back to<Thr, if then entering step 6.
6. judge whether Nin>BestNin, if it is, carry out local optimum into step 7, otherwise return to step 5, again
Progress such as randomly selects at the processing.
7. local optimum is realized using LO-RANSAC modes, including following sub-step, referring to Fig. 2：
Local at most interior points in 7.1, the number of times d=10 of initialization local optimum, initialization current iteration
LobestNin=0, local preferably interior point set lobestSub=[]；
7.2, it is the multiple of threshold value to make n=4, Nin=0, sub=[], n；Randomly selected from current interior point set sub big
IAC, the present embodiment selection slo=min (4, Nin) width image pair are estimated in point in interior point set slo of minimum sampling number s
The Hj answered, calculates IAC, and judge between the corresponding homography matrix Hj and IAC of each image according to formula (6) according to formula (5)
Apart from Dj, j=1,2,3 ..., 20.
7.3, the judgement counted in IAC is carried out according to n times of given threshold value thr, Dj is comprised determining whether<n*
Thr, * represent to be multiplied by, if it is, thinking that Hj is interior point, otherwise it is assumed that Hj is exterior point.Wherein n initial value is default multiple,
Progressively reduce in subsequent iteration.After the completion of judging each image, current interior point set sub is obtained, points in calculating are realized
Summation Nin, makes n=n-1, into next step 7.4.
Flow may be designed as, and initializes j=1, judges whether Dj<N*thr, otherwise such as Hj is exterior point, j=j+1, if then
Nin=Nin+1, Hj are interior point, then make j=j+1, sub=sub+Hj；Then j is judged whether>20, it is then n=n-1, enters
Next step 7.4, otherwise returns and continues to determine whether Dj for current j<n*thr.
7.4, judge whether n>1, if not, into step 7.5, if it is, by current interior point set is counted according to formula (5)
Calculate IAC, return to step 7.3；
7.5, judge whether Nin>LobestNin, if so, lobestNin=Nin, lobestSub=sub, into step
7.6, if not into step 7.6.
7.6, d=d-1, judge whether d<1, if otherwise return to step 7.2, if it is return to final result
lobestSub,lobestNin。
8. the result returned according to step 7, makes bestNin=lobestNin, betSub=lobestsub, according to
BestNin show that interior point, than ε bestNin/20, according to formula (4) renewal theory iterations k, judges whether i>K, if it is not, then
Step 5 is returned to, is then to enter step 9.
9. by all interior points in bestSub, for estimating IAC, the corresponding image set of exterior point is removed, uncalibrated image is completed
It is preferred that.Calibration result is further obtained according to uncalibrated image preferred result, including further obtains linear solution, using linear solution as
The initial value of camera model, it is considered to single order radial distortion, is optimized with maximum likelihood and obtains final camera model calibrating parameters.
Flow terminates, and returns to calibration result, and back projection's error.
When it is implemented, above technical scheme can realize automatic running flow using computer software technology.
For checking technique effect of the embodiment of the present invention, emulation experiment has been carried out：
If randomly selecting camera parameter to carry out emulation experiment, the influence of noise is very big, may result in non-real number
Parametric solution, thus present invention experiment using the calibration result obtained from the actual experiment data inside document [10] as camera
Parameter, the intrinsic parameter of selection is as follows：
fx=657.384416175761；
fy=658.058046335663；
u0=303.625818604402；
v0=244.843359357986.
Outer parameter is spin matrix r and translation vector t, and spin matrix r is represented with rodrigues matrix herein, is
The vector of 3-dimensional.20 secondary different spin matrixs and translation vector are represented with Tables 1 and 2.Present invention experiment is set to lens distortion
0, the Chinese chess scaling board of synthesis has 12 × 13=144 angle point, and gridiron pattern size is 30mm × 30mm.Use these parameters, sheet
Invention experiment can produce the image coordinate of the scaling board angle point of synthesis.The image coordinate of last present invention experiment synthesis and the world
Coordinate is used for emulating calibration algorithm.
The rotating vector R of table 1
The image coordinate of angle point adds average for 0, variance be with the pixel value of every step 0.1 increase from 0 to 1 pixel
Noise.Calculate the distance between IAC and homography matrix.Noise in practice is 0.2 pixel, so present invention experiment handle
Thr is set to 2*10-5.Next, present invention experiment is using variance as with every pixel of step 0.2, the noise increased from 0 to 4 pixels is added
In image coordinate.In each noise level, RANSAC and the inventive method are respectively performed 100 times.In the mean error of calibration result,
Including fxAnd fyAverage relative error, u0And v0Absolute error.It is can see by experiment in mean error, present invention side
Method acquired results are less than RANSAC, particularly u0,v0.The method of this explanation present invention can obtain more accurately demarcating effect.
In the standard deviation of calibration result, including including fxAnd fyRelative error standard deviation, u0And v0Error standard deviation.Pass through
Experiment is it can be seen that in standard deviation, the inventive method acquired results are less than RANSAC, particularly u0,v0.This explanation present invention's
Method can obtain more stable demarcation effect.
The translation vector T of table 2
For checking technique effect of the embodiment of the present invention, true experiment has been carried out：
In this section, the present invention is demarcated using the 20 width images from [10].Due to not knowing real experimental data,
The present invention (is sat world coordinate point with the pixel of camera parameter back projection obtained by calibrating to image using average back projection error
Mark, and calculate its error between the angle point that extracts) represent the precision of demarcation.Tied obtained by RANSAC and the inventive method
Fruit has run 100 times.The variance of average back projection's error and error is shown in table 3.It can see from table 3, LO-RANSAC's is flat
Equal error and variance are all smaller than RANSAC, institute can obtain in the process of the present invention one it is more accurate than RANSAC with more stable mark
Determine result.
The actual experimental results of table 3
| Calibration method | Mean reprojection error | Standard deviation of the error |
| RANSAC | 0.159679252519579 | 0.004770649100921 |
| The inventive method | 0.156598456968103 | 1.952678215459548*10-16 |
Camera calibration is the underlying issue in computer vision, particularly in vision measurement.The accuracy of measurement is very big
Precision in degree dependent on demarcation, the precision of demarcation is limited by experiment condition and scaling method.Herein, it is of the invention
The RANSAC of local optimum has been used to reject insecure image.The inventive method acquired results by define IAC and
The distance between homography matrix, carries out local optimum when RANSAC reaches best model, and local optimum can disappear
Except the randomness that RANSAC is influenceed by interior spot noise, optimal image set can be then obtained, optimal calibration result is obtained.It is imitative
True experiment and actual implementation verify bright method of the invention be it is more accurate than traditional method, it is more stable.
Described above is presently preferred embodiments of the present invention, however it is not limited to the present embodiment, it is all the present embodiment spirit and
Modification, replacement, improvement for being made within principle etc., should be included within the protection domain of this patent.
Claims (7)
1. a kind of camera calibration method based on two-dimensional calibrations thing, it is characterised in that：On the basis of progress uncalibrated image is preferred
Camera calibration is realized, is comprised the following steps,
Step 1, plane reference plate is prepared；
Step 2, with the image of each angle of camera acquisition plane scaling board, if obtaining J width images；
Step 3, angle point grid is carried out to each image, calculates the corresponding homography matrix of each image, be designated as Hj, wherein j is
Picture numbers, value is 1,2,3 ..., J；
Step 4, the parameter of uniformity is randomly selected in initialization, including takes theoretical iterations k to be ∞, initial actual iteration time
Number is i=0, and best interior point set is bestSub=[], and best interior points are bestNin=0；[] represents empty set；
Step 5, s homography matrix is randomly selected from homography matrix Hj all obtained by step 3, IAC is calculated, and judge
The distance between the corresponding homography matrix Hj of each image and IAC Dj, Dj is judged whether according to default respective threshold thr<
Thr, is to think that Hj is interior point, otherwise it is assumed that Hj is exterior point, obtains interior point set sub and the summation Nin of interior points；
Wherein s is minimum sampling number；The IAC represents K-TK-1, K is intrinsic parameter；I=i+1；
Step 6, Nin is judged whether>BestNin, if it is, carrying out local optimum, otherwise return to step 5 into step 7；
Step 7, local optimum, including following sub-step are carried out,
Step 7.1, the local at most interior points in the number of times d=10 of initialization local optimum, initialization current iteration
LobestNin=0, local preferably interior point set lobestSub=[]；
Step 7.2, n=4, Nin=0, sub=[] are made, point in slo is randomly selected from current interior point set sub and is calculated
IAC, slo are more than minimum sampling number s, judge the distance between the corresponding homography matrix Hj of each image and IAC Dj；
Step 7.3, according to n times of the given threshold value thr judgement counted in IAC, Dj is comprised determining whether<N*thr, such as
Fruit is to think that Hj is interior point, otherwise it is assumed that Hj is exterior point, after the completion of judging each image, obtains current interior point set sub
With the summation Nin of interior points, n=n-1 is made；
Step 7.4, n is judged whether>1, if otherwise entering step 7.5, IAC is if it is calculated by current interior point set sub,
Return to step 7.3；
Step 7.5, Nin is judged>LobestNin, if so, lobestNin=Nin, lobestSub=sub, into step 7.6,
If otherwise entering step 7.6；
Step 7.6, d=d-1, judges whether d<1, if otherwise return to step 7.2, if it is return to final result
lobestSub,lobestNin；
Step 8, the result returned according to step 7, makes bestNin=lobestNin, betSub=lobestsub, according to
BestNin show that interior point, than ε=bestNin/J, renewal theory iterations k, judges whether i>K, if otherwise returning to step
Rapid 5, it is then to enter step 9；
Step 9, all interior point estimation IAC in bestSub, remove the corresponding image set of exterior point, complete uncalibrated image excellent
Choosing；Calibration result is further obtained according to uncalibrated image preferred result.
2. the camera calibration method based on two-dimensional calibrations thing according to claim 1, it is characterised in that：The calculating IAC is pressed
Carried out according to following formula,
h1 TK-TK-1h2=0
h1 TK-TK-1h1=h2 TK-TK-1h2
Wherein, h1,h2It is homography matrix H column vector.
3. the camera calibration method based on two-dimensional calibrations thing according to claim 2, it is characterised in that：It is described to judge every width figure
As the distance between corresponding homography matrix Hj and IAC Dj is carried out according to the following formula,
<mfenced open = "" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>d</mi>
<mo>=</mo>
<mfrac>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<msup>
<msub>
<mi>h</mi>
<mn>1</mn>
</msub>
<mi>T</mi>
</msup>
<msub>
<mi>Bh</mi>
<mn>2</mn>
</msub>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mrow>
<msub>
<mi>Bh</mi>
<mn>1</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msub>
<mi>Bh</mi>
<mn>1</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msub>
<mi>Bh</mi>
<mn>2</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msub>
<mi>Bh</mi>
<mn>2</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>+</mo>
<mfrac>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<msup>
<msub>
<mi>h</mi>
<mn>30</mn>
</msub>
<mi>T</mi>
</msup>
<msub>
<mi>Bh</mi>
<mn>40</mn>
</msub>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mrow>
<msub>
<mi>Bh</mi>
<mn>30</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msub>
<mi>Bh</mi>
<mn>30</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msub>
<mi>Bh</mi>
<mn>40</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msub>
<mi>Bh</mi>
<mn>40</mn>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Wherein, d is distance, B=K-TK-1, h30=h1-h2, h40=h1+h2, BhiIt is B and hiProduct, i=1,2,30,40, Bhi
(1) Bh first element, Bh are representedi(2) Bh second element is represented.
4. the camera calibration method based on two-dimensional calibrations thing according to claim 1, it is characterised in that：The renewal theory changes
Generation number k is carried out according to the following formula,
<mrow>
<mi>k</mi>
<mo>=</mo>
<mfrac>
<mrow>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>&eta;</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msup>
<mi>&epsiv;</mi>
<mi>s</mi>
</msup>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
Wherein, ε is the ratio of interior point, and η is given respective threshold.
5. according to camera calibration method of the claim 1 or 2 or 3 or described based on two-dimensional calibrations thing, it is characterised in that：Thr is excellent
Elect 2*10 as-5。
6. according to camera calibration method of the claim 1 or 2 or 3 or described based on two-dimensional calibrations thing, it is characterised in that：J is 20.
7. according to camera calibration method of the claim 1 or 2 or 3 or described based on two-dimensional calibrations thing, it is characterised in that：S is 2,
Slo=min (4, Nin).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710284292.XA CN107133986B (en) | 2017-04-26 | 2017-04-26 | A kind of camera calibration method based on two-dimensional calibrations object |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710284292.XA CN107133986B (en) | 2017-04-26 | 2017-04-26 | A kind of camera calibration method based on two-dimensional calibrations object |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107133986A true CN107133986A (en) | 2017-09-05 |
| CN107133986B CN107133986B (en) | 2019-08-20 |
Family
ID=59716464
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710284292.XA Expired - Fee Related CN107133986B (en) | 2017-04-26 | 2017-04-26 | A kind of camera calibration method based on two-dimensional calibrations object |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107133986B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108234819A (en) * | 2018-01-30 | 2018-06-29 | 西安电子科技大学 | Video synchronization method based on homograph |
| CN109242910A (en) * | 2018-08-21 | 2019-01-18 | 电子科技大学 | A kind of monocular camera self-calibrating method based on any known flat shape |
| CN110345875A (en) * | 2018-04-04 | 2019-10-18 | 灵动科技（北京）有限公司 | Calibration and distance measuring method, device, electronic equipment and computer readable storage medium |
| CN113375555A (en) * | 2018-07-02 | 2021-09-10 | 广西电网有限责任公司北海供电局 | Power line clamp measuring method and system based on mobile phone image |
| CN113470117A (en) * | 2021-06-28 | 2021-10-01 | 上海交通大学 | Unit attitude three-dimensional structured light calibration system and method based on spherical reverse perspective projection |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW200905610A (en) * | 2007-05-22 | 2009-02-01 | Microsoft Corp | Camera calibration |
| WO2012034092A1 (en) * | 2010-09-09 | 2012-03-15 | Qualcomm Incorporated | Online reference generation and tracking for multi-user augmented reality |
| CN102999759A (en) * | 2012-11-07 | 2013-03-27 | 东南大学 | Light stream based vehicle motion state estimating method |
| CN104182969A (en) * | 2014-08-08 | 2014-12-03 | 河南科技大学 | Internal and external parameter calibration method of single-scanline camera |
| CN104240233A (en) * | 2014-08-19 | 2014-12-24 | 长春理工大学 | Method for solving camera homography matrix and projector homography matrix |
| US20150130951A1 (en) * | 2013-10-28 | 2015-05-14 | The Regents Of The University Of Michigan | Interactive camera calibration tool |
| CN105096328A (en) * | 2015-08-13 | 2015-11-25 | 北京理工大学 | Camera self-calibration method based on IOS-PSO (Internetwork Operating System-Particle Swarm Optimization) |
-
2017
- 2017-04-26 CN CN201710284292.XA patent/CN107133986B/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW200905610A (en) * | 2007-05-22 | 2009-02-01 | Microsoft Corp | Camera calibration |
| WO2012034092A1 (en) * | 2010-09-09 | 2012-03-15 | Qualcomm Incorporated | Online reference generation and tracking for multi-user augmented reality |
| CN102999759A (en) * | 2012-11-07 | 2013-03-27 | 东南大学 | Light stream based vehicle motion state estimating method |
| US20150130951A1 (en) * | 2013-10-28 | 2015-05-14 | The Regents Of The University Of Michigan | Interactive camera calibration tool |
| CN104182969A (en) * | 2014-08-08 | 2014-12-03 | 河南科技大学 | Internal and external parameter calibration method of single-scanline camera |
| CN104240233A (en) * | 2014-08-19 | 2014-12-24 | 长春理工大学 | Method for solving camera homography matrix and projector homography matrix |
| CN105096328A (en) * | 2015-08-13 | 2015-11-25 | 北京理工大学 | Camera self-calibration method based on IOS-PSO (Internetwork Operating System-Particle Swarm Optimization) |
Non-Patent Citations (2)
| Title |
|---|
| ONDŘEJ CHUM ET AL.: ""Locally Optimized RANSAC"", 《SRPINGERLINK》 * |
| SIMONE GASPARINI ET AL.: ""Plane-Based Calibration of Central Catadioptric Cameras"", 《IEEEXPLORE》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108234819A (en) * | 2018-01-30 | 2018-06-29 | 西安电子科技大学 | Video synchronization method based on homograph |
| CN108234819B (en) * | 2018-01-30 | 2019-08-13 | 西安电子科技大学 | Video synchronization method based on homograph |
| CN110345875A (en) * | 2018-04-04 | 2019-10-18 | 灵动科技（北京）有限公司 | Calibration and distance measuring method, device, electronic equipment and computer readable storage medium |
| CN110345875B (en) * | 2018-04-04 | 2021-04-27 | 灵动科技（北京）有限公司 | Calibration and ranging method, device, electronic equipment and computer readable storage medium |
| CN113375555A (en) * | 2018-07-02 | 2021-09-10 | 广西电网有限责任公司北海供电局 | Power line clamp measuring method and system based on mobile phone image |
| CN109242910A (en) * | 2018-08-21 | 2019-01-18 | 电子科技大学 | A kind of monocular camera self-calibrating method based on any known flat shape |
| CN109242910B (en) * | 2018-08-21 | 2021-10-08 | 电子科技大学 | Monocular camera self-calibration method based on any known plane shape |
| CN113470117A (en) * | 2021-06-28 | 2021-10-01 | 上海交通大学 | Unit attitude three-dimensional structured light calibration system and method based on spherical reverse perspective projection |
| CN113470117B (en) * | 2021-06-28 | 2022-12-27 | 上海交通大学 | Unit attitude three-dimensional structured light calibration system and method based on spherical reverse perspective projection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107133986B (en) | 2019-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107133986B (en) | A kind of camera calibration method based on two-dimensional calibrations object | |
| CN105205858B (en) | A kind of indoor scene three-dimensional rebuilding method based on single deep vision sensor | |
| Berman et al. | Air-light estimation using haze-lines | |
| CN104778688B (en) | The method for registering and device of cloud data | |
| CN104596519B (en) | Vision positioning method based on RANSAC algorithms | |
| CN109523501A (en) | One kind being based on dimensionality reduction and the matched battery open defect detection method of point cloud data | |
| CN109211198B (en) | Intelligent target detection and measurement system and method based on trinocular vision | |
| CN106485690A (en) | Cloud data based on a feature and the autoregistration fusion method of optical image | |
| US10460471B2 (en) | Camera pose estimating method and system | |
| CN101650828B (en) | Method for reducing random error of round object location in camera calibration | |
| CN105654547B (en) | Three-dimensional rebuilding method | |
| CN109118544B (en) | Synthetic aperture imaging method based on perspective transformation | |
| CN110360991A (en) | A kind of photogrammetric survey method, device and storage medium | |
| CN105335952B (en) | Matching power flow computational methods and device and parallax value calculating method and equipment | |
| CN103295239A (en) | Laser-point cloud data automatic registration method based on plane base images | |
| CN108765476A (en) | A kind of polarization image method for registering | |
| CN109493384A (en) | Camera position and orientation estimation method, system, equipment and storage medium | |
| CN114998773B (en) | Characteristic mismatching elimination method and system suitable for aerial image of unmanned aerial vehicle system | |
| CN109658366A (en) | Based on the real-time video joining method for improving RANSAC and dynamic fusion | |
| Remondino et al. | Evaluating hand-crafted and learning-based features for photogrammetric applications | |
| CN110223356A (en) | A kind of monocular camera full automatic calibration method based on energy growth | |
| CN111028257B (en) | Method and device for extracting geometric center and edge of microlens subimage | |
| Magri et al. | Bending the doming effect in structure from motion reconstructions through bundle adjustment | |
| CN107680035A (en) | A kind of parameter calibration method and device, server and readable storage medium storing program for executing | |
| CN111563895A (en) | Picture definition determining method, device, equipment and storage medium |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190820 Termination date: 20210426 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |